bpo-34523: Add _PyCoreConfig.filesystem_encoding (GH-8963)

_PyCoreConfig_Read() is now responsible to choose the filesystem
encoding and error handler. Using Py_Main(), the encoding is now
chosen even before calling Py_Initialize().

_PyCoreConfig.filesystem_encoding is now the reference, instead of
Py_FileSystemDefaultEncoding, for the Python filesystem encoding.

Changes:

* Add filesystem_encoding and filesystem_errors to _PyCoreConfig
* _PyCoreConfig_Read() now reads the locale encoding for the file
  system encoding.
* PyUnicode_EncodeFSDefault() and PyUnicode_DecodeFSDefaultAndSize()
  now use the interpreter configuration rather than
  Py_FileSystemDefaultEncoding and Py_FileSystemDefaultEncodeErrors
  global configuration variables.
* Add _Py_SetFileSystemEncoding() and _Py_ClearFileSystemEncoding()
  private functions to only modify Py_FileSystemDefaultEncoding and
  Py_FileSystemDefaultEncodeErrors in coreconfig.c.
* _Py_CoerceLegacyLocale() now takes an int rather than
  _PyCoreConfig for the warning.

Include/coreconfig.h

@@ -66,6 +66,17 @@ typedef struct {
     int coerce_c_locale;    /* PYTHONCOERCECLOCALE, -1 means unknown */
     int coerce_c_locale_warn; /* PYTHONCOERCECLOCALE=warn */
 
+    /* Python filesystem encoding and error handler: see
+       sys.getfilesystemencoding() and sys.getfilesystemencodeerrors().
+
+       Updated later by initfsencoding(). On Windows, can be updated by
+       sys._enablelegacywindowsfsencoding() at runtime.
+
+       See Py_FileSystemDefaultEncoding and Py_FileSystemDefaultEncodeErrors.
+       */
+    char *filesystem_encoding;
+    char *filesystem_errors;
+
     /* Enable UTF-8 mode?
        Set by -X utf8 command line option and PYTHONUTF8 environment variable.
        If set to -1 (default), inherit Py_UTF8Mode value. */
@@ -325,6 +336,14 @@ PyAPI_FUNC(int) _PyCoreConfig_GetEnvDup(
 #endif
 
 
+#ifdef Py_BUILD_CORE
+PyAPI_FUNC(int) _Py_SetFileSystemEncoding(
+    const char *encoding,
+    const char *errors);
+PyAPI_FUNC(void) _Py_ClearFileSystemEncoding(void);
+#endif
+
+
 #ifdef __cplusplus
 }
 #endif

Include/pylifecycle.h

@@ -175,7 +175,7 @@ PyAPI_FUNC(int) _PyOS_URandomNonblock(void *buffer, Py_ssize_t size);
 
 /* Legacy locale support */
 #ifndef Py_LIMITED_API
-PyAPI_FUNC(void) _Py_CoerceLegacyLocale(const _PyCoreConfig *config);
+PyAPI_FUNC(void) _Py_CoerceLegacyLocale(int warn);
 PyAPI_FUNC(int) _Py_LegacyLocaleDetected(void);
 PyAPI_FUNC(char *) _Py_SetLocaleFromEnv(int category);
 #endif

Lib/test/test_embed.py

@@ -251,6 +251,8 @@ class EmbeddingTests(EmbeddingTestsMixin, unittest.TestCase):
 
 class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
     maxDiff = 4096
+    UTF8_MODE_ERRORS = ('surrogatepass' if sys.platform == 'win32'
+                        else 'surrogateescape')
     DEFAULT_CONFIG = {
         'install_signal_handlers': 1,
         'use_environment': 1,
@@ -265,8 +267,12 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
         'show_alloc_count': 0,
         'dump_refs': 0,
         'malloc_stats': 0,
-        'utf8_mode': 0,
 
+        # None means that the default encoding is read at runtime:
+        # see get_locale_encoding().
+        'filesystem_encoding': None,
+        'filesystem_errors': sys.getfilesystemencodeerrors(),
+        'utf8_mode': 0,
         'coerce_c_locale': 0,
         'coerce_c_locale_warn': 0,
 
@@ -297,6 +303,7 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
         '_frozen': 0,
     }
 
+
     def get_stdio_encoding(self, env):
         code = 'import sys; print(sys.stdout.encoding, sys.stdout.errors)'
         args = (sys.executable, '-c', code)
@@ -308,6 +315,29 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
         out = proc.stdout.rstrip()
         return out.split()
 
+    def get_locale_encoding(self, isolated):
+        if sys.platform in ('win32', 'darwin') or support.is_android:
+            # Windows, macOS and Android use UTF-8
+            return "utf-8"
+
+        code = ('import codecs, locale, sys',
+                'locale.setlocale(locale.LC_CTYPE, "")',
+                'enc = locale.nl_langinfo(locale.CODESET)',
+                'enc = codecs.lookup(enc).name',
+                'print(enc)')
+        args = (sys.executable, '-c', '; '.join(code))
+        env = dict(os.environ)
+        if not isolated:
+            env['PYTHONCOERCECLOCALE'] = '0'
+            env['PYTHONUTF8'] = '0'
+        proc = subprocess.run(args, text=True, env=env,
+                              stdout=subprocess.PIPE,
+                              stderr=subprocess.PIPE)
+        if proc.returncode:
+            raise Exception(f"failed to get the locale encoding: "
+                            f"stdout={proc.stdout!r} stderr={proc.stderr!r}")
+        return proc.stdout.rstrip()
+
     def check_config(self, testname, expected):
         expected = dict(self.DEFAULT_CONFIG, **expected)
 
@@ -326,6 +356,8 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
                 expected['stdio_encoding'] = res[0]
             if expected['stdio_errors'] is None:
                 expected['stdio_errors'] = res[1]
+        if expected['filesystem_encoding'] is None:
+            expected['filesystem_encoding'] = self.get_locale_encoding(expected['isolated'])
         for key, value in expected.items():
             expected[key] = str(value)
 
@@ -357,7 +389,8 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
             'utf8_mode': 1,
             'stdio_encoding': 'utf-8',
             'stdio_errors': 'surrogateescape',
-
+            'filesystem_encoding': 'utf-8',
+            'filesystem_errors': self.UTF8_MODE_ERRORS,
             'user_site_directory': 0,
             '_frozen': 1,
         }
@@ -378,6 +411,8 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
             'utf8_mode': 1,
             'stdio_encoding': 'iso8859-1',
             'stdio_errors': 'replace',
+            'filesystem_encoding': 'utf-8',
+            'filesystem_errors': self.UTF8_MODE_ERRORS,
 
             'pycache_prefix': 'conf_pycache_prefix',
             'program_name': './conf_program_name',
@@ -409,6 +444,8 @@ class InitConfigTests(EmbeddingTestsMixin, unittest.TestCase):
             'import_time': 1,
             'malloc_stats': 1,
             'utf8_mode': 1,
+            'filesystem_encoding': 'utf-8',
+            'filesystem_errors': self.UTF8_MODE_ERRORS,
             'inspect': 1,
             'optimization_level': 2,
             'pycache_prefix': 'env_pycache_prefix',

Lib/test/test_sys.py

@@ -861,6 +861,16 @@ class SysModuleTest(unittest.TestCase):
     def test_no_duplicates_in_meta_path(self):
         self.assertEqual(len(sys.meta_path), len(set(sys.meta_path)))
 
+    @unittest.skipUnless(hasattr(sys, "_enablelegacywindowsfsencoding"),
+                         'needs sys._enablelegacywindowsfsencoding()')
+    def test__enablelegacywindowsfsencoding(self):
+        code = ('import sys',
+                'sys._enablelegacywindowsfsencoding()',
+                'print(sys.getfilesystemencoding(), sys.getfilesystemencodeerrors())')
+        rc, out, err = assert_python_ok('-c', '; '.join(code))
+        out = out.decode('ascii', 'replace').rstrip()
+        self.assertEqual(out, 'mbcs replace')
+
 
 @test.support.cpython_only
 class SizeofTest(unittest.TestCase):

Misc/NEWS.d/next/Core and Builtins/2018-08-28-01-45-01.bpo-34523.aUUkc3.rst

+The Python filesystem encoding is now read earlier during the Python
+initialization.

Modules/main.c

@@ -1339,7 +1339,7 @@ pymain_read_conf(_PyMain *pymain, _PyCoreConfig *config,
          */
         if (config->coerce_c_locale && !locale_coerced) {
             locale_coerced = 1;
-            _Py_CoerceLegacyLocale(config);
+            _Py_CoerceLegacyLocale(config->coerce_c_locale_warn);
             encoding_changed = 1;
         }
 

Objects/unicodeobject.c

@@ -3410,27 +3410,24 @@ PyUnicode_EncodeLocale(PyObject *unicode, const char *errors)
 PyObject *
 PyUnicode_EncodeFSDefault(PyObject *unicode)
 {
+    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
+    const _PyCoreConfig *config = &interp->core_config;
 #if defined(__APPLE__)
-    return _PyUnicode_AsUTF8String(unicode, Py_FileSystemDefaultEncodeErrors);
+    return _PyUnicode_AsUTF8String(unicode, config->filesystem_errors);
 #else
-    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
     /* Bootstrap check: if the filesystem codec is implemented in Python, we
        cannot use it to encode and decode filenames before it is loaded. Load
        the Python codec requires to encode at least its own filename. Use the C
-       version of the locale codec until the codec registry is initialized and
-       the Python codec is loaded.
-
-       Py_FileSystemDefaultEncoding is shared between all interpreters, we
-       cannot only rely on it: check also interp->fscodec_initialized for
-       subinterpreters. */
-    if (Py_FileSystemDefaultEncoding && interp->fscodec_initialized) {
+       implementation of the locale codec until the codec registry is
+       initialized and the Python codec is loaded. See initfsencoding(). */
+    if (interp->fscodec_initialized) {
         return PyUnicode_AsEncodedString(unicode,
-                                         Py_FileSystemDefaultEncoding,
-                                         Py_FileSystemDefaultEncodeErrors);
+                                         config->filesystem_encoding,
+                                         config->filesystem_errors);
     }
     else {
         return unicode_encode_locale(unicode,
-                                     Py_FileSystemDefaultEncodeErrors, 0);
+                                     config->filesystem_errors, 0);
     }
 #endif
 }
@@ -3636,27 +3633,24 @@ PyUnicode_DecodeFSDefault(const char *s) {
 PyObject*
 PyUnicode_DecodeFSDefaultAndSize(const char *s, Py_ssize_t size)
 {
+    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
+    const _PyCoreConfig *config = &interp->core_config;
 #if defined(__APPLE__)
-    return PyUnicode_DecodeUTF8Stateful(s, size, Py_FileSystemDefaultEncodeErrors, NULL);
+    return PyUnicode_DecodeUTF8Stateful(s, size, config->filesystem_errors, NULL);
 #else
-    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
     /* Bootstrap check: if the filesystem codec is implemented in Python, we
        cannot use it to encode and decode filenames before it is loaded. Load
        the Python codec requires to encode at least its own filename. Use the C
-       version of the locale codec until the codec registry is initialized and
-       the Python codec is loaded.
-
-       Py_FileSystemDefaultEncoding is shared between all interpreters, we
-       cannot only rely on it: check also interp->fscodec_initialized for
-       subinterpreters. */
-    if (Py_FileSystemDefaultEncoding && interp->fscodec_initialized) {
+       implementation of the locale codec until the codec registry is
+       initialized and the Python codec is loaded. See initfsencoding(). */
+    if (interp->fscodec_initialized) {
         return PyUnicode_Decode(s, size,
-                                Py_FileSystemDefaultEncoding,
-                                Py_FileSystemDefaultEncodeErrors);
+                                config->filesystem_encoding,
+                                config->filesystem_errors);
     }
     else {
         return unicode_decode_locale(s, size,
-                                     Py_FileSystemDefaultEncodeErrors, 0);
+                                     config->filesystem_errors, 0);
     }
 #endif
 }

Programs/_freeze_importlib.c

@@ -81,8 +81,15 @@ main(int argc, char *argv[])
     config.program_name = L"./_freeze_importlib";
     /* Don't install importlib, since it could execute outdated bytecode. */
     config._install_importlib = 0;
-    config.install_signal_handlers = 1;
     config._frozen = 1;
+#ifdef MS_WINDOWS
+    /* bpo-34523: initfsencoding() is not called if _install_importlib=0,
+       so interp->fscodec_initialized value remains 0.
+       PyUnicode_EncodeFSDefault() doesn't support the "surrogatepass" error
+       handler in such case, whereas it's the default error handler on Windows.
+       Force the "strict" error handler to work around this bootstrap issue. */
+    config.filesystem_errors = "strict";
+#endif
 
     _PyInitError err = _Py_InitializeFromConfig(&config);
     /* No need to call _PyCoreConfig_Clear() since we didn't allocate any

Programs/_testembed.c

@@ -328,6 +328,8 @@ dump_config(void)
     printf("dump_refs = %i\n", config->dump_refs);
     printf("malloc_stats = %i\n", config->malloc_stats);
 
+    printf("filesystem_encoding = %s\n", config->filesystem_encoding);
+    printf("filesystem_errors = %s\n", config->filesystem_errors);
     printf("coerce_c_locale = %i\n", config->coerce_c_locale);
     printf("coerce_c_locale_warn = %i\n", config->coerce_c_locale_warn);
     printf("utf8_mode = %i\n", config->utf8_mode);

Python/coreconfig.c

@@ -5,6 +5,11 @@
 #  include <langinfo.h>
 #endif
 
+#include <locale.h>     /* setlocale() */
+#ifdef HAVE_LANGINFO_H
+#include <langinfo.h>   /* nl_langinfo(CODESET) */
+#endif
+
 
 #define DECODE_LOCALE_ERR(NAME, LEN) \
     (((LEN) == -2) \
@@ -32,6 +37,8 @@ const char *Py_FileSystemDefaultEncoding = NULL; /* set by initfsencoding() */
 int Py_HasFileSystemDefaultEncoding = 0;
 #endif
 const char *Py_FileSystemDefaultEncodeErrors = "surrogateescape";
+static int _Py_HasFileSystemDefaultEncodeErrors = 1;
+
 /* UTF-8 mode (PEP 540): if equals to 1, use the UTF-8 encoding, and change
    stdin and stdout error handler to "surrogateescape". It is equal to
    -1 by default: unknown, will be set by Py_Main() */
@@ -88,6 +95,47 @@ _Py_wstrlist_copy(int len, wchar_t **list)
 }
 
 
+void
+_Py_ClearFileSystemEncoding(void)
+{
+    if (!Py_HasFileSystemDefaultEncoding && Py_FileSystemDefaultEncoding) {
+        PyMem_RawFree((char*)Py_FileSystemDefaultEncoding);
+        Py_FileSystemDefaultEncoding = NULL;
+    }
+    if (!_Py_HasFileSystemDefaultEncodeErrors && Py_FileSystemDefaultEncodeErrors) {
+        PyMem_RawFree((char*)Py_FileSystemDefaultEncodeErrors);
+        Py_FileSystemDefaultEncodeErrors = NULL;
+    }
+}
+
+
+/* Set Py_FileSystemDefaultEncoding and Py_FileSystemDefaultEncodeErrors
+   global configuration variables. */
+int
+_Py_SetFileSystemEncoding(const char *encoding, const char *errors)
+{
+    char *encoding2 = _PyMem_RawStrdup(encoding);
+    if (encoding2 == NULL) {
+        return -1;
+    }
+
+    char *errors2 = _PyMem_RawStrdup(errors);
+    if (errors2 == NULL) {
+        PyMem_RawFree(encoding2);
+        return -1;
+    }
+
+    _Py_ClearFileSystemEncoding();
+
+    Py_FileSystemDefaultEncoding = encoding2;
+    Py_HasFileSystemDefaultEncoding = 0;
+
+    Py_FileSystemDefaultEncodeErrors = errors2;
+    _Py_HasFileSystemDefaultEncodeErrors = 0;
+    return 0;
+}
+
+
 /* Helper to allow an embedding application to override the normal
  * mechanism that attempts to figure out an appropriate IO encoding
  */
@@ -209,6 +257,8 @@ _PyCoreConfig_Clear(_PyCoreConfig *config)
 #endif
     CLEAR(config->base_exec_prefix);
 
+    CLEAR(config->filesystem_encoding);
+    CLEAR(config->filesystem_errors);
     CLEAR(config->stdio_encoding);
     CLEAR(config->stdio_errors);
 #undef CLEAR
@@ -302,6 +352,8 @@ _PyCoreConfig_Copy(_PyCoreConfig *config, const _PyCoreConfig *config2)
     COPY_ATTR(quiet);
     COPY_ATTR(user_site_directory);
     COPY_ATTR(buffered_stdio);
+    COPY_STR_ATTR(filesystem_encoding);
+    COPY_STR_ATTR(filesystem_errors);
     COPY_STR_ATTR(stdio_encoding);
     COPY_STR_ATTR(stdio_errors);
 #ifdef MS_WINDOWS
@@ -312,6 +364,7 @@ _PyCoreConfig_Copy(_PyCoreConfig *config, const _PyCoreConfig *config2)
     COPY_ATTR(_frozen);
 
 #undef COPY_ATTR
+#undef COPY_STR_ATTR
 #undef COPY_WSTR_ATTR
 #undef COPY_WSTRLIST
     return 0;
@@ -976,8 +1029,8 @@ get_stdio_errors(const _PyCoreConfig *config)
 }
 
 
-_PyInitError
-_Py_get_locale_encoding(char **locale_encoding)
+static _PyInitError
+get_locale_encoding(char **locale_encoding)
 {
 #ifdef MS_WINDOWS
     char encoding[20];
@@ -1087,7 +1140,7 @@ config_init_stdio_encoding(_PyCoreConfig *config)
 
     /* Choose the default error handler based on the current locale. */
     if (config->stdio_encoding == NULL) {
-        _PyInitError err = _Py_get_locale_encoding(&config->stdio_encoding);
+        _PyInitError err = get_locale_encoding(&config->stdio_encoding);
         if (_Py_INIT_FAILED(err)) {
             return err;
         }
@@ -1104,6 +1157,81 @@ config_init_stdio_encoding(_PyCoreConfig *config)
 }
 
 
+static _PyInitError
+config_init_fs_encoding(_PyCoreConfig *config)
+{
+#ifdef MS_WINDOWS
+    if (config->legacy_windows_fs_encoding) {
+        /* Legacy Windows filesystem encoding: mbcs/replace */
+        if (config->filesystem_encoding == NULL) {
+            config->filesystem_encoding = _PyMem_RawStrdup("mbcs");
+            if (config->filesystem_encoding == NULL) {
+                return _Py_INIT_NO_MEMORY();
+            }
+        }
+        if (config->filesystem_errors == NULL) {
+            config->filesystem_errors = _PyMem_RawStrdup("replace");
+            if (config->filesystem_errors == NULL) {
+                return _Py_INIT_NO_MEMORY();
+            }
+        }
+    }
+
+    /* Windows defaults to utf-8/surrogatepass (PEP 529) */
+    if (config->filesystem_encoding == NULL) {
+        config->filesystem_encoding = _PyMem_RawStrdup("utf-8");
+        if (config->filesystem_encoding == NULL) {
+            return _Py_INIT_NO_MEMORY();
+        }
+    }
+    if (config->filesystem_errors == NULL) {
+        config->filesystem_errors = _PyMem_RawStrdup("surrogatepass");
+        if (config->filesystem_errors == NULL) {
+            return _Py_INIT_NO_MEMORY();
+        }
+    }
+#else
+    if (config->utf8_mode) {
+        /* UTF-8 Mode use: utf-8/surrogateescape */
+        if (config->filesystem_encoding == NULL) {
+            config->filesystem_encoding = _PyMem_RawStrdup("utf-8");
+            if (config->filesystem_encoding == NULL) {
+                return _Py_INIT_NO_MEMORY();
+            }
+        }
+        /* errors defaults to surrogateescape above */
+    }
+
+    if (config->filesystem_encoding == NULL) {
+        /* macOS and Android use UTF-8, other platforms use
+           the locale encoding. */
+        char *locale_encoding;
+#if defined(__APPLE__) || defined(__ANDROID__)
+        locale_encoding = "UTF-8";
+#else
+        _PyInitError err = get_locale_encoding(&locale_encoding);
+        if (_Py_INIT_FAILED(err)) {
+            return err;
+        }
+#endif
+        config->filesystem_encoding = _PyMem_RawStrdup(locale_encoding);
+        if (config->filesystem_encoding == NULL) {
+            return _Py_INIT_NO_MEMORY();
+        }
+    }
+
+    if (config->filesystem_errors == NULL) {
+        /* by default, use the "surrogateescape" error handler */
+        config->filesystem_errors = _PyMem_RawStrdup("surrogateescape");
+        if (config->filesystem_errors == NULL) {
+            return _Py_INIT_NO_MEMORY();
+        }
+    }
+#endif
+    return _Py_INIT_OK();
+}
+
+
 /* Read configuration settings from standard locations
  *
  * This function doesn't make any changes to the interpreter state - it
@@ -1216,6 +1344,13 @@ _PyCoreConfig_Read(_PyCoreConfig *config)
         config->argc = 0;
     }
 
+    if (config->filesystem_encoding == NULL && config->filesystem_errors == NULL) {
+        err = config_init_fs_encoding(config);
+        if (_Py_INIT_FAILED(err)) {
+            return err;
+        }
+    }
+
     err = config_init_stdio_encoding(config);
     if (_Py_INIT_FAILED(err)) {
         return err;
@@ -1223,6 +1358,10 @@ _PyCoreConfig_Read(_PyCoreConfig *config)
 
     assert(config->coerce_c_locale >= 0);
     assert(config->use_environment >= 0);
+    assert(config->filesystem_encoding != NULL);
+    assert(config->filesystem_errors != NULL);
+    assert(config->stdio_encoding != NULL);
+    assert(config->stdio_errors != NULL);
 
     return _Py_INIT_OK();
 }

Python/pylifecycle.c

@@ -339,7 +339,7 @@ static const char C_LOCALE_COERCION_WARNING[] =
     "or PYTHONCOERCECLOCALE=0 to disable this locale coercion behavior).\n";
 
 static void
-_coerce_default_locale_settings(const _PyCoreConfig *config, const _LocaleCoercionTarget *target)
+_coerce_default_locale_settings(int warn, const _LocaleCoercionTarget *target)
 {
     const char *newloc = target->locale_name;
 
@@ -352,7 +352,7 @@ _coerce_default_locale_settings(const _PyCoreConfig *config, const _LocaleCoerci
                 "Error setting LC_CTYPE, skipping C locale coercion\n");
         return;
     }
-    if (config->coerce_c_locale_warn) {
+    if (warn) {
         fprintf(stderr, C_LOCALE_COERCION_WARNING, newloc);
     }
 
@@ -362,7 +362,7 @@ _coerce_default_locale_settings(const _PyCoreConfig *config, const _LocaleCoerci
 #endif
 
 void
-_Py_CoerceLegacyLocale(const _PyCoreConfig *config)
+_Py_CoerceLegacyLocale(int warn)
 {
 #ifdef PY_COERCE_C_LOCALE
     const char *locale_override = getenv("LC_ALL");
@@ -385,7 +385,7 @@ defined(HAVE_LANGINFO_H) && defined(CODESET)
                 }
 #endif
                 /* Successfully configured locale, so make it the default */
-                _coerce_default_locale_settings(config, target);
+                _coerce_default_locale_settings(warn, target);
                 return;
             }
         }
@@ -1162,11 +1162,7 @@ Py_FinalizeEx(void)
     /* Cleanup Unicode implementation */
     _PyUnicode_Fini();
 
-    /* reset file system default encoding */
-    if (!Py_HasFileSystemDefaultEncoding && Py_FileSystemDefaultEncoding) {
-        PyMem_RawFree((char*)Py_FileSystemDefaultEncoding);
-        Py_FileSystemDefaultEncoding = NULL;
-    }
+    _Py_ClearFileSystemEncoding();
 
     /* XXX Still allocated:
        - various static ad-hoc pointers to interned strings
@@ -1475,59 +1471,31 @@ add_main_module(PyInterpreterState *interp)
 static _PyInitError
 initfsencoding(PyInterpreterState *interp)
 {
-    PyObject *codec;
+    _PyCoreConfig *config = &interp->core_config;
 
-#ifdef MS_WINDOWS
-    if (Py_LegacyWindowsFSEncodingFlag) {
-        Py_FileSystemDefaultEncoding = "mbcs";
-        Py_FileSystemDefaultEncodeErrors = "replace";
-    }
-    else {
-        Py_FileSystemDefaultEncoding = "utf-8";
-        Py_FileSystemDefaultEncodeErrors = "surrogatepass";
+    char *encoding = get_codec_name(config->filesystem_encoding);
+    if (encoding == NULL) {
+        /* Such error can only occurs in critical situations: no more
+           memory, import a module of the standard library failed, etc. */
+        return _Py_INIT_ERR("failed to get the Python codec "
+                            "of the filesystem encoding");
     }
-#else
-    if (Py_FileSystemDefaultEncoding == NULL) {
-        if (interp->core_config.utf8_mode) {
-            Py_FileSystemDefaultEncoding = "utf-8";
-            Py_HasFileSystemDefaultEncoding = 1;
-        }
-        else if (_Py_GetForceASCII()) {
-            Py_FileSystemDefaultEncoding = "ascii";
-            Py_HasFileSystemDefaultEncoding = 1;
-        }
-        else {
-            extern _PyInitError _Py_get_locale_encoding(char **locale_encoding);
 
-            char *locale_encoding;
-            _PyInitError err = _Py_get_locale_encoding(&locale_encoding);
-            if (_Py_INIT_FAILED(err)) {
-                return err;
-            }
-
-            Py_FileSystemDefaultEncoding = get_codec_name(locale_encoding);
-            PyMem_RawFree(locale_encoding);
-            if (Py_FileSystemDefaultEncoding == NULL) {
-                return _Py_INIT_ERR("failed to get the Python codec "
-                                    "of the locale encoding");
-            }
+    /* Update the filesystem encoding to the normalized Python codec name.
+       For example, replace "ANSI_X3.4-1968" (locale encoding) with "ascii"
+       (Python codec name). */
+    PyMem_RawFree(config->filesystem_encoding);
+    config->filesystem_encoding = encoding;
 
-            Py_HasFileSystemDefaultEncoding = 0;
-            interp->fscodec_initialized = 1;
-            return _Py_INIT_OK();
-        }
+    /* Set Py_FileSystemDefaultEncoding and Py_FileSystemDefaultEncodeErrors
+       global configuration variables. */
+    if (_Py_SetFileSystemEncoding(config->filesystem_encoding,
+                                  config->filesystem_errors) < 0) {
+        return _Py_INIT_NO_MEMORY();
     }
-#endif
 
-    /* the encoding is mbcs, utf-8 or ascii */
-    codec = _PyCodec_Lookup(Py_FileSystemDefaultEncoding);
-    if (!codec) {
-        /* Such error can only occurs in critical situations: no more
-         * memory, import a module of the standard library failed,
-         * etc. */
-        return _Py_INIT_ERR("unable to load the file system codec");
-    }
-    Py_DECREF(codec);
+    /* PyUnicode can now use the Python codec rather than C implementation
+       for the filesystem encoding */
     interp->fscodec_initialized = 1;
     return _Py_INIT_OK();
 }

Python/sysmodule.c

@@ -389,11 +389,9 @@ implementation."
 static PyObject *
 sys_getfilesystemencoding(PyObject *self, PyObject *Py_UNUSED(ignored))
 {
-    if (Py_FileSystemDefaultEncoding)
-        return PyUnicode_FromString(Py_FileSystemDefaultEncoding);
-    PyErr_SetString(PyExc_RuntimeError,
-                    "filesystem encoding is not initialized");
-    return NULL;
+    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
+    const _PyCoreConfig *config = &interp->core_config;
+    return PyUnicode_FromString(config->filesystem_encoding);
 }
 
 PyDoc_STRVAR(getfilesystemencoding_doc,
@@ -406,11 +404,9 @@ operating system filenames."
 static PyObject *
 sys_getfilesystemencodeerrors(PyObject *self, PyObject *Py_UNUSED(ignored))
 {
-    if (Py_FileSystemDefaultEncodeErrors)
-        return PyUnicode_FromString(Py_FileSystemDefaultEncodeErrors);
-    PyErr_SetString(PyExc_RuntimeError,
-        "filesystem encoding is not initialized");
-    return NULL;
+    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
+    const _PyCoreConfig *config = &interp->core_config;
+    return PyUnicode_FromString(config->filesystem_errors);
 }
 
 PyDoc_STRVAR(getfilesystemencodeerrors_doc,
@@ -1150,8 +1146,30 @@ environment variable before launching Python."
 static PyObject *
 sys_enablelegacywindowsfsencoding(PyObject *self)
 {
-    Py_FileSystemDefaultEncoding = "mbcs";
-    Py_FileSystemDefaultEncodeErrors = "replace";
+    PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE();
+    _PyCoreConfig *config = &interp->core_config;
+
+    /* Set the filesystem encoding to mbcs/replace (PEP 529) */
+    char *encoding = _PyMem_RawStrdup("mbcs");
+    char *errors = _PyMem_RawStrdup("replace");
+    if (encoding == NULL || errors == NULL) {
+        PyMem_Free(encoding);
+        PyMem_Free(errors);
+        PyErr_NoMemory();
+        return NULL;
+    }
+
+    PyMem_RawFree(config->filesystem_encoding);
+    config->filesystem_encoding = encoding;
+    PyMem_RawFree(config->filesystem_errors);
+    config->filesystem_errors = errors;
+
+    if (_Py_SetFileSystemEncoding(config->filesystem_encoding,
+                                  config->filesystem_errors) < 0) {
+        PyErr_NoMemory();
+        return NULL;
+    }
+
     Py_RETURN_NONE;
 }