updated

.venv/lib/python3.9/site-packages/_distutils_hack/__init__.py

@@ -0,0 +1,132 @@
+import sys
+import os
+import re
+import importlib
+import warnings
+
+
+is_pypy = '__pypy__' in sys.builtin_module_names
+
+
+warnings.filterwarnings('ignore',
+                        r'.+ distutils\b.+ deprecated',
+                        DeprecationWarning)
+
+
+def warn_distutils_present():
+    if 'distutils' not in sys.modules:
+        return
+    if is_pypy and sys.version_info < (3, 7):
+        # PyPy for 3.6 unconditionally imports distutils, so bypass the warning
+        # https://foss.heptapod.net/pypy/pypy/-/blob/be829135bc0d758997b3566062999ee8b23872b4/lib-python/3/site.py#L250
+        return
+    warnings.warn(
+        "Distutils was imported before Setuptools, but importing Setuptools "
+        "also replaces the `distutils` module in `sys.modules`. This may lead "
+        "to undesirable behaviors or errors. To avoid these issues, avoid "
+        "using distutils directly, ensure that setuptools is installed in the "
+        "traditional way (e.g. not an editable install), and/or make sure "
+        "that setuptools is always imported before distutils.")
+
+
+def clear_distutils():
+    if 'distutils' not in sys.modules:
+        return
+    warnings.warn("Setuptools is replacing distutils.")
+    mods = [name for name in sys.modules if re.match(r'distutils\b', name)]
+    for name in mods:
+        del sys.modules[name]
+
+
+def enabled():
+    """
+    Allow selection of distutils by environment variable.
+    """
+    which = os.environ.get('SETUPTOOLS_USE_DISTUTILS', 'stdlib')
+    return which == 'local'
+
+
+def ensure_local_distutils():
+    clear_distutils()
+
+    # With the DistutilsMetaFinder in place,
+    # perform an import to cause distutils to be
+    # loaded from setuptools._distutils. Ref #2906.
+    add_shim()
+    importlib.import_module('distutils')
+    remove_shim()
+
+    # check that submodules load as expected
+    core = importlib.import_module('distutils.core')
+    assert '_distutils' in core.__file__, core.__file__
+
+
+def do_override():
+    """
+    Ensure that the local copy of distutils is preferred over stdlib.
+
+    See https://github.com/pypa/setuptools/issues/417#issuecomment-392298401
+    for more motivation.
+    """
+    if enabled():
+        warn_distutils_present()
+        ensure_local_distutils()
+
+
+class DistutilsMetaFinder:
+    def find_spec(self, fullname, path, target=None):
+        if path is not None:
+            return
+
+        method_name = 'spec_for_{fullname}'.format(**locals())
+        method = getattr(self, method_name, lambda: None)
+        return method()
+
+    def spec_for_distutils(self):
+        import importlib.abc
+        import importlib.util
+
+        class DistutilsLoader(importlib.abc.Loader):
+
+            def create_module(self, spec):
+                return importlib.import_module('setuptools._distutils')
+
+            def exec_module(self, module):
+                pass
+
+        return importlib.util.spec_from_loader('distutils', DistutilsLoader())
+
+    def spec_for_pip(self):
+        """
+        Ensure stdlib distutils when running under pip.
+        See pypa/pip#8761 for rationale.
+        """
+        if self.pip_imported_during_build():
+            return
+        clear_distutils()
+        self.spec_for_distutils = lambda: None
+
+    @staticmethod
+    def pip_imported_during_build():
+        """
+        Detect if pip is being imported in a build script. Ref #2355.
+        """
+        import traceback
+        return any(
+            frame.f_globals['__file__'].endswith('setup.py')
+            for frame, line in traceback.walk_stack(None)
+        )
+
+
+DISTUTILS_FINDER = DistutilsMetaFinder()
+
+
+def add_shim():
+    sys.meta_path.insert(0, DISTUTILS_FINDER)
+
+
+def remove_shim():
+    try:
+        sys.meta_path.remove(DISTUTILS_FINDER)
+    except ValueError:
+        pass

.venv/lib/python3.9/site-packages/_distutils_hack/override.py

@@ -0,0 +1 @@
+__import__('_distutils_hack').do_override()

.venv/lib/python3.9/site-packages/certifi-2026.2.25.dist-info/INSTALLER

@@ -0,0 +1 @@
+pip

.venv/lib/python3.9/site-packages/certifi-2026.2.25.dist-info/METADATA

@@ -0,0 +1,78 @@
+Metadata-Version: 2.4
+Name: certifi
+Version: 2026.2.25
+Summary: Python package for providing Mozilla's CA Bundle.
+Home-page: https://github.com/certifi/python-certifi
+Author: Kenneth Reitz
+Author-email: me@kennethreitz.com
+License: MPL-2.0
+Project-URL: Source, https://github.com/certifi/python-certifi
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Mozilla Public License 2.0 (MPL 2.0)
+Classifier: Natural Language :: English
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.7
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Requires-Python: >=3.7
+License-File: LICENSE
+Dynamic: author
+Dynamic: author-email
+Dynamic: classifier
+Dynamic: description
+Dynamic: home-page
+Dynamic: license
+Dynamic: license-file
+Dynamic: project-url
+Dynamic: requires-python
+Dynamic: summary
+
+Certifi: Python SSL Certificates
+================================
+
+Certifi provides Mozilla's carefully curated collection of Root Certificates for
+validating the trustworthiness of SSL certificates while verifying the identity
+of TLS hosts. It has been extracted from the `Requests`_ project.
+
+Installation
+------------
+
+``certifi`` is available on PyPI. Simply install it with ``pip``::
+
+    $ pip install certifi
+
+Usage
+-----
+
+To reference the installed certificate authority (CA) bundle, you can use the
+built-in function::
+
+    >>> import certifi
+
+    >>> certifi.where()
+    '/usr/local/lib/python3.7/site-packages/certifi/cacert.pem'
+
+Or from the command line::
+
+    $ python -m certifi
+    /usr/local/lib/python3.7/site-packages/certifi/cacert.pem
+
+Enjoy!
+
+.. _`Requests`: https://requests.readthedocs.io/en/master/
+
+Addition/Removal of Certificates
+--------------------------------
+
+Certifi does not support any addition/removal or other modification of the
+CA trust store content. This project is intended to provide a reliable and
+highly portable root of trust to python deployments. Look to upstream projects
+for methods to use alternate trust.

.venv/lib/python3.9/site-packages/certifi-2026.2.25.dist-info/RECORD

@@ -0,0 +1,14 @@
+certifi-2026.2.25.dist-info/INSTALLER,sha256=zuuue4knoyJ-UwPPXg8fezS7VCrXJQrAP7zeNuwvFQg,4
+certifi-2026.2.25.dist-info/METADATA,sha256=4NMuGXdg_hBiRA3paKVXYcDmE3VXEBWxTvCL2xlDyPU,2474
+certifi-2026.2.25.dist-info/RECORD,,
+certifi-2026.2.25.dist-info/WHEEL,sha256=YCfwYGOYMi5Jhw2fU4yNgwErybb2IX5PEwBKV4ZbdBo,91
+certifi-2026.2.25.dist-info/licenses/LICENSE,sha256=6TcW2mucDVpKHfYP5pWzcPBpVgPSH2-D8FPkLPwQyvc,989
+certifi-2026.2.25.dist-info/top_level.txt,sha256=KMu4vUCfsjLrkPbSNdgdekS-pVJzBAJFO__nI8NF6-U,8
+certifi/__init__.py,sha256=c9eaYufv1pSLl0Q8QNcMiMLLH4WquDcxdPyKjmI4opY,94
+certifi/__main__.py,sha256=xBBoj905TUWBLRGANOcf7oi6e-3dMP4cEoG9OyMs11g,243
+certifi/__pycache__/__init__.cpython-39.pyc,,
+certifi/__pycache__/__main__.cpython-39.pyc,,
+certifi/__pycache__/core.cpython-39.pyc,,
+certifi/cacert.pem,sha256=_JFloSQDJj5-v72te-ej6sD6XTJdPHBGXyjTaQByyig,272441
+certifi/core.py,sha256=XFXycndG5pf37ayeF8N32HUuDafsyhkVMbO4BAPWHa0,3394
+certifi/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0

.venv/lib/python3.9/site-packages/certifi-2026.2.25.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (82.0.0)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

.venv/lib/python3.9/site-packages/certifi-2026.2.25.dist-info/licenses/LICENSE

@@ -0,0 +1,20 @@
+This package contains a modified version of ca-bundle.crt:
+
+ca-bundle.crt -- Bundle of CA Root Certificates
+
+This is a bundle of X.509 certificates of public Certificate Authorities
+(CA). These were automatically extracted from Mozilla's root certificates
+file (certdata.txt).  This file can be found in the mozilla source tree:
+https://hg.mozilla.org/mozilla-central/file/tip/security/nss/lib/ckfw/builtins/certdata.txt
+It contains the certificates in PEM format and therefore
+can be directly used with curl / libcurl / php_curl, or with
+an Apache+mod_ssl webserver for SSL client authentication.
+Just configure this file as the SSLCACertificateFile.#
+
+***** BEGIN LICENSE BLOCK *****
+This Source Code Form is subject to the terms of the Mozilla Public License,
+v. 2.0. If a copy of the MPL was not distributed with this file, You can obtain
+one at http://mozilla.org/MPL/2.0/.
+
+***** END LICENSE BLOCK *****
+@(#) $RCSfile: certdata.txt,v $ $Revision: 1.80 $ $Date: 2011/11/03 15:11:58 $

.venv/lib/python3.9/site-packages/certifi-2026.2.25.dist-info/top_level.txt

@@ -0,0 +1 @@
+certifi

.venv/lib/python3.9/site-packages/certifi/__init__.py

@@ -0,0 +1,4 @@
+from .core import contents, where
+
+__all__ = ["contents", "where"]
+__version__ = "2026.02.25"

.venv/lib/python3.9/site-packages/certifi/__main__.py

@@ -0,0 +1,12 @@
+import argparse
+
+from certifi import contents, where
+
+parser = argparse.ArgumentParser()
+parser.add_argument("-c", "--contents", action="store_true")
+args = parser.parse_args()
+
+if args.contents:
+    print(contents())
+else:
+    print(where())

.venv/lib/python3.9/site-packages/certifi/core.py

@@ -0,0 +1,83 @@
+"""
+certifi.py
+~~~~~~~~~~
+
+This module returns the installation location of cacert.pem or its contents.
+"""
+import sys
+import atexit
+
+def exit_cacert_ctx() -> None:
+    _CACERT_CTX.__exit__(None, None, None)  # type: ignore[union-attr]
+
+
+if sys.version_info >= (3, 11):
+
+    from importlib.resources import as_file, files
+
+    _CACERT_CTX = None
+    _CACERT_PATH = None
+
+    def where() -> str:
+        # This is slightly terrible, but we want to delay extracting the file
+        # in cases where we're inside of a zipimport situation until someone
+        # actually calls where(), but we don't want to re-extract the file
+        # on every call of where(), so we'll do it once then store it in a
+        # global variable.
+        global _CACERT_CTX
+        global _CACERT_PATH
+        if _CACERT_PATH is None:
+            # This is slightly janky, the importlib.resources API wants you to
+            # manage the cleanup of this file, so it doesn't actually return a
+            # path, it returns a context manager that will give you the path
+            # when you enter it and will do any cleanup when you leave it. In
+            # the common case of not needing a temporary file, it will just
+            # return the file system location and the __exit__() is a no-op.
+            #
+            # We also have to hold onto the actual context manager, because
+            # it will do the cleanup whenever it gets garbage collected, so
+            # we will also store that at the global level as well.
+            _CACERT_CTX = as_file(files("certifi").joinpath("cacert.pem"))
+            _CACERT_PATH = str(_CACERT_CTX.__enter__())
+            atexit.register(exit_cacert_ctx)
+
+        return _CACERT_PATH
+
+    def contents() -> str:
+        return files("certifi").joinpath("cacert.pem").read_text(encoding="ascii")
+
+else:
+
+    from importlib.resources import path as get_path, read_text
+
+    _CACERT_CTX = None
+    _CACERT_PATH = None
+
+    def where() -> str:
+        # This is slightly terrible, but we want to delay extracting the
+        # file in cases where we're inside of a zipimport situation until
+        # someone actually calls where(), but we don't want to re-extract
+        # the file on every call of where(), so we'll do it once then store
+        # it in a global variable.
+        global _CACERT_CTX
+        global _CACERT_PATH
+        if _CACERT_PATH is None:
+            # This is slightly janky, the importlib.resources API wants you
+            # to manage the cleanup of this file, so it doesn't actually
+            # return a path, it returns a context manager that will give
+            # you the path when you enter it and will do any cleanup when
+            # you leave it. In the common case of not needing a temporary
+            # file, it will just return the file system location and the
+            # __exit__() is a no-op.
+            #
+            # We also have to hold onto the actual context manager, because
+            # it will do the cleanup whenever it gets garbage collected, so
+            # we will also store that at the global level as well.
+            _CACERT_CTX = get_path("certifi", "cacert.pem")
+            _CACERT_PATH = str(_CACERT_CTX.__enter__())
+            atexit.register(exit_cacert_ctx)
+
+        return _CACERT_PATH
+
+    def contents() -> str:
+        return read_text("certifi", "cacert.pem", encoding="ascii")

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/INSTALLER

@@ -0,0 +1 @@
+pip

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/METADATA

@@ -0,0 +1,68 @@
+Metadata-Version: 2.4
+Name: cffi
+Version: 2.0.0
+Summary: Foreign Function Interface for Python calling C code.
+Author: Armin Rigo, Maciej Fijalkowski
+Maintainer: Matt Davis, Matt Clay, Matti Picus
+License-Expression: MIT
+Project-URL: Documentation, https://cffi.readthedocs.io/
+Project-URL: Changelog, https://cffi.readthedocs.io/en/latest/whatsnew.html
+Project-URL: Downloads, https://github.com/python-cffi/cffi/releases
+Project-URL: Contact, https://groups.google.com/forum/#!forum/python-cffi
+Project-URL: Source Code, https://github.com/python-cffi/cffi
+Project-URL: Issue Tracker, https://github.com/python-cffi/cffi/issues
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: Free Threading :: 2 - Beta
+Classifier: Programming Language :: Python :: Implementation :: CPython
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: AUTHORS
+Requires-Dist: pycparser; implementation_name != "PyPy"
+Dynamic: license-file
+
+[![GitHub Actions Status](https://github.com/python-cffi/cffi/actions/workflows/ci.yaml/badge.svg?branch=main)](https://github.com/python-cffi/cffi/actions/workflows/ci.yaml?query=branch%3Amain++)
+[![PyPI version](https://img.shields.io/pypi/v/cffi.svg)](https://pypi.org/project/cffi)
+[![Read the Docs](https://img.shields.io/badge/docs-latest-blue.svg)][Documentation]
+
+
+CFFI
+====
+
+Foreign Function Interface for Python calling C code.
+
+Please see the [Documentation] or uncompiled in the `doc/` subdirectory.
+
+Download
+--------
+
+[Download page](https://github.com/python-cffi/cffi/releases)
+
+Source Code
+-----------
+
+Source code is publicly available on
+[GitHub](https://github.com/python-cffi/cffi).
+
+Contact
+-------
+
+[Mailing list](https://groups.google.com/forum/#!forum/python-cffi)
+
+Testing/development tips
+------------------------
+
+After `git clone` or `wget && tar`, we will get a directory called `cffi` or `cffi-x.x.x`. we call it `repo-directory`. To run tests under CPython, run the following in the `repo-directory`:
+
+    pip install pytest
+    pip install -e .  # editable install of CFFI for local development
+    pytest src/c/ testing/
+
+[Documentation]: http://cffi.readthedocs.org/

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/RECORD

@@ -0,0 +1,49 @@
+_cffi_backend.cpython-39-x86_64-linux-gnu.so,sha256=9J_Iuv_0UXORJQsmFFKfZwHTQDTC313tRixPBAunqCU,339784
+cffi-2.0.0.dist-info/INSTALLER,sha256=zuuue4knoyJ-UwPPXg8fezS7VCrXJQrAP7zeNuwvFQg,4
+cffi-2.0.0.dist-info/METADATA,sha256=uYzn40F68Im8EtXHNBLZs7FoPM-OxzyYbDWsjJvhujk,2559
+cffi-2.0.0.dist-info/RECORD,,
+cffi-2.0.0.dist-info/WHEEL,sha256=A4SdjgbIBZ_kjTsEuOcR8fCpVKV5dez8wjeT_kpjoQs,147
+cffi-2.0.0.dist-info/entry_points.txt,sha256=y6jTxnyeuLnL-XJcDv8uML3n6wyYiGRg8MTp_QGJ9Ho,75
+cffi-2.0.0.dist-info/licenses/AUTHORS,sha256=KmemC7-zN1nWfWRf8TG45ta8TK_CMtdR_Kw-2k0xTMg,208
+cffi-2.0.0.dist-info/licenses/LICENSE,sha256=W6JN3FcGf5JJrdZEw6_EGl1tw34jQz73Wdld83Cwr2M,1123
+cffi-2.0.0.dist-info/top_level.txt,sha256=rE7WR3rZfNKxWI9-jn6hsHCAl7MDkB-FmuQbxWjFehQ,19
+cffi/__init__.py,sha256=-ksBQ7MfDzVvbBlV_ftYBWAmEqfA86ljIzMxzaZeAlI,511
+cffi/__pycache__/__init__.cpython-39.pyc,,
+cffi/__pycache__/_imp_emulation.cpython-39.pyc,,
+cffi/__pycache__/_shimmed_dist_utils.cpython-39.pyc,,
+cffi/__pycache__/api.cpython-39.pyc,,
+cffi/__pycache__/backend_ctypes.cpython-39.pyc,,
+cffi/__pycache__/cffi_opcode.cpython-39.pyc,,
+cffi/__pycache__/commontypes.cpython-39.pyc,,
+cffi/__pycache__/cparser.cpython-39.pyc,,
+cffi/__pycache__/error.cpython-39.pyc,,
+cffi/__pycache__/ffiplatform.cpython-39.pyc,,
+cffi/__pycache__/lock.cpython-39.pyc,,
+cffi/__pycache__/model.cpython-39.pyc,,
+cffi/__pycache__/pkgconfig.cpython-39.pyc,,
+cffi/__pycache__/recompiler.cpython-39.pyc,,
+cffi/__pycache__/setuptools_ext.cpython-39.pyc,,
+cffi/__pycache__/vengine_cpy.cpython-39.pyc,,
+cffi/__pycache__/vengine_gen.cpython-39.pyc,,
+cffi/__pycache__/verifier.cpython-39.pyc,,
+cffi/_cffi_errors.h,sha256=zQXt7uR_m8gUW-fI2hJg0KoSkJFwXv8RGUkEDZ177dQ,3908
+cffi/_cffi_include.h,sha256=Exhmgm9qzHWzWivjfTe0D7Xp4rPUkVxdNuwGhMTMzbw,15055
+cffi/_embedding.h,sha256=Ai33FHblE7XSpHOCp8kPcWwN5_9BV14OvN0JVa6ITpw,18786
+cffi/_imp_emulation.py,sha256=RxREG8zAbI2RPGBww90u_5fi8sWdahpdipOoPzkp7C0,2960
+cffi/_shimmed_dist_utils.py,sha256=Bjj2wm8yZbvFvWEx5AEfmqaqZyZFhYfoyLLQHkXZuao,2230
+cffi/api.py,sha256=alBv6hZQkjpmZplBphdaRn2lPO9-CORs_M7ixabvZWI,42169
+cffi/backend_ctypes.py,sha256=h5ZIzLc6BFVXnGyc9xPqZWUS7qGy7yFSDqXe68Sa8z4,42454
+cffi/cffi_opcode.py,sha256=JDV5l0R0_OadBX_uE7xPPTYtMdmpp8I9UYd6av7aiDU,5731
+cffi/commontypes.py,sha256=7N6zPtCFlvxXMWhHV08psUjdYIK2XgsN3yo5dgua_v4,2805
+cffi/cparser.py,sha256=QUTfmlL-aO-MYR8bFGlvAUHc36OQr7XYLe0WLkGFjRo,44790
+cffi/error.py,sha256=v6xTiS4U0kvDcy4h_BDRo5v39ZQuj-IMRYLv5ETddZs,877
+cffi/ffiplatform.py,sha256=avxFjdikYGJoEtmJO7ewVmwG_VEVl6EZ_WaNhZYCqv4,3584
+cffi/lock.py,sha256=l9TTdwMIMpi6jDkJGnQgE9cvTIR7CAntIJr8EGHt3pY,747
+cffi/model.py,sha256=W30UFQZE73jL5Mx5N81YT77us2W2iJjTm0XYfnwz1cg,21797
+cffi/parse_c_type.h,sha256=OdwQfwM9ktq6vlCB43exFQmxDBtj2MBNdK8LYl15tjw,5976
+cffi/pkgconfig.py,sha256=LP1w7vmWvmKwyqLaU1Z243FOWGNQMrgMUZrvgFuOlco,4374
+cffi/recompiler.py,sha256=78J6lMEEOygXNmjN9-fOFFO3j7eW-iFxSrxfvQb54bY,65509
+cffi/setuptools_ext.py,sha256=0rCwBJ1W7FHWtiMKfNXsSST88V8UXrui5oeXFlDNLG8,9411
+cffi/vengine_cpy.py,sha256=oyQKD23kpE0aChUKA8Jg0e723foPiYzLYEdb-J0MiNs,43881
+cffi/vengine_gen.py,sha256=DUlEIrDiVin1Pnhn1sfoamnS5NLqfJcOdhRoeSNeJRg,26939
+cffi/verifier.py,sha256=oX8jpaohg2Qm3aHcznidAdvrVm5N4sQYG0a3Eo5mIl4,11182

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/WHEEL

@@ -0,0 +1,6 @@
+Wheel-Version: 1.0
+Generator: setuptools (80.9.0)
+Root-Is-Purelib: false
+Tag: cp39-cp39-manylinux_2_17_x86_64
+Tag: cp39-cp39-manylinux2014_x86_64
+

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[distutils.setup_keywords]
+cffi_modules = cffi.setuptools_ext:cffi_modules

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/licenses/AUTHORS

@@ -0,0 +1,8 @@
+This package has been mostly done by Armin Rigo with help from
+Maciej Fijałkowski. The idea is heavily based (although not directly
+copied) from LuaJIT ffi by Mike Pall.
+
+
+Other contributors:
+
+  Google Inc.

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/licenses/LICENSE

@@ -0,0 +1,23 @@
+
+Except when otherwise stated (look for LICENSE files in directories or
+information at the beginning of each file) all software and
+documentation is licensed as follows: 
+
+    MIT No Attribution
+
+    Permission is hereby granted, free of charge, to any person 
+    obtaining a copy of this software and associated documentation 
+    files (the "Software"), to deal in the Software without 
+    restriction, including without limitation the rights to use, 
+    copy, modify, merge, publish, distribute, sublicense, and/or 
+    sell copies of the Software, and to permit persons to whom the 
+    Software is furnished to do so.
+
+    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 
+    OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 
+    THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 
+    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 
+    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 
+    DEALINGS IN THE SOFTWARE.
+

.venv/lib/python3.9/site-packages/cffi-2.0.0.dist-info/top_level.txt

@@ -0,0 +1,2 @@
+_cffi_backend
+cffi

.venv/lib/python3.9/site-packages/cffi/__init__.py

@@ -0,0 +1,14 @@
+__all__ = ['FFI', 'VerificationError', 'VerificationMissing', 'CDefError',
+           'FFIError']
+
+from .api import FFI
+from .error import CDefError, FFIError, VerificationError, VerificationMissing
+from .error import PkgConfigError
+
+__version__ = "2.0.0"
+__version_info__ = (2, 0, 0)
+
+# The verifier module file names are based on the CRC32 of a string that
+# contains the following version number.  It may be older than __version__
+# if nothing is clearly incompatible.
+__version_verifier_modules__ = "0.8.6"

.venv/lib/python3.9/site-packages/cffi/_cffi_errors.h

@@ -0,0 +1,149 @@
+#ifndef CFFI_MESSAGEBOX
+# ifdef _MSC_VER
+#  define CFFI_MESSAGEBOX  1
+# else
+#  define CFFI_MESSAGEBOX  0
+# endif
+#endif
+
+
+#if CFFI_MESSAGEBOX
+/* Windows only: logic to take the Python-CFFI embedding logic
+   initialization errors and display them in a background thread
+   with MessageBox.  The idea is that if the whole program closes
+   as a result of this problem, then likely it is already a console
+   program and you can read the stderr output in the console too.
+   If it is not a console program, then it will likely show its own
+   dialog to complain, or generally not abruptly close, and for this
+   case the background thread should stay alive.
+*/
+static void *volatile _cffi_bootstrap_text;
+
+static PyObject *_cffi_start_error_capture(void)
+{
+    PyObject *result = NULL;
+    PyObject *x, *m, *bi;
+
+    if (InterlockedCompareExchangePointer(&_cffi_bootstrap_text,
+            (void *)1, NULL) != NULL)
+        return (PyObject *)1;
+
+    m = PyImport_AddModule("_cffi_error_capture");
+    if (m == NULL)
+        goto error;
+
+    result = PyModule_GetDict(m);
+    if (result == NULL)
+        goto error;
+
+#if PY_MAJOR_VERSION >= 3
+    bi = PyImport_ImportModule("builtins");
+#else
+    bi = PyImport_ImportModule("__builtin__");
+#endif
+    if (bi == NULL)
+        goto error;
+    PyDict_SetItemString(result, "__builtins__", bi);
+    Py_DECREF(bi);
+
+    x = PyRun_String(
+        "import sys\n"
+        "class FileLike:\n"
+        "  def write(self, x):\n"
+        "    try:\n"
+        "      of.write(x)\n"
+        "    except: pass\n"
+        "    self.buf += x\n"
+        "  def flush(self):\n"
+        "    pass\n"
+        "fl = FileLike()\n"
+        "fl.buf = ''\n"
+        "of = sys.stderr\n"
+        "sys.stderr = fl\n"
+        "def done():\n"
+        "  sys.stderr = of\n"
+        "  return fl.buf\n",   /* make sure the returned value stays alive */
+        Py_file_input,
+        result, result);
+    Py_XDECREF(x);
+
+ error:
+    if (PyErr_Occurred())
+    {
+        PyErr_WriteUnraisable(Py_None);
+        PyErr_Clear();
+    }
+    return result;
+}
+
+#pragma comment(lib, "user32.lib")
+
+static DWORD WINAPI _cffi_bootstrap_dialog(LPVOID ignored)
+{
+    Sleep(666);    /* may be interrupted if the whole process is closing */
+#if PY_MAJOR_VERSION >= 3
+    MessageBoxW(NULL, (wchar_t *)_cffi_bootstrap_text,
+                L"Python-CFFI error",
+                MB_OK | MB_ICONERROR);
+#else
+    MessageBoxA(NULL, (char *)_cffi_bootstrap_text,
+                "Python-CFFI error",
+                MB_OK | MB_ICONERROR);
+#endif
+    _cffi_bootstrap_text = NULL;
+    return 0;
+}
+
+static void _cffi_stop_error_capture(PyObject *ecap)
+{
+    PyObject *s;
+    void *text;
+
+    if (ecap == (PyObject *)1)
+        return;
+
+    if (ecap == NULL)
+        goto error;
+
+    s = PyRun_String("done()", Py_eval_input, ecap, ecap);
+    if (s == NULL)
+        goto error;
+
+    /* Show a dialog box, but in a background thread, and
+       never show multiple dialog boxes at once. */
+#if PY_MAJOR_VERSION >= 3
+    text = PyUnicode_AsWideCharString(s, NULL);
+#else
+    text = PyString_AsString(s);
+#endif
+
+    _cffi_bootstrap_text = text;
+
+    if (text != NULL)
+    {
+        HANDLE h;
+        h = CreateThread(NULL, 0, _cffi_bootstrap_dialog,
+                         NULL, 0, NULL);
+        if (h != NULL)
+            CloseHandle(h);
+    }
+    /* decref the string, but it should stay alive as 'fl.buf'
+       in the small module above.  It will really be freed only if
+       we later get another similar error.  So it's a leak of at
+       most one copy of the small module.  That's fine for this
+       situation which is usually a "fatal error" anyway. */
+    Py_DECREF(s);
+    PyErr_Clear();
+    return;
+
+  error:
+    _cffi_bootstrap_text = NULL;
+    PyErr_Clear();
+}
+
+#else
+
+static PyObject *_cffi_start_error_capture(void) { return NULL; }
+static void _cffi_stop_error_capture(PyObject *ecap) { }
+
+#endif

.venv/lib/python3.9/site-packages/cffi/_cffi_include.h

@@ -0,0 +1,389 @@
+#define _CFFI_
+
+/* We try to define Py_LIMITED_API before including Python.h.
+
+   Mess: we can only define it if Py_DEBUG, Py_TRACE_REFS and
+   Py_REF_DEBUG are not defined.  This is a best-effort approximation:
+   we can learn about Py_DEBUG from pyconfig.h, but it is unclear if
+   the same works for the other two macros.  Py_DEBUG implies them,
+   but not the other way around.
+
+   The implementation is messy (issue #350): on Windows, with _MSC_VER,
+   we have to define Py_LIMITED_API even before including pyconfig.h.
+   In that case, we guess what pyconfig.h will do to the macros above,
+   and check our guess after the #include.
+
+   Note that on Windows, with CPython 3.x, you need >= 3.5 and virtualenv
+   version >= 16.0.0.  With older versions of either, you don't get a
+   copy of PYTHON3.DLL in the virtualenv.  We can't check the version of
+   CPython *before* we even include pyconfig.h.  ffi.set_source() puts
+   a ``#define _CFFI_NO_LIMITED_API'' at the start of this file if it is
+   running on Windows < 3.5, as an attempt at fixing it, but that's
+   arguably wrong because it may not be the target version of Python.
+   Still better than nothing I guess.  As another workaround, you can
+   remove the definition of Py_LIMITED_API here.
+
+   See also 'py_limited_api' in cffi/setuptools_ext.py.
+*/
+#if !defined(_CFFI_USE_EMBEDDING) && !defined(Py_LIMITED_API)
+#  ifdef _MSC_VER
+#    if !defined(_DEBUG) && !defined(Py_DEBUG) && !defined(Py_TRACE_REFS) && !defined(Py_REF_DEBUG) && !defined(_CFFI_NO_LIMITED_API)
+#      define Py_LIMITED_API
+#    endif
+#    include <pyconfig.h>
+     /* sanity-check: Py_LIMITED_API will cause crashes if any of these
+        are also defined.  Normally, the Python file PC/pyconfig.h does not
+        cause any of these to be defined, with the exception that _DEBUG
+        causes Py_DEBUG.  Double-check that. */
+#    ifdef Py_LIMITED_API
+#      if defined(Py_DEBUG)
+#        error "pyconfig.h unexpectedly defines Py_DEBUG, but Py_LIMITED_API is set"
+#      endif
+#      if defined(Py_TRACE_REFS)
+#        error "pyconfig.h unexpectedly defines Py_TRACE_REFS, but Py_LIMITED_API is set"
+#      endif
+#      if defined(Py_REF_DEBUG)
+#        error "pyconfig.h unexpectedly defines Py_REF_DEBUG, but Py_LIMITED_API is set"
+#      endif
+#    endif
+#  else
+#    include <pyconfig.h>
+#    if !defined(Py_DEBUG) && !defined(Py_TRACE_REFS) && !defined(Py_REF_DEBUG) && !defined(_CFFI_NO_LIMITED_API)
+#      define Py_LIMITED_API
+#    endif
+#  endif
+#endif
+
+#include <Python.h>
+#ifdef __cplusplus
+extern "C" {
+#endif
+#include <stddef.h>
+#include "parse_c_type.h"
+
+/* this block of #ifs should be kept exactly identical between
+   c/_cffi_backend.c, cffi/vengine_cpy.py, cffi/vengine_gen.py
+   and cffi/_cffi_include.h */
+#if defined(_MSC_VER)
+# include <malloc.h>   /* for alloca() */
+# if _MSC_VER < 1600   /* MSVC < 2010 */
+   typedef __int8 int8_t;
+   typedef __int16 int16_t;
+   typedef __int32 int32_t;
+   typedef __int64 int64_t;
+   typedef unsigned __int8 uint8_t;
+   typedef unsigned __int16 uint16_t;
+   typedef unsigned __int32 uint32_t;
+   typedef unsigned __int64 uint64_t;
+   typedef __int8 int_least8_t;
+   typedef __int16 int_least16_t;
+   typedef __int32 int_least32_t;
+   typedef __int64 int_least64_t;
+   typedef unsigned __int8 uint_least8_t;
+   typedef unsigned __int16 uint_least16_t;
+   typedef unsigned __int32 uint_least32_t;
+   typedef unsigned __int64 uint_least64_t;
+   typedef __int8 int_fast8_t;
+   typedef __int16 int_fast16_t;
+   typedef __int32 int_fast32_t;
+   typedef __int64 int_fast64_t;
+   typedef unsigned __int8 uint_fast8_t;
+   typedef unsigned __int16 uint_fast16_t;
+   typedef unsigned __int32 uint_fast32_t;
+   typedef unsigned __int64 uint_fast64_t;
+   typedef __int64 intmax_t;
+   typedef unsigned __int64 uintmax_t;
+# else
+#  include <stdint.h>
+# endif
+# if _MSC_VER < 1800   /* MSVC < 2013 */
+#  ifndef __cplusplus
+    typedef unsigned char _Bool;
+#  endif
+# endif
+# define _cffi_float_complex_t   _Fcomplex    /* include <complex.h> for it */
+# define _cffi_double_complex_t  _Dcomplex    /* include <complex.h> for it */
+#else
+# include <stdint.h>
+# if (defined (__SVR4) && defined (__sun)) || defined(_AIX) || defined(__hpux)
+#  include <alloca.h>
+# endif
+# define _cffi_float_complex_t   float _Complex
+# define _cffi_double_complex_t  double _Complex
+#endif
+
+#ifdef __GNUC__
+# define _CFFI_UNUSED_FN  __attribute__((unused))
+#else
+# define _CFFI_UNUSED_FN  /* nothing */
+#endif
+
+#ifdef __cplusplus
+# ifndef _Bool
+   typedef bool _Bool;   /* semi-hackish: C++ has no _Bool; bool is builtin */
+# endif
+#endif
+
+/**********  CPython-specific section  **********/
+#ifndef PYPY_VERSION
+
+
+#if PY_MAJOR_VERSION >= 3
+# define PyInt_FromLong PyLong_FromLong
+#endif
+
+#define _cffi_from_c_double PyFloat_FromDouble
+#define _cffi_from_c_float PyFloat_FromDouble
+#define _cffi_from_c_long PyInt_FromLong
+#define _cffi_from_c_ulong PyLong_FromUnsignedLong
+#define _cffi_from_c_longlong PyLong_FromLongLong
+#define _cffi_from_c_ulonglong PyLong_FromUnsignedLongLong
+#define _cffi_from_c__Bool PyBool_FromLong
+
+#define _cffi_to_c_double PyFloat_AsDouble
+#define _cffi_to_c_float PyFloat_AsDouble
+
+#define _cffi_from_c_int(x, type)                                        \
+    (((type)-1) > 0 ? /* unsigned */                                     \
+        (sizeof(type) < sizeof(long) ?                                   \
+            PyInt_FromLong((long)x) :                                    \
+         sizeof(type) == sizeof(long) ?                                  \
+            PyLong_FromUnsignedLong((unsigned long)x) :                  \
+            PyLong_FromUnsignedLongLong((unsigned long long)x)) :        \
+        (sizeof(type) <= sizeof(long) ?                                  \
+            PyInt_FromLong((long)x) :                                    \
+            PyLong_FromLongLong((long long)x)))
+
+#define _cffi_to_c_int(o, type)                                          \
+    ((type)(                                                             \
+     sizeof(type) == 1 ? (((type)-1) > 0 ? (type)_cffi_to_c_u8(o)        \
+                                         : (type)_cffi_to_c_i8(o)) :     \
+     sizeof(type) == 2 ? (((type)-1) > 0 ? (type)_cffi_to_c_u16(o)       \
+                                         : (type)_cffi_to_c_i16(o)) :    \
+     sizeof(type) == 4 ? (((type)-1) > 0 ? (type)_cffi_to_c_u32(o)       \
+                                         : (type)_cffi_to_c_i32(o)) :    \
+     sizeof(type) == 8 ? (((type)-1) > 0 ? (type)_cffi_to_c_u64(o)       \
+                                         : (type)_cffi_to_c_i64(o)) :    \
+     (Py_FatalError("unsupported size for type " #type), (type)0)))
+
+#define _cffi_to_c_i8                                                    \
+                 ((int(*)(PyObject *))_cffi_exports[1])
+#define _cffi_to_c_u8                                                    \
+                 ((int(*)(PyObject *))_cffi_exports[2])
+#define _cffi_to_c_i16                                                   \
+                 ((int(*)(PyObject *))_cffi_exports[3])
+#define _cffi_to_c_u16                                                   \
+                 ((int(*)(PyObject *))_cffi_exports[4])
+#define _cffi_to_c_i32                                                   \
+                 ((int(*)(PyObject *))_cffi_exports[5])
+#define _cffi_to_c_u32                                                   \
+                 ((unsigned int(*)(PyObject *))_cffi_exports[6])
+#define _cffi_to_c_i64                                                   \
+                 ((long long(*)(PyObject *))_cffi_exports[7])
+#define _cffi_to_c_u64                                                   \
+                 ((unsigned long long(*)(PyObject *))_cffi_exports[8])
+#define _cffi_to_c_char                                                  \
+                 ((int(*)(PyObject *))_cffi_exports[9])
+#define _cffi_from_c_pointer                                             \
+    ((PyObject *(*)(char *, struct _cffi_ctypedescr *))_cffi_exports[10])
+#define _cffi_to_c_pointer                                               \
+    ((char *(*)(PyObject *, struct _cffi_ctypedescr *))_cffi_exports[11])
+#define _cffi_get_struct_layout                                          \
+    not used any more
+#define _cffi_restore_errno                                              \
+    ((void(*)(void))_cffi_exports[13])
+#define _cffi_save_errno                                                 \
+    ((void(*)(void))_cffi_exports[14])
+#define _cffi_from_c_char                                                \
+    ((PyObject *(*)(char))_cffi_exports[15])
+#define _cffi_from_c_deref                                               \
+    ((PyObject *(*)(char *, struct _cffi_ctypedescr *))_cffi_exports[16])
+#define _cffi_to_c                                                       \
+    ((int(*)(char *, struct _cffi_ctypedescr *, PyObject *))_cffi_exports[17])
+#define _cffi_from_c_struct                                              \
+    ((PyObject *(*)(char *, struct _cffi_ctypedescr *))_cffi_exports[18])
+#define _cffi_to_c_wchar_t                                               \
+    ((_cffi_wchar_t(*)(PyObject *))_cffi_exports[19])
+#define _cffi_from_c_wchar_t                                             \
+    ((PyObject *(*)(_cffi_wchar_t))_cffi_exports[20])
+#define _cffi_to_c_long_double                                           \
+    ((long double(*)(PyObject *))_cffi_exports[21])
+#define _cffi_to_c__Bool                                                 \
+    ((_Bool(*)(PyObject *))_cffi_exports[22])
+#define _cffi_prepare_pointer_call_argument                              \
+    ((Py_ssize_t(*)(struct _cffi_ctypedescr *,                           \
+                    PyObject *, char **))_cffi_exports[23])
+#define _cffi_convert_array_from_object                                  \
+    ((int(*)(char *, struct _cffi_ctypedescr *, PyObject *))_cffi_exports[24])
+#define _CFFI_CPIDX  25
+#define _cffi_call_python                                                \
+    ((void(*)(struct _cffi_externpy_s *, char *))_cffi_exports[_CFFI_CPIDX])
+#define _cffi_to_c_wchar3216_t                                           \
+    ((int(*)(PyObject *))_cffi_exports[26])
+#define _cffi_from_c_wchar3216_t                                         \
+    ((PyObject *(*)(int))_cffi_exports[27])
+#define _CFFI_NUM_EXPORTS 28
+
+struct _cffi_ctypedescr;
+
+static void *_cffi_exports[_CFFI_NUM_EXPORTS];
+
+#define _cffi_type(index)   (                           \
+    assert((((uintptr_t)_cffi_types[index]) & 1) == 0), \
+    (struct _cffi_ctypedescr *)_cffi_types[index])
+
+static PyObject *_cffi_init(const char *module_name, Py_ssize_t version,
+                            const struct _cffi_type_context_s *ctx)
+{
+    PyObject *module, *o_arg, *new_module;
+    void *raw[] = {
+        (void *)module_name,
+        (void *)version,
+        (void *)_cffi_exports,
+        (void *)ctx,
+    };
+
+    module = PyImport_ImportModule("_cffi_backend");
+    if (module == NULL)
+        goto failure;
+
+    o_arg = PyLong_FromVoidPtr((void *)raw);
+    if (o_arg == NULL)
+        goto failure;
+
+    new_module = PyObject_CallMethod(
+        module, (char *)"_init_cffi_1_0_external_module", (char *)"O", o_arg);
+
+    Py_DECREF(o_arg);
+    Py_DECREF(module);
+    return new_module;
+
+  failure:
+    Py_XDECREF(module);
+    return NULL;
+}
+
+
+#ifdef HAVE_WCHAR_H
+typedef wchar_t _cffi_wchar_t;
+#else
+typedef uint16_t _cffi_wchar_t;   /* same random pick as _cffi_backend.c */
+#endif
+
+_CFFI_UNUSED_FN static uint16_t _cffi_to_c_char16_t(PyObject *o)
+{
+    if (sizeof(_cffi_wchar_t) == 2)
+        return (uint16_t)_cffi_to_c_wchar_t(o);
+    else
+        return (uint16_t)_cffi_to_c_wchar3216_t(o);
+}
+
+_CFFI_UNUSED_FN static PyObject *_cffi_from_c_char16_t(uint16_t x)
+{
+    if (sizeof(_cffi_wchar_t) == 2)
+        return _cffi_from_c_wchar_t((_cffi_wchar_t)x);
+    else
+        return _cffi_from_c_wchar3216_t((int)x);
+}
+
+_CFFI_UNUSED_FN static int _cffi_to_c_char32_t(PyObject *o)
+{
+    if (sizeof(_cffi_wchar_t) == 4)
+        return (int)_cffi_to_c_wchar_t(o);
+    else
+        return (int)_cffi_to_c_wchar3216_t(o);
+}
+
+_CFFI_UNUSED_FN static PyObject *_cffi_from_c_char32_t(unsigned int x)
+{
+    if (sizeof(_cffi_wchar_t) == 4)
+        return _cffi_from_c_wchar_t((_cffi_wchar_t)x);
+    else
+        return _cffi_from_c_wchar3216_t((int)x);
+}
+
+union _cffi_union_alignment_u {
+    unsigned char m_char;
+    unsigned short m_short;
+    unsigned int m_int;
+    unsigned long m_long;
+    unsigned long long m_longlong;
+    float m_float;
+    double m_double;
+    long double m_longdouble;
+};
+
+struct _cffi_freeme_s {
+    struct _cffi_freeme_s *next;
+    union _cffi_union_alignment_u alignment;
+};
+
+_CFFI_UNUSED_FN static int
+_cffi_convert_array_argument(struct _cffi_ctypedescr *ctptr, PyObject *arg,
+                             char **output_data, Py_ssize_t datasize,
+                             struct _cffi_freeme_s **freeme)
+{
+    char *p;
+    if (datasize < 0)
+        return -1;
+
+    p = *output_data;
+    if (p == NULL) {
+        struct _cffi_freeme_s *fp = (struct _cffi_freeme_s *)PyObject_Malloc(
+            offsetof(struct _cffi_freeme_s, alignment) + (size_t)datasize);
+        if (fp == NULL)
+            return -1;
+        fp->next = *freeme;
+        *freeme = fp;
+        p = *output_data = (char *)&fp->alignment;
+    }
+    memset((void *)p, 0, (size_t)datasize);
+    return _cffi_convert_array_from_object(p, ctptr, arg);
+}
+
+_CFFI_UNUSED_FN static void
+_cffi_free_array_arguments(struct _cffi_freeme_s *freeme)
+{
+    do {
+        void *p = (void *)freeme;
+        freeme = freeme->next;
+        PyObject_Free(p);
+    } while (freeme != NULL);
+}
+
+/**********  end CPython-specific section  **********/
+#else
+_CFFI_UNUSED_FN
+static void (*_cffi_call_python_org)(struct _cffi_externpy_s *, char *);
+# define _cffi_call_python  _cffi_call_python_org
+#endif
+
+
+#define _cffi_array_len(array)   (sizeof(array) / sizeof((array)[0]))
+
+#define _cffi_prim_int(size, sign)                                      \
+    ((size) == 1 ? ((sign) ? _CFFI_PRIM_INT8  : _CFFI_PRIM_UINT8)  :    \
+     (size) == 2 ? ((sign) ? _CFFI_PRIM_INT16 : _CFFI_PRIM_UINT16) :    \
+     (size) == 4 ? ((sign) ? _CFFI_PRIM_INT32 : _CFFI_PRIM_UINT32) :    \
+     (size) == 8 ? ((sign) ? _CFFI_PRIM_INT64 : _CFFI_PRIM_UINT64) :    \
+     _CFFI__UNKNOWN_PRIM)
+
+#define _cffi_prim_float(size)                                          \
+    ((size) == sizeof(float) ? _CFFI_PRIM_FLOAT :                       \
+     (size) == sizeof(double) ? _CFFI_PRIM_DOUBLE :                     \
+     (size) == sizeof(long double) ? _CFFI__UNKNOWN_LONG_DOUBLE :       \
+     _CFFI__UNKNOWN_FLOAT_PRIM)
+
+#define _cffi_check_int(got, got_nonpos, expected)      \
+    ((got_nonpos) == (expected <= 0) &&                 \
+     (got) == (unsigned long long)expected)
+
+#ifdef MS_WIN32
+# define _cffi_stdcall  __stdcall
+#else
+# define _cffi_stdcall  /* nothing */
+#endif
+
+#ifdef __cplusplus
+}
+#endif

.venv/lib/python3.9/site-packages/cffi/_embedding.h

@@ -0,0 +1,550 @@
+
+/***** Support code for embedding *****/
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+#if defined(_WIN32)
+#  define CFFI_DLLEXPORT  __declspec(dllexport)
+#elif defined(__GNUC__)
+#  define CFFI_DLLEXPORT  __attribute__((visibility("default")))
+#else
+#  define CFFI_DLLEXPORT  /* nothing */
+#endif
+
+
+/* There are two global variables of type _cffi_call_python_fnptr:
+
+   * _cffi_call_python, which we declare just below, is the one called
+     by ``extern "Python"`` implementations.
+
+   * _cffi_call_python_org, which on CPython is actually part of the
+     _cffi_exports[] array, is the function pointer copied from
+     _cffi_backend.  If _cffi_start_python() fails, then this is set
+     to NULL; otherwise, it should never be NULL.
+
+   After initialization is complete, both are equal.  However, the
+   first one remains equal to &_cffi_start_and_call_python until the
+   very end of initialization, when we are (or should be) sure that
+   concurrent threads also see a completely initialized world, and
+   only then is it changed.
+*/
+#undef _cffi_call_python
+typedef void (*_cffi_call_python_fnptr)(struct _cffi_externpy_s *, char *);
+static void _cffi_start_and_call_python(struct _cffi_externpy_s *, char *);
+static _cffi_call_python_fnptr _cffi_call_python = &_cffi_start_and_call_python;
+
+
+#ifndef _MSC_VER
+   /* --- Assuming a GCC not infinitely old --- */
+# define cffi_compare_and_swap(l,o,n)  __sync_bool_compare_and_swap(l,o,n)
+# define cffi_write_barrier()          __sync_synchronize()
+# if !defined(__amd64__) && !defined(__x86_64__) &&   \
+     !defined(__i386__) && !defined(__i386)
+#   define cffi_read_barrier()         __sync_synchronize()
+# else
+#   define cffi_read_barrier()         (void)0
+# endif
+#else
+   /* --- Windows threads version --- */
+# include <Windows.h>
+# define cffi_compare_and_swap(l,o,n) \
+                               (InterlockedCompareExchangePointer(l,n,o) == (o))
+# define cffi_write_barrier()       InterlockedCompareExchange(&_cffi_dummy,0,0)
+# define cffi_read_barrier()           (void)0
+static volatile LONG _cffi_dummy;
+#endif
+
+#ifdef WITH_THREAD
+# ifndef _MSC_VER
+#  include <pthread.h>
+   static pthread_mutex_t _cffi_embed_startup_lock;
+# else
+   static CRITICAL_SECTION _cffi_embed_startup_lock;
+# endif
+  static char _cffi_embed_startup_lock_ready = 0;
+#endif
+
+static void _cffi_acquire_reentrant_mutex(void)
+{
+    static void *volatile lock = NULL;
+
+    while (!cffi_compare_and_swap(&lock, NULL, (void *)1)) {
+        /* should ideally do a spin loop instruction here, but
+           hard to do it portably and doesn't really matter I
+           think: pthread_mutex_init() should be very fast, and
+           this is only run at start-up anyway. */
+    }
+
+#ifdef WITH_THREAD
+    if (!_cffi_embed_startup_lock_ready) {
+# ifndef _MSC_VER
+        pthread_mutexattr_t attr;
+        pthread_mutexattr_init(&attr);
+        pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
+        pthread_mutex_init(&_cffi_embed_startup_lock, &attr);
+# else
+        InitializeCriticalSection(&_cffi_embed_startup_lock);
+# endif
+        _cffi_embed_startup_lock_ready = 1;
+    }
+#endif
+
+    while (!cffi_compare_and_swap(&lock, (void *)1, NULL))
+        ;
+
+#ifndef _MSC_VER
+    pthread_mutex_lock(&_cffi_embed_startup_lock);
+#else
+    EnterCriticalSection(&_cffi_embed_startup_lock);
+#endif
+}
+
+static void _cffi_release_reentrant_mutex(void)
+{
+#ifndef _MSC_VER
+    pthread_mutex_unlock(&_cffi_embed_startup_lock);
+#else
+    LeaveCriticalSection(&_cffi_embed_startup_lock);
+#endif
+}
+
+
+/**********  CPython-specific section  **********/
+#ifndef PYPY_VERSION
+
+#include "_cffi_errors.h"
+
+
+#define _cffi_call_python_org  _cffi_exports[_CFFI_CPIDX]
+
+PyMODINIT_FUNC _CFFI_PYTHON_STARTUP_FUNC(void);   /* forward */
+
+static void _cffi_py_initialize(void)
+{
+    /* XXX use initsigs=0, which "skips initialization registration of
+       signal handlers, which might be useful when Python is
+       embedded" according to the Python docs.  But review and think
+       if it should be a user-controllable setting.
+
+       XXX we should also give a way to write errors to a buffer
+       instead of to stderr.
+
+       XXX if importing 'site' fails, CPython (any version) calls
+       exit().  Should we try to work around this behavior here?
+    */
+    Py_InitializeEx(0);
+}
+
+static int _cffi_initialize_python(void)
+{
+    /* This initializes Python, imports _cffi_backend, and then the
+       present .dll/.so is set up as a CPython C extension module.
+    */
+    int result;
+    PyGILState_STATE state;
+    PyObject *pycode=NULL, *global_dict=NULL, *x;
+    PyObject *builtins;
+
+    state = PyGILState_Ensure();
+
+    /* Call the initxxx() function from the present module.  It will
+       create and initialize us as a CPython extension module, instead
+       of letting the startup Python code do it---it might reimport
+       the same .dll/.so and get maybe confused on some platforms.
+       It might also have troubles locating the .dll/.so again for all
+       I know.
+    */
+    (void)_CFFI_PYTHON_STARTUP_FUNC();
+    if (PyErr_Occurred())
+        goto error;
+
+    /* Now run the Python code provided to ffi.embedding_init_code().
+     */
+    pycode = Py_CompileString(_CFFI_PYTHON_STARTUP_CODE,
+                              "<init code for '" _CFFI_MODULE_NAME "'>",
+                              Py_file_input);
+    if (pycode == NULL)
+        goto error;
+    global_dict = PyDict_New();
+    if (global_dict == NULL)
+        goto error;
+    builtins = PyEval_GetBuiltins();
+    if (builtins == NULL)
+        goto error;
+    if (PyDict_SetItemString(global_dict, "__builtins__", builtins) < 0)
+        goto error;
+    x = PyEval_EvalCode(
+#if PY_MAJOR_VERSION < 3
+                        (PyCodeObject *)
+#endif
+                        pycode, global_dict, global_dict);
+    if (x == NULL)
+        goto error;
+    Py_DECREF(x);
+
+    /* Done!  Now if we've been called from
+       _cffi_start_and_call_python() in an ``extern "Python"``, we can
+       only hope that the Python code did correctly set up the
+       corresponding @ffi.def_extern() function.  Otherwise, the
+       general logic of ``extern "Python"`` functions (inside the
+       _cffi_backend module) will find that the reference is still
+       missing and print an error.
+     */
+    result = 0;
+ done:
+    Py_XDECREF(pycode);
+    Py_XDECREF(global_dict);
+    PyGILState_Release(state);
+    return result;
+
+ error:;
+    {
+        /* Print as much information as potentially useful.
+           Debugging load-time failures with embedding is not fun
+        */
+        PyObject *ecap;
+        PyObject *exception, *v, *tb, *f, *modules, *mod;
+        PyErr_Fetch(&exception, &v, &tb);
+        ecap = _cffi_start_error_capture();
+        f = PySys_GetObject((char *)"stderr");
+        if (f != NULL && f != Py_None) {
+            PyFile_WriteString(
+                "Failed to initialize the Python-CFFI embedding logic:\n\n", f);
+        }
+
+        if (exception != NULL) {
+            PyErr_NormalizeException(&exception, &v, &tb);
+            PyErr_Display(exception, v, tb);
+        }
+        Py_XDECREF(exception);
+        Py_XDECREF(v);
+        Py_XDECREF(tb);
+
+        if (f != NULL && f != Py_None) {
+            PyFile_WriteString("\nFrom: " _CFFI_MODULE_NAME
+                               "\ncompiled with cffi version: 2.0.0"
+                               "\n_cffi_backend module: ", f);
+            modules = PyImport_GetModuleDict();
+            mod = PyDict_GetItemString(modules, "_cffi_backend");
+            if (mod == NULL) {
+                PyFile_WriteString("not loaded", f);
+            }
+            else {
+                v = PyObject_GetAttrString(mod, "__file__");
+                PyFile_WriteObject(v, f, 0);
+                Py_XDECREF(v);
+            }
+            PyFile_WriteString("\nsys.path: ", f);
+            PyFile_WriteObject(PySys_GetObject((char *)"path"), f, 0);
+            PyFile_WriteString("\n\n", f);
+        }
+        _cffi_stop_error_capture(ecap);
+    }
+    result = -1;
+    goto done;
+}
+
+#if PY_VERSION_HEX < 0x03080000
+PyAPI_DATA(char *) _PyParser_TokenNames[];  /* from CPython */
+#endif
+
+static int _cffi_carefully_make_gil(void)
+{
+    /* This does the basic initialization of Python.  It can be called
+       completely concurrently from unrelated threads.  It assumes
+       that we don't hold the GIL before (if it exists), and we don't
+       hold it afterwards.
+
+       (What it really does used to be completely different in Python 2
+       and Python 3, with the Python 2 solution avoiding the spin-lock
+       around the Py_InitializeEx() call.  However, after recent changes
+       to CPython 2.7 (issue #358) it no longer works.  So we use the
+       Python 3 solution everywhere.)
+
+       This initializes Python by calling Py_InitializeEx().
+       Important: this must not be called concurrently at all.
+       So we use a global variable as a simple spin lock.  This global
+       variable must be from 'libpythonX.Y.so', not from this
+       cffi-based extension module, because it must be shared from
+       different cffi-based extension modules.
+
+       In Python < 3.8, we choose
+       _PyParser_TokenNames[0] as a completely arbitrary pointer value
+       that is never written to.  The default is to point to the
+       string "ENDMARKER".  We change it temporarily to point to the
+       next character in that string.  (Yes, I know it's REALLY
+       obscure.)
+
+       In Python >= 3.8, this string array is no longer writable, so
+       instead we pick PyCapsuleType.tp_version_tag.  We can't change
+       Python < 3.8 because someone might use a mixture of cffi
+       embedded modules, some of which were compiled before this file
+       changed.
+
+       In Python >= 3.12, this stopped working because that particular
+       tp_version_tag gets modified during interpreter startup.  It's
+       arguably a bad idea before 3.12 too, but again we can't change
+       that because someone might use a mixture of cffi embedded
+       modules, and no-one reported a bug so far.  In Python >= 3.12
+       we go instead for PyCapsuleType.tp_as_buffer, which is supposed
+       to always be NULL.  We write to it temporarily a pointer to
+       a struct full of NULLs, which is semantically the same.
+    */
+
+#ifdef WITH_THREAD
+# if PY_VERSION_HEX < 0x03080000
+    char *volatile *lock = (char *volatile *)_PyParser_TokenNames;
+    char *old_value, *locked_value;
+
+    while (1) {    /* spin loop */
+        old_value = *lock;
+        locked_value = old_value + 1;
+        if (old_value[0] == 'E') {
+            assert(old_value[1] == 'N');
+            if (cffi_compare_and_swap(lock, old_value, locked_value))
+                break;
+        }
+        else {
+            assert(old_value[0] == 'N');
+            /* should ideally do a spin loop instruction here, but
+               hard to do it portably and doesn't really matter I
+               think: PyEval_InitThreads() should be very fast, and
+               this is only run at start-up anyway. */
+        }
+    }
+# else
+#  if PY_VERSION_HEX < 0x030C0000
+    int volatile *lock = (int volatile *)&PyCapsule_Type.tp_version_tag;
+    int old_value, locked_value = -42;
+    assert(!(PyCapsule_Type.tp_flags & Py_TPFLAGS_HAVE_VERSION_TAG));
+#  else
+    static struct ebp_s { PyBufferProcs buf; int mark; } empty_buffer_procs;
+    empty_buffer_procs.mark = -42;
+    PyBufferProcs *volatile *lock = (PyBufferProcs *volatile *)
+        &PyCapsule_Type.tp_as_buffer;
+    PyBufferProcs *old_value, *locked_value = &empty_buffer_procs.buf;
+#  endif
+
+    while (1) {    /* spin loop */
+        old_value = *lock;
+        if (old_value == 0) {
+            if (cffi_compare_and_swap(lock, old_value, locked_value))
+                break;
+        }
+        else {
+#  if PY_VERSION_HEX < 0x030C0000
+            assert(old_value == locked_value);
+#  else
+            /* The pointer should point to a possibly different
+               empty_buffer_procs from another C extension module */
+            assert(((struct ebp_s *)old_value)->mark == -42);
+#  endif
+            /* should ideally do a spin loop instruction here, but
+               hard to do it portably and doesn't really matter I
+               think: PyEval_InitThreads() should be very fast, and
+               this is only run at start-up anyway. */
+        }
+    }
+# endif
+#endif
+
+    /* call Py_InitializeEx() */
+    if (!Py_IsInitialized()) {
+        _cffi_py_initialize();
+#if PY_VERSION_HEX < 0x03070000
+        PyEval_InitThreads();
+#endif
+        PyEval_SaveThread();  /* release the GIL */
+        /* the returned tstate must be the one that has been stored into the
+           autoTLSkey by _PyGILState_Init() called from Py_Initialize(). */
+    }
+    else {
+#if PY_VERSION_HEX < 0x03070000
+        /* PyEval_InitThreads() is always a no-op from CPython 3.7 */
+        PyGILState_STATE state = PyGILState_Ensure();
+        PyEval_InitThreads();
+        PyGILState_Release(state);
+#endif
+    }
+
+#ifdef WITH_THREAD
+    /* release the lock */
+    while (!cffi_compare_and_swap(lock, locked_value, old_value))
+        ;
+#endif
+
+    return 0;
+}
+
+/**********  end CPython-specific section  **********/
+
+
+#else
+
+
+/**********  PyPy-specific section  **********/
+
+PyMODINIT_FUNC _CFFI_PYTHON_STARTUP_FUNC(const void *[]);   /* forward */
+
+static struct _cffi_pypy_init_s {
+    const char *name;
+    void *func;    /* function pointer */
+    const char *code;
+} _cffi_pypy_init = {
+    _CFFI_MODULE_NAME,
+    _CFFI_PYTHON_STARTUP_FUNC,
+    _CFFI_PYTHON_STARTUP_CODE,
+};
+
+extern int pypy_carefully_make_gil(const char *);
+extern int pypy_init_embedded_cffi_module(int, struct _cffi_pypy_init_s *);
+
+static int _cffi_carefully_make_gil(void)
+{
+    return pypy_carefully_make_gil(_CFFI_MODULE_NAME);
+}
+
+static int _cffi_initialize_python(void)
+{
+    return pypy_init_embedded_cffi_module(0xB011, &_cffi_pypy_init);
+}
+
+/**********  end PyPy-specific section  **********/
+
+
+#endif
+
+
+#ifdef __GNUC__
+__attribute__((noinline))
+#endif
+static _cffi_call_python_fnptr _cffi_start_python(void)
+{
+    /* Delicate logic to initialize Python.  This function can be
+       called multiple times concurrently, e.g. when the process calls
+       its first ``extern "Python"`` functions in multiple threads at
+       once.  It can also be called recursively, in which case we must
+       ignore it.  We also have to consider what occurs if several
+       different cffi-based extensions reach this code in parallel
+       threads---it is a different copy of the code, then, and we
+       can't have any shared global variable unless it comes from
+       'libpythonX.Y.so'.
+
+       Idea:
+
+       * _cffi_carefully_make_gil(): "carefully" call
+         PyEval_InitThreads() (possibly with Py_InitializeEx() first).
+
+       * then we use a (local) custom lock to make sure that a call to this
+         cffi-based extension will wait if another call to the *same*
+         extension is running the initialization in another thread.
+         It is reentrant, so that a recursive call will not block, but
+         only one from a different thread.
+
+       * then we grab the GIL and (Python 2) we call Py_InitializeEx().
+         At this point, concurrent calls to Py_InitializeEx() are not
+         possible: we have the GIL.
+
+       * do the rest of the specific initialization, which may
+         temporarily release the GIL but not the custom lock.
+         Only release the custom lock when we are done.
+    */
+    static char called = 0;
+
+    if (_cffi_carefully_make_gil() != 0)
+        return NULL;
+
+    _cffi_acquire_reentrant_mutex();
+
+    /* Here the GIL exists, but we don't have it.  We're only protected
+       from concurrency by the reentrant mutex. */
+
+    /* This file only initializes the embedded module once, the first
+       time this is called, even if there are subinterpreters. */
+    if (!called) {
+        called = 1;  /* invoke _cffi_initialize_python() only once,
+                        but don't set '_cffi_call_python' right now,
+                        otherwise concurrent threads won't call
+                        this function at all (we need them to wait) */
+        if (_cffi_initialize_python() == 0) {
+            /* now initialization is finished.  Switch to the fast-path. */
+
+            /* We would like nobody to see the new value of
+               '_cffi_call_python' without also seeing the rest of the
+               data initialized.  However, this is not possible.  But
+               the new value of '_cffi_call_python' is the function
+               'cffi_call_python()' from _cffi_backend.  So:  */
+            cffi_write_barrier();
+            /* ^^^ we put a write barrier here, and a corresponding
+               read barrier at the start of cffi_call_python().  This
+               ensures that after that read barrier, we see everything
+               done here before the write barrier.
+            */
+
+            assert(_cffi_call_python_org != NULL);
+            _cffi_call_python = (_cffi_call_python_fnptr)_cffi_call_python_org;
+        }
+        else {
+            /* initialization failed.  Reset this to NULL, even if it was
+               already set to some other value.  Future calls to
+               _cffi_start_python() are still forced to occur, and will
+               always return NULL from now on. */
+            _cffi_call_python_org = NULL;
+        }
+    }
+
+    _cffi_release_reentrant_mutex();
+
+    return (_cffi_call_python_fnptr)_cffi_call_python_org;
+}
+
+static
+void _cffi_start_and_call_python(struct _cffi_externpy_s *externpy, char *args)
+{
+    _cffi_call_python_fnptr fnptr;
+    int current_err = errno;
+#ifdef _MSC_VER
+    int current_lasterr = GetLastError();
+#endif
+    fnptr = _cffi_start_python();
+    if (fnptr == NULL) {
+        fprintf(stderr, "function %s() called, but initialization code "
+                        "failed.  Returning 0.\n", externpy->name);
+        memset(args, 0, externpy->size_of_result);
+    }
+#ifdef _MSC_VER
+    SetLastError(current_lasterr);
+#endif
+    errno = current_err;
+
+    if (fnptr != NULL)
+        fnptr(externpy, args);
+}
+
+
+/* The cffi_start_python() function makes sure Python is initialized
+   and our cffi module is set up.  It can be called manually from the
+   user C code.  The same effect is obtained automatically from any
+   dll-exported ``extern "Python"`` function.  This function returns
+   -1 if initialization failed, 0 if all is OK.  */
+_CFFI_UNUSED_FN
+static int cffi_start_python(void)
+{
+    if (_cffi_call_python == &_cffi_start_and_call_python) {
+        if (_cffi_start_python() == NULL)
+            return -1;
+    }
+    cffi_read_barrier();
+    return 0;
+}
+
+#undef cffi_compare_and_swap
+#undef cffi_write_barrier
+#undef cffi_read_barrier
+
+#ifdef __cplusplus
+}
+#endif

.venv/lib/python3.9/site-packages/cffi/_imp_emulation.py

@@ -0,0 +1,83 @@
+
+try:
+    # this works on Python < 3.12
+    from imp import *
+
+except ImportError:
+    # this is a limited emulation for Python >= 3.12.
+    # Note that this is used only for tests or for the old ffi.verify().
+    # This is copied from the source code of Python 3.11.
+
+    from _imp import (acquire_lock, release_lock,
+                      is_builtin, is_frozen)
+
+    from importlib._bootstrap import _load
+
+    from importlib import machinery
+    import os
+    import sys
+    import tokenize
+
+    SEARCH_ERROR = 0
+    PY_SOURCE = 1
+    PY_COMPILED = 2
+    C_EXTENSION = 3
+    PY_RESOURCE = 4
+    PKG_DIRECTORY = 5
+    C_BUILTIN = 6
+    PY_FROZEN = 7
+    PY_CODERESOURCE = 8
+    IMP_HOOK = 9
+
+    def get_suffixes():
+        extensions = [(s, 'rb', C_EXTENSION)
+                      for s in machinery.EXTENSION_SUFFIXES]
+        source = [(s, 'r', PY_SOURCE) for s in machinery.SOURCE_SUFFIXES]
+        bytecode = [(s, 'rb', PY_COMPILED) for s in machinery.BYTECODE_SUFFIXES]
+        return extensions + source + bytecode
+
+    def find_module(name, path=None):
+        if not isinstance(name, str):
+            raise TypeError("'name' must be a str, not {}".format(type(name)))
+        elif not isinstance(path, (type(None), list)):
+            # Backwards-compatibility
+            raise RuntimeError("'path' must be None or a list, "
+                               "not {}".format(type(path)))
+
+        if path is None:
+            if is_builtin(name):
+                return None, None, ('', '', C_BUILTIN)
+            elif is_frozen(name):
+                return None, None, ('', '', PY_FROZEN)
+            else:
+                path = sys.path
+
+        for entry in path:
+            package_directory = os.path.join(entry, name)
+            for suffix in ['.py', machinery.BYTECODE_SUFFIXES[0]]:
+                package_file_name = '__init__' + suffix
+                file_path = os.path.join(package_directory, package_file_name)
+                if os.path.isfile(file_path):
+                    return None, package_directory, ('', '', PKG_DIRECTORY)
+            for suffix, mode, type_ in get_suffixes():
+                file_name = name + suffix
+                file_path = os.path.join(entry, file_name)
+                if os.path.isfile(file_path):
+                    break
+            else:
+                continue
+            break  # Break out of outer loop when breaking out of inner loop.
+        else:
+            raise ImportError(name, name=name)
+
+        encoding = None
+        if 'b' not in mode:
+            with open(file_path, 'rb') as file:
+                encoding = tokenize.detect_encoding(file.readline)[0]
+        file = open(file_path, mode, encoding=encoding)
+        return file, file_path, (suffix, mode, type_)
+
+    def load_dynamic(name, path, file=None):
+        loader = machinery.ExtensionFileLoader(name, path)
+        spec = machinery.ModuleSpec(name=name, loader=loader, origin=path)
+        return _load(spec)

.venv/lib/python3.9/site-packages/cffi/_shimmed_dist_utils.py

@@ -0,0 +1,45 @@
+"""
+Temporary shim module to indirect the bits of distutils we need from setuptools/distutils while providing useful
+error messages beyond `No module named 'distutils' on Python >= 3.12, or when setuptools' vendored distutils is broken.
+
+This is a compromise to avoid a hard-dep on setuptools for Python >= 3.12, since many users don't need runtime compilation support from CFFI.
+"""
+import sys
+
+try:
+    # import setuptools first; this is the most robust way to ensure its embedded distutils is available
+    # (the .pth shim should usually work, but this is even more robust)
+    import setuptools
+except Exception as ex:
+    if sys.version_info >= (3, 12):
+        # Python 3.12 has no built-in distutils to fall back on, so any import problem is fatal
+        raise Exception("This CFFI feature requires setuptools on Python >= 3.12. The setuptools module is missing or non-functional.") from ex
+
+    # silently ignore on older Pythons (support fallback to stdlib distutils where available)
+else:
+    del setuptools
+
+try:
+    # bring in just the bits of distutils we need, whether they really came from setuptools or stdlib-embedded distutils
+    from distutils import log, sysconfig
+    from distutils.ccompiler import CCompiler
+    from distutils.command.build_ext import build_ext
+    from distutils.core import Distribution, Extension
+    from distutils.dir_util import mkpath
+    from distutils.errors import DistutilsSetupError, CompileError, LinkError
+    from distutils.log import set_threshold, set_verbosity
+
+    if sys.platform == 'win32':
+        try:
+            # FUTURE: msvc9compiler module was removed in setuptools 74; consider removing, as it's only used by an ancient patch in `recompiler`
+            from distutils.msvc9compiler import MSVCCompiler
+        except ImportError:
+            MSVCCompiler = None
+except Exception as ex:
+    if sys.version_info >= (3, 12):
+        raise Exception("This CFFI feature requires setuptools on Python >= 3.12. Please install the setuptools package.") from ex
+
+    # anything older, just let the underlying distutils import error fly
+    raise Exception("This CFFI feature requires distutils. Please install the distutils or setuptools package.") from ex
+
+del sys

.venv/lib/python3.9/site-packages/cffi/api.py

@@ -0,0 +1,967 @@
+import sys, types
+from .lock import allocate_lock
+from .error import CDefError
+from . import model
+
+try:
+    callable
+except NameError:
+    # Python 3.1
+    from collections import Callable
+    callable = lambda x: isinstance(x, Callable)
+
+try:
+    basestring
+except NameError:
+    # Python 3.x
+    basestring = str
+
+_unspecified = object()
+
+
+
+class FFI(object):
+    r'''
+    The main top-level class that you instantiate once, or once per module.
+
+    Example usage:
+
+        ffi = FFI()
+        ffi.cdef("""
+            int printf(const char *, ...);
+        """)
+
+        C = ffi.dlopen(None)   # standard library
+        -or-
+        C = ffi.verify()  # use a C compiler: verify the decl above is right
+
+        C.printf("hello, %s!\n", ffi.new("char[]", "world"))
+    '''
+
+    def __init__(self, backend=None):
+        """Create an FFI instance.  The 'backend' argument is used to
+        select a non-default backend, mostly for tests.
+        """
+        if backend is None:
+            # You need PyPy (>= 2.0 beta), or a CPython (>= 2.6) with
+            # _cffi_backend.so compiled.
+            import _cffi_backend as backend
+            from . import __version__
+            if backend.__version__ != __version__:
+                # bad version!  Try to be as explicit as possible.
+                if hasattr(backend, '__file__'):
+                    # CPython
+                    raise Exception("Version mismatch: this is the 'cffi' package version %s, located in %r.  When we import the top-level '_cffi_backend' extension module, we get version %s, located in %r.  The two versions should be equal; check your installation." % (
+                        __version__, __file__,
+                        backend.__version__, backend.__file__))
+                else:
+                    # PyPy
+                    raise Exception("Version mismatch: this is the 'cffi' package version %s, located in %r.  This interpreter comes with a built-in '_cffi_backend' module, which is version %s.  The two versions should be equal; check your installation." % (
+                        __version__, __file__, backend.__version__))
+            # (If you insist you can also try to pass the option
+            # 'backend=backend_ctypes.CTypesBackend()', but don't
+            # rely on it!  It's probably not going to work well.)
+
+        from . import cparser
+        self._backend = backend
+        self._lock = allocate_lock()
+        self._parser = cparser.Parser()
+        self._cached_btypes = {}
+        self._parsed_types = types.ModuleType('parsed_types').__dict__
+        self._new_types = types.ModuleType('new_types').__dict__
+        self._function_caches = []
+        self._libraries = []
+        self._cdefsources = []
+        self._included_ffis = []
+        self._windows_unicode = None
+        self._init_once_cache = {}
+        self._cdef_version = None
+        self._embedding = None
+        self._typecache = model.get_typecache(backend)
+        if hasattr(backend, 'set_ffi'):
+            backend.set_ffi(self)
+        for name in list(backend.__dict__):
+            if name.startswith('RTLD_'):
+                setattr(self, name, getattr(backend, name))
+        #
+        with self._lock:
+            self.BVoidP = self._get_cached_btype(model.voidp_type)
+            self.BCharA = self._get_cached_btype(model.char_array_type)
+        if isinstance(backend, types.ModuleType):
+            # _cffi_backend: attach these constants to the class
+            if not hasattr(FFI, 'NULL'):
+                FFI.NULL = self.cast(self.BVoidP, 0)
+                FFI.CData, FFI.CType = backend._get_types()
+        else:
+            # ctypes backend: attach these constants to the instance
+            self.NULL = self.cast(self.BVoidP, 0)
+            self.CData, self.CType = backend._get_types()
+        self.buffer = backend.buffer
+
+    def cdef(self, csource, override=False, packed=False, pack=None):
+        """Parse the given C source.  This registers all declared functions,
+        types, and global variables.  The functions and global variables can
+        then be accessed via either 'ffi.dlopen()' or 'ffi.verify()'.
+        The types can be used in 'ffi.new()' and other functions.
+        If 'packed' is specified as True, all structs declared inside this
+        cdef are packed, i.e. laid out without any field alignment at all.
+        Alternatively, 'pack' can be a small integer, and requests for
+        alignment greater than that are ignored (pack=1 is equivalent to
+        packed=True).
+        """
+        self._cdef(csource, override=override, packed=packed, pack=pack)
+
+    def embedding_api(self, csource, packed=False, pack=None):
+        self._cdef(csource, packed=packed, pack=pack, dllexport=True)
+        if self._embedding is None:
+            self._embedding = ''
+
+    def _cdef(self, csource, override=False, **options):
+        if not isinstance(csource, str):    # unicode, on Python 2
+            if not isinstance(csource, basestring):
+                raise TypeError("cdef() argument must be a string")
+            csource = csource.encode('ascii')
+        with self._lock:
+            self._cdef_version = object()
+            self._parser.parse(csource, override=override, **options)
+            self._cdefsources.append(csource)
+            if override:
+                for cache in self._function_caches:
+                    cache.clear()
+            finishlist = self._parser._recomplete
+            if finishlist:
+                self._parser._recomplete = []
+                for tp in finishlist:
+                    tp.finish_backend_type(self, finishlist)
+
+    def dlopen(self, name, flags=0):
+        """Load and return a dynamic library identified by 'name'.
+        The standard C library can be loaded by passing None.
+        Note that functions and types declared by 'ffi.cdef()' are not
+        linked to a particular library, just like C headers; in the
+        library we only look for the actual (untyped) symbols.
+        """
+        if not (isinstance(name, basestring) or
+                name is None or
+                isinstance(name, self.CData)):
+            raise TypeError("dlopen(name): name must be a file name, None, "
+                            "or an already-opened 'void *' handle")
+        with self._lock:
+            lib, function_cache = _make_ffi_library(self, name, flags)
+            self._function_caches.append(function_cache)
+            self._libraries.append(lib)
+        return lib
+
+    def dlclose(self, lib):
+        """Close a library obtained with ffi.dlopen().  After this call,
+        access to functions or variables from the library will fail
+        (possibly with a segmentation fault).
+        """
+        type(lib).__cffi_close__(lib)
+
+    def _typeof_locked(self, cdecl):
+        # call me with the lock!
+        key = cdecl
+        if key in self._parsed_types:
+            return self._parsed_types[key]
+        #
+        if not isinstance(cdecl, str):    # unicode, on Python 2
+            cdecl = cdecl.encode('ascii')
+        #
+        type = self._parser.parse_type(cdecl)
+        really_a_function_type = type.is_raw_function
+        if really_a_function_type:
+            type = type.as_function_pointer()
+        btype = self._get_cached_btype(type)
+        result = btype, really_a_function_type
+        self._parsed_types[key] = result
+        return result
+
+    def _typeof(self, cdecl, consider_function_as_funcptr=False):
+        # string -> ctype object
+        try:
+            result = self._parsed_types[cdecl]
+        except KeyError:
+            with self._lock:
+                result = self._typeof_locked(cdecl)
+        #
+        btype, really_a_function_type = result
+        if really_a_function_type and not consider_function_as_funcptr:
+            raise CDefError("the type %r is a function type, not a "
+                            "pointer-to-function type" % (cdecl,))
+        return btype
+
+    def typeof(self, cdecl):
+        """Parse the C type given as a string and return the
+        corresponding <ctype> object.
+        It can also be used on 'cdata' instance to get its C type.
+        """
+        if isinstance(cdecl, basestring):
+            return self._typeof(cdecl)
+        if isinstance(cdecl, self.CData):
+            return self._backend.typeof(cdecl)
+        if isinstance(cdecl, types.BuiltinFunctionType):
+            res = _builtin_function_type(cdecl)
+            if res is not None:
+                return res
+        if (isinstance(cdecl, types.FunctionType)
+                and hasattr(cdecl, '_cffi_base_type')):
+            with self._lock:
+                return self._get_cached_btype(cdecl._cffi_base_type)
+        raise TypeError(type(cdecl))
+
+    def sizeof(self, cdecl):
+        """Return the size in bytes of the argument.  It can be a
+        string naming a C type, or a 'cdata' instance.
+        """
+        if isinstance(cdecl, basestring):
+            BType = self._typeof(cdecl)
+            return self._backend.sizeof(BType)
+        else:
+            return self._backend.sizeof(cdecl)
+
+    def alignof(self, cdecl):
+        """Return the natural alignment size in bytes of the C type
+        given as a string.
+        """
+        if isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl)
+        return self._backend.alignof(cdecl)
+
+    def offsetof(self, cdecl, *fields_or_indexes):
+        """Return the offset of the named field inside the given
+        structure or array, which must be given as a C type name.
+        You can give several field names in case of nested structures.
+        You can also give numeric values which correspond to array
+        items, in case of an array type.
+        """
+        if isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl)
+        return self._typeoffsetof(cdecl, *fields_or_indexes)[1]
+
+    def new(self, cdecl, init=None):
+        """Allocate an instance according to the specified C type and
+        return a pointer to it.  The specified C type must be either a
+        pointer or an array: ``new('X *')`` allocates an X and returns
+        a pointer to it, whereas ``new('X[n]')`` allocates an array of
+        n X'es and returns an array referencing it (which works
+        mostly like a pointer, like in C).  You can also use
+        ``new('X[]', n)`` to allocate an array of a non-constant
+        length n.
+
+        The memory is initialized following the rules of declaring a
+        global variable in C: by default it is zero-initialized, but
+        an explicit initializer can be given which can be used to
+        fill all or part of the memory.
+
+        When the returned <cdata> object goes out of scope, the memory
+        is freed.  In other words the returned <cdata> object has
+        ownership of the value of type 'cdecl' that it points to.  This
+        means that the raw data can be used as long as this object is
+        kept alive, but must not be used for a longer time.  Be careful
+        about that when copying the pointer to the memory somewhere
+        else, e.g. into another structure.
+        """
+        if isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl)
+        return self._backend.newp(cdecl, init)
+
+    def new_allocator(self, alloc=None, free=None,
+                      should_clear_after_alloc=True):
+        """Return a new allocator, i.e. a function that behaves like ffi.new()
+        but uses the provided low-level 'alloc' and 'free' functions.
+
+        'alloc' is called with the size as argument.  If it returns NULL, a
+        MemoryError is raised.  'free' is called with the result of 'alloc'
+        as argument.  Both can be either Python function or directly C
+        functions.  If 'free' is None, then no free function is called.
+        If both 'alloc' and 'free' are None, the default is used.
+
+        If 'should_clear_after_alloc' is set to False, then the memory
+        returned by 'alloc' is assumed to be already cleared (or you are
+        fine with garbage); otherwise CFFI will clear it.
+        """
+        compiled_ffi = self._backend.FFI()
+        allocator = compiled_ffi.new_allocator(alloc, free,
+                                               should_clear_after_alloc)
+        def allocate(cdecl, init=None):
+            if isinstance(cdecl, basestring):
+                cdecl = self._typeof(cdecl)
+            return allocator(cdecl, init)
+        return allocate
+
+    def cast(self, cdecl, source):
+        """Similar to a C cast: returns an instance of the named C
+        type initialized with the given 'source'.  The source is
+        casted between integers or pointers of any type.
+        """
+        if isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl)
+        return self._backend.cast(cdecl, source)
+
+    def string(self, cdata, maxlen=-1):
+        """Return a Python string (or unicode string) from the 'cdata'.
+        If 'cdata' is a pointer or array of characters or bytes, returns
+        the null-terminated string.  The returned string extends until
+        the first null character, or at most 'maxlen' characters.  If
+        'cdata' is an array then 'maxlen' defaults to its length.
+
+        If 'cdata' is a pointer or array of wchar_t, returns a unicode
+        string following the same rules.
+
+        If 'cdata' is a single character or byte or a wchar_t, returns
+        it as a string or unicode string.
+
+        If 'cdata' is an enum, returns the value of the enumerator as a
+        string, or 'NUMBER' if the value is out of range.
+        """
+        return self._backend.string(cdata, maxlen)
+
+    def unpack(self, cdata, length):
+        """Unpack an array of C data of the given length,
+        returning a Python string/unicode/list.
+
+        If 'cdata' is a pointer to 'char', returns a byte string.
+        It does not stop at the first null.  This is equivalent to:
+        ffi.buffer(cdata, length)[:]
+
+        If 'cdata' is a pointer to 'wchar_t', returns a unicode string.
+        'length' is measured in wchar_t's; it is not the size in bytes.
+
+        If 'cdata' is a pointer to anything else, returns a list of
+        'length' items.  This is a faster equivalent to:
+        [cdata[i] for i in range(length)]
+        """
+        return self._backend.unpack(cdata, length)
+
+   #def buffer(self, cdata, size=-1):
+   #    """Return a read-write buffer object that references the raw C data
+   #    pointed to by the given 'cdata'.  The 'cdata' must be a pointer or
+   #    an array.  Can be passed to functions expecting a buffer, or directly
+   #    manipulated with:
+   #
+   #        buf[:]          get a copy of it in a regular string, or
+   #        buf[idx]        as a single character
+   #        buf[:] = ...
+   #        buf[idx] = ...  change the content
+   #    """
+   #    note that 'buffer' is a type, set on this instance by __init__
+
+    def from_buffer(self, cdecl, python_buffer=_unspecified,
+                    require_writable=False):
+        """Return a cdata of the given type pointing to the data of the
+        given Python object, which must support the buffer interface.
+        Note that this is not meant to be used on the built-in types
+        str or unicode (you can build 'char[]' arrays explicitly)
+        but only on objects containing large quantities of raw data
+        in some other format, like 'array.array' or numpy arrays.
+
+        The first argument is optional and default to 'char[]'.
+        """
+        if python_buffer is _unspecified:
+            cdecl, python_buffer = self.BCharA, cdecl
+        elif isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl)
+        return self._backend.from_buffer(cdecl, python_buffer,
+                                         require_writable)
+
+    def memmove(self, dest, src, n):
+        """ffi.memmove(dest, src, n) copies n bytes of memory from src to dest.
+
+        Like the C function memmove(), the memory areas may overlap;
+        apart from that it behaves like the C function memcpy().
+
+        'src' can be any cdata ptr or array, or any Python buffer object.
+        'dest' can be any cdata ptr or array, or a writable Python buffer
+        object.  The size to copy, 'n', is always measured in bytes.
+
+        Unlike other methods, this one supports all Python buffer including
+        byte strings and bytearrays---but it still does not support
+        non-contiguous buffers.
+        """
+        return self._backend.memmove(dest, src, n)
+
+    def callback(self, cdecl, python_callable=None, error=None, onerror=None):
+        """Return a callback object or a decorator making such a
+        callback object.  'cdecl' must name a C function pointer type.
+        The callback invokes the specified 'python_callable' (which may
+        be provided either directly or via a decorator).  Important: the
+        callback object must be manually kept alive for as long as the
+        callback may be invoked from the C level.
+        """
+        def callback_decorator_wrap(python_callable):
+            if not callable(python_callable):
+                raise TypeError("the 'python_callable' argument "
+                                "is not callable")
+            return self._backend.callback(cdecl, python_callable,
+                                          error, onerror)
+        if isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl, consider_function_as_funcptr=True)
+        if python_callable is None:
+            return callback_decorator_wrap                # decorator mode
+        else:
+            return callback_decorator_wrap(python_callable)  # direct mode
+
+    def getctype(self, cdecl, replace_with=''):
+        """Return a string giving the C type 'cdecl', which may be itself
+        a string or a <ctype> object.  If 'replace_with' is given, it gives
+        extra text to append (or insert for more complicated C types), like
+        a variable name, or '*' to get actually the C type 'pointer-to-cdecl'.
+        """
+        if isinstance(cdecl, basestring):
+            cdecl = self._typeof(cdecl)
+        replace_with = replace_with.strip()
+        if (replace_with.startswith('*')
+                and '&[' in self._backend.getcname(cdecl, '&')):
+            replace_with = '(%s)' % replace_with
+        elif replace_with and not replace_with[0] in '[(':
+            replace_with = ' ' + replace_with
+        return self._backend.getcname(cdecl, replace_with)
+
+    def gc(self, cdata, destructor, size=0):
+        """Return a new cdata object that points to the same
+        data.  Later, when this new cdata object is garbage-collected,
+        'destructor(old_cdata_object)' will be called.
+
+        The optional 'size' gives an estimate of the size, used to
+        trigger the garbage collection more eagerly.  So far only used
+        on PyPy.  It tells the GC that the returned object keeps alive
+        roughly 'size' bytes of external memory.
+        """
+        return self._backend.gcp(cdata, destructor, size)
+
+    def _get_cached_btype(self, type):
+        assert self._lock.acquire(False) is False
+        # call me with the lock!
+        try:
+            BType = self._cached_btypes[type]
+        except KeyError:
+            finishlist = []
+            BType = type.get_cached_btype(self, finishlist)
+            for type in finishlist:
+                type.finish_backend_type(self, finishlist)
+        return BType
+
+    def verify(self, source='', tmpdir=None, **kwargs):
+        """Verify that the current ffi signatures compile on this
+        machine, and return a dynamic library object.  The dynamic
+        library can be used to call functions and access global
+        variables declared in this 'ffi'.  The library is compiled
+        by the C compiler: it gives you C-level API compatibility
+        (including calling macros).  This is unlike 'ffi.dlopen()',
+        which requires binary compatibility in the signatures.
+        """
+        from .verifier import Verifier, _caller_dir_pycache
+        #
+        # If set_unicode(True) was called, insert the UNICODE and
+        # _UNICODE macro declarations
+        if self._windows_unicode:
+            self._apply_windows_unicode(kwargs)
+        #
+        # Set the tmpdir here, and not in Verifier.__init__: it picks
+        # up the caller's directory, which we want to be the caller of
+        # ffi.verify(), as opposed to the caller of Veritier().
+        tmpdir = tmpdir or _caller_dir_pycache()
+        #
+        # Make a Verifier() and use it to load the library.
+        self.verifier = Verifier(self, source, tmpdir, **kwargs)
+        lib = self.verifier.load_library()
+        #
+        # Save the loaded library for keep-alive purposes, even
+        # if the caller doesn't keep it alive itself (it should).
+        self._libraries.append(lib)
+        return lib
+
+    def _get_errno(self):
+        return self._backend.get_errno()
+    def _set_errno(self, errno):
+        self._backend.set_errno(errno)
+    errno = property(_get_errno, _set_errno, None,
+                     "the value of 'errno' from/to the C calls")
+
+    def getwinerror(self, code=-1):
+        return self._backend.getwinerror(code)
+
+    def _pointer_to(self, ctype):
+        with self._lock:
+            return model.pointer_cache(self, ctype)
+
+    def addressof(self, cdata, *fields_or_indexes):
+        """Return the address of a <cdata 'struct-or-union'>.
+        If 'fields_or_indexes' are given, returns the address of that
+        field or array item in the structure or array, recursively in
+        case of nested structures.
+        """
+        try:
+            ctype = self._backend.typeof(cdata)
+        except TypeError:
+            if '__addressof__' in type(cdata).__dict__:
+                return type(cdata).__addressof__(cdata, *fields_or_indexes)
+            raise
+        if fields_or_indexes:
+            ctype, offset = self._typeoffsetof(ctype, *fields_or_indexes)
+        else:
+            if ctype.kind == "pointer":
+                raise TypeError("addressof(pointer)")
+            offset = 0
+        ctypeptr = self._pointer_to(ctype)
+        return self._backend.rawaddressof(ctypeptr, cdata, offset)
+
+    def _typeoffsetof(self, ctype, field_or_index, *fields_or_indexes):
+        ctype, offset = self._backend.typeoffsetof(ctype, field_or_index)
+        for field1 in fields_or_indexes:
+            ctype, offset1 = self._backend.typeoffsetof(ctype, field1, 1)
+            offset += offset1
+        return ctype, offset
+
+    def include(self, ffi_to_include):
+        """Includes the typedefs, structs, unions and enums defined
+        in another FFI instance.  Usage is similar to a #include in C,
+        where a part of the program might include types defined in
+        another part for its own usage.  Note that the include()
+        method has no effect on functions, constants and global
+        variables, which must anyway be accessed directly from the
+        lib object returned by the original FFI instance.
+        """
+        if not isinstance(ffi_to_include, FFI):
+            raise TypeError("ffi.include() expects an argument that is also of"
+                            " type cffi.FFI, not %r" % (
+                                type(ffi_to_include).__name__,))
+        if ffi_to_include is self:
+            raise ValueError("self.include(self)")
+        with ffi_to_include._lock:
+            with self._lock:
+                self._parser.include(ffi_to_include._parser)
+                self._cdefsources.append('[')
+                self._cdefsources.extend(ffi_to_include._cdefsources)
+                self._cdefsources.append(']')
+                self._included_ffis.append(ffi_to_include)
+
+    def new_handle(self, x):
+        return self._backend.newp_handle(self.BVoidP, x)
+
+    def from_handle(self, x):
+        return self._backend.from_handle(x)
+
+    def release(self, x):
+        self._backend.release(x)
+
+    def set_unicode(self, enabled_flag):
+        """Windows: if 'enabled_flag' is True, enable the UNICODE and
+        _UNICODE defines in C, and declare the types like TCHAR and LPTCSTR
+        to be (pointers to) wchar_t.  If 'enabled_flag' is False,
+        declare these types to be (pointers to) plain 8-bit characters.
+        This is mostly for backward compatibility; you usually want True.
+        """
+        if self._windows_unicode is not None:
+            raise ValueError("set_unicode() can only be called once")
+        enabled_flag = bool(enabled_flag)
+        if enabled_flag:
+            self.cdef("typedef wchar_t TBYTE;"
+                      "typedef wchar_t TCHAR;"
+                      "typedef const wchar_t *LPCTSTR;"
+                      "typedef const wchar_t *PCTSTR;"
+                      "typedef wchar_t *LPTSTR;"
+                      "typedef wchar_t *PTSTR;"
+                      "typedef TBYTE *PTBYTE;"
+                      "typedef TCHAR *PTCHAR;")
+        else:
+            self.cdef("typedef char TBYTE;"
+                      "typedef char TCHAR;"
+                      "typedef const char *LPCTSTR;"
+                      "typedef const char *PCTSTR;"
+                      "typedef char *LPTSTR;"
+                      "typedef char *PTSTR;"
+                      "typedef TBYTE *PTBYTE;"
+                      "typedef TCHAR *PTCHAR;")
+        self._windows_unicode = enabled_flag
+
+    def _apply_windows_unicode(self, kwds):
+        defmacros = kwds.get('define_macros', ())
+        if not isinstance(defmacros, (list, tuple)):
+            raise TypeError("'define_macros' must be a list or tuple")
+        defmacros = list(defmacros) + [('UNICODE', '1'),
+                                       ('_UNICODE', '1')]
+        kwds['define_macros'] = defmacros
+
+    def _apply_embedding_fix(self, kwds):
+        # must include an argument like "-lpython2.7" for the compiler
+        def ensure(key, value):
+            lst = kwds.setdefault(key, [])
+            if value not in lst:
+                lst.append(value)
+        #
+        if '__pypy__' in sys.builtin_module_names:
+            import os
+            if sys.platform == "win32":
+                # we need 'libpypy-c.lib'.  Current distributions of
+                # pypy (>= 4.1) contain it as 'libs/python27.lib'.
+                pythonlib = "python{0[0]}{0[1]}".format(sys.version_info)
+                if hasattr(sys, 'prefix'):
+                    ensure('library_dirs', os.path.join(sys.prefix, 'libs'))
+            else:
+                # we need 'libpypy-c.{so,dylib}', which should be by
+                # default located in 'sys.prefix/bin' for installed
+                # systems.
+                if sys.version_info < (3,):
+                    pythonlib = "pypy-c"
+                else:
+                    pythonlib = "pypy3-c"
+                if hasattr(sys, 'prefix'):
+                    ensure('library_dirs', os.path.join(sys.prefix, 'bin'))
+            # On uninstalled pypy's, the libpypy-c is typically found in
+            # .../pypy/goal/.
+            if hasattr(sys, 'prefix'):
+                ensure('library_dirs', os.path.join(sys.prefix, 'pypy', 'goal'))
+        else:
+            if sys.platform == "win32":
+                template = "python%d%d"
+                if hasattr(sys, 'gettotalrefcount'):
+                    template += '_d'
+            else:
+                try:
+                    import sysconfig
+                except ImportError:    # 2.6
+                    from cffi._shimmed_dist_utils import sysconfig
+                template = "python%d.%d"
+                if sysconfig.get_config_var('DEBUG_EXT'):
+                    template += sysconfig.get_config_var('DEBUG_EXT')
+            pythonlib = (template %
+                    (sys.hexversion >> 24, (sys.hexversion >> 16) & 0xff))
+            if hasattr(sys, 'abiflags'):
+                pythonlib += sys.abiflags
+        ensure('libraries', pythonlib)
+        if sys.platform == "win32":
+            ensure('extra_link_args', '/MANIFEST')
+
+    def set_source(self, module_name, source, source_extension='.c', **kwds):
+        import os
+        if hasattr(self, '_assigned_source'):
+            raise ValueError("set_source() cannot be called several times "
+                             "per ffi object")
+        if not isinstance(module_name, basestring):
+            raise TypeError("'module_name' must be a string")
+        if os.sep in module_name or (os.altsep and os.altsep in module_name):
+            raise ValueError("'module_name' must not contain '/': use a dotted "
+                             "name to make a 'package.module' location")
+        self._assigned_source = (str(module_name), source,
+                                 source_extension, kwds)
+
+    def set_source_pkgconfig(self, module_name, pkgconfig_libs, source,
+                             source_extension='.c', **kwds):
+        from . import pkgconfig
+        if not isinstance(pkgconfig_libs, list):
+            raise TypeError("the pkgconfig_libs argument must be a list "
+                            "of package names")
+        kwds2 = pkgconfig.flags_from_pkgconfig(pkgconfig_libs)
+        pkgconfig.merge_flags(kwds, kwds2)
+        self.set_source(module_name, source, source_extension, **kwds)
+
+    def distutils_extension(self, tmpdir='build', verbose=True):
+        from cffi._shimmed_dist_utils import mkpath
+        from .recompiler import recompile
+        #
+        if not hasattr(self, '_assigned_source'):
+            if hasattr(self, 'verifier'):     # fallback, 'tmpdir' ignored
+                return self.verifier.get_extension()
+            raise ValueError("set_source() must be called before"
+                             " distutils_extension()")
+        module_name, source, source_extension, kwds = self._assigned_source
+        if source is None:
+            raise TypeError("distutils_extension() is only for C extension "
+                            "modules, not for dlopen()-style pure Python "
+                            "modules")
+        mkpath(tmpdir)
+        ext, updated = recompile(self, module_name,
+                                 source, tmpdir=tmpdir, extradir=tmpdir,
+                                 source_extension=source_extension,
+                                 call_c_compiler=False, **kwds)
+        if verbose:
+            if updated:
+                sys.stderr.write("regenerated: %r\n" % (ext.sources[0],))
+            else:
+                sys.stderr.write("not modified: %r\n" % (ext.sources[0],))
+        return ext
+
+    def emit_c_code(self, filename):
+        from .recompiler import recompile
+        #
+        if not hasattr(self, '_assigned_source'):
+            raise ValueError("set_source() must be called before emit_c_code()")
+        module_name, source, source_extension, kwds = self._assigned_source
+        if source is None:
+            raise TypeError("emit_c_code() is only for C extension modules, "
+                            "not for dlopen()-style pure Python modules")
+        recompile(self, module_name, source,
+                  c_file=filename, call_c_compiler=False,
+                  uses_ffiplatform=False, **kwds)
+
+    def emit_python_code(self, filename):
+        from .recompiler import recompile
+        #
+        if not hasattr(self, '_assigned_source'):
+            raise ValueError("set_source() must be called before emit_c_code()")
+        module_name, source, source_extension, kwds = self._assigned_source
+        if source is not None:
+            raise TypeError("emit_python_code() is only for dlopen()-style "
+                            "pure Python modules, not for C extension modules")
+        recompile(self, module_name, source,
+                  c_file=filename, call_c_compiler=False,
+                  uses_ffiplatform=False, **kwds)
+
+    def compile(self, tmpdir='.', verbose=0, target=None, debug=None):
+        """The 'target' argument gives the final file name of the
+        compiled DLL.  Use '*' to force distutils' choice, suitable for
+        regular CPython C API modules.  Use a file name ending in '.*'
+        to ask for the system's default extension for dynamic libraries
+        (.so/.dll/.dylib).
+
+        The default is '*' when building a non-embedded C API extension,
+        and (module_name + '.*') when building an embedded library.
+        """
+        from .recompiler import recompile
+        #
+        if not hasattr(self, '_assigned_source'):
+            raise ValueError("set_source() must be called before compile()")
+        module_name, source, source_extension, kwds = self._assigned_source
+        return recompile(self, module_name, source, tmpdir=tmpdir,
+                         target=target, source_extension=source_extension,
+                         compiler_verbose=verbose, debug=debug, **kwds)
+
+    def init_once(self, func, tag):
+        # Read _init_once_cache[tag], which is either (False, lock) if
+        # we're calling the function now in some thread, or (True, result).
+        # Don't call setdefault() in most cases, to avoid allocating and
+        # immediately freeing a lock; but still use setdefaut() to avoid
+        # races.
+        try:
+            x = self._init_once_cache[tag]
+        except KeyError:
+            x = self._init_once_cache.setdefault(tag, (False, allocate_lock()))
+        # Common case: we got (True, result), so we return the result.
+        if x[0]:
+            return x[1]
+        # Else, it's a lock.  Acquire it to serialize the following tests.
+        with x[1]:
+            # Read again from _init_once_cache the current status.
+            x = self._init_once_cache[tag]
+            if x[0]:
+                return x[1]
+            # Call the function and store the result back.
+            result = func()
+            self._init_once_cache[tag] = (True, result)
+        return result
+
+    def embedding_init_code(self, pysource):
+        if self._embedding:
+            raise ValueError("embedding_init_code() can only be called once")
+        # fix 'pysource' before it gets dumped into the C file:
+        # - remove empty lines at the beginning, so it starts at "line 1"
+        # - dedent, if all non-empty lines are indented
+        # - check for SyntaxErrors
+        import re
+        match = re.match(r'\s*\n', pysource)
+        if match:
+            pysource = pysource[match.end():]
+        lines = pysource.splitlines() or ['']
+        prefix = re.match(r'\s*', lines[0]).group()
+        for i in range(1, len(lines)):
+            line = lines[i]
+            if line.rstrip():
+                while not line.startswith(prefix):
+                    prefix = prefix[:-1]
+        i = len(prefix)
+        lines = [line[i:]+'\n' for line in lines]
+        pysource = ''.join(lines)
+        #
+        compile(pysource, "cffi_init", "exec")
+        #
+        self._embedding = pysource
+
+    def def_extern(self, *args, **kwds):
+        raise ValueError("ffi.def_extern() is only available on API-mode FFI "
+                         "objects")
+
+    def list_types(self):
+        """Returns the user type names known to this FFI instance.
+        This returns a tuple containing three lists of names:
+        (typedef_names, names_of_structs, names_of_unions)
+        """
+        typedefs = []
+        structs = []
+        unions = []
+        for key in self._parser._declarations:
+            if key.startswith('typedef '):
+                typedefs.append(key[8:])
+            elif key.startswith('struct '):
+                structs.append(key[7:])
+            elif key.startswith('union '):
+                unions.append(key[6:])
+        typedefs.sort()
+        structs.sort()
+        unions.sort()
+        return (typedefs, structs, unions)
+
+
+def _load_backend_lib(backend, name, flags):
+    import os
+    if not isinstance(name, basestring):
+        if sys.platform != "win32" or name is not None:
+            return backend.load_library(name, flags)
+        name = "c"    # Windows: load_library(None) fails, but this works
+                      # on Python 2 (backward compatibility hack only)
+    first_error = None
+    if '.' in name or '/' in name or os.sep in name:
+        try:
+            return backend.load_library(name, flags)
+        except OSError as e:
+            first_error = e
+    import ctypes.util
+    path = ctypes.util.find_library(name)
+    if path is None:
+        if name == "c" and sys.platform == "win32" and sys.version_info >= (3,):
+            raise OSError("dlopen(None) cannot work on Windows for Python 3 "
+                          "(see http://bugs.python.org/issue23606)")
+        msg = ("ctypes.util.find_library() did not manage "
+               "to locate a library called %r" % (name,))
+        if first_error is not None:
+            msg = "%s.  Additionally, %s" % (first_error, msg)
+        raise OSError(msg)
+    return backend.load_library(path, flags)
+
+def _make_ffi_library(ffi, libname, flags):
+    backend = ffi._backend
+    backendlib = _load_backend_lib(backend, libname, flags)
+    #
+    def accessor_function(name):
+        key = 'function ' + name
+        tp, _ = ffi._parser._declarations[key]
+        BType = ffi._get_cached_btype(tp)
+        value = backendlib.load_function(BType, name)
+        library.__dict__[name] = value
+    #
+    def accessor_variable(name):
+        key = 'variable ' + name
+        tp, _ = ffi._parser._declarations[key]
+        BType = ffi._get_cached_btype(tp)
+        read_variable = backendlib.read_variable
+        write_variable = backendlib.write_variable
+        setattr(FFILibrary, name, property(
+            lambda self: read_variable(BType, name),
+            lambda self, value: write_variable(BType, name, value)))
+    #
+    def addressof_var(name):
+        try:
+            return addr_variables[name]
+        except KeyError:
+            with ffi._lock:
+                if name not in addr_variables:
+                    key = 'variable ' + name
+                    tp, _ = ffi._parser._declarations[key]
+                    BType = ffi._get_cached_btype(tp)
+                    if BType.kind != 'array':
+                        BType = model.pointer_cache(ffi, BType)
+                    p = backendlib.load_function(BType, name)
+                    addr_variables[name] = p
+            return addr_variables[name]
+    #
+    def accessor_constant(name):
+        raise NotImplementedError("non-integer constant '%s' cannot be "
+                                  "accessed from a dlopen() library" % (name,))
+    #
+    def accessor_int_constant(name):
+        library.__dict__[name] = ffi._parser._int_constants[name]
+    #
+    accessors = {}
+    accessors_version = [False]
+    addr_variables = {}
+    #
+    def update_accessors():
+        if accessors_version[0] is ffi._cdef_version:
+            return
+        #
+        for key, (tp, _) in ffi._parser._declarations.items():
+            if not isinstance(tp, model.EnumType):
+                tag, name = key.split(' ', 1)
+                if tag == 'function':
+                    accessors[name] = accessor_function
+                elif tag == 'variable':
+                    accessors[name] = accessor_variable
+                elif tag == 'constant':
+                    accessors[name] = accessor_constant
+            else:
+                for i, enumname in enumerate(tp.enumerators):
+                    def accessor_enum(name, tp=tp, i=i):
+                        tp.check_not_partial()
+                        library.__dict__[name] = tp.enumvalues[i]
+                    accessors[enumname] = accessor_enum
+        for name in ffi._parser._int_constants:
+            accessors.setdefault(name, accessor_int_constant)
+        accessors_version[0] = ffi._cdef_version
+    #
+    def make_accessor(name):
+        with ffi._lock:
+            if name in library.__dict__ or name in FFILibrary.__dict__:
+                return    # added by another thread while waiting for the lock
+            if name not in accessors:
+                update_accessors()
+                if name not in accessors:
+                    raise AttributeError(name)
+            accessors[name](name)
+    #
+    class FFILibrary(object):
+        def __getattr__(self, name):
+            make_accessor(name)
+            return getattr(self, name)
+        def __setattr__(self, name, value):
+            try:
+                property = getattr(self.__class__, name)
+            except AttributeError:
+                make_accessor(name)
+                setattr(self, name, value)
+            else:
+                property.__set__(self, value)
+        def __dir__(self):
+            with ffi._lock:
+                update_accessors()
+                return accessors.keys()
+        def __addressof__(self, name):
+            if name in library.__dict__:
+                return library.__dict__[name]
+            if name in FFILibrary.__dict__:
+                return addressof_var(name)
+            make_accessor(name)
+            if name in library.__dict__:
+                return library.__dict__[name]
+            if name in FFILibrary.__dict__:
+                return addressof_var(name)
+            raise AttributeError("cffi library has no function or "
+                                 "global variable named '%s'" % (name,))
+        def __cffi_close__(self):
+            backendlib.close_lib()
+            self.__dict__.clear()
+    #
+    if isinstance(libname, basestring):
+        try:
+            if not isinstance(libname, str):    # unicode, on Python 2
+                libname = libname.encode('utf-8')
+            FFILibrary.__name__ = 'FFILibrary_%s' % libname
+        except UnicodeError:
+            pass
+    library = FFILibrary()
+    return library, library.__dict__
+
+def _builtin_function_type(func):
+    # a hack to make at least ffi.typeof(builtin_function) work,
+    # if the builtin function was obtained by 'vengine_cpy'.
+    import sys
+    try:
+        module = sys.modules[func.__module__]
+        ffi = module._cffi_original_ffi
+        types_of_builtin_funcs = module._cffi_types_of_builtin_funcs
+        tp = types_of_builtin_funcs[func]
+    except (KeyError, AttributeError, TypeError):
+        return None
+    else:
+        with ffi._lock:
+            return ffi._get_cached_btype(tp)

.venv/lib/python3.9/site-packages/cffi/cffi_opcode.py

@@ -0,0 +1,187 @@
+from .error import VerificationError
+
+class CffiOp(object):
+    def __init__(self, op, arg):
+        self.op = op
+        self.arg = arg
+
+    def as_c_expr(self):
+        if self.op is None:
+            assert isinstance(self.arg, str)
+            return '(_cffi_opcode_t)(%s)' % (self.arg,)
+        classname = CLASS_NAME[self.op]
+        return '_CFFI_OP(_CFFI_OP_%s, %s)' % (classname, self.arg)
+
+    def as_python_bytes(self):
+        if self.op is None and self.arg.isdigit():
+            value = int(self.arg)     # non-negative: '-' not in self.arg
+            if value >= 2**31:
+                raise OverflowError("cannot emit %r: limited to 2**31-1"
+                                    % (self.arg,))
+            return format_four_bytes(value)
+        if isinstance(self.arg, str):
+            raise VerificationError("cannot emit to Python: %r" % (self.arg,))
+        return format_four_bytes((self.arg << 8) | self.op)
+
+    def __str__(self):
+        classname = CLASS_NAME.get(self.op, self.op)
+        return '(%s %s)' % (classname, self.arg)
+
+def format_four_bytes(num):
+    return '\\x%02X\\x%02X\\x%02X\\x%02X' % (
+        (num >> 24) & 0xFF,
+        (num >> 16) & 0xFF,
+        (num >>  8) & 0xFF,
+        (num      ) & 0xFF)
+
+OP_PRIMITIVE       = 1
+OP_POINTER         = 3
+OP_ARRAY           = 5
+OP_OPEN_ARRAY      = 7
+OP_STRUCT_UNION    = 9
+OP_ENUM            = 11
+OP_FUNCTION        = 13
+OP_FUNCTION_END    = 15
+OP_NOOP            = 17
+OP_BITFIELD        = 19
+OP_TYPENAME        = 21
+OP_CPYTHON_BLTN_V  = 23   # varargs
+OP_CPYTHON_BLTN_N  = 25   # noargs
+OP_CPYTHON_BLTN_O  = 27   # O  (i.e. a single arg)
+OP_CONSTANT        = 29
+OP_CONSTANT_INT    = 31
+OP_GLOBAL_VAR      = 33
+OP_DLOPEN_FUNC     = 35
+OP_DLOPEN_CONST    = 37
+OP_GLOBAL_VAR_F    = 39
+OP_EXTERN_PYTHON   = 41
+
+PRIM_VOID          = 0
+PRIM_BOOL          = 1
+PRIM_CHAR          = 2
+PRIM_SCHAR         = 3
+PRIM_UCHAR         = 4
+PRIM_SHORT         = 5
+PRIM_USHORT        = 6
+PRIM_INT           = 7
+PRIM_UINT          = 8
+PRIM_LONG          = 9
+PRIM_ULONG         = 10
+PRIM_LONGLONG      = 11
+PRIM_ULONGLONG     = 12
+PRIM_FLOAT         = 13
+PRIM_DOUBLE        = 14
+PRIM_LONGDOUBLE    = 15
+
+PRIM_WCHAR         = 16
+PRIM_INT8          = 17
+PRIM_UINT8         = 18
+PRIM_INT16         = 19
+PRIM_UINT16        = 20
+PRIM_INT32         = 21
+PRIM_UINT32        = 22
+PRIM_INT64         = 23
+PRIM_UINT64        = 24
+PRIM_INTPTR        = 25
+PRIM_UINTPTR       = 26
+PRIM_PTRDIFF       = 27
+PRIM_SIZE          = 28
+PRIM_SSIZE         = 29
+PRIM_INT_LEAST8    = 30
+PRIM_UINT_LEAST8   = 31
+PRIM_INT_LEAST16   = 32
+PRIM_UINT_LEAST16  = 33
+PRIM_INT_LEAST32   = 34
+PRIM_UINT_LEAST32  = 35
+PRIM_INT_LEAST64   = 36
+PRIM_UINT_LEAST64  = 37
+PRIM_INT_FAST8     = 38
+PRIM_UINT_FAST8    = 39
+PRIM_INT_FAST16    = 40
+PRIM_UINT_FAST16   = 41
+PRIM_INT_FAST32    = 42
+PRIM_UINT_FAST32   = 43
+PRIM_INT_FAST64    = 44
+PRIM_UINT_FAST64   = 45
+PRIM_INTMAX        = 46
+PRIM_UINTMAX       = 47
+PRIM_FLOATCOMPLEX  = 48
+PRIM_DOUBLECOMPLEX = 49
+PRIM_CHAR16        = 50
+PRIM_CHAR32        = 51
+
+_NUM_PRIM          = 52
+_UNKNOWN_PRIM          = -1
+_UNKNOWN_FLOAT_PRIM    = -2
+_UNKNOWN_LONG_DOUBLE   = -3
+
+_IO_FILE_STRUCT        = -1
+
+PRIMITIVE_TO_INDEX = {
+    'char':               PRIM_CHAR,
+    'short':              PRIM_SHORT,
+    'int':                PRIM_INT,
+    'long':               PRIM_LONG,
+    'long long':          PRIM_LONGLONG,
+    'signed char':        PRIM_SCHAR,
+    'unsigned char':      PRIM_UCHAR,
+    'unsigned short':     PRIM_USHORT,
+    'unsigned int':       PRIM_UINT,
+    'unsigned long':      PRIM_ULONG,
+    'unsigned long long': PRIM_ULONGLONG,
+    'float':              PRIM_FLOAT,
+    'double':             PRIM_DOUBLE,
+    'long double':        PRIM_LONGDOUBLE,
+    '_cffi_float_complex_t': PRIM_FLOATCOMPLEX,
+    '_cffi_double_complex_t': PRIM_DOUBLECOMPLEX,
+    '_Bool':              PRIM_BOOL,
+    'wchar_t':            PRIM_WCHAR,
+    'char16_t':           PRIM_CHAR16,
+    'char32_t':           PRIM_CHAR32,
+    'int8_t':             PRIM_INT8,
+    'uint8_t':            PRIM_UINT8,
+    'int16_t':            PRIM_INT16,
+    'uint16_t':           PRIM_UINT16,
+    'int32_t':            PRIM_INT32,
+    'uint32_t':           PRIM_UINT32,
+    'int64_t':            PRIM_INT64,
+    'uint64_t':           PRIM_UINT64,
+    'intptr_t':           PRIM_INTPTR,
+    'uintptr_t':          PRIM_UINTPTR,
+    'ptrdiff_t':          PRIM_PTRDIFF,
+    'size_t':             PRIM_SIZE,
+    'ssize_t':            PRIM_SSIZE,
+    'int_least8_t':       PRIM_INT_LEAST8,
+    'uint_least8_t':      PRIM_UINT_LEAST8,
+    'int_least16_t':      PRIM_INT_LEAST16,
+    'uint_least16_t':     PRIM_UINT_LEAST16,
+    'int_least32_t':      PRIM_INT_LEAST32,
+    'uint_least32_t':     PRIM_UINT_LEAST32,
+    'int_least64_t':      PRIM_INT_LEAST64,
+    'uint_least64_t':     PRIM_UINT_LEAST64,
+    'int_fast8_t':        PRIM_INT_FAST8,
+    'uint_fast8_t':       PRIM_UINT_FAST8,
+    'int_fast16_t':       PRIM_INT_FAST16,
+    'uint_fast16_t':      PRIM_UINT_FAST16,
+    'int_fast32_t':       PRIM_INT_FAST32,
+    'uint_fast32_t':      PRIM_UINT_FAST32,
+    'int_fast64_t':       PRIM_INT_FAST64,
+    'uint_fast64_t':      PRIM_UINT_FAST64,
+    'intmax_t':           PRIM_INTMAX,
+    'uintmax_t':          PRIM_UINTMAX,
+    }
+
+F_UNION         = 0x01
+F_CHECK_FIELDS  = 0x02
+F_PACKED        = 0x04
+F_EXTERNAL      = 0x08
+F_OPAQUE        = 0x10
+
+G_FLAGS = dict([('_CFFI_' + _key, globals()[_key])
+                for _key in ['F_UNION', 'F_CHECK_FIELDS', 'F_PACKED',
+                             'F_EXTERNAL', 'F_OPAQUE']])
+
+CLASS_NAME = {}
+for _name, _value in list(globals().items()):
+    if _name.startswith('OP_') and isinstance(_value, int):
+        CLASS_NAME[_value] = _name[3:]

.venv/lib/python3.9/site-packages/cffi/commontypes.py

@@ -0,0 +1,82 @@
+import sys
+from . import model
+from .error import FFIError
+
+
+COMMON_TYPES = {}
+
+try:
+    # fetch "bool" and all simple Windows types
+    from _cffi_backend import _get_common_types
+    _get_common_types(COMMON_TYPES)
+except ImportError:
+    pass
+
+COMMON_TYPES['FILE'] = model.unknown_type('FILE', '_IO_FILE')
+COMMON_TYPES['bool'] = '_Bool'    # in case we got ImportError above
+COMMON_TYPES['float _Complex'] = '_cffi_float_complex_t'
+COMMON_TYPES['double _Complex'] = '_cffi_double_complex_t'
+
+for _type in model.PrimitiveType.ALL_PRIMITIVE_TYPES:
+    if _type.endswith('_t'):
+        COMMON_TYPES[_type] = _type
+del _type
+
+_CACHE = {}
+
+def resolve_common_type(parser, commontype):
+    try:
+        return _CACHE[commontype]
+    except KeyError:
+        cdecl = COMMON_TYPES.get(commontype, commontype)
+        if not isinstance(cdecl, str):
+            result, quals = cdecl, 0    # cdecl is already a BaseType
+        elif cdecl in model.PrimitiveType.ALL_PRIMITIVE_TYPES:
+            result, quals = model.PrimitiveType(cdecl), 0
+        elif cdecl == 'set-unicode-needed':
+            raise FFIError("The Windows type %r is only available after "
+                           "you call ffi.set_unicode()" % (commontype,))
+        else:
+            if commontype == cdecl:
+                raise FFIError(
+                    "Unsupported type: %r.  Please look at "
+        "http://cffi.readthedocs.io/en/latest/cdef.html#ffi-cdef-limitations "
+                    "and file an issue if you think this type should really "
+                    "be supported." % (commontype,))
+            result, quals = parser.parse_type_and_quals(cdecl)   # recursive
+
+        assert isinstance(result, model.BaseTypeByIdentity)
+        _CACHE[commontype] = result, quals
+        return result, quals
+
+
+# ____________________________________________________________
+# extra types for Windows (most of them are in commontypes.c)
+
+
+def win_common_types():
+    return {
+        "UNICODE_STRING": model.StructType(
+            "_UNICODE_STRING",
+            ["Length",
+             "MaximumLength",
+             "Buffer"],
+            [model.PrimitiveType("unsigned short"),
+             model.PrimitiveType("unsigned short"),
+             model.PointerType(model.PrimitiveType("wchar_t"))],
+            [-1, -1, -1]),
+        "PUNICODE_STRING": "UNICODE_STRING *",
+        "PCUNICODE_STRING": "const UNICODE_STRING *",
+
+        "TBYTE": "set-unicode-needed",
+        "TCHAR": "set-unicode-needed",
+        "LPCTSTR": "set-unicode-needed",
+        "PCTSTR": "set-unicode-needed",
+        "LPTSTR": "set-unicode-needed",
+        "PTSTR": "set-unicode-needed",
+        "PTBYTE": "set-unicode-needed",
+        "PTCHAR": "set-unicode-needed",
+        }
+
+if sys.platform == 'win32':
+    COMMON_TYPES.update(win_common_types())

.venv/lib/python3.9/site-packages/cffi/error.py

@@ -0,0 +1,31 @@
+
+class FFIError(Exception):
+    __module__ = 'cffi'
+
+class CDefError(Exception):
+    __module__ = 'cffi'
+    def __str__(self):
+        try:
+            current_decl = self.args[1]
+            filename = current_decl.coord.file
+            linenum = current_decl.coord.line
+            prefix = '%s:%d: ' % (filename, linenum)
+        except (AttributeError, TypeError, IndexError):
+            prefix = ''
+        return '%s%s' % (prefix, self.args[0])
+
+class VerificationError(Exception):
+    """ An error raised when verification fails
+    """
+    __module__ = 'cffi'
+
+class VerificationMissing(Exception):
+    """ An error raised when incomplete structures are passed into
+    cdef, but no verification has been done
+    """
+    __module__ = 'cffi'
+
+class PkgConfigError(Exception):
+    """ An error raised for missing modules in pkg-config
+    """
+    __module__ = 'cffi'

.venv/lib/python3.9/site-packages/cffi/ffiplatform.py

@@ -0,0 +1,113 @@
+import sys, os
+from .error import VerificationError
+
+
+LIST_OF_FILE_NAMES = ['sources', 'include_dirs', 'library_dirs',
+                      'extra_objects', 'depends']
+
+def get_extension(srcfilename, modname, sources=(), **kwds):
+    from cffi._shimmed_dist_utils import Extension
+    allsources = [srcfilename]
+    for src in sources:
+        allsources.append(os.path.normpath(src))
+    return Extension(name=modname, sources=allsources, **kwds)
+
+def compile(tmpdir, ext, compiler_verbose=0, debug=None):
+    """Compile a C extension module using distutils."""
+
+    saved_environ = os.environ.copy()
+    try:
+        outputfilename = _build(tmpdir, ext, compiler_verbose, debug)
+        outputfilename = os.path.abspath(outputfilename)
+    finally:
+        # workaround for a distutils bugs where some env vars can
+        # become longer and longer every time it is used
+        for key, value in saved_environ.items():
+            if os.environ.get(key) != value:
+                os.environ[key] = value
+    return outputfilename
+
+def _build(tmpdir, ext, compiler_verbose=0, debug=None):
+    # XXX compact but horrible :-(
+    from cffi._shimmed_dist_utils import Distribution, CompileError, LinkError, set_threshold, set_verbosity
+
+    dist = Distribution({'ext_modules': [ext]})
+    dist.parse_config_files()
+    options = dist.get_option_dict('build_ext')
+    if debug is None:
+        debug = sys.flags.debug
+    options['debug'] = ('ffiplatform', debug)
+    options['force'] = ('ffiplatform', True)
+    options['build_lib'] = ('ffiplatform', tmpdir)
+    options['build_temp'] = ('ffiplatform', tmpdir)
+    #
+    try:
+        old_level = set_threshold(0) or 0
+        try:
+            set_verbosity(compiler_verbose)
+            dist.run_command('build_ext')
+            cmd_obj = dist.get_command_obj('build_ext')
+            [soname] = cmd_obj.get_outputs()
+        finally:
+            set_threshold(old_level)
+    except (CompileError, LinkError) as e:
+        raise VerificationError('%s: %s' % (e.__class__.__name__, e))
+    #
+    return soname
+
+try:
+    from os.path import samefile
+except ImportError:
+    def samefile(f1, f2):
+        return os.path.abspath(f1) == os.path.abspath(f2)
+
+def maybe_relative_path(path):
+    if not os.path.isabs(path):
+        return path      # already relative
+    dir = path
+    names = []
+    while True:
+        prevdir = dir
+        dir, name = os.path.split(prevdir)
+        if dir == prevdir or not dir:
+            return path     # failed to make it relative
+        names.append(name)
+        try:
+            if samefile(dir, os.curdir):
+                names.reverse()
+                return os.path.join(*names)
+        except OSError:
+            pass
+
+# ____________________________________________________________
+
+try:
+    int_or_long = (int, long)
+    import cStringIO
+except NameError:
+    int_or_long = int      # Python 3
+    import io as cStringIO
+
+def _flatten(x, f):
+    if isinstance(x, str):
+        f.write('%ds%s' % (len(x), x))
+    elif isinstance(x, dict):
+        keys = sorted(x.keys())
+        f.write('%dd' % len(keys))
+        for key in keys:
+            _flatten(key, f)
+            _flatten(x[key], f)
+    elif isinstance(x, (list, tuple)):
+        f.write('%dl' % len(x))
+        for value in x:
+            _flatten(value, f)
+    elif isinstance(x, int_or_long):
+        f.write('%di' % (x,))
+    else:
+        raise TypeError(
+            "the keywords to verify() contains unsupported object %r" % (x,))
+
+def flatten(x):
+    f = cStringIO.StringIO()
+    _flatten(x, f)
+    return f.getvalue()

.venv/lib/python3.9/site-packages/cffi/lock.py

@@ -0,0 +1,30 @@
+import sys
+
+if sys.version_info < (3,):
+    try:
+        from thread import allocate_lock
+    except ImportError:
+        from dummy_thread import allocate_lock
+else:
+    try:
+        from _thread import allocate_lock
+    except ImportError:
+        from _dummy_thread import allocate_lock
+
+
+##import sys
+##l1 = allocate_lock
+
+##class allocate_lock(object):
+##    def __init__(self):
+##        self._real = l1()
+##    def __enter__(self):
+##        for i in range(4, 0, -1):
+##            print sys._getframe(i).f_code
+##        print
+##        return self._real.__enter__()
+##    def __exit__(self, *args):
+##        return self._real.__exit__(*args)
+##    def acquire(self, f):
+##        assert f is False
+##        return self._real.acquire(f)

.venv/lib/python3.9/site-packages/cffi/model.py

@@ -0,0 +1,618 @@
+import types
+import weakref
+
+from .lock import allocate_lock
+from .error import CDefError, VerificationError, VerificationMissing
+
+# type qualifiers
+Q_CONST    = 0x01
+Q_RESTRICT = 0x02
+Q_VOLATILE = 0x04
+
+def qualify(quals, replace_with):
+    if quals & Q_CONST:
+        replace_with = ' const ' + replace_with.lstrip()
+    if quals & Q_VOLATILE:
+        replace_with = ' volatile ' + replace_with.lstrip()
+    if quals & Q_RESTRICT:
+        # It seems that __restrict is supported by gcc and msvc.
+        # If you hit some different compiler, add a #define in
+        # _cffi_include.h for it (and in its copies, documented there)
+        replace_with = ' __restrict ' + replace_with.lstrip()
+    return replace_with
+
+
+class BaseTypeByIdentity(object):
+    is_array_type = False
+    is_raw_function = False
+
+    def get_c_name(self, replace_with='', context='a C file', quals=0):
+        result = self.c_name_with_marker
+        assert result.count('&') == 1
+        # some logic duplication with ffi.getctype()... :-(
+        replace_with = replace_with.strip()
+        if replace_with:
+            if replace_with.startswith('*') and '&[' in result:
+                replace_with = '(%s)' % replace_with
+            elif not replace_with[0] in '[(':
+                replace_with = ' ' + replace_with
+        replace_with = qualify(quals, replace_with)
+        result = result.replace('&', replace_with)
+        if '$' in result:
+            raise VerificationError(
+                "cannot generate '%s' in %s: unknown type name"
+                % (self._get_c_name(), context))
+        return result
+
+    def _get_c_name(self):
+        return self.c_name_with_marker.replace('&', '')
+
+    def has_c_name(self):
+        return '$' not in self._get_c_name()
+
+    def is_integer_type(self):
+        return False
+
+    def get_cached_btype(self, ffi, finishlist, can_delay=False):
+        try:
+            BType = ffi._cached_btypes[self]
+        except KeyError:
+            BType = self.build_backend_type(ffi, finishlist)
+            BType2 = ffi._cached_btypes.setdefault(self, BType)
+            assert BType2 is BType
+        return BType
+
+    def __repr__(self):
+        return '<%s>' % (self._get_c_name(),)
+
+    def _get_items(self):
+        return [(name, getattr(self, name)) for name in self._attrs_]
+
+
+class BaseType(BaseTypeByIdentity):
+
+    def __eq__(self, other):
+        return (self.__class__ == other.__class__ and
+                self._get_items() == other._get_items())
+
+    def __ne__(self, other):
+        return not self == other
+
+    def __hash__(self):
+        return hash((self.__class__, tuple(self._get_items())))
+
+
+class VoidType(BaseType):
+    _attrs_ = ()
+
+    def __init__(self):
+        self.c_name_with_marker = 'void&'
+
+    def build_backend_type(self, ffi, finishlist):
+        return global_cache(self, ffi, 'new_void_type')
+
+void_type = VoidType()
+
+
+class BasePrimitiveType(BaseType):
+    def is_complex_type(self):
+        return False
+
+
+class PrimitiveType(BasePrimitiveType):
+    _attrs_ = ('name',)
+
+    ALL_PRIMITIVE_TYPES = {
+        'char':               'c',
+        'short':              'i',
+        'int':                'i',
+        'long':               'i',
+        'long long':          'i',
+        'signed char':        'i',
+        'unsigned char':      'i',
+        'unsigned short':     'i',
+        'unsigned int':       'i',
+        'unsigned long':      'i',
+        'unsigned long long': 'i',
+        'float':              'f',
+        'double':             'f',
+        'long double':        'f',
+        '_cffi_float_complex_t': 'j',
+        '_cffi_double_complex_t': 'j',
+        '_Bool':              'i',
+        # the following types are not primitive in the C sense
+        'wchar_t':            'c',
+        'char16_t':           'c',
+        'char32_t':           'c',
+        'int8_t':             'i',
+        'uint8_t':            'i',
+        'int16_t':            'i',
+        'uint16_t':           'i',
+        'int32_t':            'i',
+        'uint32_t':           'i',
+        'int64_t':            'i',
+        'uint64_t':           'i',
+        'int_least8_t':       'i',
+        'uint_least8_t':      'i',
+        'int_least16_t':      'i',
+        'uint_least16_t':     'i',
+        'int_least32_t':      'i',
+        'uint_least32_t':     'i',
+        'int_least64_t':      'i',
+        'uint_least64_t':     'i',
+        'int_fast8_t':        'i',
+        'uint_fast8_t':       'i',
+        'int_fast16_t':       'i',
+        'uint_fast16_t':      'i',
+        'int_fast32_t':       'i',
+        'uint_fast32_t':      'i',
+        'int_fast64_t':       'i',
+        'uint_fast64_t':      'i',
+        'intptr_t':           'i',
+        'uintptr_t':          'i',
+        'intmax_t':           'i',
+        'uintmax_t':          'i',
+        'ptrdiff_t':          'i',
+        'size_t':             'i',
+        'ssize_t':            'i',
+        }
+
+    def __init__(self, name):
+        assert name in self.ALL_PRIMITIVE_TYPES
+        self.name = name
+        self.c_name_with_marker = name + '&'
+
+    def is_char_type(self):
+        return self.ALL_PRIMITIVE_TYPES[self.name] == 'c'
+    def is_integer_type(self):
+        return self.ALL_PRIMITIVE_TYPES[self.name] == 'i'
+    def is_float_type(self):
+        return self.ALL_PRIMITIVE_TYPES[self.name] == 'f'
+    def is_complex_type(self):
+        return self.ALL_PRIMITIVE_TYPES[self.name] == 'j'
+
+    def build_backend_type(self, ffi, finishlist):
+        return global_cache(self, ffi, 'new_primitive_type', self.name)
+
+
+class UnknownIntegerType(BasePrimitiveType):
+    _attrs_ = ('name',)
+
+    def __init__(self, name):
+        self.name = name
+        self.c_name_with_marker = name + '&'
+
+    def is_integer_type(self):
+        return True
+
+    def build_backend_type(self, ffi, finishlist):
+        raise NotImplementedError("integer type '%s' can only be used after "
+                                  "compilation" % self.name)
+
+class UnknownFloatType(BasePrimitiveType):
+    _attrs_ = ('name', )
+
+    def __init__(self, name):
+        self.name = name
+        self.c_name_with_marker = name + '&'
+
+    def build_backend_type(self, ffi, finishlist):
+        raise NotImplementedError("float type '%s' can only be used after "
+                                  "compilation" % self.name)
+
+
+class BaseFunctionType(BaseType):
+    _attrs_ = ('args', 'result', 'ellipsis', 'abi')
+
+    def __init__(self, args, result, ellipsis, abi=None):
+        self.args = args
+        self.result = result
+        self.ellipsis = ellipsis
+        self.abi = abi
+        #
+        reprargs = [arg._get_c_name() for arg in self.args]
+        if self.ellipsis:
+            reprargs.append('...')
+        reprargs = reprargs or ['void']
+        replace_with = self._base_pattern % (', '.join(reprargs),)
+        if abi is not None:
+            replace_with = replace_with[:1] + abi + ' ' + replace_with[1:]
+        self.c_name_with_marker = (
+            self.result.c_name_with_marker.replace('&', replace_with))
+
+
+class RawFunctionType(BaseFunctionType):
+    # Corresponds to a C type like 'int(int)', which is the C type of
+    # a function, but not a pointer-to-function.  The backend has no
+    # notion of such a type; it's used temporarily by parsing.
+    _base_pattern = '(&)(%s)'
+    is_raw_function = True
+
+    def build_backend_type(self, ffi, finishlist):
+        raise CDefError("cannot render the type %r: it is a function "
+                        "type, not a pointer-to-function type" % (self,))
+
+    def as_function_pointer(self):
+        return FunctionPtrType(self.args, self.result, self.ellipsis, self.abi)
+
+
+class FunctionPtrType(BaseFunctionType):
+    _base_pattern = '(*&)(%s)'
+
+    def build_backend_type(self, ffi, finishlist):
+        result = self.result.get_cached_btype(ffi, finishlist)
+        args = []
+        for tp in self.args:
+            args.append(tp.get_cached_btype(ffi, finishlist))
+        abi_args = ()
+        if self.abi == "__stdcall":
+            if not self.ellipsis:    # __stdcall ignored for variadic funcs
+                try:
+                    abi_args = (ffi._backend.FFI_STDCALL,)
+                except AttributeError:
+                    pass
+        return global_cache(self, ffi, 'new_function_type',
+                            tuple(args), result, self.ellipsis, *abi_args)
+
+    def as_raw_function(self):
+        return RawFunctionType(self.args, self.result, self.ellipsis, self.abi)
+
+
+class PointerType(BaseType):
+    _attrs_ = ('totype', 'quals')
+
+    def __init__(self, totype, quals=0):
+        self.totype = totype
+        self.quals = quals
+        extra = " *&"
+        if totype.is_array_type:
+            extra = "(%s)" % (extra.lstrip(),)
+        extra = qualify(quals, extra)
+        self.c_name_with_marker = totype.c_name_with_marker.replace('&', extra)
+
+    def build_backend_type(self, ffi, finishlist):
+        BItem = self.totype.get_cached_btype(ffi, finishlist, can_delay=True)
+        return global_cache(self, ffi, 'new_pointer_type', BItem)
+
+voidp_type = PointerType(void_type)
+
+def ConstPointerType(totype):
+    return PointerType(totype, Q_CONST)
+
+const_voidp_type = ConstPointerType(void_type)
+
+
+class NamedPointerType(PointerType):
+    _attrs_ = ('totype', 'name')
+
+    def __init__(self, totype, name, quals=0):
+        PointerType.__init__(self, totype, quals)
+        self.name = name
+        self.c_name_with_marker = name + '&'
+
+
+class ArrayType(BaseType):
+    _attrs_ = ('item', 'length')
+    is_array_type = True
+
+    def __init__(self, item, length):
+        self.item = item
+        self.length = length
+        #
+        if length is None:
+            brackets = '&[]'
+        elif length == '...':
+            brackets = '&[/*...*/]'
+        else:
+            brackets = '&[%s]' % length
+        self.c_name_with_marker = (
+            self.item.c_name_with_marker.replace('&', brackets))
+
+    def length_is_unknown(self):
+        return isinstance(self.length, str)
+
+    def resolve_length(self, newlength):
+        return ArrayType(self.item, newlength)
+
+    def build_backend_type(self, ffi, finishlist):
+        if self.length_is_unknown():
+            raise CDefError("cannot render the type %r: unknown length" %
+                            (self,))
+        self.item.get_cached_btype(ffi, finishlist)   # force the item BType
+        BPtrItem = PointerType(self.item).get_cached_btype(ffi, finishlist)
+        return global_cache(self, ffi, 'new_array_type', BPtrItem, self.length)
+
+char_array_type = ArrayType(PrimitiveType('char'), None)
+
+
+class StructOrUnionOrEnum(BaseTypeByIdentity):
+    _attrs_ = ('name',)
+    forcename = None
+
+    def build_c_name_with_marker(self):
+        name = self.forcename or '%s %s' % (self.kind, self.name)
+        self.c_name_with_marker = name + '&'
+
+    def force_the_name(self, forcename):
+        self.forcename = forcename
+        self.build_c_name_with_marker()
+
+    def get_official_name(self):
+        assert self.c_name_with_marker.endswith('&')
+        return self.c_name_with_marker[:-1]
+
+
+class StructOrUnion(StructOrUnionOrEnum):
+    fixedlayout = None
+    completed = 0
+    partial = False
+    packed = 0
+
+    def __init__(self, name, fldnames, fldtypes, fldbitsize, fldquals=None):
+        self.name = name
+        self.fldnames = fldnames
+        self.fldtypes = fldtypes
+        self.fldbitsize = fldbitsize
+        self.fldquals = fldquals
+        self.build_c_name_with_marker()
+
+    def anonymous_struct_fields(self):
+        if self.fldtypes is not None:
+            for name, type in zip(self.fldnames, self.fldtypes):
+                if name == '' and isinstance(type, StructOrUnion):
+                    yield type
+
+    def enumfields(self, expand_anonymous_struct_union=True):
+        fldquals = self.fldquals
+        if fldquals is None:
+            fldquals = (0,) * len(self.fldnames)
+        for name, type, bitsize, quals in zip(self.fldnames, self.fldtypes,
+                                              self.fldbitsize, fldquals):
+            if (name == '' and isinstance(type, StructOrUnion)
+                    and expand_anonymous_struct_union):
+                # nested anonymous struct/union
+                for result in type.enumfields():
+                    yield result
+            else:
+                yield (name, type, bitsize, quals)
+
+    def force_flatten(self):
+        # force the struct or union to have a declaration that lists
+        # directly all fields returned by enumfields(), flattening
+        # nested anonymous structs/unions.
+        names = []
+        types = []
+        bitsizes = []
+        fldquals = []
+        for name, type, bitsize, quals in self.enumfields():
+            names.append(name)
+            types.append(type)
+            bitsizes.append(bitsize)
+            fldquals.append(quals)
+        self.fldnames = tuple(names)
+        self.fldtypes = tuple(types)
+        self.fldbitsize = tuple(bitsizes)
+        self.fldquals = tuple(fldquals)
+
+    def get_cached_btype(self, ffi, finishlist, can_delay=False):
+        BType = StructOrUnionOrEnum.get_cached_btype(self, ffi, finishlist,
+                                                     can_delay)
+        if not can_delay:
+            self.finish_backend_type(ffi, finishlist)
+        return BType
+
+    def finish_backend_type(self, ffi, finishlist):
+        if self.completed:
+            if self.completed != 2:
+                raise NotImplementedError("recursive structure declaration "
+                                          "for '%s'" % (self.name,))
+            return
+        BType = ffi._cached_btypes[self]
+        #
+        self.completed = 1
+        #
+        if self.fldtypes is None:
+            pass    # not completing it: it's an opaque struct
+            #
+        elif self.fixedlayout is None:
+            fldtypes = [tp.get_cached_btype(ffi, finishlist)
+                        for tp in self.fldtypes]
+            lst = list(zip(self.fldnames, fldtypes, self.fldbitsize))
+            extra_flags = ()
+            if self.packed:
+                if self.packed == 1:
+                    extra_flags = (8,)    # SF_PACKED
+                else:
+                    extra_flags = (0, self.packed)
+            ffi._backend.complete_struct_or_union(BType, lst, self,
+                                                  -1, -1, *extra_flags)
+            #
+        else:
+            fldtypes = []
+            fieldofs, fieldsize, totalsize, totalalignment = self.fixedlayout
+            for i in range(len(self.fldnames)):
+                fsize = fieldsize[i]
+                ftype = self.fldtypes[i]
+                #
+                if isinstance(ftype, ArrayType) and ftype.length_is_unknown():
+                    # fix the length to match the total size
+                    BItemType = ftype.item.get_cached_btype(ffi, finishlist)
+                    nlen, nrest = divmod(fsize, ffi.sizeof(BItemType))
+                    if nrest != 0:
+                        self._verification_error(
+                            "field '%s.%s' has a bogus size?" % (
+                            self.name, self.fldnames[i] or '{}'))
+                    ftype = ftype.resolve_length(nlen)
+                    self.fldtypes = (self.fldtypes[:i] + (ftype,) +
+                                     self.fldtypes[i+1:])
+                #
+                BFieldType = ftype.get_cached_btype(ffi, finishlist)
+                if isinstance(ftype, ArrayType) and ftype.length is None:
+                    assert fsize == 0
+                else:
+                    bitemsize = ffi.sizeof(BFieldType)
+                    if bitemsize != fsize:
+                        self._verification_error(
+                            "field '%s.%s' is declared as %d bytes, but is "
+                            "really %d bytes" % (self.name,
+                                                 self.fldnames[i] or '{}',
+                                                 bitemsize, fsize))
+                fldtypes.append(BFieldType)
+            #
+            lst = list(zip(self.fldnames, fldtypes, self.fldbitsize, fieldofs))
+            ffi._backend.complete_struct_or_union(BType, lst, self,
+                                                  totalsize, totalalignment)
+        self.completed = 2
+
+    def _verification_error(self, msg):
+        raise VerificationError(msg)
+
+    def check_not_partial(self):
+        if self.partial and self.fixedlayout is None:
+            raise VerificationMissing(self._get_c_name())
+
+    def build_backend_type(self, ffi, finishlist):
+        self.check_not_partial()
+        finishlist.append(self)
+        #
+        return global_cache(self, ffi, 'new_%s_type' % self.kind,
+                            self.get_official_name(), key=self)
+
+
+class StructType(StructOrUnion):
+    kind = 'struct'
+
+
+class UnionType(StructOrUnion):
+    kind = 'union'
+
+
+class EnumType(StructOrUnionOrEnum):
+    kind = 'enum'
+    partial = False
+    partial_resolved = False
+
+    def __init__(self, name, enumerators, enumvalues, baseinttype=None):
+        self.name = name
+        self.enumerators = enumerators
+        self.enumvalues = enumvalues
+        self.baseinttype = baseinttype
+        self.build_c_name_with_marker()
+
+    def force_the_name(self, forcename):
+        StructOrUnionOrEnum.force_the_name(self, forcename)
+        if self.forcename is None:
+            name = self.get_official_name()
+            self.forcename = '$' + name.replace(' ', '_')
+
+    def check_not_partial(self):
+        if self.partial and not self.partial_resolved:
+            raise VerificationMissing(self._get_c_name())
+
+    def build_backend_type(self, ffi, finishlist):
+        self.check_not_partial()
+        base_btype = self.build_baseinttype(ffi, finishlist)
+        return global_cache(self, ffi, 'new_enum_type',
+                            self.get_official_name(),
+                            self.enumerators, self.enumvalues,
+                            base_btype, key=self)
+
+    def build_baseinttype(self, ffi, finishlist):
+        if self.baseinttype is not None:
+            return self.baseinttype.get_cached_btype(ffi, finishlist)
+        #
+        if self.enumvalues:
+            smallest_value = min(self.enumvalues)
+            largest_value = max(self.enumvalues)
+        else:
+            import warnings
+            try:
+                # XXX!  The goal is to ensure that the warnings.warn()
+                # will not suppress the warning.  We want to get it
+                # several times if we reach this point several times.
+                __warningregistry__.clear()
+            except NameError:
+                pass
+            warnings.warn("%r has no values explicitly defined; "
+                          "guessing that it is equivalent to 'unsigned int'"
+                          % self._get_c_name())
+            smallest_value = largest_value = 0
+        if smallest_value < 0:   # needs a signed type
+            sign = 1
+            candidate1 = PrimitiveType("int")
+            candidate2 = PrimitiveType("long")
+        else:
+            sign = 0
+            candidate1 = PrimitiveType("unsigned int")
+            candidate2 = PrimitiveType("unsigned long")
+        btype1 = candidate1.get_cached_btype(ffi, finishlist)
+        btype2 = candidate2.get_cached_btype(ffi, finishlist)
+        size1 = ffi.sizeof(btype1)
+        size2 = ffi.sizeof(btype2)
+        if (smallest_value >= ((-1) << (8*size1-1)) and
+            largest_value < (1 << (8*size1-sign))):
+            return btype1
+        if (smallest_value >= ((-1) << (8*size2-1)) and
+            largest_value < (1 << (8*size2-sign))):
+            return btype2
+        raise CDefError("%s values don't all fit into either 'long' "
+                        "or 'unsigned long'" % self._get_c_name())
+
+def unknown_type(name, structname=None):
+    if structname is None:
+        structname = '$%s' % name
+    tp = StructType(structname, None, None, None)
+    tp.force_the_name(name)
+    tp.origin = "unknown_type"
+    return tp
+
+def unknown_ptr_type(name, structname=None):
+    if structname is None:
+        structname = '$$%s' % name
+    tp = StructType(structname, None, None, None)
+    return NamedPointerType(tp, name)
+
+
+global_lock = allocate_lock()
+_typecache_cffi_backend = weakref.WeakValueDictionary()
+
+def get_typecache(backend):
+    # returns _typecache_cffi_backend if backend is the _cffi_backend
+    # module, or type(backend).__typecache if backend is an instance of
+    # CTypesBackend (or some FakeBackend class during tests)
+    if isinstance(backend, types.ModuleType):
+        return _typecache_cffi_backend
+    with global_lock:
+        if not hasattr(type(backend), '__typecache'):
+            type(backend).__typecache = weakref.WeakValueDictionary()
+        return type(backend).__typecache
+
+def global_cache(srctype, ffi, funcname, *args, **kwds):
+    key = kwds.pop('key', (funcname, args))
+    assert not kwds
+    try:
+        return ffi._typecache[key]
+    except KeyError:
+        pass
+    try:
+        res = getattr(ffi._backend, funcname)(*args)
+    except NotImplementedError as e:
+        raise NotImplementedError("%s: %r: %s" % (funcname, srctype, e))
+    # note that setdefault() on WeakValueDictionary is not atomic
+    # and contains a rare bug (http://bugs.python.org/issue19542);
+    # we have to use a lock and do it ourselves
+    cache = ffi._typecache
+    with global_lock:
+        res1 = cache.get(key)
+        if res1 is None:
+            cache[key] = res
+            return res
+        else:
+            return res1
+
+def pointer_cache(ffi, BType):
+    return global_cache('?', ffi, 'new_pointer_type', BType)
+
+def attach_exception_info(e, name):
+    if e.args and type(e.args[0]) is str:
+        e.args = ('%s: %s' % (name, e.args[0]),) + e.args[1:]

.venv/lib/python3.9/site-packages/cffi/parse_c_type.h

@@ -0,0 +1,181 @@
+
+/* This part is from file 'cffi/parse_c_type.h'.  It is copied at the
+   beginning of C sources generated by CFFI's ffi.set_source(). */
+
+typedef void *_cffi_opcode_t;
+
+#define _CFFI_OP(opcode, arg)   (_cffi_opcode_t)(opcode | (((uintptr_t)(arg)) << 8))
+#define _CFFI_GETOP(cffi_opcode)    ((unsigned char)(uintptr_t)cffi_opcode)
+#define _CFFI_GETARG(cffi_opcode)   (((intptr_t)cffi_opcode) >> 8)
+
+#define _CFFI_OP_PRIMITIVE       1
+#define _CFFI_OP_POINTER         3
+#define _CFFI_OP_ARRAY           5
+#define _CFFI_OP_OPEN_ARRAY      7
+#define _CFFI_OP_STRUCT_UNION    9
+#define _CFFI_OP_ENUM           11
+#define _CFFI_OP_FUNCTION       13
+#define _CFFI_OP_FUNCTION_END   15
+#define _CFFI_OP_NOOP           17
+#define _CFFI_OP_BITFIELD       19
+#define _CFFI_OP_TYPENAME       21
+#define _CFFI_OP_CPYTHON_BLTN_V 23   // varargs
+#define _CFFI_OP_CPYTHON_BLTN_N 25   // noargs
+#define _CFFI_OP_CPYTHON_BLTN_O 27   // O  (i.e. a single arg)
+#define _CFFI_OP_CONSTANT       29
+#define _CFFI_OP_CONSTANT_INT   31
+#define _CFFI_OP_GLOBAL_VAR     33
+#define _CFFI_OP_DLOPEN_FUNC    35
+#define _CFFI_OP_DLOPEN_CONST   37
+#define _CFFI_OP_GLOBAL_VAR_F   39
+#define _CFFI_OP_EXTERN_PYTHON  41
+
+#define _CFFI_PRIM_VOID          0
+#define _CFFI_PRIM_BOOL          1
+#define _CFFI_PRIM_CHAR          2
+#define _CFFI_PRIM_SCHAR         3
+#define _CFFI_PRIM_UCHAR         4
+#define _CFFI_PRIM_SHORT         5
+#define _CFFI_PRIM_USHORT        6
+#define _CFFI_PRIM_INT           7
+#define _CFFI_PRIM_UINT          8
+#define _CFFI_PRIM_LONG          9
+#define _CFFI_PRIM_ULONG        10
+#define _CFFI_PRIM_LONGLONG     11
+#define _CFFI_PRIM_ULONGLONG    12
+#define _CFFI_PRIM_FLOAT        13
+#define _CFFI_PRIM_DOUBLE       14
+#define _CFFI_PRIM_LONGDOUBLE   15
+
+#define _CFFI_PRIM_WCHAR        16
+#define _CFFI_PRIM_INT8         17
+#define _CFFI_PRIM_UINT8        18
+#define _CFFI_PRIM_INT16        19
+#define _CFFI_PRIM_UINT16       20
+#define _CFFI_PRIM_INT32        21
+#define _CFFI_PRIM_UINT32       22
+#define _CFFI_PRIM_INT64        23
+#define _CFFI_PRIM_UINT64       24
+#define _CFFI_PRIM_INTPTR       25
+#define _CFFI_PRIM_UINTPTR      26
+#define _CFFI_PRIM_PTRDIFF      27
+#define _CFFI_PRIM_SIZE         28
+#define _CFFI_PRIM_SSIZE        29
+#define _CFFI_PRIM_INT_LEAST8   30
+#define _CFFI_PRIM_UINT_LEAST8  31
+#define _CFFI_PRIM_INT_LEAST16  32
+#define _CFFI_PRIM_UINT_LEAST16 33
+#define _CFFI_PRIM_INT_LEAST32  34
+#define _CFFI_PRIM_UINT_LEAST32 35
+#define _CFFI_PRIM_INT_LEAST64  36
+#define _CFFI_PRIM_UINT_LEAST64 37
+#define _CFFI_PRIM_INT_FAST8    38
+#define _CFFI_PRIM_UINT_FAST8   39
+#define _CFFI_PRIM_INT_FAST16   40
+#define _CFFI_PRIM_UINT_FAST16  41
+#define _CFFI_PRIM_INT_FAST32   42
+#define _CFFI_PRIM_UINT_FAST32  43
+#define _CFFI_PRIM_INT_FAST64   44
+#define _CFFI_PRIM_UINT_FAST64  45
+#define _CFFI_PRIM_INTMAX       46
+#define _CFFI_PRIM_UINTMAX      47
+#define _CFFI_PRIM_FLOATCOMPLEX 48
+#define _CFFI_PRIM_DOUBLECOMPLEX 49
+#define _CFFI_PRIM_CHAR16       50
+#define _CFFI_PRIM_CHAR32       51
+
+#define _CFFI__NUM_PRIM         52
+#define _CFFI__UNKNOWN_PRIM           (-1)
+#define _CFFI__UNKNOWN_FLOAT_PRIM     (-2)
+#define _CFFI__UNKNOWN_LONG_DOUBLE    (-3)
+
+#define _CFFI__IO_FILE_STRUCT         (-1)
+
+
+struct _cffi_global_s {
+    const char *name;
+    void *address;
+    _cffi_opcode_t type_op;
+    void *size_or_direct_fn;  // OP_GLOBAL_VAR: size, or 0 if unknown
+                              // OP_CPYTHON_BLTN_*: addr of direct function
+};
+
+struct _cffi_getconst_s {
+    unsigned long long value;
+    const struct _cffi_type_context_s *ctx;
+    int gindex;
+};
+
+struct _cffi_struct_union_s {
+    const char *name;
+    int type_index;          // -> _cffi_types, on a OP_STRUCT_UNION
+    int flags;               // _CFFI_F_* flags below
+    size_t size;
+    int alignment;
+    int first_field_index;   // -> _cffi_fields array
+    int num_fields;
+};
+#define _CFFI_F_UNION         0x01   // is a union, not a struct
+#define _CFFI_F_CHECK_FIELDS  0x02   // complain if fields are not in the
+                                     // "standard layout" or if some are missing
+#define _CFFI_F_PACKED        0x04   // for CHECK_FIELDS, assume a packed struct
+#define _CFFI_F_EXTERNAL      0x08   // in some other ffi.include()
+#define _CFFI_F_OPAQUE        0x10   // opaque
+
+struct _cffi_field_s {
+    const char *name;
+    size_t field_offset;
+    size_t field_size;
+    _cffi_opcode_t field_type_op;
+};
+
+struct _cffi_enum_s {
+    const char *name;
+    int type_index;          // -> _cffi_types, on a OP_ENUM
+    int type_prim;           // _CFFI_PRIM_xxx
+    const char *enumerators; // comma-delimited string
+};
+
+struct _cffi_typename_s {
+    const char *name;
+    int type_index;   /* if opaque, points to a possibly artificial
+                         OP_STRUCT which is itself opaque */
+};
+
+struct _cffi_type_context_s {
+    _cffi_opcode_t *types;
+    const struct _cffi_global_s *globals;
+    const struct _cffi_field_s *fields;
+    const struct _cffi_struct_union_s *struct_unions;
+    const struct _cffi_enum_s *enums;
+    const struct _cffi_typename_s *typenames;
+    int num_globals;
+    int num_struct_unions;
+    int num_enums;
+    int num_typenames;
+    const char *const *includes;
+    int num_types;
+    int flags;      /* future extension */
+};
+
+struct _cffi_parse_info_s {
+    const struct _cffi_type_context_s *ctx;
+    _cffi_opcode_t *output;
+    unsigned int output_size;
+    size_t error_location;
+    const char *error_message;
+};
+
+struct _cffi_externpy_s {
+    const char *name;
+    size_t size_of_result;
+    void *reserved1, *reserved2;
+};
+
+#ifdef _CFFI_INTERNAL
+static int parse_c_type(struct _cffi_parse_info_s *info, const char *input);
+static int search_in_globals(const struct _cffi_type_context_s *ctx,
+                             const char *search, size_t search_len);
+static int search_in_struct_unions(const struct _cffi_type_context_s *ctx,
+                                   const char *search, size_t search_len);
+#endif

.venv/lib/python3.9/site-packages/cffi/pkgconfig.py

@@ -0,0 +1,121 @@
+# pkg-config, https://www.freedesktop.org/wiki/Software/pkg-config/ integration for cffi
+import sys, os, subprocess
+
+from .error import PkgConfigError
+
+
+def merge_flags(cfg1, cfg2):
+    """Merge values from cffi config flags cfg2 to cf1
+
+    Example:
+        merge_flags({"libraries": ["one"]}, {"libraries": ["two"]})
+        {"libraries": ["one", "two"]}
+    """
+    for key, value in cfg2.items():
+        if key not in cfg1:
+            cfg1[key] = value
+        else:
+            if not isinstance(cfg1[key], list):
+                raise TypeError("cfg1[%r] should be a list of strings" % (key,))
+            if not isinstance(value, list):
+                raise TypeError("cfg2[%r] should be a list of strings" % (key,))
+            cfg1[key].extend(value)
+    return cfg1
+
+
+def call(libname, flag, encoding=sys.getfilesystemencoding()):
+    """Calls pkg-config and returns the output if found
+    """
+    a = ["pkg-config", "--print-errors"]
+    a.append(flag)
+    a.append(libname)
+    try:
+        pc = subprocess.Popen(a, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    except EnvironmentError as e:
+        raise PkgConfigError("cannot run pkg-config: %s" % (str(e).strip(),))
+
+    bout, berr = pc.communicate()
+    if pc.returncode != 0:
+        try:
+            berr = berr.decode(encoding)
+        except Exception:
+            pass
+        raise PkgConfigError(berr.strip())
+
+    if sys.version_info >= (3,) and not isinstance(bout, str):   # Python 3.x
+        try:
+            bout = bout.decode(encoding)
+        except UnicodeDecodeError:
+            raise PkgConfigError("pkg-config %s %s returned bytes that cannot "
+                                 "be decoded with encoding %r:\n%r" %
+                                 (flag, libname, encoding, bout))
+
+    if os.altsep != '\\' and '\\' in bout:
+        raise PkgConfigError("pkg-config %s %s returned an unsupported "
+                             "backslash-escaped output:\n%r" %
+                             (flag, libname, bout))
+    return bout
+
+
+def flags_from_pkgconfig(libs):
+    r"""Return compiler line flags for FFI.set_source based on pkg-config output
+
+    Usage
+        ...
+        ffibuilder.set_source("_foo", pkgconfig = ["libfoo", "libbar >= 1.8.3"])
+
+    If pkg-config is installed on build machine, then arguments include_dirs,
+    library_dirs, libraries, define_macros, extra_compile_args and
+    extra_link_args are extended with an output of pkg-config for libfoo and
+    libbar.
+
+    Raises PkgConfigError in case the pkg-config call fails.
+    """
+
+    def get_include_dirs(string):
+        return [x[2:] for x in string.split() if x.startswith("-I")]
+
+    def get_library_dirs(string):
+        return [x[2:] for x in string.split() if x.startswith("-L")]
+
+    def get_libraries(string):
+        return [x[2:] for x in string.split() if x.startswith("-l")]
+
+    # convert -Dfoo=bar to list of tuples [("foo", "bar")] expected by distutils
+    def get_macros(string):
+        def _macro(x):
+            x = x[2:]    # drop "-D"
+            if '=' in x:
+                return tuple(x.split("=", 1))  # "-Dfoo=bar" => ("foo", "bar")
+            else:
+                return (x, None)               # "-Dfoo" => ("foo", None)
+        return [_macro(x) for x in string.split() if x.startswith("-D")]
+
+    def get_other_cflags(string):
+        return [x for x in string.split() if not x.startswith("-I") and
+                                             not x.startswith("-D")]
+
+    def get_other_libs(string):
+        return [x for x in string.split() if not x.startswith("-L") and
+                                             not x.startswith("-l")]
+
+    # return kwargs for given libname
+    def kwargs(libname):
+        fse = sys.getfilesystemencoding()
+        all_cflags = call(libname, "--cflags")
+        all_libs = call(libname, "--libs")
+        return {
+            "include_dirs": get_include_dirs(all_cflags),
+            "library_dirs": get_library_dirs(all_libs),
+            "libraries": get_libraries(all_libs),
+            "define_macros": get_macros(all_cflags),
+            "extra_compile_args": get_other_cflags(all_cflags),
+            "extra_link_args": get_other_libs(all_libs),
+            }
+
+    # merge all arguments together
+    ret = {}
+    for libname in libs:
+        lib_flags = kwargs(libname)
+        merge_flags(ret, lib_flags)
+    return ret

.venv/lib/python3.9/site-packages/cffi/setuptools_ext.py

@@ -0,0 +1,229 @@
+import os
+import sys
+import sysconfig
+
+try:
+    basestring
+except NameError:
+    # Python 3.x
+    basestring = str
+
+def error(msg):
+    from cffi._shimmed_dist_utils import DistutilsSetupError
+    raise DistutilsSetupError(msg)
+
+
+def execfile(filename, glob):
+    # We use execfile() (here rewritten for Python 3) instead of
+    # __import__() to load the build script.  The problem with
+    # a normal import is that in some packages, the intermediate
+    # __init__.py files may already try to import the file that
+    # we are generating.
+    with open(filename) as f:
+        src = f.read()
+    src += '\n'      # Python 2.6 compatibility
+    code = compile(src, filename, 'exec')
+    exec(code, glob, glob)
+
+
+def add_cffi_module(dist, mod_spec):
+    from cffi.api import FFI
+
+    if not isinstance(mod_spec, basestring):
+        error("argument to 'cffi_modules=...' must be a str or a list of str,"
+              " not %r" % (type(mod_spec).__name__,))
+    mod_spec = str(mod_spec)
+    try:
+        build_file_name, ffi_var_name = mod_spec.split(':')
+    except ValueError:
+        error("%r must be of the form 'path/build.py:ffi_variable'" %
+              (mod_spec,))
+    if not os.path.exists(build_file_name):
+        ext = ''
+        rewritten = build_file_name.replace('.', '/') + '.py'
+        if os.path.exists(rewritten):
+            ext = ' (rewrite cffi_modules to [%r])' % (
+                rewritten + ':' + ffi_var_name,)
+        error("%r does not name an existing file%s" % (build_file_name, ext))
+
+    mod_vars = {'__name__': '__cffi__', '__file__': build_file_name}
+    execfile(build_file_name, mod_vars)
+
+    try:
+        ffi = mod_vars[ffi_var_name]
+    except KeyError:
+        error("%r: object %r not found in module" % (mod_spec,
+                                                     ffi_var_name))
+    if not isinstance(ffi, FFI):
+        ffi = ffi()      # maybe it's a function instead of directly an ffi
+    if not isinstance(ffi, FFI):
+        error("%r is not an FFI instance (got %r)" % (mod_spec,
+                                                      type(ffi).__name__))
+    if not hasattr(ffi, '_assigned_source'):
+        error("%r: the set_source() method was not called" % (mod_spec,))
+    module_name, source, source_extension, kwds = ffi._assigned_source
+    if ffi._windows_unicode:
+        kwds = kwds.copy()
+        ffi._apply_windows_unicode(kwds)
+
+    if source is None:
+        _add_py_module(dist, ffi, module_name)
+    else:
+        _add_c_module(dist, ffi, module_name, source, source_extension, kwds)
+
+def _set_py_limited_api(Extension, kwds):
+    """
+    Add py_limited_api to kwds if setuptools >= 26 is in use.
+    Do not alter the setting if it already exists.
+    Setuptools takes care of ignoring the flag on Python 2 and PyPy.
+
+    CPython itself should ignore the flag in a debugging version
+    (by not listing .abi3.so in the extensions it supports), but
+    it doesn't so far, creating troubles.  That's why we check
+    for "not hasattr(sys, 'gettotalrefcount')" (the 2.7 compatible equivalent
+    of 'd' not in sys.abiflags). (http://bugs.python.org/issue28401)
+
+    On Windows, with CPython <= 3.4, it's better not to use py_limited_api
+    because virtualenv *still* doesn't copy PYTHON3.DLL on these versions.
+    Recently (2020) we started shipping only >= 3.5 wheels, though.  So
+    we'll give it another try and set py_limited_api on Windows >= 3.5.
+    """
+    from cffi._shimmed_dist_utils import log
+    from cffi import recompiler
+
+    if ('py_limited_api' not in kwds and not hasattr(sys, 'gettotalrefcount')
+            and recompiler.USE_LIMITED_API):
+        import setuptools
+        try:
+            setuptools_major_version = int(setuptools.__version__.partition('.')[0])
+            if setuptools_major_version >= 26:
+                kwds['py_limited_api'] = True
+        except ValueError:  # certain development versions of setuptools
+            # If we don't know the version number of setuptools, we
+            # try to set 'py_limited_api' anyway.  At worst, we get a
+            # warning.
+            kwds['py_limited_api'] = True
+
+    if sysconfig.get_config_var("Py_GIL_DISABLED"):
+        if kwds.get('py_limited_api'):
+            log.info("Ignoring py_limited_api=True for free-threaded build.")
+
+        kwds['py_limited_api'] = False
+
+    if kwds.get('py_limited_api') is False:
+        # avoid setting Py_LIMITED_API if py_limited_api=False
+        # which _cffi_include.h does unless _CFFI_NO_LIMITED_API is defined
+        kwds.setdefault("define_macros", []).append(("_CFFI_NO_LIMITED_API", None))
+    return kwds
+
+def _add_c_module(dist, ffi, module_name, source, source_extension, kwds):
+    # We are a setuptools extension. Need this build_ext for py_limited_api.
+    from setuptools.command.build_ext import build_ext
+    from cffi._shimmed_dist_utils import Extension, log, mkpath
+    from cffi import recompiler
+
+    allsources = ['$PLACEHOLDER']
+    allsources.extend(kwds.pop('sources', []))
+    kwds = _set_py_limited_api(Extension, kwds)
+    ext = Extension(name=module_name, sources=allsources, **kwds)
+
+    def make_mod(tmpdir, pre_run=None):
+        c_file = os.path.join(tmpdir, module_name + source_extension)
+        log.info("generating cffi module %r" % c_file)
+        mkpath(tmpdir)
+        # a setuptools-only, API-only hook: called with the "ext" and "ffi"
+        # arguments just before we turn the ffi into C code.  To use it,
+        # subclass the 'distutils.command.build_ext.build_ext' class and
+        # add a method 'def pre_run(self, ext, ffi)'.
+        if pre_run is not None:
+            pre_run(ext, ffi)
+        updated = recompiler.make_c_source(ffi, module_name, source, c_file)
+        if not updated:
+            log.info("already up-to-date")
+        return c_file
+
+    if dist.ext_modules is None:
+        dist.ext_modules = []
+    dist.ext_modules.append(ext)
+
+    base_class = dist.cmdclass.get('build_ext', build_ext)
+    class build_ext_make_mod(base_class):
+        def run(self):
+            if ext.sources[0] == '$PLACEHOLDER':
+                pre_run = getattr(self, 'pre_run', None)
+                ext.sources[0] = make_mod(self.build_temp, pre_run)
+            base_class.run(self)
+    dist.cmdclass['build_ext'] = build_ext_make_mod
+    # NB. multiple runs here will create multiple 'build_ext_make_mod'
+    # classes.  Even in this case the 'build_ext' command should be
+    # run once; but just in case, the logic above does nothing if
+    # called again.
+
+
+def _add_py_module(dist, ffi, module_name):
+    from setuptools.command.build_py import build_py
+    from setuptools.command.build_ext import build_ext
+    from cffi._shimmed_dist_utils import log, mkpath
+    from cffi import recompiler
+
+    def generate_mod(py_file):
+        log.info("generating cffi module %r" % py_file)
+        mkpath(os.path.dirname(py_file))
+        updated = recompiler.make_py_source(ffi, module_name, py_file)
+        if not updated:
+            log.info("already up-to-date")
+
+    base_class = dist.cmdclass.get('build_py', build_py)
+    class build_py_make_mod(base_class):
+        def run(self):
+            base_class.run(self)
+            module_path = module_name.split('.')
+            module_path[-1] += '.py'
+            generate_mod(os.path.join(self.build_lib, *module_path))
+        def get_source_files(self):
+            # This is called from 'setup.py sdist' only.  Exclude
+            # the generate .py module in this case.
+            saved_py_modules = self.py_modules
+            try:
+                if saved_py_modules:
+                    self.py_modules = [m for m in saved_py_modules
+                                         if m != module_name]
+                return base_class.get_source_files(self)
+            finally:
+                self.py_modules = saved_py_modules
+    dist.cmdclass['build_py'] = build_py_make_mod
+
+    # distutils and setuptools have no notion I could find of a
+    # generated python module.  If we don't add module_name to
+    # dist.py_modules, then things mostly work but there are some
+    # combination of options (--root and --record) that will miss
+    # the module.  So we add it here, which gives a few apparently
+    # harmless warnings about not finding the file outside the
+    # build directory.
+    # Then we need to hack more in get_source_files(); see above.
+    if dist.py_modules is None:
+        dist.py_modules = []
+    dist.py_modules.append(module_name)
+
+    # the following is only for "build_ext -i"
+    base_class_2 = dist.cmdclass.get('build_ext', build_ext)
+    class build_ext_make_mod(base_class_2):
+        def run(self):
+            base_class_2.run(self)
+            if self.inplace:
+                # from get_ext_fullpath() in distutils/command/build_ext.py
+                module_path = module_name.split('.')
+                package = '.'.join(module_path[:-1])
+                build_py = self.get_finalized_command('build_py')
+                package_dir = build_py.get_package_dir(package)
+                file_name = module_path[-1] + '.py'
+                generate_mod(os.path.join(package_dir, file_name))
+    dist.cmdclass['build_ext'] = build_ext_make_mod
+
+def cffi_modules(dist, attr, value):
+    assert attr == 'cffi_modules'
+    if isinstance(value, basestring):
+        value = [value]
+
+    for cffi_module in value:
+        add_cffi_module(dist, cffi_module)

.venv/lib/python3.9/site-packages/cffi/vengine_gen.py

@@ -0,0 +1,679 @@
+#
+# DEPRECATED: implementation for ffi.verify()
+#
+import sys, os
+import types
+
+from . import model
+from .error import VerificationError
+
+
+class VGenericEngine(object):
+    _class_key = 'g'
+    _gen_python_module = False
+
+    def __init__(self, verifier):
+        self.verifier = verifier
+        self.ffi = verifier.ffi
+        self.export_symbols = []
+        self._struct_pending_verification = {}
+
+    def patch_extension_kwds(self, kwds):
+        # add 'export_symbols' to the dictionary.  Note that we add the
+        # list before filling it.  When we fill it, it will thus also show
+        # up in kwds['export_symbols'].
+        kwds.setdefault('export_symbols', self.export_symbols)
+
+    def find_module(self, module_name, path, so_suffixes):
+        for so_suffix in so_suffixes:
+            basename = module_name + so_suffix
+            if path is None:
+                path = sys.path
+            for dirname in path:
+                filename = os.path.join(dirname, basename)
+                if os.path.isfile(filename):
+                    return filename
+
+    def collect_types(self):
+        pass      # not needed in the generic engine
+
+    def _prnt(self, what=''):
+        self._f.write(what + '\n')
+
+    def write_source_to_f(self):
+        prnt = self._prnt
+        # first paste some standard set of lines that are mostly '#include'
+        prnt(cffimod_header)
+        # then paste the C source given by the user, verbatim.
+        prnt(self.verifier.preamble)
+        #
+        # call generate_gen_xxx_decl(), for every xxx found from
+        # ffi._parser._declarations.  This generates all the functions.
+        self._generate('decl')
+        #
+        # on Windows, distutils insists on putting init_cffi_xyz in
+        # 'export_symbols', so instead of fighting it, just give up and
+        # give it one
+        if sys.platform == 'win32':
+            if sys.version_info >= (3,):
+                prefix = 'PyInit_'
+            else:
+                prefix = 'init'
+            modname = self.verifier.get_module_name()
+            prnt("void %s%s(void) { }\n" % (prefix, modname))
+
+    def load_library(self, flags=0):
+        # import it with the CFFI backend
+        backend = self.ffi._backend
+        # needs to make a path that contains '/', on Posix
+        filename = os.path.join(os.curdir, self.verifier.modulefilename)
+        module = backend.load_library(filename, flags)
+        #
+        # call loading_gen_struct() to get the struct layout inferred by
+        # the C compiler
+        self._load(module, 'loading')
+
+        # build the FFILibrary class and instance, this is a module subclass
+        # because modules are expected to have usually-constant-attributes and
+        # in PyPy this means the JIT is able to treat attributes as constant,
+        # which we want.
+        class FFILibrary(types.ModuleType):
+            _cffi_generic_module = module
+            _cffi_ffi = self.ffi
+            _cffi_dir = []
+            def __dir__(self):
+                return FFILibrary._cffi_dir
+        library = FFILibrary("")
+        #
+        # finally, call the loaded_gen_xxx() functions.  This will set
+        # up the 'library' object.
+        self._load(module, 'loaded', library=library)
+        return library
+
+    def _get_declarations(self):
+        lst = [(key, tp) for (key, (tp, qual)) in
+                                self.ffi._parser._declarations.items()]
+        lst.sort()
+        return lst
+
+    def _generate(self, step_name):
+        for name, tp in self._get_declarations():
+            kind, realname = name.split(' ', 1)
+            try:
+                method = getattr(self, '_generate_gen_%s_%s' % (kind,
+                                                                step_name))
+            except AttributeError:
+                raise VerificationError(
+                    "not implemented in verify(): %r" % name)
+            try:
+                method(tp, realname)
+            except Exception as e:
+                model.attach_exception_info(e, name)
+                raise
+
+    def _load(self, module, step_name, **kwds):
+        for name, tp in self._get_declarations():
+            kind, realname = name.split(' ', 1)
+            method = getattr(self, '_%s_gen_%s' % (step_name, kind))
+            try:
+                method(tp, realname, module, **kwds)
+            except Exception as e:
+                model.attach_exception_info(e, name)
+                raise
+
+    def _generate_nothing(self, tp, name):
+        pass
+
+    def _loaded_noop(self, tp, name, module, **kwds):
+        pass
+
+    # ----------
+    # typedefs: generates no code so far
+
+    _generate_gen_typedef_decl   = _generate_nothing
+    _loading_gen_typedef         = _loaded_noop
+    _loaded_gen_typedef          = _loaded_noop
+
+    # ----------
+    # function declarations
+
+    def _generate_gen_function_decl(self, tp, name):
+        assert isinstance(tp, model.FunctionPtrType)
+        if tp.ellipsis:
+            # cannot support vararg functions better than this: check for its
+            # exact type (including the fixed arguments), and build it as a
+            # constant function pointer (no _cffi_f_%s wrapper)
+            self._generate_gen_const(False, name, tp)
+            return
+        prnt = self._prnt
+        numargs = len(tp.args)
+        argnames = []
+        for i, type in enumerate(tp.args):
+            indirection = ''
+            if isinstance(type, model.StructOrUnion):
+                indirection = '*'
+            argnames.append('%sx%d' % (indirection, i))
+        context = 'argument of %s' % name
+        arglist = [type.get_c_name(' %s' % arg, context)
+                   for type, arg in zip(tp.args, argnames)]
+        tpresult = tp.result
+        if isinstance(tpresult, model.StructOrUnion):
+            arglist.insert(0, tpresult.get_c_name(' *r', context))
+            tpresult = model.void_type
+        arglist = ', '.join(arglist) or 'void'
+        wrappername = '_cffi_f_%s' % name
+        self.export_symbols.append(wrappername)
+        if tp.abi:
+            abi = tp.abi + ' '
+        else:
+            abi = ''
+        funcdecl = ' %s%s(%s)' % (abi, wrappername, arglist)
+        context = 'result of %s' % name
+        prnt(tpresult.get_c_name(funcdecl, context))
+        prnt('{')
+        #
+        if isinstance(tp.result, model.StructOrUnion):
+            result_code = '*r = '
+        elif not isinstance(tp.result, model.VoidType):
+            result_code = 'return '
+        else:
+            result_code = ''
+        prnt('  %s%s(%s);' % (result_code, name, ', '.join(argnames)))
+        prnt('}')
+        prnt()
+
+    _loading_gen_function = _loaded_noop
+
+    def _loaded_gen_function(self, tp, name, module, library):
+        assert isinstance(tp, model.FunctionPtrType)
+        if tp.ellipsis:
+            newfunction = self._load_constant(False, tp, name, module)
+        else:
+            indirections = []
+            base_tp = tp
+            if (any(isinstance(typ, model.StructOrUnion) for typ in tp.args)
+                    or isinstance(tp.result, model.StructOrUnion)):
+                indirect_args = []
+                for i, typ in enumerate(tp.args):
+                    if isinstance(typ, model.StructOrUnion):
+                        typ = model.PointerType(typ)
+                        indirections.append((i, typ))
+                    indirect_args.append(typ)
+                indirect_result = tp.result
+                if isinstance(indirect_result, model.StructOrUnion):
+                    if indirect_result.fldtypes is None:
+                        raise TypeError("'%s' is used as result type, "
+                                        "but is opaque" % (
+                                            indirect_result._get_c_name(),))
+                    indirect_result = model.PointerType(indirect_result)
+                    indirect_args.insert(0, indirect_result)
+                    indirections.insert(0, ("result", indirect_result))
+                    indirect_result = model.void_type
+                tp = model.FunctionPtrType(tuple(indirect_args),
+                                           indirect_result, tp.ellipsis)
+            BFunc = self.ffi._get_cached_btype(tp)
+            wrappername = '_cffi_f_%s' % name
+            newfunction = module.load_function(BFunc, wrappername)
+            for i, typ in indirections:
+                newfunction = self._make_struct_wrapper(newfunction, i, typ,
+                                                        base_tp)
+        setattr(library, name, newfunction)
+        type(library)._cffi_dir.append(name)
+
+    def _make_struct_wrapper(self, oldfunc, i, tp, base_tp):
+        backend = self.ffi._backend
+        BType = self.ffi._get_cached_btype(tp)
+        if i == "result":
+            ffi = self.ffi
+            def newfunc(*args):
+                res = ffi.new(BType)
+                oldfunc(res, *args)
+                return res[0]
+        else:
+            def newfunc(*args):
+                args = args[:i] + (backend.newp(BType, args[i]),) + args[i+1:]
+                return oldfunc(*args)
+        newfunc._cffi_base_type = base_tp
+        return newfunc
+
+    # ----------
+    # named structs
+
+    def _generate_gen_struct_decl(self, tp, name):
+        assert name == tp.name
+        self._generate_struct_or_union_decl(tp, 'struct', name)
+
+    def _loading_gen_struct(self, tp, name, module):
+        self._loading_struct_or_union(tp, 'struct', name, module)
+
+    def _loaded_gen_struct(self, tp, name, module, **kwds):
+        self._loaded_struct_or_union(tp)
+
+    def _generate_gen_union_decl(self, tp, name):
+        assert name == tp.name
+        self._generate_struct_or_union_decl(tp, 'union', name)
+
+    def _loading_gen_union(self, tp, name, module):
+        self._loading_struct_or_union(tp, 'union', name, module)
+
+    def _loaded_gen_union(self, tp, name, module, **kwds):
+        self._loaded_struct_or_union(tp)
+
+    def _generate_struct_or_union_decl(self, tp, prefix, name):
+        if tp.fldnames is None:
+            return     # nothing to do with opaque structs
+        checkfuncname = '_cffi_check_%s_%s' % (prefix, name)
+        layoutfuncname = '_cffi_layout_%s_%s' % (prefix, name)
+        cname = ('%s %s' % (prefix, name)).strip()
+        #
+        prnt = self._prnt
+        prnt('static void %s(%s *p)' % (checkfuncname, cname))
+        prnt('{')
+        prnt('  /* only to generate compile-time warnings or errors */')
+        prnt('  (void)p;')
+        for fname, ftype, fbitsize, fqual in tp.enumfields():
+            if (isinstance(ftype, model.PrimitiveType)
+                and ftype.is_integer_type()) or fbitsize >= 0:
+                # accept all integers, but complain on float or double
+                prnt('  (void)((p->%s) << 1);' % fname)
+            else:
+                # only accept exactly the type declared.
+                try:
+                    prnt('  { %s = &p->%s; (void)tmp; }' % (
+                        ftype.get_c_name('*tmp', 'field %r'%fname, quals=fqual),
+                        fname))
+                except VerificationError as e:
+                    prnt('  /* %s */' % str(e))   # cannot verify it, ignore
+        prnt('}')
+        self.export_symbols.append(layoutfuncname)
+        prnt('intptr_t %s(intptr_t i)' % (layoutfuncname,))
+        prnt('{')
+        prnt('  struct _cffi_aligncheck { char x; %s y; };' % cname)
+        prnt('  static intptr_t nums[] = {')
+        prnt('    sizeof(%s),' % cname)
+        prnt('    offsetof(struct _cffi_aligncheck, y),')
+        for fname, ftype, fbitsize, fqual in tp.enumfields():
+            if fbitsize >= 0:
+                continue      # xxx ignore fbitsize for now
+            prnt('    offsetof(%s, %s),' % (cname, fname))
+            if isinstance(ftype, model.ArrayType) and ftype.length is None:
+                prnt('    0,  /* %s */' % ftype._get_c_name())
+            else:
+                prnt('    sizeof(((%s *)0)->%s),' % (cname, fname))
+        prnt('    -1')
+        prnt('  };')
+        prnt('  return nums[i];')
+        prnt('  /* the next line is not executed, but compiled */')
+        prnt('  %s(0);' % (checkfuncname,))
+        prnt('}')
+        prnt()
+
+    def _loading_struct_or_union(self, tp, prefix, name, module):
+        if tp.fldnames is None:
+            return     # nothing to do with opaque structs
+        layoutfuncname = '_cffi_layout_%s_%s' % (prefix, name)
+        #
+        BFunc = self.ffi._typeof_locked("intptr_t(*)(intptr_t)")[0]
+        function = module.load_function(BFunc, layoutfuncname)
+        layout = []
+        num = 0
+        while True:
+            x = function(num)
+            if x < 0: break
+            layout.append(x)
+            num += 1
+        if isinstance(tp, model.StructOrUnion) and tp.partial:
+            # use the function()'s sizes and offsets to guide the
+            # layout of the struct
+            totalsize = layout[0]
+            totalalignment = layout[1]
+            fieldofs = layout[2::2]
+            fieldsize = layout[3::2]
+            tp.force_flatten()
+            assert len(fieldofs) == len(fieldsize) == len(tp.fldnames)
+            tp.fixedlayout = fieldofs, fieldsize, totalsize, totalalignment
+        else:
+            cname = ('%s %s' % (prefix, name)).strip()
+            self._struct_pending_verification[tp] = layout, cname
+
+    def _loaded_struct_or_union(self, tp):
+        if tp.fldnames is None:
+            return     # nothing to do with opaque structs
+        self.ffi._get_cached_btype(tp)   # force 'fixedlayout' to be considered
+
+        if tp in self._struct_pending_verification:
+            # check that the layout sizes and offsets match the real ones
+            def check(realvalue, expectedvalue, msg):
+                if realvalue != expectedvalue:
+                    raise VerificationError(
+                        "%s (we have %d, but C compiler says %d)"
+                        % (msg, expectedvalue, realvalue))
+            ffi = self.ffi
+            BStruct = ffi._get_cached_btype(tp)
+            layout, cname = self._struct_pending_verification.pop(tp)
+            check(layout[0], ffi.sizeof(BStruct), "wrong total size")
+            check(layout[1], ffi.alignof(BStruct), "wrong total alignment")
+            i = 2
+            for fname, ftype, fbitsize, fqual in tp.enumfields():
+                if fbitsize >= 0:
+                    continue        # xxx ignore fbitsize for now
+                check(layout[i], ffi.offsetof(BStruct, fname),
+                      "wrong offset for field %r" % (fname,))
+                if layout[i+1] != 0:
+                    BField = ffi._get_cached_btype(ftype)
+                    check(layout[i+1], ffi.sizeof(BField),
+                          "wrong size for field %r" % (fname,))
+                i += 2
+            assert i == len(layout)
+
+    # ----------
+    # 'anonymous' declarations.  These are produced for anonymous structs
+    # or unions; the 'name' is obtained by a typedef.
+
+    def _generate_gen_anonymous_decl(self, tp, name):
+        if isinstance(tp, model.EnumType):
+            self._generate_gen_enum_decl(tp, name, '')
+        else:
+            self._generate_struct_or_union_decl(tp, '', name)
+
+    def _loading_gen_anonymous(self, tp, name, module):
+        if isinstance(tp, model.EnumType):
+            self._loading_gen_enum(tp, name, module, '')
+        else:
+            self._loading_struct_or_union(tp, '', name, module)
+
+    def _loaded_gen_anonymous(self, tp, name, module, **kwds):
+        if isinstance(tp, model.EnumType):
+            self._loaded_gen_enum(tp, name, module, **kwds)
+        else:
+            self._loaded_struct_or_union(tp)
+
+    # ----------
+    # constants, likely declared with '#define'
+
+    def _generate_gen_const(self, is_int, name, tp=None, category='const',
+                            check_value=None):
+        prnt = self._prnt
+        funcname = '_cffi_%s_%s' % (category, name)
+        self.export_symbols.append(funcname)
+        if check_value is not None:
+            assert is_int
+            assert category == 'const'
+            prnt('int %s(char *out_error)' % funcname)
+            prnt('{')
+            self._check_int_constant_value(name, check_value)
+            prnt('  return 0;')
+            prnt('}')
+        elif is_int:
+            assert category == 'const'
+            prnt('int %s(long long *out_value)' % funcname)
+            prnt('{')
+            prnt('  *out_value = (long long)(%s);' % (name,))
+            prnt('  return (%s) <= 0;' % (name,))
+            prnt('}')
+        else:
+            assert tp is not None
+            assert check_value is None
+            if category == 'var':
+                ampersand = '&'
+            else:
+                ampersand = ''
+            extra = ''
+            if category == 'const' and isinstance(tp, model.StructOrUnion):
+                extra = 'const *'
+                ampersand = '&'
+            prnt(tp.get_c_name(' %s%s(void)' % (extra, funcname), name))
+            prnt('{')
+            prnt('  return (%s%s);' % (ampersand, name))
+            prnt('}')
+        prnt()
+
+    def _generate_gen_constant_decl(self, tp, name):
+        is_int = isinstance(tp, model.PrimitiveType) and tp.is_integer_type()
+        self._generate_gen_const(is_int, name, tp)
+
+    _loading_gen_constant = _loaded_noop
+
+    def _load_constant(self, is_int, tp, name, module, check_value=None):
+        funcname = '_cffi_const_%s' % name
+        if check_value is not None:
+            assert is_int
+            self._load_known_int_constant(module, funcname)
+            value = check_value
+        elif is_int:
+            BType = self.ffi._typeof_locked("long long*")[0]
+            BFunc = self.ffi._typeof_locked("int(*)(long long*)")[0]
+            function = module.load_function(BFunc, funcname)
+            p = self.ffi.new(BType)
+            negative = function(p)
+            value = int(p[0])
+            if value < 0 and not negative:
+                BLongLong = self.ffi._typeof_locked("long long")[0]
+                value += (1 << (8*self.ffi.sizeof(BLongLong)))
+        else:
+            assert check_value is None
+            fntypeextra = '(*)(void)'
+            if isinstance(tp, model.StructOrUnion):
+                fntypeextra = '*' + fntypeextra
+            BFunc = self.ffi._typeof_locked(tp.get_c_name(fntypeextra, name))[0]
+            function = module.load_function(BFunc, funcname)
+            value = function()
+            if isinstance(tp, model.StructOrUnion):
+                value = value[0]
+        return value
+
+    def _loaded_gen_constant(self, tp, name, module, library):
+        is_int = isinstance(tp, model.PrimitiveType) and tp.is_integer_type()
+        value = self._load_constant(is_int, tp, name, module)
+        setattr(library, name, value)
+        type(library)._cffi_dir.append(name)
+
+    # ----------
+    # enums
+
+    def _check_int_constant_value(self, name, value):
+        prnt = self._prnt
+        if value <= 0:
+            prnt('  if ((%s) > 0 || (long)(%s) != %dL) {' % (
+                name, name, value))
+        else:
+            prnt('  if ((%s) <= 0 || (unsigned long)(%s) != %dUL) {' % (
+                name, name, value))
+        prnt('    char buf[64];')
+        prnt('    if ((%s) <= 0)' % name)
+        prnt('        sprintf(buf, "%%ld", (long)(%s));' % name)
+        prnt('    else')
+        prnt('        sprintf(buf, "%%lu", (unsigned long)(%s));' %
+             name)
+        prnt('    sprintf(out_error, "%s has the real value %s, not %s",')
+        prnt('            "%s", buf, "%d");' % (name[:100], value))
+        prnt('    return -1;')
+        prnt('  }')
+
+    def _load_known_int_constant(self, module, funcname):
+        BType = self.ffi._typeof_locked("char[]")[0]
+        BFunc = self.ffi._typeof_locked("int(*)(char*)")[0]
+        function = module.load_function(BFunc, funcname)
+        p = self.ffi.new(BType, 256)
+        if function(p) < 0:
+            error = self.ffi.string(p)
+            if sys.version_info >= (3,):
+                error = str(error, 'utf-8')
+            raise VerificationError(error)
+
+    def _enum_funcname(self, prefix, name):
+        # "$enum_$1" => "___D_enum____D_1"
+        name = name.replace('$', '___D_')
+        return '_cffi_e_%s_%s' % (prefix, name)
+
+    def _generate_gen_enum_decl(self, tp, name, prefix='enum'):
+        if tp.partial:
+            for enumerator in tp.enumerators:
+                self._generate_gen_const(True, enumerator)
+            return
+        #
+        funcname = self._enum_funcname(prefix, name)
+        self.export_symbols.append(funcname)
+        prnt = self._prnt
+        prnt('int %s(char *out_error)' % funcname)
+        prnt('{')
+        for enumerator, enumvalue in zip(tp.enumerators, tp.enumvalues):
+            self._check_int_constant_value(enumerator, enumvalue)
+        prnt('  return 0;')
+        prnt('}')
+        prnt()
+
+    def _loading_gen_enum(self, tp, name, module, prefix='enum'):
+        if tp.partial:
+            enumvalues = [self._load_constant(True, tp, enumerator, module)
+                          for enumerator in tp.enumerators]
+            tp.enumvalues = tuple(enumvalues)
+            tp.partial_resolved = True
+        else:
+            funcname = self._enum_funcname(prefix, name)
+            self._load_known_int_constant(module, funcname)
+
+    def _loaded_gen_enum(self, tp, name, module, library):
+        for enumerator, enumvalue in zip(tp.enumerators, tp.enumvalues):
+            setattr(library, enumerator, enumvalue)
+            type(library)._cffi_dir.append(enumerator)
+
+    # ----------
+    # macros: for now only for integers
+
+    def _generate_gen_macro_decl(self, tp, name):
+        if tp == '...':
+            check_value = None
+        else:
+            check_value = tp     # an integer
+        self._generate_gen_const(True, name, check_value=check_value)
+
+    _loading_gen_macro = _loaded_noop
+
+    def _loaded_gen_macro(self, tp, name, module, library):
+        if tp == '...':
+            check_value = None
+        else:
+            check_value = tp     # an integer
+        value = self._load_constant(True, tp, name, module,
+                                    check_value=check_value)
+        setattr(library, name, value)
+        type(library)._cffi_dir.append(name)
+
+    # ----------
+    # global variables
+
+    def _generate_gen_variable_decl(self, tp, name):
+        if isinstance(tp, model.ArrayType):
+            if tp.length_is_unknown():
+                prnt = self._prnt
+                funcname = '_cffi_sizeof_%s' % (name,)
+                self.export_symbols.append(funcname)
+                prnt("size_t %s(void)" % funcname)
+                prnt("{")
+                prnt("  return sizeof(%s);" % (name,))
+                prnt("}")
+            tp_ptr = model.PointerType(tp.item)
+            self._generate_gen_const(False, name, tp_ptr)
+        else:
+            tp_ptr = model.PointerType(tp)
+            self._generate_gen_const(False, name, tp_ptr, category='var')
+
+    _loading_gen_variable = _loaded_noop
+
+    def _loaded_gen_variable(self, tp, name, module, library):
+        if isinstance(tp, model.ArrayType):   # int a[5] is "constant" in the
+                                              # sense that "a=..." is forbidden
+            if tp.length_is_unknown():
+                funcname = '_cffi_sizeof_%s' % (name,)
+                BFunc = self.ffi._typeof_locked('size_t(*)(void)')[0]
+                function = module.load_function(BFunc, funcname)
+                size = function()
+                BItemType = self.ffi._get_cached_btype(tp.item)
+                length, rest = divmod(size, self.ffi.sizeof(BItemType))
+                if rest != 0:
+                    raise VerificationError(
+                        "bad size: %r does not seem to be an array of %s" %
+                        (name, tp.item))
+                tp = tp.resolve_length(length)
+            tp_ptr = model.PointerType(tp.item)
+            value = self._load_constant(False, tp_ptr, name, module)
+            # 'value' is a <cdata 'type *'> which we have to replace with
+            # a <cdata 'type[N]'> if the N is actually known
+            if tp.length is not None:
+                BArray = self.ffi._get_cached_btype(tp)
+                value = self.ffi.cast(BArray, value)
+            setattr(library, name, value)
+            type(library)._cffi_dir.append(name)
+            return
+        # remove ptr=<cdata 'int *'> from the library instance, and replace
+        # it by a property on the class, which reads/writes into ptr[0].
+        funcname = '_cffi_var_%s' % name
+        BFunc = self.ffi._typeof_locked(tp.get_c_name('*(*)(void)', name))[0]
+        function = module.load_function(BFunc, funcname)
+        ptr = function()
+        def getter(library):
+            return ptr[0]
+        def setter(library, value):
+            ptr[0] = value
+        setattr(type(library), name, property(getter, setter))
+        type(library)._cffi_dir.append(name)
+
+cffimod_header = r'''
+#include <stdio.h>
+#include <stddef.h>
+#include <stdarg.h>
+#include <errno.h>
+#include <sys/types.h>   /* XXX for ssize_t on some platforms */
+
+/* this block of #ifs should be kept exactly identical between
+   c/_cffi_backend.c, cffi/vengine_cpy.py, cffi/vengine_gen.py
+   and cffi/_cffi_include.h */
+#if defined(_MSC_VER)
+# include <malloc.h>   /* for alloca() */
+# if _MSC_VER < 1600   /* MSVC < 2010 */
+   typedef __int8 int8_t;
+   typedef __int16 int16_t;
+   typedef __int32 int32_t;
+   typedef __int64 int64_t;
+   typedef unsigned __int8 uint8_t;
+   typedef unsigned __int16 uint16_t;
+   typedef unsigned __int32 uint32_t;
+   typedef unsigned __int64 uint64_t;
+   typedef __int8 int_least8_t;
+   typedef __int16 int_least16_t;
+   typedef __int32 int_least32_t;
+   typedef __int64 int_least64_t;
+   typedef unsigned __int8 uint_least8_t;
+   typedef unsigned __int16 uint_least16_t;
+   typedef unsigned __int32 uint_least32_t;
+   typedef unsigned __int64 uint_least64_t;
+   typedef __int8 int_fast8_t;
+   typedef __int16 int_fast16_t;
+   typedef __int32 int_fast32_t;
+   typedef __int64 int_fast64_t;
+   typedef unsigned __int8 uint_fast8_t;
+   typedef unsigned __int16 uint_fast16_t;
+   typedef unsigned __int32 uint_fast32_t;
+   typedef unsigned __int64 uint_fast64_t;
+   typedef __int64 intmax_t;
+   typedef unsigned __int64 uintmax_t;
+# else
+#  include <stdint.h>
+# endif
+# if _MSC_VER < 1800   /* MSVC < 2013 */
+#  ifndef __cplusplus
+    typedef unsigned char _Bool;
+#  endif
+# endif
+# define _cffi_float_complex_t   _Fcomplex    /* include <complex.h> for it */
+# define _cffi_double_complex_t  _Dcomplex    /* include <complex.h> for it */
+#else
+# include <stdint.h>
+# if (defined (__SVR4) && defined (__sun)) || defined(_AIX) || defined(__hpux)
+#  include <alloca.h>
+# endif
+# define _cffi_float_complex_t   float _Complex
+# define _cffi_double_complex_t  double _Complex
+#endif
+'''

.venv/lib/python3.9/site-packages/cffi/verifier.py

@@ -0,0 +1,306 @@
+#
+# DEPRECATED: implementation for ffi.verify()
+#
+import sys, os, binascii, shutil, io
+from . import __version_verifier_modules__
+from . import ffiplatform
+from .error import VerificationError
+
+if sys.version_info >= (3, 3):
+    import importlib.machinery
+    def _extension_suffixes():
+        return importlib.machinery.EXTENSION_SUFFIXES[:]
+else:
+    import imp
+    def _extension_suffixes():
+        return [suffix for suffix, _, type in imp.get_suffixes()
+                if type == imp.C_EXTENSION]
+
+
+if sys.version_info >= (3,):
+    NativeIO = io.StringIO
+else:
+    class NativeIO(io.BytesIO):
+        def write(self, s):
+            if isinstance(s, unicode):
+                s = s.encode('ascii')
+            super(NativeIO, self).write(s)
+
+
+class Verifier(object):
+
+    def __init__(self, ffi, preamble, tmpdir=None, modulename=None,
+                 ext_package=None, tag='', force_generic_engine=False,
+                 source_extension='.c', flags=None, relative_to=None, **kwds):
+        if ffi._parser._uses_new_feature:
+            raise VerificationError(
+                "feature not supported with ffi.verify(), but only "
+                "with ffi.set_source(): %s" % (ffi._parser._uses_new_feature,))
+        self.ffi = ffi
+        self.preamble = preamble
+        if not modulename:
+            flattened_kwds = ffiplatform.flatten(kwds)
+        vengine_class = _locate_engine_class(ffi, force_generic_engine)
+        self._vengine = vengine_class(self)
+        self._vengine.patch_extension_kwds(kwds)
+        self.flags = flags
+        self.kwds = self.make_relative_to(kwds, relative_to)
+        #
+        if modulename:
+            if tag:
+                raise TypeError("can't specify both 'modulename' and 'tag'")
+        else:
+            key = '\x00'.join(['%d.%d' % sys.version_info[:2],
+                               __version_verifier_modules__,
+                               preamble, flattened_kwds] +
+                              ffi._cdefsources)
+            if sys.version_info >= (3,):
+                key = key.encode('utf-8')
+            k1 = hex(binascii.crc32(key[0::2]) & 0xffffffff)
+            k1 = k1.lstrip('0x').rstrip('L')
+            k2 = hex(binascii.crc32(key[1::2]) & 0xffffffff)
+            k2 = k2.lstrip('0').rstrip('L')
+            modulename = '_cffi_%s_%s%s%s' % (tag, self._vengine._class_key,
+                                              k1, k2)
+        suffix = _get_so_suffixes()[0]
+        self.tmpdir = tmpdir or _caller_dir_pycache()
+        self.sourcefilename = os.path.join(self.tmpdir, modulename + source_extension)
+        self.modulefilename = os.path.join(self.tmpdir, modulename + suffix)
+        self.ext_package = ext_package
+        self._has_source = False
+        self._has_module = False
+
+    def write_source(self, file=None):
+        """Write the C source code.  It is produced in 'self.sourcefilename',
+        which can be tweaked beforehand."""
+        with self.ffi._lock:
+            if self._has_source and file is None:
+                raise VerificationError(
+                    "source code already written")
+            self._write_source(file)
+
+    def compile_module(self):
+        """Write the C source code (if not done already) and compile it.
+        This produces a dynamic link library in 'self.modulefilename'."""
+        with self.ffi._lock:
+            if self._has_module:
+                raise VerificationError("module already compiled")
+            if not self._has_source:
+                self._write_source()
+            self._compile_module()
+
+    def load_library(self):
+        """Get a C module from this Verifier instance.
+        Returns an instance of a FFILibrary class that behaves like the
+        objects returned by ffi.dlopen(), but that delegates all
+        operations to the C module.  If necessary, the C code is written
+        and compiled first.
+        """
+        with self.ffi._lock:
+            if not self._has_module:
+                self._locate_module()
+                if not self._has_module:
+                    if not self._has_source:
+                        self._write_source()
+                    self._compile_module()
+            return self._load_library()
+
+    def get_module_name(self):
+        basename = os.path.basename(self.modulefilename)
+        # kill both the .so extension and the other .'s, as introduced
+        # by Python 3: 'basename.cpython-33m.so'
+        basename = basename.split('.', 1)[0]
+        # and the _d added in Python 2 debug builds --- but try to be
+        # conservative and not kill a legitimate _d
+        if basename.endswith('_d') and hasattr(sys, 'gettotalrefcount'):
+            basename = basename[:-2]
+        return basename
+
+    def get_extension(self):
+        if not self._has_source:
+            with self.ffi._lock:
+                if not self._has_source:
+                    self._write_source()
+        sourcename = ffiplatform.maybe_relative_path(self.sourcefilename)
+        modname = self.get_module_name()
+        return ffiplatform.get_extension(sourcename, modname, **self.kwds)
+
+    def generates_python_module(self):
+        return self._vengine._gen_python_module
+
+    def make_relative_to(self, kwds, relative_to):
+        if relative_to and os.path.dirname(relative_to):
+            dirname = os.path.dirname(relative_to)
+            kwds = kwds.copy()
+            for key in ffiplatform.LIST_OF_FILE_NAMES:
+                if key in kwds:
+                    lst = kwds[key]
+                    if not isinstance(lst, (list, tuple)):
+                        raise TypeError("keyword '%s' should be a list or tuple"
+                                        % (key,))
+                    lst = [os.path.join(dirname, fn) for fn in lst]
+                    kwds[key] = lst
+        return kwds
+
+    # ----------
+
+    def _locate_module(self):
+        if not os.path.isfile(self.modulefilename):
+            if self.ext_package:
+                try:
+                    pkg = __import__(self.ext_package, None, None, ['__doc__'])
+                except ImportError:
+                    return      # cannot import the package itself, give up
+                    # (e.g. it might be called differently before installation)
+                path = pkg.__path__
+            else:
+                path = None
+            filename = self._vengine.find_module(self.get_module_name(), path,
+                                                 _get_so_suffixes())
+            if filename is None:
+                return
+            self.modulefilename = filename
+        self._vengine.collect_types()
+        self._has_module = True
+
+    def _write_source_to(self, file):
+        self._vengine._f = file
+        try:
+            self._vengine.write_source_to_f()
+        finally:
+            del self._vengine._f
+
+    def _write_source(self, file=None):
+        if file is not None:
+            self._write_source_to(file)
+        else:
+            # Write our source file to an in memory file.
+            f = NativeIO()
+            self._write_source_to(f)
+            source_data = f.getvalue()
+
+            # Determine if this matches the current file
+            if os.path.exists(self.sourcefilename):
+                with open(self.sourcefilename, "r") as fp:
+                    needs_written = not (fp.read() == source_data)
+            else:
+                needs_written = True
+
+            # Actually write the file out if it doesn't match
+            if needs_written:
+                _ensure_dir(self.sourcefilename)
+                with open(self.sourcefilename, "w") as fp:
+                    fp.write(source_data)
+
+            # Set this flag
+            self._has_source = True
+
+    def _compile_module(self):
+        # compile this C source
+        tmpdir = os.path.dirname(self.sourcefilename)
+        outputfilename = ffiplatform.compile(tmpdir, self.get_extension())
+        try:
+            same = ffiplatform.samefile(outputfilename, self.modulefilename)
+        except OSError:
+            same = False
+        if not same:
+            _ensure_dir(self.modulefilename)
+            shutil.move(outputfilename, self.modulefilename)
+        self._has_module = True
+
+    def _load_library(self):
+        assert self._has_module
+        if self.flags is not None:
+            return self._vengine.load_library(self.flags)
+        else:
+            return self._vengine.load_library()
+
+# ____________________________________________________________
+
+_FORCE_GENERIC_ENGINE = False      # for tests
+
+def _locate_engine_class(ffi, force_generic_engine):
+    if _FORCE_GENERIC_ENGINE:
+        force_generic_engine = True
+    if not force_generic_engine:
+        if '__pypy__' in sys.builtin_module_names:
+            force_generic_engine = True
+        else:
+            try:
+                import _cffi_backend
+            except ImportError:
+                _cffi_backend = '?'
+            if ffi._backend is not _cffi_backend:
+                force_generic_engine = True
+    if force_generic_engine:
+        from . import vengine_gen
+        return vengine_gen.VGenericEngine
+    else:
+        from . import vengine_cpy
+        return vengine_cpy.VCPythonEngine
+
+# ____________________________________________________________
+
+_TMPDIR = None
+
+def _caller_dir_pycache():
+    if _TMPDIR:
+        return _TMPDIR
+    result = os.environ.get('CFFI_TMPDIR')
+    if result:
+        return result
+    filename = sys._getframe(2).f_code.co_filename
+    return os.path.abspath(os.path.join(os.path.dirname(filename),
+                           '__pycache__'))
+
+def set_tmpdir(dirname):
+    """Set the temporary directory to use instead of __pycache__."""
+    global _TMPDIR
+    _TMPDIR = dirname
+
+def cleanup_tmpdir(tmpdir=None, keep_so=False):
+    """Clean up the temporary directory by removing all files in it
+    called `_cffi_*.{c,so}` as well as the `build` subdirectory."""
+    tmpdir = tmpdir or _caller_dir_pycache()
+    try:
+        filelist = os.listdir(tmpdir)
+    except OSError:
+        return
+    if keep_so:
+        suffix = '.c'   # only remove .c files
+    else:
+        suffix = _get_so_suffixes()[0].lower()
+    for fn in filelist:
+        if fn.lower().startswith('_cffi_') and (
+                fn.lower().endswith(suffix) or fn.lower().endswith('.c')):
+            try:
+                os.unlink(os.path.join(tmpdir, fn))
+            except OSError:
+                pass
+    clean_dir = [os.path.join(tmpdir, 'build')]
+    for dir in clean_dir:
+        try:
+            for fn in os.listdir(dir):
+                fn = os.path.join(dir, fn)
+                if os.path.isdir(fn):
+                    clean_dir.append(fn)
+                else:
+                    os.unlink(fn)
+        except OSError:
+            pass
+
+def _get_so_suffixes():
+    suffixes = _extension_suffixes()
+    if not suffixes:
+        # bah, no C_EXTENSION available.  Occurs on pypy without cpyext
+        if sys.platform == 'win32':
+            suffixes = [".pyd"]
+        else:
+            suffixes = [".so"]
+
+    return suffixes
+
+def _ensure_dir(filename):
+    dirname = os.path.dirname(filename)
+    if dirname and not os.path.isdir(dirname):
+        os.makedirs(dirname)

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/INSTALLER

@@ -0,0 +1 @@
+pip

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/METADATA

@@ -0,0 +1,808 @@
+Metadata-Version: 2.4
+Name: charset-normalizer
+Version: 3.4.7
+Summary: The Real First Universal Charset Detector. Open, modern and actively maintained alternative to Chardet.
+Author-email: "Ahmed R. TAHRI" <tahri.ahmed@proton.me>
+Maintainer-email: "Ahmed R. TAHRI" <tahri.ahmed@proton.me>
+License: MIT
+Project-URL: Changelog, https://github.com/jawah/charset_normalizer/blob/master/CHANGELOG.md
+Project-URL: Documentation, https://charset-normalizer.readthedocs.io/
+Project-URL: Code, https://github.com/jawah/charset_normalizer
+Project-URL: Issue tracker, https://github.com/jawah/charset_normalizer/issues
+Keywords: encoding,charset,charset-detector,detector,normalization,unicode,chardet,detect
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.7
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: Implementation :: CPython
+Classifier: Programming Language :: Python :: Implementation :: PyPy
+Classifier: Programming Language :: Python :: Free Threading :: 4 - Resilient
+Classifier: Topic :: Text Processing :: Linguistic
+Classifier: Topic :: Utilities
+Classifier: Typing :: Typed
+Requires-Python: >=3.7
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: unicode-backport
+Dynamic: license-file
+
+<h1 align="center">Charset Detection, for Everyone 👋</h1>
+
+<p align="center">
+  <sup>The Real First Universal Charset Detector</sup><br>
+  <a href="https://pypi.org/project/charset-normalizer">
+    <img src="https://img.shields.io/pypi/pyversions/charset_normalizer.svg?orange=blue" />
+  </a>
+  <a href="https://pepy.tech/project/charset-normalizer/">
+    <img alt="Download Count Total" src="https://static.pepy.tech/badge/charset-normalizer/month" />
+  </a>
+  <a href="https://bestpractices.coreinfrastructure.org/projects/7297">
+    <img src="https://bestpractices.coreinfrastructure.org/projects/7297/badge">
+  </a>
+</p>
+<p align="center">
+  <sup><i>Featured Packages</i></sup><br>
+  <a href="https://github.com/jawah/niquests">
+   <img alt="Static Badge" src="https://img.shields.io/badge/Niquests-Most_Advanced_HTTP_Client-cyan">
+  </a>
+  <a href="https://github.com/jawah/wassima">
+   <img alt="Static Badge" src="https://img.shields.io/badge/Wassima-Certifi_Replacement-cyan">
+  </a>
+</p>
+<p align="center">
+  <sup><i>In other language (unofficial port - by the community)</i></sup><br>
+  <a href="https://github.com/nickspring/charset-normalizer-rs">
+   <img alt="Static Badge" src="https://img.shields.io/badge/Rust-red">
+  </a>
+</p>
+
+> A library that helps you read text from an unknown charset encoding.<br /> Motivated by `chardet`,
+> I'm trying to resolve the issue by taking a new approach.
+> All IANA character set names for which the Python core library provides codecs are supported.
+> You can also register your own set of codecs, and yes, it would work as-is.
+
+<p align="center">
+  >>>>> <a href="https://charsetnormalizerweb.ousret.now.sh" target="_blank">👉 Try Me Online Now, Then Adopt Me 👈 </a> <<<<<
+</p>
+
+This project offers you an alternative to **Universal Charset Encoding Detector**, also known as **Chardet**.
+
+| Feature                                          | [Chardet](https://github.com/chardet/chardet) |                                       Charset Normalizer                                        | [cChardet](https://github.com/PyYoshi/cChardet) |
+|--------------------------------------------------|:---------------------------------------------:|:-----------------------------------------------------------------------------------------------:|:-----------------------------------------------:|
+| `Fast`                                           |                       ✅                       |                                                ✅                                                |                        ✅                        |
+| `Universal`[^1]                                  |                       ❌                       |                                                ✅                                                |                        ❌                        |
+| `Reliable` **without** distinguishable standards |                       ✅                       |                                                ✅                                                |                        ✅                        |
+| `Reliable` **with** distinguishable standards    |                       ✅                       |                                                ✅                                                |                        ✅                        |
+| `License`                                        |        _Disputed_[^2]<br>_restrictive_        |                                               MIT                                               |            MPL-1.1<br>_restrictive_             |
+| `Native Python`                                  |                       ✅                       |                                                ✅                                                |                        ❌                        |
+| `Detect spoken language`                         |                       ✅                       |                                                ✅                                                |                       N/A                       |
+| `UnicodeDecodeError Safety`                      |                       ✅                       |                                                ✅                                                |                        ❌                        |
+| `Whl Size (min)`                                 |                    500 kB                     |                                             150 kB                                              |                     ~200 kB                     |
+| `Supported Encoding`                             |                      99                       | [99](https://charset-normalizer.readthedocs.io/en/latest/user/support.html#supported-encodings) |                       40                        |
+| `Can register custom encoding`                   |                       ❌                       |                                                ✅                                                |                        ❌                        |
+
+<p align="center">
+<img src="https://i.imgflip.com/373iay.gif" alt="Reading Normalized Text" width="226"/><img src="https://media.tenor.com/images/c0180f70732a18b4965448d33adba3d0/tenor.gif" alt="Cat Reading Text" width="200"/>
+</p>
+
+[^1]: They are clearly using specific code for a specific encoding even if covering most of used one.
+[^2]: Chardet 7.0+ was relicensed from LGPL-2.1 to MIT following an AI-assisted rewrite. This relicensing is disputed on two independent grounds: **(a)** the original author [contests](https://github.com/chardet/chardet/issues/327) that the maintainer had the right to relicense, arguing the rewrite is a derivative work of the LGPL-licensed codebase since it was not a clean room implementation; **(b)** the copyright claim itself is [questionable](https://github.com/chardet/chardet/issues/334) given the code was primarily generated by an LLM, and AI-generated output may not be copyrightable under most jurisdictions. Either issue alone could undermine the MIT license. Beyond licensing, the rewrite raises questions about responsible use of AI in open source: key architectural ideas pioneered by charset-normalizer - notably decode-first validity filtering (our foundational approach since v1) and encoding pairwise similarity with the same algorithm and threshold — surfaced in chardet 7 without acknowledgment. The project also imported test files from charset-normalizer to train and benchmark against it, then claimed superior accuracy on those very files. Charset-normalizer has always been MIT-licensed, encoding-agnostic by design, and built on a verifiable human-authored history.
+
+## ⚡ Performance
+
+This package offer better performances (99th, and 95th) against Chardet. Here are some numbers.
+
+| Package                                           | Accuracy | Mean per file (ms) | File per sec (est) |
+|---------------------------------------------------|:--------:|:------------------:|:------------------:|
+| [chardet 7.1](https://github.com/chardet/chardet) |   89 %   |        3 ms        |    333 file/sec    |
+| charset-normalizer                                | **97 %** |        3 ms        |    333 file/sec    |
+
+| Package                                           | 99th percentile | 95th percentile | 50th percentile |
+|---------------------------------------------------|:---------------:|:---------------:|:---------------:|
+| [chardet 7.1](https://github.com/chardet/chardet) |      32 ms      |      17 ms      |     < 1 ms      |
+| charset-normalizer                                |      16 ms      |      10 ms      |      1 ms       |
+
+_updated as of March 2026 using CPython 3.12, Charset-Normalizer 3.4.6, and Chardet 7.1.0_
+
+~Chardet's performance on larger file (1MB+) are very poor. Expect huge difference on large payload.~ No longer the case since Chardet 7.0+
+
+> Stats are generated using 400+ files using default parameters. More details on used files, see GHA workflows.
+> And yes, these results might change at any time. The dataset can be updated to include more files.
+> The actual delays heavily depends on your CPU capabilities. The factors should remain the same.
+> Chardet claims on his documentation to have a greater accuracy than us based on the dataset they trained Chardet on(...)
+> Well, it's normal, the opposite would have been worrying. Whereas charset-normalizer don't train on anything, our solution
+> is based on a completely different algorithm, still heuristic through, it does not need weights across every encoding tables.
+
+## ✨ Installation
+
+Using pip:
+
+```sh
+pip install charset-normalizer -U
+```
+
+## 🚀 Basic Usage
+
+### CLI
+This package comes with a CLI.
+
+```
+usage: normalizer [-h] [-v] [-a] [-n] [-m] [-r] [-f] [-t THRESHOLD]
+                  file [file ...]
+
+The Real First Universal Charset Detector. Discover originating encoding used
+on text file. Normalize text to unicode.
+
+positional arguments:
+  files                 File(s) to be analysed
+
+optional arguments:
+  -h, --help            show this help message and exit
+  -v, --verbose         Display complementary information about file if any.
+                        Stdout will contain logs about the detection process.
+  -a, --with-alternative
+                        Output complementary possibilities if any. Top-level
+                        JSON WILL be a list.
+  -n, --normalize       Permit to normalize input file. If not set, program
+                        does not write anything.
+  -m, --minimal         Only output the charset detected to STDOUT. Disabling
+                        JSON output.
+  -r, --replace         Replace file when trying to normalize it instead of
+                        creating a new one.
+  -f, --force           Replace file without asking if you are sure, use this
+                        flag with caution.
+  -t THRESHOLD, --threshold THRESHOLD
+                        Define a custom maximum amount of chaos allowed in
+                        decoded content. 0. <= chaos <= 1.
+  --version             Show version information and exit.
+```
+
+```bash
+normalizer ./data/sample.1.fr.srt
+```
+
+or
+
+```bash
+python -m charset_normalizer ./data/sample.1.fr.srt
+```
+
+🎉 Since version 1.4.0 the CLI produce easily usable stdout result in JSON format.
+
+```json
+{
+    "path": "/home/default/projects/charset_normalizer/data/sample.1.fr.srt",
+    "encoding": "cp1252",
+    "encoding_aliases": [
+        "1252",
+        "windows_1252"
+    ],
+    "alternative_encodings": [
+        "cp1254",
+        "cp1256",
+        "cp1258",
+        "iso8859_14",
+        "iso8859_15",
+        "iso8859_16",
+        "iso8859_3",
+        "iso8859_9",
+        "latin_1",
+        "mbcs"
+    ],
+    "language": "French",
+    "alphabets": [
+        "Basic Latin",
+        "Latin-1 Supplement"
+    ],
+    "has_sig_or_bom": false,
+    "chaos": 0.149,
+    "coherence": 97.152,
+    "unicode_path": null,
+    "is_preferred": true
+}
+```
+
+### Python
+*Just print out normalized text*
+```python
+from charset_normalizer import from_path
+
+results = from_path('./my_subtitle.srt')
+
+print(str(results.best()))
+```
+
+*Upgrade your code without effort*
+```python
+from charset_normalizer import detect
+```
+
+The above code will behave the same as **chardet**. We ensure that we offer the best (reasonable) BC result possible.
+
+See the docs for advanced usage : [readthedocs.io](https://charset-normalizer.readthedocs.io/en/latest/)
+
+## 😇 Why
+
+When I started using Chardet, I noticed that it was not suited to my expectations, and I wanted to propose a
+reliable alternative using a completely different method. Also! I never back down on a good challenge!
+
+I **don't care** about the **originating charset** encoding, because **two different tables** can
+produce **two identical rendered string.**
+What I want is to get readable text, the best I can.
+
+In a way, **I'm brute forcing text decoding.** How cool is that ? 😎
+
+Don't confuse package **ftfy** with charset-normalizer or chardet. ftfy goal is to repair Unicode string whereas charset-normalizer to convert raw file in unknown encoding to unicode.
+
+## 🍰 How
+
+  - Discard all charset encoding table that could not fit the binary content.
+  - Measure noise, or the mess once opened (by chunks) with a corresponding charset encoding.
+  - Extract matches with the lowest mess detected.
+  - Additionally, we measure coherence / probe for a language.
+
+**Wait a minute**, what is noise/mess and coherence according to **YOU ?**
+
+*Noise :* I opened hundred of text files, **written by humans**, with the wrong encoding table. **I observed**, then
+**I established** some ground rules about **what is obvious** when **it seems like** a mess (aka. defining noise in rendered text).
+ I know that my interpretation of what is noise is probably incomplete, feel free to contribute in order to
+ improve or rewrite it.
+
+*Coherence :* For each language there is on earth, we have computed ranked letter appearance occurrences (the best we can). So I thought
+that intel is worth something here. So I use those records against decoded text to check if I can detect intelligent design.
+
+## ⚡ Known limitations
+
+  - Language detection is unreliable when text contains two or more languages sharing identical letters. (eg. HTML (english tags) + Turkish content (Sharing Latin characters))
+  - Every charset detector heavily depends on sufficient content. In common cases, do not bother run detection on very tiny content.
+
+## ⚠️ About Python EOLs
+
+**If you are running:**
+
+- Python >=2.7,<3.5: Unsupported
+- Python 3.5: charset-normalizer < 2.1
+- Python 3.6: charset-normalizer < 3.1
+
+Upgrade your Python interpreter as soon as possible.
+
+## 👤 Contributing
+
+Contributions, issues and feature requests are very much welcome.<br />
+Feel free to check [issues page](https://github.com/ousret/charset_normalizer/issues) if you want to contribute.
+
+## 📝 License
+
+Copyright © [Ahmed TAHRI @Ousret](https://github.com/Ousret).<br />
+This project is [MIT](https://github.com/Ousret/charset_normalizer/blob/master/LICENSE) licensed.
+
+Characters frequencies used in this project © 2012 [Denny Vrandečić](http://simia.net/letters/)
+
+## 💼 For Enterprise
+
+Professional support for charset-normalizer is available as part of the [Tidelift
+Subscription][1]. Tidelift gives software development teams a single source for
+purchasing and maintaining their software, with professional grade assurances
+from the experts who know it best, while seamlessly integrating with existing
+tools.
+
+[1]: https://tidelift.com/subscription/pkg/pypi-charset-normalizer?utm_source=pypi-charset-normalizer&utm_medium=readme
+
+[![OpenSSF Best Practices](https://www.bestpractices.dev/projects/7297/badge)](https://www.bestpractices.dev/projects/7297)
+
+# Changelog
+All notable changes to charset-normalizer will be documented in this file. This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
+
+## [3.4.7](https://github.com/Ousret/charset_normalizer/compare/3.4.6...3.4.7) (2026-04-02)
+
+### Changed
+- Pre-built optimized version using mypy[c] v1.20.
+- Relax `setuptools` constraint to `setuptools>=68,<82.1`.
+
+### Fixed
+- Correctly remove SIG remnant in utf-7 decoded string. (#718) (#716)
+
+## [3.4.6](https://github.com/Ousret/charset_normalizer/compare/3.4.5...3.4.6) (2026-03-15)
+
+### Changed
+- Flattened the logic in `charset_normalizer.md` for higher performance. Removed `eligible(..)` and `feed(...)`
+  in favor of `feed_info(...)`.
+- Raised upper bound for mypy[c] to 1.20, for our optimized version.
+- Updated `UNICODE_RANGES_COMBINED` using Unicode blocks v17.
+
+### Fixed
+- Edge case where noise difference between two candidates can be almost insignificant. (#672)
+- CLI `--normalize` writing to wrong path when passing multiple files in. (#702)
+
+### Misc
+- Freethreaded pre-built wheels now shipped in PyPI starting with 3.14t. (#616)
+
+## [3.4.5](https://github.com/Ousret/charset_normalizer/compare/3.4.4...3.4.5) (2026-03-06)
+
+### Changed
+- Update `setuptools` constraint to `setuptools>=68,<=82`.
+- Raised upper bound of mypyc for the optional pre-built extension to v1.19.1
+
+### Fixed
+- Add explicit link to lib math in our optimized build. (#692)
+- Logger level not restored correctly for empty byte sequences. (#701)
+- TypeError when passing bytearray to from_bytes. (#703)
+
+### Misc
+- Applied safe micro-optimizations in both our noise detector and language detector.
+- Rewrote the `query_yes_no` function (inside CLI) to avoid using ambiguous licensed code.
+- Added `cd.py` submodule into mypyc optional compilation to reduce further the performance impact.
+
+## [3.4.4](https://github.com/Ousret/charset_normalizer/compare/3.4.2...3.4.4) (2025-10-13)
+
+### Changed
+- Bound `setuptools` to a specific constraint `setuptools>=68,<=81`.
+- Raised upper bound of mypyc for the optional pre-built extension to v1.18.2
+
+### Removed
+- `setuptools-scm` as a build dependency.
+
+### Misc
+- Enforced hashes in `dev-requirements.txt` and created `ci-requirements.txt` for security purposes.
+- Additional pre-built wheels for riscv64, s390x, and armv7l architectures.
+- Restore ` multiple.intoto.jsonl` in GitHub releases in addition to individual attestation file per wheel.
+
+## [3.4.3](https://github.com/Ousret/charset_normalizer/compare/3.4.2...3.4.3) (2025-08-09)
+
+### Changed
+- mypy(c) is no longer a required dependency at build time if `CHARSET_NORMALIZER_USE_MYPYC` isn't set to `1`. (#595) (#583)
+- automatically lower confidence on small bytes samples that are not Unicode in `detect` output legacy function. (#391)
+
+### Added
+- Custom build backend to overcome inability to mark mypy as an optional dependency in the build phase.
+- Support for Python 3.14
+
+### Fixed
+- sdist archive contained useless directories.
+- automatically fallback on valid UTF-16 or UTF-32 even if the md says it's noisy. (#633)
+
+### Misc
+- SBOM are automatically published to the relevant GitHub release to comply with regulatory changes.
+  Each published wheel comes with its SBOM. We choose CycloneDX as the format.
+- Prebuilt optimized wheel are no longer distributed by default for CPython 3.7 due to a change in cibuildwheel.
+
+## [3.4.2](https://github.com/Ousret/charset_normalizer/compare/3.4.1...3.4.2) (2025-05-02)
+
+### Fixed
+- Addressed the DeprecationWarning in our CLI regarding `argparse.FileType` by backporting the target class into the package. (#591)
+- Improved the overall reliability of the detector with CJK Ideographs. (#605) (#587)
+
+### Changed
+- Optional mypyc compilation upgraded to version 1.15 for Python >= 3.8
+
+## [3.4.1](https://github.com/Ousret/charset_normalizer/compare/3.4.0...3.4.1) (2024-12-24)
+
+### Changed
+- Project metadata are now stored using `pyproject.toml` instead of `setup.cfg` using setuptools as the build backend.
+- Enforce annotation delayed loading for a simpler and consistent types in the project.
+- Optional mypyc compilation upgraded to version 1.14 for Python >= 3.8
+
+### Added
+- pre-commit configuration.
+- noxfile.
+
+### Removed
+- `build-requirements.txt` as per using `pyproject.toml` native build configuration.
+- `bin/integration.py` and `bin/serve.py` in favor of downstream integration test (see noxfile).
+- `setup.cfg` in favor of `pyproject.toml` metadata configuration.
+- Unused `utils.range_scan` function.
+
+### Fixed
+- Converting content to Unicode bytes may insert `utf_8` instead of preferred `utf-8`. (#572)
+- Deprecation warning "'count' is passed as positional argument" when converting to Unicode bytes on Python 3.13+
+
+## [3.4.0](https://github.com/Ousret/charset_normalizer/compare/3.3.2...3.4.0) (2024-10-08)
+
+### Added
+- Argument `--no-preemptive` in the CLI to prevent the detector to search for hints.
+- Support for Python 3.13 (#512)
+
+### Fixed
+- Relax the TypeError exception thrown when trying to compare a CharsetMatch with anything else than a CharsetMatch.
+- Improved the general reliability of the detector based on user feedbacks. (#520) (#509) (#498) (#407) (#537)
+- Declared charset in content (preemptive detection) not changed when converting to utf-8 bytes. (#381)
+
+## [3.3.2](https://github.com/Ousret/charset_normalizer/compare/3.3.1...3.3.2) (2023-10-31)
+
+### Fixed
+- Unintentional memory usage regression when using large payload that match several encoding (#376)
+- Regression on some detection case showcased in the documentation (#371)
+
+### Added
+- Noise (md) probe that identify malformed arabic representation due to the presence of letters in isolated form (credit to my wife)
+
+## [3.3.1](https://github.com/Ousret/charset_normalizer/compare/3.3.0...3.3.1) (2023-10-22)
+
+### Changed
+- Optional mypyc compilation upgraded to version 1.6.1 for Python >= 3.8
+- Improved the general detection reliability based on reports from the community
+
+## [3.3.0](https://github.com/Ousret/charset_normalizer/compare/3.2.0...3.3.0) (2023-09-30)
+
+### Added
+- Allow to execute the CLI (e.g. normalizer) through `python -m charset_normalizer.cli` or `python -m charset_normalizer`
+- Support for 9 forgotten encoding that are supported by Python but unlisted in `encoding.aliases` as they have no alias (#323)
+
+### Removed
+- (internal) Redundant utils.is_ascii function and unused function is_private_use_only
+- (internal) charset_normalizer.assets is moved inside charset_normalizer.constant
+
+### Changed
+- (internal) Unicode code blocks in constants are updated using the latest v15.0.0 definition to improve detection
+- Optional mypyc compilation upgraded to version 1.5.1 for Python >= 3.8
+
+### Fixed
+- Unable to properly sort CharsetMatch when both chaos/noise and coherence were close due to an unreachable condition in \_\_lt\_\_ (#350)
+
+## [3.2.0](https://github.com/Ousret/charset_normalizer/compare/3.1.0...3.2.0) (2023-06-07)
+
+### Changed
+- Typehint for function `from_path` no longer enforce `PathLike` as its first argument
+- Minor improvement over the global detection reliability
+
+### Added
+- Introduce function `is_binary` that relies on main capabilities, and optimized to detect binaries
+- Propagate `enable_fallback` argument throughout `from_bytes`, `from_path`, and `from_fp` that allow a deeper control over the detection (default True)
+- Explicit support for Python 3.12
+
+### Fixed
+- Edge case detection failure where a file would contain 'very-long' camel cased word (Issue #289)
+
+## [3.1.0](https://github.com/Ousret/charset_normalizer/compare/3.0.1...3.1.0) (2023-03-06)
+
+### Added
+- Argument `should_rename_legacy` for legacy function `detect` and disregard any new arguments without errors (PR #262)
+
+### Removed
+- Support for Python 3.6 (PR #260)
+
+### Changed
+- Optional speedup provided by mypy/c 1.0.1
+
+## [3.0.1](https://github.com/Ousret/charset_normalizer/compare/3.0.0...3.0.1) (2022-11-18)
+
+### Fixed
+- Multi-bytes cutter/chunk generator did not always cut correctly (PR #233)
+
+### Changed
+- Speedup provided by mypy/c 0.990 on Python >= 3.7
+
+## [3.0.0](https://github.com/Ousret/charset_normalizer/compare/2.1.1...3.0.0) (2022-10-20)
+
+### Added
+- Extend the capability of explain=True when cp_isolation contains at most two entries (min one), will log in details of the Mess-detector results
+- Support for alternative language frequency set in charset_normalizer.assets.FREQUENCIES
+- Add parameter `language_threshold` in `from_bytes`, `from_path` and `from_fp` to adjust the minimum expected coherence ratio
+- `normalizer --version` now specify if current version provide extra speedup (meaning mypyc compilation whl)
+
+### Changed
+- Build with static metadata using 'build' frontend
+- Make the language detection stricter
+- Optional: Module `md.py` can be compiled using Mypyc to provide an extra speedup up to 4x faster than v2.1
+
+### Fixed
+- CLI with opt --normalize fail when using full path for files
+- TooManyAccentuatedPlugin induce false positive on the mess detection when too few alpha character have been fed to it
+- Sphinx warnings when generating the documentation
+
+### Removed
+- Coherence detector no longer return 'Simple English' instead return 'English'
+- Coherence detector no longer return 'Classical Chinese' instead return 'Chinese'
+- Breaking: Method `first()` and `best()` from CharsetMatch
+- UTF-7 will no longer appear as "detected" without a recognized SIG/mark (is unreliable/conflict with ASCII)
+- Breaking: Class aliases CharsetDetector, CharsetDoctor, CharsetNormalizerMatch and CharsetNormalizerMatches
+- Breaking: Top-level function `normalize`
+- Breaking: Properties `chaos_secondary_pass`, `coherence_non_latin` and `w_counter` from CharsetMatch
+- Support for the backport `unicodedata2`
+
+## [3.0.0rc1](https://github.com/Ousret/charset_normalizer/compare/3.0.0b2...3.0.0rc1) (2022-10-18)
+
+### Added
+- Extend the capability of explain=True when cp_isolation contains at most two entries (min one), will log in details of the Mess-detector results
+- Support for alternative language frequency set in charset_normalizer.assets.FREQUENCIES
+- Add parameter `language_threshold` in `from_bytes`, `from_path` and `from_fp` to adjust the minimum expected coherence ratio
+
+### Changed
+- Build with static metadata using 'build' frontend
+- Make the language detection stricter
+
+### Fixed
+- CLI with opt --normalize fail when using full path for files
+- TooManyAccentuatedPlugin induce false positive on the mess detection when too few alpha character have been fed to it
+
+### Removed
+- Coherence detector no longer return 'Simple English' instead return 'English'
+- Coherence detector no longer return 'Classical Chinese' instead return 'Chinese'
+
+## [3.0.0b2](https://github.com/Ousret/charset_normalizer/compare/3.0.0b1...3.0.0b2) (2022-08-21)
+
+### Added
+- `normalizer --version` now specify if current version provide extra speedup (meaning mypyc compilation whl)
+
+### Removed
+- Breaking: Method `first()` and `best()` from CharsetMatch
+- UTF-7 will no longer appear as "detected" without a recognized SIG/mark (is unreliable/conflict with ASCII)
+
+### Fixed
+- Sphinx warnings when generating the documentation
+
+## [3.0.0b1](https://github.com/Ousret/charset_normalizer/compare/2.1.0...3.0.0b1) (2022-08-15)
+
+### Changed
+- Optional: Module `md.py` can be compiled using Mypyc to provide an extra speedup up to 4x faster than v2.1
+
+### Removed
+- Breaking: Class aliases CharsetDetector, CharsetDoctor, CharsetNormalizerMatch and CharsetNormalizerMatches
+- Breaking: Top-level function `normalize`
+- Breaking: Properties `chaos_secondary_pass`, `coherence_non_latin` and `w_counter` from CharsetMatch
+- Support for the backport `unicodedata2`
+
+## [2.1.1](https://github.com/Ousret/charset_normalizer/compare/2.1.0...2.1.1) (2022-08-19)
+
+### Deprecated
+- Function `normalize` scheduled for removal in 3.0
+
+### Changed
+- Removed useless call to decode in fn is_unprintable (#206)
+
+### Fixed
+- Third-party library (i18n xgettext) crashing not recognizing utf_8 (PEP 263) with underscore from [@aleksandernovikov](https://github.com/aleksandernovikov) (#204)
+
+## [2.1.0](https://github.com/Ousret/charset_normalizer/compare/2.0.12...2.1.0) (2022-06-19)
+
+### Added
+- Output the Unicode table version when running the CLI with `--version` (PR #194)
+
+### Changed
+- Re-use decoded buffer for single byte character sets from [@nijel](https://github.com/nijel) (PR #175)
+- Fixing some performance bottlenecks from [@deedy5](https://github.com/deedy5) (PR #183)
+
+### Fixed
+- Workaround potential bug in cpython with Zero Width No-Break Space located in Arabic Presentation Forms-B, Unicode 1.1 not acknowledged as space (PR #175)
+- CLI default threshold aligned with the API threshold from [@oleksandr-kuzmenko](https://github.com/oleksandr-kuzmenko) (PR #181)
+
+### Removed
+- Support for Python 3.5 (PR #192)
+
+### Deprecated
+- Use of backport unicodedata from `unicodedata2` as Python is quickly catching up, scheduled for removal in 3.0 (PR #194)
+
+## [2.0.12](https://github.com/Ousret/charset_normalizer/compare/2.0.11...2.0.12) (2022-02-12)
+
+### Fixed
+- ASCII miss-detection on rare cases (PR #170)
+
+## [2.0.11](https://github.com/Ousret/charset_normalizer/compare/2.0.10...2.0.11) (2022-01-30)
+
+### Added
+- Explicit support for Python 3.11 (PR #164)
+
+### Changed
+- The logging behavior have been completely reviewed, now using only TRACE and DEBUG levels (PR #163 #165)
+
+## [2.0.10](https://github.com/Ousret/charset_normalizer/compare/2.0.9...2.0.10) (2022-01-04)
+
+### Fixed
+- Fallback match entries might lead to UnicodeDecodeError for large bytes sequence (PR #154)
+
+### Changed
+- Skipping the language-detection (CD) on ASCII (PR #155)
+
+## [2.0.9](https://github.com/Ousret/charset_normalizer/compare/2.0.8...2.0.9) (2021-12-03)
+
+### Changed
+- Moderating the logging impact (since 2.0.8) for specific environments (PR #147)
+
+### Fixed
+- Wrong logging level applied when setting kwarg `explain` to True (PR #146)
+
+## [2.0.8](https://github.com/Ousret/charset_normalizer/compare/2.0.7...2.0.8) (2021-11-24)
+### Changed
+- Improvement over Vietnamese detection (PR #126)
+- MD improvement on trailing data and long foreign (non-pure latin) data (PR #124)
+- Efficiency improvements in cd/alphabet_languages from [@adbar](https://github.com/adbar) (PR #122)
+- call sum() without an intermediary list following PEP 289 recommendations from [@adbar](https://github.com/adbar) (PR #129)
+- Code style as refactored by Sourcery-AI (PR #131)
+- Minor adjustment on the MD around european words (PR #133)
+- Remove and replace SRTs from assets / tests (PR #139)
+- Initialize the library logger with a `NullHandler` by default from [@nmaynes](https://github.com/nmaynes) (PR #135)
+- Setting kwarg `explain` to True will add provisionally (bounded to function lifespan) a specific stream handler (PR #135)
+
+### Fixed
+- Fix large (misleading) sequence giving UnicodeDecodeError (PR #137)
+- Avoid using too insignificant chunk (PR #137)
+
+### Added
+- Add and expose function `set_logging_handler` to configure a specific StreamHandler from [@nmaynes](https://github.com/nmaynes) (PR #135)
+- Add `CHANGELOG.md` entries, format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) (PR #141)
+
+## [2.0.7](https://github.com/Ousret/charset_normalizer/compare/2.0.6...2.0.7) (2021-10-11)
+### Added
+- Add support for Kazakh (Cyrillic) language detection (PR #109)
+
+### Changed
+- Further, improve inferring the language from a given single-byte code page (PR #112)
+- Vainly trying to leverage PEP263 when PEP3120 is not supported (PR #116)
+- Refactoring for potential performance improvements in loops from [@adbar](https://github.com/adbar) (PR #113)
+- Various detection improvement (MD+CD) (PR #117)
+
+### Removed
+- Remove redundant logging entry about detected language(s) (PR #115)
+
+### Fixed
+- Fix a minor inconsistency between Python 3.5 and other versions regarding language detection (PR #117 #102)
+
+## [2.0.6](https://github.com/Ousret/charset_normalizer/compare/2.0.5...2.0.6) (2021-09-18)
+### Fixed
+- Unforeseen regression with the loss of the backward-compatibility with some older minor of Python 3.5.x (PR #100)
+- Fix CLI crash when using --minimal output in certain cases (PR #103)
+
+### Changed
+- Minor improvement to the detection efficiency (less than 1%) (PR #106 #101)
+
+## [2.0.5](https://github.com/Ousret/charset_normalizer/compare/2.0.4...2.0.5) (2021-09-14)
+### Changed
+- The project now comply with: flake8, mypy, isort and black to ensure a better overall quality (PR #81)
+- The BC-support with v1.x was improved, the old staticmethods are restored (PR #82)
+- The Unicode detection is slightly improved (PR #93)
+- Add syntax sugar \_\_bool\_\_ for results CharsetMatches list-container (PR #91)
+
+### Removed
+- The project no longer raise warning on tiny content given for detection, will be simply logged as warning instead (PR #92)
+
+### Fixed
+- In some rare case, the chunks extractor could cut in the middle of a multi-byte character and could mislead the mess detection (PR #95)
+- Some rare 'space' characters could trip up the UnprintablePlugin/Mess detection (PR #96)
+- The MANIFEST.in was not exhaustive (PR #78)
+
+## [2.0.4](https://github.com/Ousret/charset_normalizer/compare/2.0.3...2.0.4) (2021-07-30)
+### Fixed
+- The CLI no longer raise an unexpected exception when no encoding has been found (PR #70)
+- Fix accessing the 'alphabets' property when the payload contains surrogate characters (PR #68)
+- The logger could mislead (explain=True) on detected languages and the impact of one MBCS match (PR #72)
+- Submatch factoring could be wrong in rare edge cases (PR #72)
+- Multiple files given to the CLI were ignored when publishing results to STDOUT. (After the first path) (PR #72)
+- Fix line endings from CRLF to LF for certain project files (PR #67)
+
+### Changed
+- Adjust the MD to lower the sensitivity, thus improving the global detection reliability (PR #69 #76)
+- Allow fallback on specified encoding if any (PR #71)
+
+## [2.0.3](https://github.com/Ousret/charset_normalizer/compare/2.0.2...2.0.3) (2021-07-16)
+### Changed
+- Part of the detection mechanism has been improved to be less sensitive, resulting in more accurate detection results. Especially ASCII. (PR #63)
+- According to the community wishes, the detection will fall back on ASCII or UTF-8 in a last-resort case. (PR #64)
+
+## [2.0.2](https://github.com/Ousret/charset_normalizer/compare/2.0.1...2.0.2) (2021-07-15)
+### Fixed
+- Empty/Too small JSON payload miss-detection fixed. Report from [@tseaver](https://github.com/tseaver) (PR #59)
+
+### Changed
+- Don't inject unicodedata2 into sys.modules from [@akx](https://github.com/akx) (PR #57)
+
+## [2.0.1](https://github.com/Ousret/charset_normalizer/compare/2.0.0...2.0.1) (2021-07-13)
+### Fixed
+- Make it work where there isn't a filesystem available, dropping assets frequencies.json. Report from [@sethmlarson](https://github.com/sethmlarson). (PR #55)
+- Using explain=False permanently disable the verbose output in the current runtime (PR #47)
+- One log entry (language target preemptive) was not show in logs when using explain=True (PR #47)
+- Fix undesired exception (ValueError) on getitem of instance CharsetMatches (PR #52)
+
+### Changed
+- Public function normalize default args values were not aligned with from_bytes (PR #53)
+
+### Added
+- You may now use charset aliases in cp_isolation and cp_exclusion arguments (PR #47)
+
+## [2.0.0](https://github.com/Ousret/charset_normalizer/compare/1.4.1...2.0.0) (2021-07-02)
+### Changed
+- 4x to 5 times faster than the previous 1.4.0 release. At least 2x faster than Chardet.
+- Accent has been made on UTF-8 detection, should perform rather instantaneous.
+- The backward compatibility with Chardet has been greatly improved. The legacy detect function returns an identical charset name whenever possible.
+- The detection mechanism has been slightly improved, now Turkish content is detected correctly (most of the time)
+- The program has been rewritten to ease the readability and maintainability. (+Using static typing)+
+- utf_7 detection has been reinstated.
+
+### Removed
+- This package no longer require anything when used with Python 3.5 (Dropped cached_property)
+- Removed support for these languages: Catalan, Esperanto, Kazakh, Baque, Volapük, Azeri, Galician, Nynorsk, Macedonian, and Serbocroatian.
+- The exception hook on UnicodeDecodeError has been removed.
+
+### Deprecated
+- Methods coherence_non_latin, w_counter, chaos_secondary_pass of the class CharsetMatch are now deprecated and scheduled for removal in v3.0
+
+### Fixed
+- The CLI output used the relative path of the file(s). Should be absolute.
+
+## [1.4.1](https://github.com/Ousret/charset_normalizer/compare/1.4.0...1.4.1) (2021-05-28)
+### Fixed
+- Logger configuration/usage no longer conflict with others (PR #44)
+
+## [1.4.0](https://github.com/Ousret/charset_normalizer/compare/1.3.9...1.4.0) (2021-05-21)
+### Removed
+- Using standard logging instead of using the package loguru.
+- Dropping nose test framework in favor of the maintained pytest.
+- Choose to not use dragonmapper package to help with gibberish Chinese/CJK text.
+- Require cached_property only for Python 3.5 due to constraint. Dropping for every other interpreter version.
+- Stop support for UTF-7 that does not contain a SIG.
+- Dropping PrettyTable, replaced with pure JSON output in CLI.
+
+### Fixed
+- BOM marker in a CharsetNormalizerMatch instance could be False in rare cases even if obviously present. Due to the sub-match factoring process.
+- Not searching properly for the BOM when trying utf32/16 parent codec.
+
+### Changed
+- Improving the package final size by compressing frequencies.json.
+- Huge improvement over the larges payload.
+
+### Added
+- CLI now produces JSON consumable output.
+- Return ASCII if given sequences fit. Given reasonable confidence.
+
+## [1.3.9](https://github.com/Ousret/charset_normalizer/compare/1.3.8...1.3.9) (2021-05-13)
+
+### Fixed
+- In some very rare cases, you may end up getting encode/decode errors due to a bad bytes payload (PR #40)
+
+## [1.3.8](https://github.com/Ousret/charset_normalizer/compare/1.3.7...1.3.8) (2021-05-12)
+
+### Fixed
+- Empty given payload for detection may cause an exception if trying to access the `alphabets` property. (PR #39)
+
+## [1.3.7](https://github.com/Ousret/charset_normalizer/compare/1.3.6...1.3.7) (2021-05-12)
+
+### Fixed
+- The legacy detect function should return UTF-8-SIG if sig is present in the payload. (PR #38)
+
+## [1.3.6](https://github.com/Ousret/charset_normalizer/compare/1.3.5...1.3.6) (2021-02-09)
+
+### Changed
+- Amend the previous release to allow prettytable 2.0 (PR #35)
+
+## [1.3.5](https://github.com/Ousret/charset_normalizer/compare/1.3.4...1.3.5) (2021-02-08)
+
+### Fixed
+- Fix error while using the package with a python pre-release interpreter (PR #33)
+
+### Changed
+- Dependencies refactoring, constraints revised.
+
+### Added
+- Add python 3.9 and 3.10 to the supported interpreters
+
+MIT License
+
+Copyright (c) 2025 TAHRI Ahmed R.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/RECORD

@@ -0,0 +1,36 @@
+../../../bin/normalizer,sha256=SLSYxbOgNj-jB0bWbAPZMfKoeEan16B2x0xmbPcd3lw,273
+81d243bd2c585b0f4821__mypyc.cpython-39-x86_64-linux-gnu.so,sha256=Ujxa5iPb6AesmIlxxCdsF1WEp8tSMlzb-rW_532i3cU,410832
+charset_normalizer-3.4.7.dist-info/INSTALLER,sha256=zuuue4knoyJ-UwPPXg8fezS7VCrXJQrAP7zeNuwvFQg,4
+charset_normalizer-3.4.7.dist-info/METADATA,sha256=K8lK8L8LaZ1YmKvWLt3zEkpIxiCOC58xNhzFQrfQJxQ,40931
+charset_normalizer-3.4.7.dist-info/RECORD,,
+charset_normalizer-3.4.7.dist-info/WHEEL,sha256=OZLCzpIUHEX42a9RmtxH4_auVuxdXhghS0dmIdhxCRQ,184
+charset_normalizer-3.4.7.dist-info/entry_points.txt,sha256=ADSTKrkXZ3hhdOVFi6DcUEHQRS0xfxDIE_pEz4wLIXA,65
+charset_normalizer-3.4.7.dist-info/licenses/LICENSE,sha256=bQ1Bv-FwrGx9wkjJpj4lTQ-0WmDVCoJX0K-SxuJJuIc,1071
+charset_normalizer-3.4.7.dist-info/top_level.txt,sha256=c_vZbitqecT2GfK3zdxSTLCn8C-6pGnHQY5o_5Y32M0,47
+charset_normalizer/__init__.py,sha256=OKRxRv2Zhnqk00tqkN0c1BtJjm165fWXLydE52IKuHc,1590
+charset_normalizer/__main__.py,sha256=yzYxMR-IhKRHYwcSlavEv8oGdwxsR89mr2X09qXGdps,109
+charset_normalizer/__pycache__/__init__.cpython-39.pyc,,
+charset_normalizer/__pycache__/__main__.cpython-39.pyc,,
+charset_normalizer/__pycache__/api.cpython-39.pyc,,
+charset_normalizer/__pycache__/cd.cpython-39.pyc,,
+charset_normalizer/__pycache__/constant.cpython-39.pyc,,
+charset_normalizer/__pycache__/legacy.cpython-39.pyc,,
+charset_normalizer/__pycache__/md.cpython-39.pyc,,
+charset_normalizer/__pycache__/models.cpython-39.pyc,,
+charset_normalizer/__pycache__/utils.cpython-39.pyc,,
+charset_normalizer/__pycache__/version.cpython-39.pyc,,
+charset_normalizer/api.py,sha256=387F3n23MlMu-xfSbFULW2DLGsBmVrZVGhnkiGXeKBo,38844
+charset_normalizer/cd.cpython-39-x86_64-linux-gnu.so,sha256=q9uObNt5v2-3JR3GmbMsHY7cJIiCz9Eez4Hmvs0IR3M,15912
+charset_normalizer/cd.py,sha256=v0iPJweGsRegXywrM1LzUgqW9bJ1KFvIblQHP1jm5FQ,15174
+charset_normalizer/cli/__init__.py,sha256=D8I86lFk2-py45JvqxniTirSj_sFyE6sjaY_0-G1shc,136
+charset_normalizer/cli/__main__.py,sha256=E9FFSV1E2iOE_B2B1tJHQT9ExJqc60Ks_c-08sNawh8,11940
+charset_normalizer/cli/__pycache__/__init__.cpython-39.pyc,,
+charset_normalizer/cli/__pycache__/__main__.cpython-39.pyc,,
+charset_normalizer/constant.py,sha256=yvLAWDrdSC743Cu4amhwHLIO-FGuRTOTZouCzZKGikc,44431
+charset_normalizer/legacy.py,sha256=yBIFMNABNPE5JkdKOWyVo36fZtV9nm8bf37LrDWulz8,2661
+charset_normalizer/md.cpython-39-x86_64-linux-gnu.so,sha256=MvlOADQzK-FTt2QLGo1y7joPU_7evdj0LunrWj6uzXs,15912
+charset_normalizer/md.py,sha256=AYCdfDX79FrgoId3zXqmbCuDcbGr1NRuGqgJN94Rx9Q,30441
+charset_normalizer/models.py,sha256=FbaQnI6ECmVmyHRSvVM5fHNeMAQ3KSGdwLjGcQqWDws,12821
+charset_normalizer/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+charset_normalizer/utils.py,sha256=9cpi-_0-vC9pGDfuoarhC6VlF_Jxwx5Jsa_8I4w2D8k,12282
+charset_normalizer/version.py,sha256=2LxFuGp3BBuIwt95cp64y7v8bCNHcMAi08IfXt_47Co,115

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/WHEEL

@@ -0,0 +1,7 @@
+Wheel-Version: 1.0
+Generator: setuptools (82.0.1)
+Root-Is-Purelib: false
+Tag: cp39-cp39-manylinux_2_17_x86_64
+Tag: cp39-cp39-manylinux2014_x86_64
+Tag: cp39-cp39-manylinux_2_28_x86_64
+

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+normalizer = charset_normalizer.cli:cli_detect

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 TAHRI Ahmed R.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

.venv/lib/python3.9/site-packages/charset_normalizer-3.4.7.dist-info/top_level.txt

@@ -0,0 +1,2 @@
+81d243bd2c585b0f4821__mypyc
+charset_normalizer

.venv/lib/python3.9/site-packages/charset_normalizer/__init__.py

@@ -0,0 +1,48 @@
+"""
+Charset-Normalizer
+~~~~~~~~~~~~~~
+The Real First Universal Charset Detector.
+A library that helps you read text from an unknown charset encoding.
+Motivated by chardet, This package is trying to resolve the issue by taking a new approach.
+All IANA character set names for which the Python core library provides codecs are supported.
+
+Basic usage:
+   >>> from charset_normalizer import from_bytes
+   >>> results = from_bytes('Bсеки човек има право на образование. Oбразованието!'.encode('utf_8'))
+   >>> best_guess = results.best()
+   >>> str(best_guess)
+   'Bсеки човек има право на образование. Oбразованието!'
+
+Others methods and usages are available - see the full documentation
+at <https://github.com/Ousret/charset_normalizer>.
+:copyright: (c) 2021 by Ahmed TAHRI
+:license: MIT, see LICENSE for more details.
+"""
+
+from __future__ import annotations
+
+import logging
+
+from .api import from_bytes, from_fp, from_path, is_binary
+from .legacy import detect
+from .models import CharsetMatch, CharsetMatches
+from .utils import set_logging_handler
+from .version import VERSION, __version__
+
+__all__ = (
+    "from_fp",
+    "from_path",
+    "from_bytes",
+    "is_binary",
+    "detect",
+    "CharsetMatch",
+    "CharsetMatches",
+    "__version__",
+    "VERSION",
+    "set_logging_handler",
+)
+
+# Attach a NullHandler to the top level logger by default
+# https://docs.python.org/3.3/howto/logging.html#configuring-logging-for-a-library
+
+logging.getLogger("charset_normalizer").addHandler(logging.NullHandler())

.venv/lib/python3.9/site-packages/charset_normalizer/__main__.py

@@ -0,0 +1,6 @@
+from __future__ import annotations
+
+from .cli import cli_detect
+
+if __name__ == "__main__":
+    cli_detect()

.venv/lib/python3.9/site-packages/charset_normalizer/api.py

@@ -0,0 +1,988 @@
+from __future__ import annotations
+
+import logging
+from os import PathLike
+from typing import BinaryIO
+
+from .cd import (
+    coherence_ratio,
+    encoding_languages,
+    mb_encoding_languages,
+    merge_coherence_ratios,
+)
+from .constant import (
+    IANA_SUPPORTED,
+    IANA_SUPPORTED_SIMILAR,
+    TOO_BIG_SEQUENCE,
+    TOO_SMALL_SEQUENCE,
+    TRACE,
+)
+from .md import mess_ratio
+from .models import CharsetMatch, CharsetMatches
+from .utils import (
+    any_specified_encoding,
+    cut_sequence_chunks,
+    iana_name,
+    identify_sig_or_bom,
+    is_multi_byte_encoding,
+    should_strip_sig_or_bom,
+)
+
+logger = logging.getLogger("charset_normalizer")
+explain_handler = logging.StreamHandler()
+explain_handler.setFormatter(
+    logging.Formatter("%(asctime)s | %(levelname)s | %(message)s")
+)
+
+# Pre-compute a reordered encoding list: multibyte first, then single-byte.
+# This allows the mb_definitive_match optimization to fire earlier, skipping
+# all single-byte encodings for genuine CJK content. Multibyte codecs
+# hard-fail (UnicodeDecodeError) on single-byte data almost instantly, so
+# testing them first costs negligible time for non-CJK files.
+_mb_supported: list[str] = []
+_sb_supported: list[str] = []
+
+for _supported_enc in IANA_SUPPORTED:
+    try:
+        if is_multi_byte_encoding(_supported_enc):
+            _mb_supported.append(_supported_enc)
+        else:
+            _sb_supported.append(_supported_enc)
+    except ImportError:
+        _sb_supported.append(_supported_enc)
+
+IANA_SUPPORTED_MB_FIRST: list[str] = _mb_supported + _sb_supported
+
+
+def from_bytes(
+    sequences: bytes | bytearray,
+    steps: int = 5,
+    chunk_size: int = 512,
+    threshold: float = 0.2,
+    cp_isolation: list[str] | None = None,
+    cp_exclusion: list[str] | None = None,
+    preemptive_behaviour: bool = True,
+    explain: bool = False,
+    language_threshold: float = 0.1,
+    enable_fallback: bool = True,
+) -> CharsetMatches:
+    """
+    Given a raw bytes sequence, return the best possibles charset usable to render str objects.
+    If there is no results, it is a strong indicator that the source is binary/not text.
+    By default, the process will extract 5 blocks of 512o each to assess the mess and coherence of a given sequence.
+    And will give up a particular code page after 20% of measured mess. Those criteria are customizable at will.
+
+    The preemptive behavior DOES NOT replace the traditional detection workflow, it prioritize a particular code page
+    but never take it for granted. Can improve the performance.
+
+    You may want to focus your attention to some code page or/and not others, use cp_isolation and cp_exclusion for that
+    purpose.
+
+    This function will strip the SIG in the payload/sequence every time except on UTF-16, UTF-32.
+    By default the library does not setup any handler other than the NullHandler, if you choose to set the 'explain'
+    toggle to True it will alter the logger configuration to add a StreamHandler that is suitable for debugging.
+    Custom logging format and handler can be set manually.
+    """
+
+    if not isinstance(sequences, (bytearray, bytes)):
+        raise TypeError(
+            "Expected object of type bytes or bytearray, got: {}".format(
+                type(sequences)
+            )
+        )
+
+    if explain:
+        previous_logger_level: int = logger.level
+        logger.addHandler(explain_handler)
+        logger.setLevel(TRACE)
+
+    length: int = len(sequences)
+
+    if length == 0:
+        logger.debug("Encoding detection on empty bytes, assuming utf_8 intention.")
+        if explain:  # Defensive: ensure exit path clean handler
+            logger.removeHandler(explain_handler)
+            logger.setLevel(previous_logger_level)
+        return CharsetMatches([CharsetMatch(sequences, "utf_8", 0.0, False, [], "")])
+
+    if cp_isolation is not None:
+        logger.log(
+            TRACE,
+            "cp_isolation is set. use this flag for debugging purpose. "
+            "limited list of encoding allowed : %s.",
+            ", ".join(cp_isolation),
+        )
+        cp_isolation = [iana_name(cp, False) for cp in cp_isolation]
+    else:
+        cp_isolation = []
+
+    if cp_exclusion is not None:
+        logger.log(
+            TRACE,
+            "cp_exclusion is set. use this flag for debugging purpose. "
+            "limited list of encoding excluded : %s.",
+            ", ".join(cp_exclusion),
+        )
+        cp_exclusion = [iana_name(cp, False) for cp in cp_exclusion]
+    else:
+        cp_exclusion = []
+
+    if length <= (chunk_size * steps):
+        logger.log(
+            TRACE,
+            "override steps (%i) and chunk_size (%i) as content does not fit (%i byte(s) given) parameters.",
+            steps,
+            chunk_size,
+            length,
+        )
+        steps = 1
+        chunk_size = length
+
+    if steps > 1 and length / steps < chunk_size:
+        chunk_size = int(length / steps)
+
+    is_too_small_sequence: bool = len(sequences) < TOO_SMALL_SEQUENCE
+    is_too_large_sequence: bool = len(sequences) >= TOO_BIG_SEQUENCE
+
+    if is_too_small_sequence:
+        logger.log(
+            TRACE,
+            "Trying to detect encoding from a tiny portion of ({}) byte(s).".format(
+                length
+            ),
+        )
+    elif is_too_large_sequence:
+        logger.log(
+            TRACE,
+            "Using lazy str decoding because the payload is quite large, ({}) byte(s).".format(
+                length
+            ),
+        )
+
+    prioritized_encodings: list[str] = []
+
+    specified_encoding: str | None = (
+        any_specified_encoding(sequences) if preemptive_behaviour else None
+    )
+
+    if specified_encoding is not None:
+        prioritized_encodings.append(specified_encoding)
+        logger.log(
+            TRACE,
+            "Detected declarative mark in sequence. Priority +1 given for %s.",
+            specified_encoding,
+        )
+
+    tested: set[str] = set()
+    tested_but_hard_failure: list[str] = []
+    tested_but_soft_failure: list[str] = []
+    soft_failure_skip: set[str] = set()
+    success_fast_tracked: set[str] = set()
+
+    # Cache for decoded payload deduplication: hash(decoded_payload) -> (mean_mess_ratio, cd_ratios_merged, passed)
+    # When multiple encodings decode to the exact same string, we can skip the expensive
+    # mess_ratio and coherence_ratio analysis and reuse the results from the first encoding.
+    payload_result_cache: dict[int, tuple[float, list[tuple[str, float]], bool]] = {}
+
+    # When a definitive result (chaos=0.0 and good coherence) is found after testing
+    # the prioritized encodings (ascii, utf_8), we can significantly reduce the remaining
+    # work. Encodings that target completely different language families (e.g., Cyrillic
+    # when the definitive match is Latin) are skipped entirely.
+    # Additionally, for same-family encodings that pass chaos probing, we reuse the
+    # definitive match's coherence ratios instead of recomputing them — a major savings
+    # since coherence_ratio accounts for ~30% of total time on slow Latin files.
+    definitive_match_found: bool = False
+    definitive_target_languages: set[str] = set()
+    # After the definitive match fires, we cap the number of additional same-family
+    # single-byte encodings that pass chaos probing. Once we've accumulated enough
+    # good candidates (N), further same-family SB encodings are unlikely to produce
+    # a better best() result and just waste mess_ratio + coherence_ratio time.
+    # The first encoding to trigger the definitive match is NOT counted (it's already in).
+    post_definitive_sb_success_count: int = 0
+    POST_DEFINITIVE_SB_CAP: int = 7
+
+    # When a non-UTF multibyte encoding passes chaos probing with significant multibyte
+    # content (decoded length < 98% of raw length), skip all remaining single-byte encodings.
+    # Rationale: multi-byte decoders (CJK) have strict byte-sequence validation — if they
+    # decode without error AND pass chaos probing with substantial multibyte content, the
+    # data is genuinely multibyte encoded. Single-byte encodings will always decode (every
+    # byte maps to something) but waste time on mess_ratio before failing.
+    # The 98% threshold prevents false triggers on files that happen to have a few valid
+    # multibyte pairs (e.g., cp424/_ude_1.txt where big5 decodes with 99% ratio).
+    mb_definitive_match_found: bool = False
+
+    fallback_ascii: CharsetMatch | None = None
+    fallback_u8: CharsetMatch | None = None
+    fallback_specified: CharsetMatch | None = None
+
+    results: CharsetMatches = CharsetMatches()
+
+    early_stop_results: CharsetMatches = CharsetMatches()
+
+    sig_encoding, sig_payload = identify_sig_or_bom(sequences)
+
+    if sig_encoding is not None:
+        prioritized_encodings.append(sig_encoding)
+        logger.log(
+            TRACE,
+            "Detected a SIG or BOM mark on first %i byte(s). Priority +1 given for %s.",
+            len(sig_payload),
+            sig_encoding,
+        )
+
+    prioritized_encodings.append("ascii")
+
+    if "utf_8" not in prioritized_encodings:
+        prioritized_encodings.append("utf_8")
+
+    for encoding_iana in prioritized_encodings + IANA_SUPPORTED_MB_FIRST:
+        if cp_isolation and encoding_iana not in cp_isolation:
+            continue
+
+        if cp_exclusion and encoding_iana in cp_exclusion:
+            continue
+
+        if encoding_iana in tested:
+            continue
+
+        tested.add(encoding_iana)
+
+        decoded_payload: str | None = None
+        bom_or_sig_available: bool = sig_encoding == encoding_iana
+        strip_sig_or_bom: bool = bom_or_sig_available and should_strip_sig_or_bom(
+            encoding_iana
+        )
+
+        if encoding_iana in {"utf_16", "utf_32"} and not bom_or_sig_available:
+            logger.log(
+                TRACE,
+                "Encoding %s won't be tested as-is because it require a BOM. Will try some sub-encoder LE/BE.",
+                encoding_iana,
+            )
+            continue
+        if encoding_iana in {"utf_7"} and not bom_or_sig_available:
+            logger.log(
+                TRACE,
+                "Encoding %s won't be tested as-is because detection is unreliable without BOM/SIG.",
+                encoding_iana,
+            )
+            continue
+
+        # Skip encodings similar to ones that already soft-failed (high mess ratio).
+        # Checked BEFORE the expensive decode attempt.
+        if encoding_iana in soft_failure_skip:
+            logger.log(
+                TRACE,
+                "%s is deemed too similar to a code page that was already considered unsuited. Continuing!",
+                encoding_iana,
+            )
+            continue
+
+        # Skip encodings that were already fast-tracked from a similar successful encoding.
+        if encoding_iana in success_fast_tracked:
+            logger.log(
+                TRACE,
+                "Skipping %s: already fast-tracked from a similar successful encoding.",
+                encoding_iana,
+            )
+            continue
+
+        try:
+            is_multi_byte_decoder: bool = is_multi_byte_encoding(encoding_iana)
+        except (ModuleNotFoundError, ImportError):  # Defensive:
+            logger.log(
+                TRACE,
+                "Encoding %s does not provide an IncrementalDecoder",
+                encoding_iana,
+            )
+            continue
+
+        # When we've already found a definitive match (chaos=0.0 with good coherence)
+        # after testing the prioritized encodings, skip encodings that target
+        # completely different language families. This avoids running expensive
+        # mess_ratio + coherence_ratio on clearly unrelated candidates (e.g., Cyrillic
+        # when the definitive match is Latin-based).
+        if definitive_match_found:
+            if not is_multi_byte_decoder:
+                enc_languages = set(encoding_languages(encoding_iana))
+            else:
+                enc_languages = set(mb_encoding_languages(encoding_iana))
+            if not enc_languages.intersection(definitive_target_languages):
+                logger.log(
+                    TRACE,
+                    "Skipping %s: definitive match already found, this encoding targets different languages (%s vs %s).",
+                    encoding_iana,
+                    enc_languages,
+                    definitive_target_languages,
+                )
+                continue
+
+        # After the definitive match, cap the number of additional same-family
+        # single-byte encodings that pass chaos probing. This avoids testing the
+        # tail of rare, low-value same-family encodings (mac_iceland, cp860, etc.)
+        # that almost never change best() but each cost ~1-2ms of mess_ratio + coherence.
+        if (
+            definitive_match_found
+            and not is_multi_byte_decoder
+            and post_definitive_sb_success_count >= POST_DEFINITIVE_SB_CAP
+        ):
+            logger.log(
+                TRACE,
+                "Skipping %s: already accumulated %d same-family results after definitive match (cap=%d).",
+                encoding_iana,
+                post_definitive_sb_success_count,
+                POST_DEFINITIVE_SB_CAP,
+            )
+            continue
+
+        # When a multibyte encoding with significant multibyte content has already
+        # passed chaos probing, skip all single-byte encodings. They will either fail
+        # chaos probing (wasting mess_ratio time) or produce inferior results.
+        if mb_definitive_match_found and not is_multi_byte_decoder:
+            logger.log(
+                TRACE,
+                "Skipping single-byte %s: multi-byte definitive match already found.",
+                encoding_iana,
+            )
+            continue
+
+        try:
+            if is_too_large_sequence and is_multi_byte_decoder is False:
+                str(
+                    (
+                        sequences[: int(50e4)]
+                        if strip_sig_or_bom is False
+                        else sequences[len(sig_payload) : int(50e4)]
+                    ),
+                    encoding=encoding_iana,
+                )
+            else:
+                # UTF-7 BOM is encoded in modified Base64 whose byte boundary
+                # can overlap with the next character. Stripping raw SIG bytes
+                # before decoding may leave stray bytes that decode as garbage.
+                # Decode the full sequence and remove the leading BOM char instead.
+                # see https://github.com/jawah/charset_normalizer/issues/718
+                # and https://github.com/jawah/charset_normalizer/issues/716
+                if encoding_iana == "utf_7" and bom_or_sig_available:
+                    decoded_payload = str(
+                        sequences,
+                        encoding=encoding_iana,
+                    )
+                    if decoded_payload and decoded_payload[0] == "\ufeff":
+                        decoded_payload = decoded_payload[1:]
+                else:
+                    decoded_payload = str(
+                        (
+                            sequences
+                            if strip_sig_or_bom is False
+                            else sequences[len(sig_payload) :]
+                        ),
+                        encoding=encoding_iana,
+                    )
+        except (UnicodeDecodeError, LookupError) as e:
+            if not isinstance(e, LookupError):
+                logger.log(
+                    TRACE,
+                    "Code page %s does not fit given bytes sequence at ALL. %s",
+                    encoding_iana,
+                    str(e),
+                )
+            tested_but_hard_failure.append(encoding_iana)
+            continue
+
+        r_ = range(
+            0 if not bom_or_sig_available else len(sig_payload),
+            length,
+            int(length / steps),
+        )
+
+        multi_byte_bonus: bool = (
+            is_multi_byte_decoder
+            and decoded_payload is not None
+            and len(decoded_payload) < length
+        )
+
+        if multi_byte_bonus:
+            logger.log(
+                TRACE,
+                "Code page %s is a multi byte encoding table and it appear that at least one character "
+                "was encoded using n-bytes.",
+                encoding_iana,
+            )
+
+        # Payload-hash deduplication: if another encoding already decoded to the
+        # exact same string, reuse its mess_ratio and coherence results entirely.
+        # This is strictly more general than the old IANA_SUPPORTED_SIMILAR approach
+        # because it catches ALL identical decoding, not just pre-mapped ones.
+        if decoded_payload is not None and not is_multi_byte_decoder:
+            payload_hash: int = hash(decoded_payload)
+            cached = payload_result_cache.get(payload_hash)
+            if cached is not None:
+                cached_mess, cached_cd, cached_passed = cached
+                if cached_passed:
+                    # The previous encoding with identical output passed chaos probing.
+                    fast_match = CharsetMatch(
+                        sequences,
+                        encoding_iana,
+                        cached_mess,
+                        bom_or_sig_available,
+                        cached_cd,
+                        (
+                            decoded_payload
+                            if (
+                                is_too_large_sequence is False
+                                or encoding_iana
+                                in [specified_encoding, "ascii", "utf_8"]
+                            )
+                            else None
+                        ),
+                        preemptive_declaration=specified_encoding,
+                    )
+                    results.append(fast_match)
+                    success_fast_tracked.add(encoding_iana)
+                    logger.log(
+                        TRACE,
+                        "%s fast-tracked (identical decoded payload to a prior encoding, chaos=%f %%).",
+                        encoding_iana,
+                        round(cached_mess * 100, ndigits=3),
+                    )
+
+                    if (
+                        encoding_iana in [specified_encoding, "ascii", "utf_8"]
+                        and cached_mess < 0.1
+                    ):
+                        if cached_mess == 0.0:
+                            logger.debug(
+                                "Encoding detection: %s is most likely the one.",
+                                fast_match.encoding,
+                            )
+                            if explain:
+                                logger.removeHandler(explain_handler)
+                                logger.setLevel(previous_logger_level)
+                            return CharsetMatches([fast_match])
+                        early_stop_results.append(fast_match)
+
+                    if (
+                        len(early_stop_results)
+                        and (specified_encoding is None or specified_encoding in tested)
+                        and "ascii" in tested
+                        and "utf_8" in tested
+                    ):
+                        probable_result: CharsetMatch = early_stop_results.best()  # type: ignore[assignment]
+                        logger.debug(
+                            "Encoding detection: %s is most likely the one.",
+                            probable_result.encoding,
+                        )
+                        if explain:
+                            logger.removeHandler(explain_handler)
+                            logger.setLevel(previous_logger_level)
+                        return CharsetMatches([probable_result])
+
+                    continue
+                else:
+                    # The previous encoding with identical output failed chaos probing.
+                    tested_but_soft_failure.append(encoding_iana)
+                    logger.log(
+                        TRACE,
+                        "%s fast-skipped (identical decoded payload to a prior encoding that failed chaos probing).",
+                        encoding_iana,
+                    )
+                    # Prepare fallbacks for special encodings even when skipped.
+                    if enable_fallback and encoding_iana in [
+                        "ascii",
+                        "utf_8",
+                        specified_encoding,
+                        "utf_16",
+                        "utf_32",
+                    ]:
+                        fallback_entry = CharsetMatch(
+                            sequences,
+                            encoding_iana,
+                            threshold,
+                            bom_or_sig_available,
+                            [],
+                            decoded_payload,
+                            preemptive_declaration=specified_encoding,
+                        )
+                        if encoding_iana == specified_encoding:
+                            fallback_specified = fallback_entry
+                        elif encoding_iana == "ascii":
+                            fallback_ascii = fallback_entry
+                        else:
+                            fallback_u8 = fallback_entry
+                    continue
+
+        max_chunk_gave_up: int = int(len(r_) / 4)
+
+        max_chunk_gave_up = max(max_chunk_gave_up, 2)
+        early_stop_count: int = 0
+        lazy_str_hard_failure = False
+
+        md_chunks: list[str] = []
+        md_ratios = []
+
+        try:
+            for chunk in cut_sequence_chunks(
+                sequences,
+                encoding_iana,
+                r_,
+                chunk_size,
+                bom_or_sig_available,
+                strip_sig_or_bom,
+                sig_payload,
+                is_multi_byte_decoder,
+                decoded_payload,
+            ):
+                md_chunks.append(chunk)
+
+                md_ratios.append(
+                    mess_ratio(
+                        chunk,
+                        threshold,
+                        explain is True and 1 <= len(cp_isolation) <= 2,
+                    )
+                )
+
+                if md_ratios[-1] >= threshold:
+                    early_stop_count += 1
+
+                if (early_stop_count >= max_chunk_gave_up) or (
+                    bom_or_sig_available and strip_sig_or_bom is False
+                ):
+                    break
+        except (
+            UnicodeDecodeError
+        ) as e:  # Lazy str loading may have missed something there
+            logger.log(
+                TRACE,
+                "LazyStr Loading: After MD chunk decode, code page %s does not fit given bytes sequence at ALL. %s",
+                encoding_iana,
+                str(e),
+            )
+            early_stop_count = max_chunk_gave_up
+            lazy_str_hard_failure = True
+
+        # We might want to check the sequence again with the whole content
+        # Only if initial MD tests passes
+        if (
+            not lazy_str_hard_failure
+            and is_too_large_sequence
+            and not is_multi_byte_decoder
+        ):
+            try:
+                sequences[int(50e3) :].decode(encoding_iana, errors="strict")
+            except UnicodeDecodeError as e:
+                logger.log(
+                    TRACE,
+                    "LazyStr Loading: After final lookup, code page %s does not fit given bytes sequence at ALL. %s",
+                    encoding_iana,
+                    str(e),
+                )
+                tested_but_hard_failure.append(encoding_iana)
+                continue
+
+        mean_mess_ratio: float = sum(md_ratios) / len(md_ratios) if md_ratios else 0.0
+        if mean_mess_ratio >= threshold or early_stop_count >= max_chunk_gave_up:
+            tested_but_soft_failure.append(encoding_iana)
+            if encoding_iana in IANA_SUPPORTED_SIMILAR:
+                soft_failure_skip.update(IANA_SUPPORTED_SIMILAR[encoding_iana])
+            # Cache this soft-failure so identical decoding from other encodings
+            # can be skipped immediately.
+            if decoded_payload is not None and not is_multi_byte_decoder:
+                payload_result_cache.setdefault(
+                    hash(decoded_payload), (mean_mess_ratio, [], False)
+                )
+            logger.log(
+                TRACE,
+                "%s was excluded because of initial chaos probing. Gave up %i time(s). "
+                "Computed mean chaos is %f %%.",
+                encoding_iana,
+                early_stop_count,
+                round(mean_mess_ratio * 100, ndigits=3),
+            )
+            # Preparing those fallbacks in case we got nothing.
+            if (
+                enable_fallback
+                and encoding_iana
+                in ["ascii", "utf_8", specified_encoding, "utf_16", "utf_32"]
+                and not lazy_str_hard_failure
+            ):
+                fallback_entry = CharsetMatch(
+                    sequences,
+                    encoding_iana,
+                    threshold,
+                    bom_or_sig_available,
+                    [],
+                    decoded_payload,
+                    preemptive_declaration=specified_encoding,
+                )
+                if encoding_iana == specified_encoding:
+                    fallback_specified = fallback_entry
+                elif encoding_iana == "ascii":
+                    fallback_ascii = fallback_entry
+                else:
+                    fallback_u8 = fallback_entry
+            continue
+
+        logger.log(
+            TRACE,
+            "%s passed initial chaos probing. Mean measured chaos is %f %%",
+            encoding_iana,
+            round(mean_mess_ratio * 100, ndigits=3),
+        )
+
+        if not is_multi_byte_decoder:
+            target_languages: list[str] = encoding_languages(encoding_iana)
+        else:
+            target_languages = mb_encoding_languages(encoding_iana)
+
+        if target_languages:
+            logger.log(
+                TRACE,
+                "{} should target any language(s) of {}".format(
+                    encoding_iana, str(target_languages)
+                ),
+            )
+
+        cd_ratios = []
+
+        # Run coherence detection on all chunks. We previously tried limiting to
+        # 1-2 chunks for post-definitive encodings to save time, but this caused
+        # coverage regressions by producing unrepresentative coherence scores.
+        # The SB cap and language-family skip optimizations provide sufficient
+        # speedup without sacrificing coherence accuracy.
+        if encoding_iana != "ascii":
+            # We shall skip the CD when its about ASCII
+            # Most of the time its not relevant to run "language-detection" on it.
+            for chunk in md_chunks:
+                chunk_languages = coherence_ratio(
+                    chunk,
+                    language_threshold,
+                    ",".join(target_languages) if target_languages else None,
+                )
+
+                cd_ratios.append(chunk_languages)
+            cd_ratios_merged = merge_coherence_ratios(cd_ratios)
+        else:
+            cd_ratios_merged = merge_coherence_ratios(cd_ratios)
+
+        if cd_ratios_merged:
+            logger.log(
+                TRACE,
+                "We detected language {} using {}".format(
+                    cd_ratios_merged, encoding_iana
+                ),
+            )
+
+        current_match = CharsetMatch(
+            sequences,
+            encoding_iana,
+            mean_mess_ratio,
+            bom_or_sig_available,
+            cd_ratios_merged,
+            (
+                decoded_payload
+                if (
+                    is_too_large_sequence is False
+                    or encoding_iana in [specified_encoding, "ascii", "utf_8"]
+                )
+                else None
+            ),
+            preemptive_declaration=specified_encoding,
+        )
+
+        results.append(current_match)
+
+        # Cache the successful result for payload-hash deduplication.
+        if decoded_payload is not None and not is_multi_byte_decoder:
+            payload_result_cache.setdefault(
+                hash(decoded_payload),
+                (mean_mess_ratio, cd_ratios_merged, True),
+            )
+
+        # Count post-definitive same-family SB successes for the early termination cap.
+        # Only count low-mess encodings (< 2%) toward the cap. High-mess encodings are
+        # marginal results that shouldn't prevent better-quality candidates from being
+        # tested. For example, iso8859_4 (mess=0%) should not be skipped just because
+        # 7 high-mess Latin encodings (cp1252 at 8%, etc.) were tried first.
+        if (
+            definitive_match_found
+            and not is_multi_byte_decoder
+            and mean_mess_ratio < 0.02
+        ):
+            post_definitive_sb_success_count += 1
+
+        if (
+            encoding_iana in [specified_encoding, "ascii", "utf_8"]
+            and mean_mess_ratio < 0.1
+        ):
+            # If md says nothing to worry about, then... stop immediately!
+            if mean_mess_ratio == 0.0:
+                logger.debug(
+                    "Encoding detection: %s is most likely the one.",
+                    current_match.encoding,
+                )
+                if explain:  # Defensive: ensure exit path clean handler
+                    logger.removeHandler(explain_handler)
+                    logger.setLevel(previous_logger_level)
+                return CharsetMatches([current_match])
+
+            early_stop_results.append(current_match)
+
+        if (
+            len(early_stop_results)
+            and (specified_encoding is None or specified_encoding in tested)
+            and "ascii" in tested
+            and "utf_8" in tested
+        ):
+            probable_result = early_stop_results.best()  # type: ignore[assignment]
+            logger.debug(
+                "Encoding detection: %s is most likely the one.",
+                probable_result.encoding,  # type: ignore[union-attr]
+            )
+            if explain:  # Defensive: ensure exit path clean handler
+                logger.removeHandler(explain_handler)
+                logger.setLevel(previous_logger_level)
+
+            return CharsetMatches([probable_result])
+
+        # Once we find a result with good coherence (>= 0.5) after testing the
+        # prioritized encodings (ascii, utf_8), activate "definitive mode": skip
+        # encodings that target completely different language families. This avoids
+        # running expensive mess_ratio + coherence_ratio on clearly unrelated
+        # candidates (e.g., Cyrillic encodings when the match is Latin-based).
+        # We require coherence >= 0.5 to avoid false positives (e.g., cp1251 decoding
+        # Hebrew text with 0.0 chaos but wrong language detection at coherence 0.33).
+        if not definitive_match_found and not is_multi_byte_decoder:
+            best_coherence = (
+                max((v for _, v in cd_ratios_merged), default=0.0)
+                if cd_ratios_merged
+                else 0.0
+            )
+            if best_coherence >= 0.5 and "ascii" in tested and "utf_8" in tested:
+                definitive_match_found = True
+                definitive_target_languages.update(target_languages)
+                logger.log(
+                    TRACE,
+                    "Definitive match found: %s (chaos=%.3f, coherence=%.2f). Encodings targeting different language families will be skipped.",
+                    encoding_iana,
+                    mean_mess_ratio,
+                    best_coherence,
+                )
+
+        # When a non-UTF multibyte encoding passes chaos probing with significant
+        # multibyte content (decoded < 98% of raw), activate mb_definitive_match.
+        # This skips all remaining single-byte encodings which would either soft-fail
+        # (running expensive mess_ratio for nothing) or produce inferior results.
+        if (
+            not mb_definitive_match_found
+            and is_multi_byte_decoder
+            and multi_byte_bonus
+            and decoded_payload is not None
+            and len(decoded_payload) < length * 0.98
+            and encoding_iana
+            not in {
+                "utf_8",
+                "utf_8_sig",
+                "utf_16",
+                "utf_16_be",
+                "utf_16_le",
+                "utf_32",
+                "utf_32_be",
+                "utf_32_le",
+                "utf_7",
+            }
+            and "ascii" in tested
+            and "utf_8" in tested
+        ):
+            mb_definitive_match_found = True
+            logger.log(
+                TRACE,
+                "Multi-byte definitive match: %s (chaos=%.3f, decoded=%d/%d=%.1f%%). Single-byte encodings will be skipped.",
+                encoding_iana,
+                mean_mess_ratio,
+                len(decoded_payload),
+                length,
+                len(decoded_payload) / length * 100,
+            )
+
+        if encoding_iana == sig_encoding:
+            logger.debug(
+                "Encoding detection: %s is most likely the one as we detected a BOM or SIG within "
+                "the beginning of the sequence.",
+                encoding_iana,
+            )
+            if explain:  # Defensive: ensure exit path clean handler
+                logger.removeHandler(explain_handler)
+                logger.setLevel(previous_logger_level)
+            return CharsetMatches([results[encoding_iana]])
+
+    if len(results) == 0:
+        if fallback_u8 or fallback_ascii or fallback_specified:
+            logger.log(
+                TRACE,
+                "Nothing got out of the detection process. Using ASCII/UTF-8/Specified fallback.",
+            )
+
+        if fallback_specified:
+            logger.debug(
+                "Encoding detection: %s will be used as a fallback match",
+                fallback_specified.encoding,
+            )
+            results.append(fallback_specified)
+        elif (
+            (fallback_u8 and fallback_ascii is None)
+            or (
+                fallback_u8
+                and fallback_ascii
+                and fallback_u8.fingerprint != fallback_ascii.fingerprint
+            )
+            or (fallback_u8 is not None)
+        ):
+            logger.debug("Encoding detection: utf_8 will be used as a fallback match")
+            results.append(fallback_u8)
+        elif fallback_ascii:
+            logger.debug("Encoding detection: ascii will be used as a fallback match")
+            results.append(fallback_ascii)
+
+    if results:
+        logger.debug(
+            "Encoding detection: Found %s as plausible (best-candidate) for content. With %i alternatives.",
+            results.best().encoding,  # type: ignore
+            len(results) - 1,
+        )
+    else:
+        logger.debug("Encoding detection: Unable to determine any suitable charset.")
+
+    if explain:
+        logger.removeHandler(explain_handler)
+        logger.setLevel(previous_logger_level)
+
+    return results
+
+
+def from_fp(
+    fp: BinaryIO,
+    steps: int = 5,
+    chunk_size: int = 512,
+    threshold: float = 0.20,
+    cp_isolation: list[str] | None = None,
+    cp_exclusion: list[str] | None = None,
+    preemptive_behaviour: bool = True,
+    explain: bool = False,
+    language_threshold: float = 0.1,
+    enable_fallback: bool = True,
+) -> CharsetMatches:
+    """
+    Same thing than the function from_bytes but using a file pointer that is already ready.
+    Will not close the file pointer.
+    """
+    return from_bytes(
+        fp.read(),
+        steps,
+        chunk_size,
+        threshold,
+        cp_isolation,
+        cp_exclusion,
+        preemptive_behaviour,
+        explain,
+        language_threshold,
+        enable_fallback,
+    )
+
+
+def from_path(
+    path: str | bytes | PathLike,  # type: ignore[type-arg]
+    steps: int = 5,
+    chunk_size: int = 512,
+    threshold: float = 0.20,
+    cp_isolation: list[str] | None = None,
+    cp_exclusion: list[str] | None = None,
+    preemptive_behaviour: bool = True,
+    explain: bool = False,
+    language_threshold: float = 0.1,
+    enable_fallback: bool = True,
+) -> CharsetMatches:
+    """
+    Same thing than the function from_bytes but with one extra step. Opening and reading given file path in binary mode.
+    Can raise IOError.
+    """
+    with open(path, "rb") as fp:
+        return from_fp(
+            fp,
+            steps,
+            chunk_size,
+            threshold,
+            cp_isolation,
+            cp_exclusion,
+            preemptive_behaviour,
+            explain,
+            language_threshold,
+            enable_fallback,
+        )
+
+
+def is_binary(
+    fp_or_path_or_payload: PathLike | str | BinaryIO | bytes,  # type: ignore[type-arg]
+    steps: int = 5,
+    chunk_size: int = 512,
+    threshold: float = 0.20,
+    cp_isolation: list[str] | None = None,
+    cp_exclusion: list[str] | None = None,
+    preemptive_behaviour: bool = True,
+    explain: bool = False,
+    language_threshold: float = 0.1,
+    enable_fallback: bool = False,
+) -> bool:
+    """
+    Detect if the given input (file, bytes, or path) points to a binary file. aka. not a string.
+    Based on the same main heuristic algorithms and default kwargs at the sole exception that fallbacks match
+    are disabled to be stricter around ASCII-compatible but unlikely to be a string.
+    """
+    if isinstance(fp_or_path_or_payload, (str, PathLike)):
+        guesses = from_path(
+            fp_or_path_or_payload,
+            steps=steps,
+            chunk_size=chunk_size,
+            threshold=threshold,
+            cp_isolation=cp_isolation,
+            cp_exclusion=cp_exclusion,
+            preemptive_behaviour=preemptive_behaviour,
+            explain=explain,
+            language_threshold=language_threshold,
+            enable_fallback=enable_fallback,
+        )
+    elif isinstance(
+        fp_or_path_or_payload,
+        (
+            bytes,
+            bytearray,
+        ),
+    ):
+        guesses = from_bytes(
+            fp_or_path_or_payload,
+            steps=steps,
+            chunk_size=chunk_size,
+            threshold=threshold,
+            cp_isolation=cp_isolation,
+            cp_exclusion=cp_exclusion,
+            preemptive_behaviour=preemptive_behaviour,
+            explain=explain,
+            language_threshold=language_threshold,
+            enable_fallback=enable_fallback,
+        )
+    else:
+        guesses = from_fp(
+            fp_or_path_or_payload,
+            steps=steps,
+            chunk_size=chunk_size,
+            threshold=threshold,
+            cp_isolation=cp_isolation,
+            cp_exclusion=cp_exclusion,
+            preemptive_behaviour=preemptive_behaviour,
+            explain=explain,
+            language_threshold=language_threshold,
+            enable_fallback=enable_fallback,
+        )
+
+    return not guesses

.venv/lib/python3.9/site-packages/charset_normalizer/cd.py

@@ -0,0 +1,454 @@
+from __future__ import annotations
+
+import importlib
+from codecs import IncrementalDecoder
+from collections import Counter
+from functools import lru_cache
+from typing import Counter as TypeCounter
+
+from .constant import (
+    FREQUENCIES,
+    KO_NAMES,
+    LANGUAGE_SUPPORTED_COUNT,
+    TOO_SMALL_SEQUENCE,
+    ZH_NAMES,
+    _FREQUENCIES_SET,
+    _FREQUENCIES_RANK,
+)
+from .md import is_suspiciously_successive_range
+from .models import CoherenceMatches
+from .utils import (
+    is_accentuated,
+    is_latin,
+    is_multi_byte_encoding,
+    is_unicode_range_secondary,
+    unicode_range,
+)
+
+
+def encoding_unicode_range(iana_name: str) -> list[str]:
+    """
+    Return associated unicode ranges in a single byte code page.
+    """
+    if is_multi_byte_encoding(iana_name):
+        raise OSError(  # Defensive:
+            "Function not supported on multi-byte code page"
+        )
+
+    decoder = importlib.import_module(f"encodings.{iana_name}").IncrementalDecoder
+
+    p: IncrementalDecoder = decoder(errors="ignore")
+    seen_ranges: dict[str, int] = {}
+    character_count: int = 0
+
+    for i in range(0x40, 0xFF):
+        chunk: str = p.decode(bytes([i]))
+
+        if chunk:
+            character_range: str | None = unicode_range(chunk)
+
+            if character_range is None:
+                continue
+
+            if is_unicode_range_secondary(character_range) is False:
+                if character_range not in seen_ranges:
+                    seen_ranges[character_range] = 0
+                seen_ranges[character_range] += 1
+                character_count += 1
+
+    return sorted(
+        [
+            character_range
+            for character_range in seen_ranges
+            if seen_ranges[character_range] / character_count >= 0.15
+        ]
+    )
+
+
+def unicode_range_languages(primary_range: str) -> list[str]:
+    """
+    Return inferred languages used with a unicode range.
+    """
+    languages: list[str] = []
+
+    for language, characters in FREQUENCIES.items():
+        for character in characters:
+            if unicode_range(character) == primary_range:
+                languages.append(language)
+                break
+
+    return languages
+
+
+@lru_cache()
+def encoding_languages(iana_name: str) -> list[str]:
+    """
+    Single-byte encoding language association. Some code page are heavily linked to particular language(s).
+    This function does the correspondence.
+    """
+    unicode_ranges: list[str] = encoding_unicode_range(iana_name)
+    primary_range: str | None = None
+
+    for specified_range in unicode_ranges:
+        if "Latin" not in specified_range:
+            primary_range = specified_range
+            break
+
+    if primary_range is None:
+        return ["Latin Based"]
+
+    return unicode_range_languages(primary_range)
+
+
+@lru_cache()
+def mb_encoding_languages(iana_name: str) -> list[str]:
+    """
+    Multi-byte encoding language association. Some code page are heavily linked to particular language(s).
+    This function does the correspondence.
+    """
+    if (
+        iana_name.startswith("shift_")
+        or iana_name.startswith("iso2022_jp")
+        or iana_name.startswith("euc_j")
+        or iana_name == "cp932"
+    ):
+        return ["Japanese"]
+    if iana_name.startswith("gb") or iana_name in ZH_NAMES:
+        return ["Chinese"]
+    if iana_name.startswith("iso2022_kr") or iana_name in KO_NAMES:
+        return ["Korean"]
+
+    return []
+
+
+@lru_cache(maxsize=LANGUAGE_SUPPORTED_COUNT)
+def get_target_features(language: str) -> tuple[bool, bool]:
+    """
+    Determine main aspects from a supported language if it contains accents and if is pure Latin.
+    """
+    target_have_accents: bool = False
+    target_pure_latin: bool = True
+
+    for character in FREQUENCIES[language]:
+        if not target_have_accents and is_accentuated(character):
+            target_have_accents = True
+        if target_pure_latin and is_latin(character) is False:
+            target_pure_latin = False
+
+    return target_have_accents, target_pure_latin
+
+
+def alphabet_languages(
+    characters: list[str], ignore_non_latin: bool = False
+) -> list[str]:
+    """
+    Return associated languages associated to given characters.
+    """
+    languages: list[tuple[str, float]] = []
+
+    characters_set: frozenset[str] = frozenset(characters)
+    source_have_accents = any(is_accentuated(character) for character in characters)
+
+    for language, language_characters in FREQUENCIES.items():
+        target_have_accents, target_pure_latin = get_target_features(language)
+
+        if ignore_non_latin and target_pure_latin is False:
+            continue
+
+        if target_have_accents is False and source_have_accents:
+            continue
+
+        character_count: int = len(language_characters)
+
+        character_match_count: int = len(_FREQUENCIES_SET[language] & characters_set)
+
+        ratio: float = character_match_count / character_count
+
+        if ratio >= 0.2:
+            languages.append((language, ratio))
+
+    languages = sorted(languages, key=lambda x: x[1], reverse=True)
+
+    return [compatible_language[0] for compatible_language in languages]
+
+
+def characters_popularity_compare(
+    language: str, ordered_characters: list[str]
+) -> float:
+    """
+    Determine if a ordered characters list (by occurrence from most appearance to rarest) match a particular language.
+    The result is a ratio between 0. (absolutely no correspondence) and 1. (near perfect fit).
+    Beware that is function is not strict on the match in order to ease the detection. (Meaning close match is 1.)
+    """
+    if language not in FREQUENCIES:
+        raise ValueError(f"{language} not available")  # Defensive:
+
+    character_approved_count: int = 0
+    frequencies_language_set: frozenset[str] = _FREQUENCIES_SET[language]
+    lang_rank: dict[str, int] = _FREQUENCIES_RANK[language]
+
+    ordered_characters_count: int = len(ordered_characters)
+    target_language_characters_count: int = len(FREQUENCIES[language])
+
+    large_alphabet: bool = target_language_characters_count > 26
+
+    expected_projection_ratio: float = (
+        target_language_characters_count / ordered_characters_count
+    )
+
+    # Pre-built rank dict for ordered_characters (avoids repeated list slicing).
+    ordered_rank: dict[str, int] = {
+        char: rank for rank, char in enumerate(ordered_characters)
+    }
+
+    # Pre-compute characters common to both orderings.
+    # Avoids repeated `c in ordered_rank` dict lookups in the inner counts.
+    common_chars: list[tuple[int, int]] = [
+        (lr, ordered_rank[c]) for c, lr in lang_rank.items() if c in ordered_rank
+    ]
+
+    # Pre-extract lr and orr arrays for faster iteration in the inner loop.
+    # Plain integer loops with local arrays are much faster under mypyc than
+    # generator expression sums over a list of tuples.
+    common_count: int = len(common_chars)
+    common_lr: list[int] = [p[0] for p in common_chars]
+    common_orr: list[int] = [p[1] for p in common_chars]
+
+    for character, character_rank in zip(
+        ordered_characters, range(0, ordered_characters_count)
+    ):
+        if character not in frequencies_language_set:
+            continue
+
+        character_rank_in_language: int = lang_rank[character]
+        character_rank_projection: int = int(character_rank * expected_projection_ratio)
+
+        if (
+            large_alphabet is False
+            and abs(character_rank_projection - character_rank_in_language) > 4
+        ):
+            continue
+
+        if (
+            large_alphabet is True
+            and abs(character_rank_projection - character_rank_in_language)
+            < target_language_characters_count / 3
+        ):
+            character_approved_count += 1
+            continue
+
+        # Count how many characters appear "before" in both orderings,
+        # and how many appear "at or after" in both orderings.
+        # Single pass over pre-extracted arrays — much faster under mypyc
+        # than two generator expression sums.
+        before_match_count: int = 0
+        after_match_count: int = 0
+        for i in range(common_count):
+            lr_i: int = common_lr[i]
+            orr_i: int = common_orr[i]
+            if lr_i < character_rank_in_language:
+                if orr_i < character_rank:
+                    before_match_count += 1
+            else:
+                if orr_i >= character_rank:
+                    after_match_count += 1
+
+        after_len: int = target_language_characters_count - character_rank_in_language
+
+        if character_rank_in_language == 0 and before_match_count <= 4:
+            character_approved_count += 1
+            continue
+
+        if after_len == 0 and after_match_count <= 4:
+            character_approved_count += 1
+            continue
+
+        if (
+            character_rank_in_language > 0
+            and before_match_count / character_rank_in_language >= 0.4
+        ) or (after_len > 0 and after_match_count / after_len >= 0.4):
+            character_approved_count += 1
+            continue
+
+    return character_approved_count / len(ordered_characters)
+
+
+def alpha_unicode_split(decoded_sequence: str) -> list[str]:
+    """
+    Given a decoded text sequence, return a list of str. Unicode range / alphabet separation.
+    Ex. a text containing English/Latin with a bit a Hebrew will return two items in the resulting list;
+    One containing the latin letters and the other hebrew.
+    """
+    layers: dict[str, list[str]] = {}
+
+    # Fast path: track single-layer key to skip dict iteration for single-script text.
+    single_layer_key: str | None = None
+    multi_layer: bool = False
+
+    # Cache the last character_range and its resolved layer to avoid repeated
+    # is_suspiciously_successive_range calls for consecutive same-range chars.
+    prev_character_range: str | None = None
+    prev_layer_target: str | None = None
+
+    for character in decoded_sequence:
+        if character.isalpha() is False:
+            continue
+
+        # ASCII fast-path: a-z and A-Z are always "Basic Latin".
+        # Avoids unicode_range() function call overhead for the most common case.
+        character_ord: int = ord(character)
+        if character_ord < 128:
+            character_range: str | None = "Basic Latin"
+        else:
+            character_range = unicode_range(character)
+
+        if character_range is None:
+            continue
+
+        # Fast path: same range as previous character → reuse cached layer target.
+        if character_range == prev_character_range:
+            if prev_layer_target is not None:
+                layers[prev_layer_target].append(character)
+            continue
+
+        layer_target_range: str | None = None
+
+        if multi_layer:
+            for discovered_range in layers:
+                if (
+                    is_suspiciously_successive_range(discovered_range, character_range)
+                    is False
+                ):
+                    layer_target_range = discovered_range
+                    break
+        elif single_layer_key is not None:
+            if (
+                is_suspiciously_successive_range(single_layer_key, character_range)
+                is False
+            ):
+                layer_target_range = single_layer_key
+
+        if layer_target_range is None:
+            layer_target_range = character_range
+
+        if layer_target_range not in layers:
+            layers[layer_target_range] = []
+            if single_layer_key is None:
+                single_layer_key = layer_target_range
+            else:
+                multi_layer = True
+
+        layers[layer_target_range].append(character)
+
+        # Cache for next iteration
+        prev_character_range = character_range
+        prev_layer_target = layer_target_range
+
+    return ["".join(chars).lower() for chars in layers.values()]
+
+
+def merge_coherence_ratios(results: list[CoherenceMatches]) -> CoherenceMatches:
+    """
+    This function merge results previously given by the function coherence_ratio.
+    The return type is the same as coherence_ratio.
+    """
+    per_language_ratios: dict[str, list[float]] = {}
+    for result in results:
+        for sub_result in result:
+            language, ratio = sub_result
+            if language not in per_language_ratios:
+                per_language_ratios[language] = [ratio]
+                continue
+            per_language_ratios[language].append(ratio)
+
+    merge = [
+        (
+            language,
+            round(
+                sum(per_language_ratios[language]) / len(per_language_ratios[language]),
+                4,
+            ),
+        )
+        for language in per_language_ratios
+    ]
+
+    return sorted(merge, key=lambda x: x[1], reverse=True)
+
+
+def filter_alt_coherence_matches(results: CoherenceMatches) -> CoherenceMatches:
+    """
+    We shall NOT return "English—" in CoherenceMatches because it is an alternative
+    of "English". This function only keeps the best match and remove the em-dash in it.
+    """
+    index_results: dict[str, list[float]] = dict()
+
+    for result in results:
+        language, ratio = result
+        no_em_name: str = language.replace("—", "")
+
+        if no_em_name not in index_results:
+            index_results[no_em_name] = []
+
+        index_results[no_em_name].append(ratio)
+
+    if any(len(index_results[e]) > 1 for e in index_results):
+        filtered_results: CoherenceMatches = []
+
+        for language in index_results:
+            filtered_results.append((language, max(index_results[language])))
+
+        return filtered_results
+
+    return results
+
+
+@lru_cache(maxsize=2048)
+def coherence_ratio(
+    decoded_sequence: str, threshold: float = 0.1, lg_inclusion: str | None = None
+) -> CoherenceMatches:
+    """
+    Detect ANY language that can be identified in given sequence. The sequence will be analysed by layers.
+    A layer = Character extraction by alphabets/ranges.
+    """
+
+    results: list[tuple[str, float]] = []
+    ignore_non_latin: bool = False
+
+    sufficient_match_count: int = 0
+
+    lg_inclusion_list = lg_inclusion.split(",") if lg_inclusion is not None else []
+    if "Latin Based" in lg_inclusion_list:
+        ignore_non_latin = True
+        lg_inclusion_list.remove("Latin Based")
+
+    for layer in alpha_unicode_split(decoded_sequence):
+        sequence_frequencies: TypeCounter[str] = Counter(layer)
+        most_common = sequence_frequencies.most_common()
+
+        character_count: int = len(layer)
+
+        if character_count <= TOO_SMALL_SEQUENCE:
+            continue
+
+        popular_character_ordered: list[str] = [c for c, o in most_common]
+
+        for language in lg_inclusion_list or alphabet_languages(
+            popular_character_ordered, ignore_non_latin
+        ):
+            ratio: float = characters_popularity_compare(
+                language, popular_character_ordered
+            )
+
+            if ratio < threshold:
+                continue
+            elif ratio >= 0.8:
+                sufficient_match_count += 1
+
+            results.append((language, round(ratio, 4)))
+
+            if sufficient_match_count >= 3:
+                break
+
+    return sorted(
+        filter_alt_coherence_matches(results), key=lambda x: x[1], reverse=True
+    )

.venv/lib/python3.9/site-packages/charset_normalizer/cli/__init__.py

@@ -0,0 +1,8 @@
+from __future__ import annotations
+
+from .__main__ import cli_detect, query_yes_no
+
+__all__ = (
+    "cli_detect",
+    "query_yes_no",
+)

.venv/lib/python3.9/site-packages/charset_normalizer/cli/__main__.py

@@ -0,0 +1,362 @@
+from __future__ import annotations
+
+import argparse
+import sys
+import typing
+from json import dumps
+from os.path import abspath, basename, dirname, join, realpath
+from platform import python_version
+from unicodedata import unidata_version
+
+import charset_normalizer.md as md_module
+from charset_normalizer import from_fp
+from charset_normalizer.models import CliDetectionResult
+from charset_normalizer.version import __version__
+
+
+def query_yes_no(question: str, default: str = "yes") -> bool:  # Defensive:
+    """Ask a yes/no question via input() and return the answer as a bool."""
+    prompt = " [Y/n] " if default == "yes" else " [y/N] "
+
+    while True:
+        choice = input(question + prompt).strip().lower()
+        if not choice:
+            return default == "yes"
+        if choice in ("y", "yes"):
+            return True
+        if choice in ("n", "no"):
+            return False
+        print("Please respond with 'y' or 'n'.")
+
+
+class FileType:
+    """Factory for creating file object types
+
+    Instances of FileType are typically passed as type= arguments to the
+    ArgumentParser add_argument() method.
+
+    Keyword Arguments:
+        - mode -- A string indicating how the file is to be opened. Accepts the
+            same values as the builtin open() function.
+        - bufsize -- The file's desired buffer size. Accepts the same values as
+            the builtin open() function.
+        - encoding -- The file's encoding. Accepts the same values as the
+            builtin open() function.
+        - errors -- A string indicating how encoding and decoding errors are to
+            be handled. Accepts the same value as the builtin open() function.
+
+    Backported from CPython 3.12
+    """
+
+    def __init__(
+        self,
+        mode: str = "r",
+        bufsize: int = -1,
+        encoding: str | None = None,
+        errors: str | None = None,
+    ):
+        self._mode = mode
+        self._bufsize = bufsize
+        self._encoding = encoding
+        self._errors = errors
+
+    def __call__(self, string: str) -> typing.IO:  # type: ignore[type-arg]
+        # the special argument "-" means sys.std{in,out}
+        if string == "-":
+            if "r" in self._mode:
+                return sys.stdin.buffer if "b" in self._mode else sys.stdin
+            elif any(c in self._mode for c in "wax"):
+                return sys.stdout.buffer if "b" in self._mode else sys.stdout
+            else:
+                msg = f'argument "-" with mode {self._mode}'
+                raise ValueError(msg)
+
+        # all other arguments are used as file names
+        try:
+            return open(string, self._mode, self._bufsize, self._encoding, self._errors)
+        except OSError as e:
+            message = f"can't open '{string}': {e}"
+            raise argparse.ArgumentTypeError(message)
+
+    def __repr__(self) -> str:
+        args = self._mode, self._bufsize
+        kwargs = [("encoding", self._encoding), ("errors", self._errors)]
+        args_str = ", ".join(
+            [repr(arg) for arg in args if arg != -1]
+            + [f"{kw}={arg!r}" for kw, arg in kwargs if arg is not None]
+        )
+        return f"{type(self).__name__}({args_str})"
+
+
+def cli_detect(argv: list[str] | None = None) -> int:
+    """
+    CLI assistant using ARGV and ArgumentParser
+    :param argv:
+    :return: 0 if everything is fine, anything else equal trouble
+    """
+    parser = argparse.ArgumentParser(
+        description="The Real First Universal Charset Detector. "
+        "Discover originating encoding used on text file. "
+        "Normalize text to unicode."
+    )
+
+    parser.add_argument(
+        "files", type=FileType("rb"), nargs="+", help="File(s) to be analysed"
+    )
+    parser.add_argument(
+        "-v",
+        "--verbose",
+        action="store_true",
+        default=False,
+        dest="verbose",
+        help="Display complementary information about file if any. "
+        "Stdout will contain logs about the detection process.",
+    )
+    parser.add_argument(
+        "-a",
+        "--with-alternative",
+        action="store_true",
+        default=False,
+        dest="alternatives",
+        help="Output complementary possibilities if any. Top-level JSON WILL be a list.",
+    )
+    parser.add_argument(
+        "-n",
+        "--normalize",
+        action="store_true",
+        default=False,
+        dest="normalize",
+        help="Permit to normalize input file. If not set, program does not write anything.",
+    )
+    parser.add_argument(
+        "-m",
+        "--minimal",
+        action="store_true",
+        default=False,
+        dest="minimal",
+        help="Only output the charset detected to STDOUT. Disabling JSON output.",
+    )
+    parser.add_argument(
+        "-r",
+        "--replace",
+        action="store_true",
+        default=False,
+        dest="replace",
+        help="Replace file when trying to normalize it instead of creating a new one.",
+    )
+    parser.add_argument(
+        "-f",
+        "--force",
+        action="store_true",
+        default=False,
+        dest="force",
+        help="Replace file without asking if you are sure, use this flag with caution.",
+    )
+    parser.add_argument(
+        "-i",
+        "--no-preemptive",
+        action="store_true",
+        default=False,
+        dest="no_preemptive",
+        help="Disable looking at a charset declaration to hint the detector.",
+    )
+    parser.add_argument(
+        "-t",
+        "--threshold",
+        action="store",
+        default=0.2,
+        type=float,
+        dest="threshold",
+        help="Define a custom maximum amount of noise allowed in decoded content. 0. <= noise <= 1.",
+    )
+    parser.add_argument(
+        "--version",
+        action="version",
+        version="Charset-Normalizer {} - Python {} - Unicode {} - SpeedUp {}".format(
+            __version__,
+            python_version(),
+            unidata_version,
+            "OFF" if md_module.__file__.lower().endswith(".py") else "ON",
+        ),
+        help="Show version information and exit.",
+    )
+
+    args = parser.parse_args(argv)
+
+    if args.replace is True and args.normalize is False:
+        if args.files:
+            for my_file in args.files:
+                my_file.close()
+        print("Use --replace in addition of --normalize only.", file=sys.stderr)
+        return 1
+
+    if args.force is True and args.replace is False:
+        if args.files:
+            for my_file in args.files:
+                my_file.close()
+        print("Use --force in addition of --replace only.", file=sys.stderr)
+        return 1
+
+    if args.threshold < 0.0 or args.threshold > 1.0:
+        if args.files:
+            for my_file in args.files:
+                my_file.close()
+        print("--threshold VALUE should be between 0. AND 1.", file=sys.stderr)
+        return 1
+
+    x_ = []
+
+    for my_file in args.files:
+        matches = from_fp(
+            my_file,
+            threshold=args.threshold,
+            explain=args.verbose,
+            preemptive_behaviour=args.no_preemptive is False,
+        )
+
+        best_guess = matches.best()
+
+        if best_guess is None:
+            print(
+                'Unable to identify originating encoding for "{}". {}'.format(
+                    my_file.name,
+                    (
+                        "Maybe try increasing maximum amount of chaos."
+                        if args.threshold < 1.0
+                        else ""
+                    ),
+                ),
+                file=sys.stderr,
+            )
+            x_.append(
+                CliDetectionResult(
+                    abspath(my_file.name),
+                    None,
+                    [],
+                    [],
+                    "Unknown",
+                    [],
+                    False,
+                    1.0,
+                    0.0,
+                    None,
+                    True,
+                )
+            )
+        else:
+            cli_result = CliDetectionResult(
+                abspath(my_file.name),
+                best_guess.encoding,
+                best_guess.encoding_aliases,
+                [
+                    cp
+                    for cp in best_guess.could_be_from_charset
+                    if cp != best_guess.encoding
+                ],
+                best_guess.language,
+                best_guess.alphabets,
+                best_guess.bom,
+                best_guess.percent_chaos,
+                best_guess.percent_coherence,
+                None,
+                True,
+            )
+            x_.append(cli_result)
+
+            if len(matches) > 1 and args.alternatives:
+                for el in matches:
+                    if el != best_guess:
+                        x_.append(
+                            CliDetectionResult(
+                                abspath(my_file.name),
+                                el.encoding,
+                                el.encoding_aliases,
+                                [
+                                    cp
+                                    for cp in el.could_be_from_charset
+                                    if cp != el.encoding
+                                ],
+                                el.language,
+                                el.alphabets,
+                                el.bom,
+                                el.percent_chaos,
+                                el.percent_coherence,
+                                None,
+                                False,
+                            )
+                        )
+
+            if args.normalize is True:
+                if best_guess.encoding.startswith("utf") is True:
+                    print(
+                        '"{}" file does not need to be normalized, as it already came from unicode.'.format(
+                            my_file.name
+                        ),
+                        file=sys.stderr,
+                    )
+                    if my_file.closed is False:
+                        my_file.close()
+                    continue
+
+                dir_path = dirname(realpath(my_file.name))
+                file_name = basename(realpath(my_file.name))
+
+                o_: list[str] = file_name.split(".")
+
+                if args.replace is False:
+                    o_.insert(-1, best_guess.encoding)
+                    if my_file.closed is False:
+                        my_file.close()
+                elif (
+                    args.force is False
+                    and query_yes_no(
+                        'Are you sure to normalize "{}" by replacing it ?'.format(
+                            my_file.name
+                        ),
+                        "no",
+                    )
+                    is False
+                ):
+                    if my_file.closed is False:
+                        my_file.close()
+                    continue
+
+                try:
+                    cli_result.unicode_path = join(dir_path, ".".join(o_))
+
+                    with open(cli_result.unicode_path, "wb") as fp:
+                        fp.write(best_guess.output())
+                except OSError as e:  # Defensive:
+                    print(str(e), file=sys.stderr)
+                    if my_file.closed is False:
+                        my_file.close()
+                    return 2
+
+        if my_file.closed is False:
+            my_file.close()
+
+    if args.minimal is False:
+        print(
+            dumps(
+                [el.__dict__ for el in x_] if len(x_) > 1 else x_[0].__dict__,
+                ensure_ascii=True,
+                indent=4,
+            )
+        )
+    else:
+        for my_file in args.files:
+            print(
+                ", ".join(
+                    [
+                        el.encoding or "undefined"
+                        for el in x_
+                        if el.path == abspath(my_file.name)
+                    ]
+                )
+            )
+
+    return 0
+
+
+if __name__ == "__main__":  # Defensive:
+    cli_detect()

.venv/lib/python3.9/site-packages/charset_normalizer/legacy.py

@@ -0,0 +1,79 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+from warnings import warn
+
+from .api import from_bytes
+from .constant import CHARDET_CORRESPONDENCE, TOO_SMALL_SEQUENCE
+
+if TYPE_CHECKING:
+    from typing import TypedDict
+
+    class ResultDict(TypedDict):
+        encoding: str | None
+        language: str
+        confidence: float | None
+
+
+def detect(
+    byte_str: bytes, should_rename_legacy: bool = False, **kwargs: Any
+) -> ResultDict:
+    """
+    chardet legacy method
+    Detect the encoding of the given byte string. It should be mostly backward-compatible.
+    Encoding name will match Chardet own writing whenever possible. (Not on encoding name unsupported by it)
+    This function is deprecated and should be used to migrate your project easily, consult the documentation for
+    further information. Not planned for removal.
+
+    :param byte_str:     The byte sequence to examine.
+    :param should_rename_legacy:  Should we rename legacy encodings
+                                  to their more modern equivalents?
+    """
+    if len(kwargs):
+        warn(
+            f"charset-normalizer disregard arguments '{','.join(list(kwargs.keys()))}' in legacy function detect()"
+        )
+
+    if not isinstance(byte_str, (bytearray, bytes)):
+        raise TypeError(  # pragma: nocover
+            f"Expected object of type bytes or bytearray, got: {type(byte_str)}"
+        )
+
+    if isinstance(byte_str, bytearray):
+        byte_str = bytes(byte_str)
+
+    r = from_bytes(byte_str).best()
+
+    encoding = r.encoding if r is not None else None
+    language = r.language if r is not None and r.language != "Unknown" else ""
+    confidence = 1.0 - r.chaos if r is not None else None
+
+    # automatically lower confidence
+    # on small bytes samples.
+    # https://github.com/jawah/charset_normalizer/issues/391
+    if (
+        confidence is not None
+        and confidence >= 0.9
+        and encoding
+        not in {
+            "utf_8",
+            "ascii",
+        }
+        and r.bom is False  # type: ignore[union-attr]
+        and len(byte_str) < TOO_SMALL_SEQUENCE
+    ):
+        confidence -= 0.2
+
+    # Note: CharsetNormalizer does not return 'UTF-8-SIG' as the sig get stripped in the detection/normalization process
+    # but chardet does return 'utf-8-sig' and it is a valid codec name.
+    if r is not None and encoding == "utf_8" and r.bom:
+        encoding += "_sig"
+
+    if should_rename_legacy is False and encoding in CHARDET_CORRESPONDENCE:
+        encoding = CHARDET_CORRESPONDENCE[encoding]
+
+    return {
+        "encoding": encoding,
+        "language": language,
+        "confidence": confidence,
+    }

.venv/lib/python3.9/site-packages/charset_normalizer/md.py

@@ -0,0 +1,936 @@
+from __future__ import annotations
+
+import sys
+from functools import lru_cache
+from logging import getLogger
+
+if sys.version_info >= (3, 8):
+    from typing import final
+else:
+    try:
+        from typing_extensions import final
+    except ImportError:
+
+        def final(cls):  # type: ignore[misc,no-untyped-def]
+            return cls
+
+
+from .constant import (
+    COMMON_CJK_CHARACTERS,
+    COMMON_SAFE_ASCII_CHARACTERS,
+    TRACE,
+    UNICODE_SECONDARY_RANGE_KEYWORD,
+    _ACCENTUATED,
+    _ARABIC,
+    _ARABIC_ISOLATED_FORM,
+    _CJK,
+    _HANGUL,
+    _HIRAGANA,
+    _KATAKANA,
+    _LATIN,
+    _THAI,
+)
+from .utils import (
+    _character_flags,
+    is_emoticon,
+    is_punctuation,
+    is_separator,
+    is_symbol,
+    remove_accent,
+    unicode_range,
+)
+
+# Combined bitmask for CJK/Hangul/Katakana/Hiragana/Thai glyph detection.
+_GLYPH_MASK: int = _CJK | _HANGUL | _KATAKANA | _HIRAGANA | _THAI
+
+
+@final
+class CharInfo:
+    """Pre-computed character properties shared across all detectors.
+
+    Instantiated once and reused via :meth:`update` on every character
+    in the hot loop so that redundant calls to str methods
+    (``isalpha``, ``isupper``, …) and cached utility functions
+    (``_character_flags``, ``is_punctuation``, …) are avoided when
+    several plugins need the same information.
+    """
+
+    __slots__ = (
+        "character",
+        "printable",
+        "alpha",
+        "upper",
+        "lower",
+        "space",
+        "digit",
+        "is_ascii",
+        "case_variable",
+        "flags",
+        "accentuated",
+        "latin",
+        "is_cjk",
+        "is_arabic",
+        "is_glyph",
+        "punct",
+        "sym",
+    )
+
+    def __init__(self) -> None:
+        self.character: str = ""
+        self.printable: bool = False
+        self.alpha: bool = False
+        self.upper: bool = False
+        self.lower: bool = False
+        self.space: bool = False
+        self.digit: bool = False
+        self.is_ascii: bool = False
+        self.case_variable: bool = False
+        self.flags: int = 0
+        self.accentuated: bool = False
+        self.latin: bool = False
+        self.is_cjk: bool = False
+        self.is_arabic: bool = False
+        self.is_glyph: bool = False
+        self.punct: bool = False
+        self.sym: bool = False
+
+    def update(self, character: str) -> None:
+        """Update all properties for *character* (called once per character)."""
+        self.character = character
+
+        # ASCII fast-path: for characters with ord < 128, we can skip
+        # _character_flags() entirely and derive most properties from ord.
+        o: int = ord(character)
+        if o < 128:
+            self.is_ascii = True
+            self.accentuated = False
+            self.is_cjk = False
+            self.is_arabic = False
+            self.is_glyph = False
+            # ASCII alpha: a-z (97-122) or A-Z (65-90)
+            if 65 <= o <= 90:
+                # Uppercase ASCII letter
+                self.alpha = True
+                self.upper = True
+                self.lower = False
+                self.space = False
+                self.digit = False
+                self.printable = True
+                self.case_variable = True
+                self.flags = _LATIN
+                self.latin = True
+                self.punct = False
+                self.sym = False
+            elif 97 <= o <= 122:
+                # Lowercase ASCII letter
+                self.alpha = True
+                self.upper = False
+                self.lower = True
+                self.space = False
+                self.digit = False
+                self.printable = True
+                self.case_variable = True
+                self.flags = _LATIN
+                self.latin = True
+                self.punct = False
+                self.sym = False
+            elif 48 <= o <= 57:
+                # ASCII digit 0-9
+                self.alpha = False
+                self.upper = False
+                self.lower = False
+                self.space = False
+                self.digit = True
+                self.printable = True
+                self.case_variable = False
+                self.flags = 0
+                self.latin = False
+                self.punct = False
+                self.sym = False
+            elif o == 32 or (9 <= o <= 13):
+                # Space, tab, newline, etc.
+                self.alpha = False
+                self.upper = False
+                self.lower = False
+                self.space = True
+                self.digit = False
+                self.printable = o == 32
+                self.case_variable = False
+                self.flags = 0
+                self.latin = False
+                self.punct = False
+                self.sym = False
+            else:
+                # Other ASCII (punctuation, symbols, control chars)
+                self.printable = character.isprintable()
+                self.alpha = False
+                self.upper = False
+                self.lower = False
+                self.space = False
+                self.digit = False
+                self.case_variable = False
+                self.flags = 0
+                self.latin = False
+                self.punct = is_punctuation(character) if self.printable else False
+                self.sym = is_symbol(character) if self.printable else False
+        else:
+            # Non-ASCII path
+            self.is_ascii = False
+            self.printable = character.isprintable()
+            self.alpha = character.isalpha()
+            self.upper = character.isupper()
+            self.lower = character.islower()
+            self.space = character.isspace()
+            self.digit = character.isdigit()
+            self.case_variable = self.lower != self.upper
+
+            # Flag-based classification (single unicodedata.name() call, lru-cached)
+            flags: int
+            if self.alpha:
+                flags = _character_flags(character)
+            else:
+                flags = 0
+            self.flags = flags
+            self.accentuated = bool(flags & _ACCENTUATED)
+            self.latin = bool(flags & _LATIN)
+            self.is_cjk = bool(flags & _CJK)
+            self.is_arabic = bool(flags & _ARABIC)
+            self.is_glyph = bool(flags & _GLYPH_MASK)
+
+            # Eagerly compute punct and sym (avoids property dispatch overhead
+            # on 300K+ accesses in the hot loop).
+            self.punct = is_punctuation(character) if self.printable else False
+            self.sym = is_symbol(character) if self.printable else False
+
+
+class MessDetectorPlugin:
+    """
+    Base abstract class used for mess detection plugins.
+    All detectors MUST extend and implement given methods.
+    """
+
+    __slots__ = ()
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """
+        The main routine to be executed upon character.
+        Insert the logic in witch the text would be considered chaotic.
+        """
+        raise NotImplementedError  # Defensive:
+
+    def reset(self) -> None:  # Defensive:
+        """
+        Permit to reset the plugin to the initial state.
+        """
+        raise NotImplementedError
+
+    @property
+    def ratio(self) -> float:
+        """
+        Compute the chaos ratio based on what your feed() has seen.
+        Must NOT be lower than 0.; No restriction gt 0.
+        """
+        raise NotImplementedError  # Defensive:
+
+
+@final
+class TooManySymbolOrPunctuationPlugin(MessDetectorPlugin):
+    __slots__ = (
+        "_punctuation_count",
+        "_symbol_count",
+        "_character_count",
+        "_last_printable_char",
+        "_frenzy_symbol_in_word",
+    )
+
+    def __init__(self) -> None:
+        self._punctuation_count: int = 0
+        self._symbol_count: int = 0
+        self._character_count: int = 0
+
+        self._last_printable_char: str | None = None
+        self._frenzy_symbol_in_word: bool = False
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        self._character_count += 1
+
+        if (
+            character != self._last_printable_char
+            and character not in COMMON_SAFE_ASCII_CHARACTERS
+        ):
+            if info.punct:
+                self._punctuation_count += 1
+            elif not info.digit and info.sym and not is_emoticon(character):
+                self._symbol_count += 2
+
+        self._last_printable_char = character
+
+    def reset(self) -> None:  # Abstract
+        self._punctuation_count = 0
+        self._character_count = 0
+        self._symbol_count = 0
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count == 0:
+            return 0.0
+
+        ratio_of_punctuation: float = (
+            self._punctuation_count + self._symbol_count
+        ) / self._character_count
+
+        return ratio_of_punctuation if ratio_of_punctuation >= 0.3 else 0.0
+
+
+@final
+class TooManyAccentuatedPlugin(MessDetectorPlugin):
+    __slots__ = ("_character_count", "_accentuated_count")
+
+    def __init__(self) -> None:
+        self._character_count: int = 0
+        self._accentuated_count: int = 0
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        self._character_count += 1
+
+        if info.accentuated:
+            self._accentuated_count += 1
+
+    def reset(self) -> None:  # Abstract
+        self._character_count = 0
+        self._accentuated_count = 0
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count < 8:
+            return 0.0
+
+        ratio_of_accentuation: float = self._accentuated_count / self._character_count
+        return ratio_of_accentuation if ratio_of_accentuation >= 0.35 else 0.0
+
+
+@final
+class UnprintablePlugin(MessDetectorPlugin):
+    __slots__ = ("_unprintable_count", "_character_count")
+
+    def __init__(self) -> None:
+        self._unprintable_count: int = 0
+        self._character_count: int = 0
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        if (
+            not info.space
+            and not info.printable
+            and character != "\x1a"
+            and character != "\ufeff"
+        ):
+            self._unprintable_count += 1
+        self._character_count += 1
+
+    def reset(self) -> None:  # Abstract
+        self._unprintable_count = 0
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count == 0:  # Defensive:
+            return 0.0
+
+        return (self._unprintable_count * 8) / self._character_count
+
+
+@final
+class SuspiciousDuplicateAccentPlugin(MessDetectorPlugin):
+    __slots__ = (
+        "_successive_count",
+        "_character_count",
+        "_last_latin_character",
+        "_last_was_accentuated",
+    )
+
+    def __init__(self) -> None:
+        self._successive_count: int = 0
+        self._character_count: int = 0
+
+        self._last_latin_character: str | None = None
+        self._last_was_accentuated: bool = False
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        self._character_count += 1
+        if (
+            self._last_latin_character is not None
+            and info.accentuated
+            and self._last_was_accentuated
+        ):
+            if info.upper and self._last_latin_character.isupper():
+                self._successive_count += 1
+            if remove_accent(character) == remove_accent(self._last_latin_character):
+                self._successive_count += 1
+        self._last_latin_character = character
+        self._last_was_accentuated = info.accentuated
+
+    def reset(self) -> None:  # Abstract
+        self._successive_count = 0
+        self._character_count = 0
+        self._last_latin_character = None
+        self._last_was_accentuated = False
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count == 0:
+            return 0.0
+
+        return (self._successive_count * 2) / self._character_count
+
+
+@final
+class SuspiciousRange(MessDetectorPlugin):
+    __slots__ = (
+        "_suspicious_successive_range_count",
+        "_character_count",
+        "_last_printable_seen",
+        "_last_printable_range",
+    )
+
+    def __init__(self) -> None:
+        self._suspicious_successive_range_count: int = 0
+        self._character_count: int = 0
+        self._last_printable_seen: str | None = None
+        self._last_printable_range: str | None = None
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        self._character_count += 1
+
+        if info.space or info.punct or character in COMMON_SAFE_ASCII_CHARACTERS:
+            self._last_printable_seen = None
+            self._last_printable_range = None
+            return
+
+        if self._last_printable_seen is None:
+            self._last_printable_seen = character
+            self._last_printable_range = unicode_range(character)
+            return
+
+        unicode_range_a: str | None = self._last_printable_range
+        unicode_range_b: str | None = unicode_range(character)
+
+        if is_suspiciously_successive_range(unicode_range_a, unicode_range_b):
+            self._suspicious_successive_range_count += 1
+
+        self._last_printable_seen = character
+        self._last_printable_range = unicode_range_b
+
+    def reset(self) -> None:  # Abstract
+        self._character_count = 0
+        self._suspicious_successive_range_count = 0
+        self._last_printable_seen = None
+        self._last_printable_range = None
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count <= 13:
+            return 0.0
+
+        ratio_of_suspicious_range_usage: float = (
+            self._suspicious_successive_range_count * 2
+        ) / self._character_count
+
+        return ratio_of_suspicious_range_usage
+
+
+@final
+class SuperWeirdWordPlugin(MessDetectorPlugin):
+    __slots__ = (
+        "_word_count",
+        "_bad_word_count",
+        "_foreign_long_count",
+        "_is_current_word_bad",
+        "_foreign_long_watch",
+        "_character_count",
+        "_bad_character_count",
+        "_buffer_length",
+        "_buffer_last_char",
+        "_buffer_last_char_accentuated",
+        "_buffer_accent_count",
+        "_buffer_glyph_count",
+        "_buffer_upper_count",
+    )
+
+    def __init__(self) -> None:
+        self._word_count: int = 0
+        self._bad_word_count: int = 0
+        self._foreign_long_count: int = 0
+
+        self._is_current_word_bad: bool = False
+        self._foreign_long_watch: bool = False
+
+        self._character_count: int = 0
+        self._bad_character_count: int = 0
+
+        self._buffer_length: int = 0
+        self._buffer_last_char: str | None = None
+        self._buffer_last_char_accentuated: bool = False
+        self._buffer_accent_count: int = 0
+        self._buffer_glyph_count: int = 0
+        self._buffer_upper_count: int = 0
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        if info.alpha:
+            self._buffer_length += 1
+            self._buffer_last_char = character
+
+            if info.upper:
+                self._buffer_upper_count += 1
+
+            self._buffer_last_char_accentuated = info.accentuated
+
+            if info.accentuated:
+                self._buffer_accent_count += 1
+            if (
+                not self._foreign_long_watch
+                and (not info.latin or info.accentuated)
+                and not info.is_glyph
+            ):
+                self._foreign_long_watch = True
+            if info.is_glyph:
+                self._buffer_glyph_count += 1
+            return
+        if not self._buffer_length:
+            return
+        if info.space or info.punct or is_separator(character):
+            self._word_count += 1
+            buffer_length: int = self._buffer_length
+
+            self._character_count += buffer_length
+
+            if buffer_length >= 4:
+                if self._buffer_accent_count / buffer_length >= 0.5:
+                    self._is_current_word_bad = True
+                elif (
+                    self._buffer_last_char_accentuated
+                    and self._buffer_last_char.isupper()  # type: ignore[union-attr]
+                    and self._buffer_upper_count != buffer_length
+                ):
+                    self._foreign_long_count += 1
+                    self._is_current_word_bad = True
+                elif self._buffer_glyph_count == 1:
+                    self._is_current_word_bad = True
+                    self._foreign_long_count += 1
+            if buffer_length >= 24 and self._foreign_long_watch:
+                probable_camel_cased: bool = (
+                    self._buffer_upper_count > 0
+                    and self._buffer_upper_count / buffer_length <= 0.3
+                )
+
+                if not probable_camel_cased:
+                    self._foreign_long_count += 1
+                    self._is_current_word_bad = True
+
+            if self._is_current_word_bad:
+                self._bad_word_count += 1
+                self._bad_character_count += buffer_length
+                self._is_current_word_bad = False
+
+            self._foreign_long_watch = False
+            self._buffer_length = 0
+            self._buffer_last_char = None
+            self._buffer_last_char_accentuated = False
+            self._buffer_accent_count = 0
+            self._buffer_glyph_count = 0
+            self._buffer_upper_count = 0
+        elif (
+            character not in {"<", ">", "-", "=", "~", "|", "_"}
+            and not info.digit
+            and info.sym
+        ):
+            self._is_current_word_bad = True
+            self._buffer_length += 1
+            self._buffer_last_char = character
+            self._buffer_last_char_accentuated = False
+
+    def reset(self) -> None:  # Abstract
+        self._buffer_length = 0
+        self._buffer_last_char = None
+        self._buffer_last_char_accentuated = False
+        self._is_current_word_bad = False
+        self._foreign_long_watch = False
+        self._bad_word_count = 0
+        self._word_count = 0
+        self._character_count = 0
+        self._bad_character_count = 0
+        self._foreign_long_count = 0
+        self._buffer_accent_count = 0
+        self._buffer_glyph_count = 0
+        self._buffer_upper_count = 0
+
+    @property
+    def ratio(self) -> float:
+        if self._word_count <= 10 and self._foreign_long_count == 0:
+            return 0.0
+
+        return self._bad_character_count / self._character_count
+
+
+@final
+class CjkUncommonPlugin(MessDetectorPlugin):
+    """
+    Detect messy CJK text that probably means nothing.
+    """
+
+    __slots__ = ("_character_count", "_uncommon_count")
+
+    def __init__(self) -> None:
+        self._character_count: int = 0
+        self._uncommon_count: int = 0
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        self._character_count += 1
+
+        if character not in COMMON_CJK_CHARACTERS:
+            self._uncommon_count += 1
+
+    def reset(self) -> None:  # Abstract
+        self._character_count = 0
+        self._uncommon_count = 0
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count < 8:
+            return 0.0
+
+        uncommon_form_usage: float = self._uncommon_count / self._character_count
+
+        # we can be pretty sure it's garbage when uncommon characters are widely
+        # used. otherwise it could just be traditional chinese for example.
+        return uncommon_form_usage / 10 if uncommon_form_usage > 0.5 else 0.0
+
+
+@final
+class ArchaicUpperLowerPlugin(MessDetectorPlugin):
+    __slots__ = (
+        "_buf",
+        "_character_count_since_last_sep",
+        "_successive_upper_lower_count",
+        "_successive_upper_lower_count_final",
+        "_character_count",
+        "_last_alpha_seen",
+        "_last_alpha_seen_upper",
+        "_last_alpha_seen_lower",
+        "_current_ascii_only",
+    )
+
+    def __init__(self) -> None:
+        self._buf: bool = False
+
+        self._character_count_since_last_sep: int = 0
+
+        self._successive_upper_lower_count: int = 0
+        self._successive_upper_lower_count_final: int = 0
+
+        self._character_count: int = 0
+
+        self._last_alpha_seen: str | None = None
+        self._last_alpha_seen_upper: bool = False
+        self._last_alpha_seen_lower: bool = False
+        self._current_ascii_only: bool = True
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        is_concerned: bool = info.alpha and info.case_variable
+        chunk_sep: bool = not is_concerned
+
+        if chunk_sep and self._character_count_since_last_sep > 0:
+            if (
+                self._character_count_since_last_sep <= 64
+                and not info.digit
+                and not self._current_ascii_only
+            ):
+                self._successive_upper_lower_count_final += (
+                    self._successive_upper_lower_count
+                )
+
+            self._successive_upper_lower_count = 0
+            self._character_count_since_last_sep = 0
+            self._last_alpha_seen = None
+            self._buf = False
+            self._character_count += 1
+            self._current_ascii_only = True
+
+            return
+
+        if self._current_ascii_only and not info.is_ascii:
+            self._current_ascii_only = False
+
+        if self._last_alpha_seen is not None:
+            if (info.upper and self._last_alpha_seen_lower) or (
+                info.lower and self._last_alpha_seen_upper
+            ):
+                if self._buf:
+                    self._successive_upper_lower_count += 2
+                    self._buf = False
+                else:
+                    self._buf = True
+            else:
+                self._buf = False
+
+        self._character_count += 1
+        self._character_count_since_last_sep += 1
+        self._last_alpha_seen = character
+        self._last_alpha_seen_upper = info.upper
+        self._last_alpha_seen_lower = info.lower
+
+    def reset(self) -> None:  # Abstract
+        self._character_count = 0
+        self._character_count_since_last_sep = 0
+        self._successive_upper_lower_count = 0
+        self._successive_upper_lower_count_final = 0
+        self._last_alpha_seen = None
+        self._last_alpha_seen_upper = False
+        self._last_alpha_seen_lower = False
+        self._buf = False
+        self._current_ascii_only = True
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count == 0:  # Defensive:
+            return 0.0
+
+        return self._successive_upper_lower_count_final / self._character_count
+
+
+@final
+class ArabicIsolatedFormPlugin(MessDetectorPlugin):
+    __slots__ = ("_character_count", "_isolated_form_count")
+
+    def __init__(self) -> None:
+        self._character_count: int = 0
+        self._isolated_form_count: int = 0
+
+    def reset(self) -> None:  # Abstract
+        self._character_count = 0
+        self._isolated_form_count = 0
+
+    def feed_info(self, character: str, info: CharInfo) -> None:
+        """Optimized feed using pre-computed character info."""
+        self._character_count += 1
+
+        if info.flags & _ARABIC_ISOLATED_FORM:
+            self._isolated_form_count += 1
+
+    @property
+    def ratio(self) -> float:
+        if self._character_count < 8:
+            return 0.0
+
+        isolated_form_usage: float = self._isolated_form_count / self._character_count
+
+        return isolated_form_usage
+
+
+@lru_cache(maxsize=1024)
+def is_suspiciously_successive_range(
+    unicode_range_a: str | None, unicode_range_b: str | None
+) -> bool:
+    """
+    Determine if two Unicode range seen next to each other can be considered as suspicious.
+    """
+    if unicode_range_a is None or unicode_range_b is None:
+        return True
+
+    if unicode_range_a == unicode_range_b:
+        return False
+
+    if "Latin" in unicode_range_a and "Latin" in unicode_range_b:
+        return False
+
+    if "Emoticons" in unicode_range_a or "Emoticons" in unicode_range_b:
+        return False
+
+    # Latin characters can be accompanied with a combining diacritical mark
+    # eg. Vietnamese.
+    if ("Latin" in unicode_range_a or "Latin" in unicode_range_b) and (
+        "Combining" in unicode_range_a or "Combining" in unicode_range_b
+    ):
+        return False
+
+    keywords_range_a, keywords_range_b = (
+        unicode_range_a.split(" "),
+        unicode_range_b.split(" "),
+    )
+
+    for el in keywords_range_a:
+        if el in UNICODE_SECONDARY_RANGE_KEYWORD:
+            continue
+        if el in keywords_range_b:
+            return False
+
+    # Japanese Exception
+    range_a_jp_chars, range_b_jp_chars = (
+        unicode_range_a
+        in (
+            "Hiragana",
+            "Katakana",
+        ),
+        unicode_range_b in ("Hiragana", "Katakana"),
+    )
+    if (range_a_jp_chars or range_b_jp_chars) and (
+        "CJK" in unicode_range_a or "CJK" in unicode_range_b
+    ):
+        return False
+    if range_a_jp_chars and range_b_jp_chars:
+        return False
+
+    if "Hangul" in unicode_range_a or "Hangul" in unicode_range_b:
+        if "CJK" in unicode_range_a or "CJK" in unicode_range_b:
+            return False
+        if unicode_range_a == "Basic Latin" or unicode_range_b == "Basic Latin":
+            return False
+
+    # Chinese/Japanese use dedicated range for punctuation and/or separators.
+    if ("CJK" in unicode_range_a or "CJK" in unicode_range_b) or (
+        unicode_range_a in ["Katakana", "Hiragana"]
+        and unicode_range_b in ["Katakana", "Hiragana"]
+    ):
+        if "Punctuation" in unicode_range_a or "Punctuation" in unicode_range_b:
+            return False
+        if "Forms" in unicode_range_a or "Forms" in unicode_range_b:
+            return False
+        if unicode_range_a == "Basic Latin" or unicode_range_b == "Basic Latin":
+            return False
+
+    return True
+
+
+@lru_cache(maxsize=2048)
+def mess_ratio(
+    decoded_sequence: str, maximum_threshold: float = 0.2, debug: bool = False
+) -> float:
+    """
+    Compute a mess ratio given a decoded bytes sequence. The maximum threshold does stop the computation earlier.
+    """
+
+    seq_len: int = len(decoded_sequence)
+
+    if seq_len < 511:
+        step: int = 32
+    elif seq_len < 1024:
+        step = 64
+    else:
+        step = 128
+
+    # Create each detector as a named local variable (unrolled from the generic loop).
+    # This eliminates per-character iteration over the detector list and
+    # per-character eligible() virtual dispatch, while keeping every plugin class
+    # intact and fully readable.
+    d_sp: TooManySymbolOrPunctuationPlugin = TooManySymbolOrPunctuationPlugin()
+    d_ta: TooManyAccentuatedPlugin = TooManyAccentuatedPlugin()
+    d_up: UnprintablePlugin = UnprintablePlugin()
+    d_sda: SuspiciousDuplicateAccentPlugin = SuspiciousDuplicateAccentPlugin()
+    d_sr: SuspiciousRange = SuspiciousRange()
+    d_sw: SuperWeirdWordPlugin = SuperWeirdWordPlugin()
+    d_cu: CjkUncommonPlugin = CjkUncommonPlugin()
+    d_au: ArchaicUpperLowerPlugin = ArchaicUpperLowerPlugin()
+    d_ai: ArabicIsolatedFormPlugin = ArabicIsolatedFormPlugin()
+
+    # Local references for feed_info methods called in the hot loop.
+    d_sp_feed = d_sp.feed_info
+    d_ta_feed = d_ta.feed_info
+    d_up_feed = d_up.feed_info
+    d_sda_feed = d_sda.feed_info
+    d_sr_feed = d_sr.feed_info
+    d_sw_feed = d_sw.feed_info
+    d_cu_feed = d_cu.feed_info
+    d_au_feed = d_au.feed_info
+    d_ai_feed = d_ai.feed_info
+
+    # Single reusable CharInfo object (avoids per-character allocation).
+    info: CharInfo = CharInfo()
+    info_update = info.update
+
+    mean_mess_ratio: float
+
+    for block_start in range(0, seq_len, step):
+        for character in decoded_sequence[block_start : block_start + step]:
+            # Pre-compute all character properties once (shared across all plugins).
+            info_update(character)
+
+            # Detectors with eligible() == always True
+            d_up_feed(character, info)
+            d_sw_feed(character, info)
+            d_au_feed(character, info)
+
+            # Detectors with eligible() == isprintable
+            if info.printable:
+                d_sp_feed(character, info)
+                d_sr_feed(character, info)
+
+            # Detectors with eligible() == isalpha
+            if info.alpha:
+                d_ta_feed(character, info)
+                # SuspiciousDuplicateAccent: isalpha() and is_latin()
+                if info.latin:
+                    d_sda_feed(character, info)
+                # CjkUncommon: is_cjk()
+                if info.is_cjk:
+                    d_cu_feed(character, info)
+                # ArabicIsolatedForm: is_arabic()
+                if info.is_arabic:
+                    d_ai_feed(character, info)
+
+        mean_mess_ratio = (
+            d_sp.ratio
+            + d_ta.ratio
+            + d_up.ratio
+            + d_sda.ratio
+            + d_sr.ratio
+            + d_sw.ratio
+            + d_cu.ratio
+            + d_au.ratio
+            + d_ai.ratio
+        )
+
+        if mean_mess_ratio >= maximum_threshold:
+            break
+    else:
+        # Flush last word buffer in SuperWeirdWordPlugin via trailing newline.
+        info_update("\n")
+        d_sw_feed("\n", info)
+        d_au_feed("\n", info)
+        d_up_feed("\n", info)
+
+        mean_mess_ratio = (
+            d_sp.ratio
+            + d_ta.ratio
+            + d_up.ratio
+            + d_sda.ratio
+            + d_sr.ratio
+            + d_sw.ratio
+            + d_cu.ratio
+            + d_au.ratio
+            + d_ai.ratio
+        )
+
+    if debug:  # Defensive:
+        logger = getLogger("charset_normalizer")
+
+        logger.log(
+            TRACE,
+            "Mess-detector extended-analysis start. "
+            f"intermediary_mean_mess_ratio_calc={step} mean_mess_ratio={mean_mess_ratio} "
+            f"maximum_threshold={maximum_threshold}",
+        )
+
+        if seq_len > 16:
+            logger.log(TRACE, f"Starting with: {decoded_sequence[:16]}")
+            logger.log(TRACE, f"Ending with: {decoded_sequence[-16::]}")
+
+        for dt in [d_sp, d_ta, d_up, d_sda, d_sr, d_sw, d_cu, d_au, d_ai]:
+            logger.log(TRACE, f"{dt.__class__}: {dt.ratio}")
+
+    return round(mean_mess_ratio, 3)

.venv/lib/python3.9/site-packages/charset_normalizer/models.py

@@ -0,0 +1,369 @@
+from __future__ import annotations
+
+from encodings.aliases import aliases
+from json import dumps
+from re import sub
+from typing import Any, Iterator, List, Tuple
+
+from .constant import RE_POSSIBLE_ENCODING_INDICATION, TOO_BIG_SEQUENCE
+from .utils import iana_name, is_multi_byte_encoding, unicode_range
+
+
+class CharsetMatch:
+    def __init__(
+        self,
+        payload: bytes | bytearray,
+        guessed_encoding: str,
+        mean_mess_ratio: float,
+        has_sig_or_bom: bool,
+        languages: CoherenceMatches,
+        decoded_payload: str | None = None,
+        preemptive_declaration: str | None = None,
+    ):
+        self._payload: bytes | bytearray = payload
+
+        self._encoding: str = guessed_encoding
+        self._mean_mess_ratio: float = mean_mess_ratio
+        self._languages: CoherenceMatches = languages
+        self._has_sig_or_bom: bool = has_sig_or_bom
+        self._unicode_ranges: list[str] | None = None
+
+        self._leaves: list[CharsetMatch] = []
+        self._mean_coherence_ratio: float = 0.0
+
+        self._output_payload: bytes | None = None
+        self._output_encoding: str | None = None
+
+        self._string: str | None = decoded_payload
+
+        self._preemptive_declaration: str | None = preemptive_declaration
+
+    def __eq__(self, other: object) -> bool:
+        if not isinstance(other, CharsetMatch):
+            if isinstance(other, str):
+                return iana_name(other) == self.encoding
+            return False
+        return self.encoding == other.encoding and self.fingerprint == other.fingerprint
+
+    def __lt__(self, other: object) -> bool:
+        """
+        Implemented to make sorted available upon CharsetMatches items.
+        """
+        if not isinstance(other, CharsetMatch):
+            raise ValueError
+
+        chaos_difference: float = abs(self.chaos - other.chaos)
+        coherence_difference: float = abs(self.coherence - other.coherence)
+
+        # Below 0.5% difference --> Use Coherence
+        if chaos_difference < 0.005 and coherence_difference > 0.02:
+            return self.coherence > other.coherence
+        elif chaos_difference < 0.005 and coherence_difference <= 0.02:
+            # When having a difficult decision, use the result that decoded as many multi-byte as possible.
+            # preserve RAM usage!
+            if len(self._payload) >= TOO_BIG_SEQUENCE:
+                return self.chaos < other.chaos
+            return self.multi_byte_usage > other.multi_byte_usage
+
+        return self.chaos < other.chaos
+
+    @property
+    def multi_byte_usage(self) -> float:
+        return 1.0 - (len(str(self)) / len(self.raw))
+
+    def __str__(self) -> str:
+        # Lazy Str Loading
+        if self._string is None:
+            self._string = str(self._payload, self._encoding, "strict")
+            # UTF-7 BOM is encoded in modified Base64 whose byte boundary
+            # can overlap with the next character, so raw-byte stripping
+            # is unreliable. Strip the decoded BOM character instead.
+            if (
+                self._has_sig_or_bom
+                and self._encoding == "utf_7"
+                and self._string
+                and self._string[0] == "\ufeff"
+            ):
+                self._string = self._string[1:]
+        return self._string
+
+    def __repr__(self) -> str:
+        return f"<CharsetMatch '{self.encoding}' fp({self.fingerprint})>"
+
+    def add_submatch(self, other: CharsetMatch) -> None:
+        if not isinstance(other, CharsetMatch) or other == self:
+            raise ValueError(
+                "Unable to add instance <{}> as a submatch of a CharsetMatch".format(
+                    other.__class__
+                )
+            )
+
+        other._string = None  # Unload RAM usage; dirty trick.
+        self._leaves.append(other)
+
+    @property
+    def encoding(self) -> str:
+        return self._encoding
+
+    @property
+    def encoding_aliases(self) -> list[str]:
+        """
+        Encoding name are known by many name, using this could help when searching for IBM855 when it's listed as CP855.
+        """
+        also_known_as: list[str] = []
+        for u, p in aliases.items():
+            if self.encoding == u:
+                also_known_as.append(p)
+            elif self.encoding == p:
+                also_known_as.append(u)
+        return also_known_as
+
+    @property
+    def bom(self) -> bool:
+        return self._has_sig_or_bom
+
+    @property
+    def byte_order_mark(self) -> bool:
+        return self._has_sig_or_bom
+
+    @property
+    def languages(self) -> list[str]:
+        """
+        Return the complete list of possible languages found in decoded sequence.
+        Usually not really useful. Returned list may be empty even if 'language' property return something != 'Unknown'.
+        """
+        return [e[0] for e in self._languages]
+
+    @property
+    def language(self) -> str:
+        """
+        Most probable language found in decoded sequence. If none were detected or inferred, the property will return
+        "Unknown".
+        """
+        if not self._languages:
+            # Trying to infer the language based on the given encoding
+            # Its either English or we should not pronounce ourselves in certain cases.
+            if "ascii" in self.could_be_from_charset:
+                return "English"
+
+            # doing it there to avoid circular import
+            from charset_normalizer.cd import encoding_languages, mb_encoding_languages
+
+            languages = (
+                mb_encoding_languages(self.encoding)
+                if is_multi_byte_encoding(self.encoding)
+                else encoding_languages(self.encoding)
+            )
+
+            if len(languages) == 0 or "Latin Based" in languages:
+                return "Unknown"
+
+            return languages[0]
+
+        return self._languages[0][0]
+
+    @property
+    def chaos(self) -> float:
+        return self._mean_mess_ratio
+
+    @property
+    def coherence(self) -> float:
+        if not self._languages:
+            return 0.0
+        return self._languages[0][1]
+
+    @property
+    def percent_chaos(self) -> float:
+        return round(self.chaos * 100, ndigits=3)
+
+    @property
+    def percent_coherence(self) -> float:
+        return round(self.coherence * 100, ndigits=3)
+
+    @property
+    def raw(self) -> bytes | bytearray:
+        """
+        Original untouched bytes.
+        """
+        return self._payload
+
+    @property
+    def submatch(self) -> list[CharsetMatch]:
+        return self._leaves
+
+    @property
+    def has_submatch(self) -> bool:
+        return len(self._leaves) > 0
+
+    @property
+    def alphabets(self) -> list[str]:
+        if self._unicode_ranges is not None:
+            return self._unicode_ranges
+        # list detected ranges
+        detected_ranges: list[str | None] = [unicode_range(char) for char in str(self)]
+        # filter and sort
+        self._unicode_ranges = sorted(list({r for r in detected_ranges if r}))
+        return self._unicode_ranges
+
+    @property
+    def could_be_from_charset(self) -> list[str]:
+        """
+        The complete list of encoding that output the exact SAME str result and therefore could be the originating
+        encoding.
+        This list does include the encoding available in property 'encoding'.
+        """
+        return [self._encoding] + [m.encoding for m in self._leaves]
+
+    def output(self, encoding: str = "utf_8") -> bytes:
+        """
+        Method to get re-encoded bytes payload using given target encoding. Default to UTF-8.
+        Any errors will be simply ignored by the encoder NOT replaced.
+        """
+        if self._output_encoding is None or self._output_encoding != encoding:
+            self._output_encoding = encoding
+            decoded_string = str(self)
+            if (
+                self._preemptive_declaration is not None
+                and self._preemptive_declaration.lower()
+                not in ["utf-8", "utf8", "utf_8"]
+            ):
+                patched_header = sub(
+                    RE_POSSIBLE_ENCODING_INDICATION,
+                    lambda m: m.string[m.span()[0] : m.span()[1]].replace(
+                        m.groups()[0],
+                        iana_name(self._output_encoding).replace("_", "-"),  # type: ignore[arg-type]
+                    ),
+                    decoded_string[:8192],
+                    count=1,
+                )
+
+                decoded_string = patched_header + decoded_string[8192:]
+
+            self._output_payload = decoded_string.encode(encoding, "replace")
+
+        return self._output_payload  # type: ignore
+
+    @property
+    def fingerprint(self) -> int:
+        """
+        Retrieve a hash fingerprint of the decoded payload, used for deduplication.
+        """
+        return hash(str(self))
+
+
+class CharsetMatches:
+    """
+    Container with every CharsetMatch items ordered by default from most probable to the less one.
+    Act like a list(iterable) but does not implements all related methods.
+    """
+
+    def __init__(self, results: list[CharsetMatch] | None = None):
+        self._results: list[CharsetMatch] = sorted(results) if results else []
+
+    def __iter__(self) -> Iterator[CharsetMatch]:
+        yield from self._results
+
+    def __getitem__(self, item: int | str) -> CharsetMatch:
+        """
+        Retrieve a single item either by its position or encoding name (alias may be used here).
+        Raise KeyError upon invalid index or encoding not present in results.
+        """
+        if isinstance(item, int):
+            return self._results[item]
+        if isinstance(item, str):
+            item = iana_name(item, False)
+            for result in self._results:
+                if item in result.could_be_from_charset:
+                    return result
+        raise KeyError
+
+    def __len__(self) -> int:
+        return len(self._results)
+
+    def __bool__(self) -> bool:
+        return len(self._results) > 0
+
+    def append(self, item: CharsetMatch) -> None:
+        """
+        Insert a single match. Will be inserted accordingly to preserve sort.
+        Can be inserted as a submatch.
+        """
+        if not isinstance(item, CharsetMatch):
+            raise ValueError(
+                "Cannot append instance '{}' to CharsetMatches".format(
+                    str(item.__class__)
+                )
+            )
+        # We should disable the submatch factoring when the input file is too heavy (conserve RAM usage)
+        if len(item.raw) < TOO_BIG_SEQUENCE:
+            for match in self._results:
+                if match.fingerprint == item.fingerprint and match.chaos == item.chaos:
+                    match.add_submatch(item)
+                    return
+        self._results.append(item)
+        self._results = sorted(self._results)
+
+    def best(self) -> CharsetMatch | None:
+        """
+        Simply return the first match. Strict equivalent to matches[0].
+        """
+        if not self._results:
+            return None
+        return self._results[0]
+
+    def first(self) -> CharsetMatch | None:
+        """
+        Redundant method, call the method best(). Kept for BC reasons.
+        """
+        return self.best()
+
+
+CoherenceMatch = Tuple[str, float]
+CoherenceMatches = List[CoherenceMatch]
+
+
+class CliDetectionResult:
+    def __init__(
+        self,
+        path: str,
+        encoding: str | None,
+        encoding_aliases: list[str],
+        alternative_encodings: list[str],
+        language: str,
+        alphabets: list[str],
+        has_sig_or_bom: bool,
+        chaos: float,
+        coherence: float,
+        unicode_path: str | None,
+        is_preferred: bool,
+    ):
+        self.path: str = path
+        self.unicode_path: str | None = unicode_path
+        self.encoding: str | None = encoding
+        self.encoding_aliases: list[str] = encoding_aliases
+        self.alternative_encodings: list[str] = alternative_encodings
+        self.language: str = language
+        self.alphabets: list[str] = alphabets
+        self.has_sig_or_bom: bool = has_sig_or_bom
+        self.chaos: float = chaos
+        self.coherence: float = coherence
+        self.is_preferred: bool = is_preferred
+
+    @property
+    def __dict__(self) -> dict[str, Any]:  # type: ignore
+        return {
+            "path": self.path,
+            "encoding": self.encoding,
+            "encoding_aliases": self.encoding_aliases,
+            "alternative_encodings": self.alternative_encodings,
+            "language": self.language,
+            "alphabets": self.alphabets,
+            "has_sig_or_bom": self.has_sig_or_bom,
+            "chaos": self.chaos,
+            "coherence": self.coherence,
+            "unicode_path": self.unicode_path,
+            "is_preferred": self.is_preferred,
+        }
+
+    def to_json(self) -> str:
+        return dumps(self.__dict__, ensure_ascii=True, indent=4)