bpo-45395: Make custom frozen modules additions instead of replacements. (gh-28778)

Currently custom modules (the array set on PyImport_FrozenModules) replace all the frozen stdlib modules. That can be problematic and is unlikely to be what the user wants. This change treats the custom frozen modules as additions instead. They take precedence over all other frozen modules except for those needed to bootstrap the import system. If the "code" field of an entry in the custom array is NULL then that frozen module is treated as disabled, which allows a custom entry to disable a frozen stdlib module.

This change allows us to get rid of is_essential_frozen_module() and simplifies the logic for which frozen modules should be ignored.

https://bugs.python.org/issue45395

Doc/library/ctypes.rst

@@ -1095,7 +1095,7 @@ We have defined the :c:type:`struct _frozen` data type, so we can get the pointe
 to the table::
 
    >>> FrozenTable = POINTER(struct_frozen)
-   >>> table = FrozenTable.in_dll(pythonapi, "PyImport_FrozenModules")
+   >>> table = FrozenTable.in_dll(pythonapi, "_PyImport_FrozenBootstrap")
    >>>
 
 Since ``table`` is a ``pointer`` to the array of ``struct_frozen`` records, we
@@ -1111,9 +1111,7 @@ hit the ``NULL`` entry::
    ...
    _frozen_importlib 31764
    _frozen_importlib_external 41499
-   __hello__ 161
+   zipimport 12345
-   __phello__ -161
-   __phello__.spam 161
    >>>
 
 The fact that standard Python has a frozen module and a frozen package

Include/internal/pycore_import.h

@@ -15,6 +15,9 @@ struct _module_alias {
     const char *orig;                 /* ASCII encoded string */
 };
 
+PyAPI_DATA(const struct _frozen *) _PyImport_FrozenBootstrap;
+PyAPI_DATA(const struct _frozen *) _PyImport_FrozenStdlib;
+PyAPI_DATA(const struct _frozen *) _PyImport_FrozenTest;
 extern const struct _module_alias * _PyImport_FrozenAliases;
 
 #ifdef __cplusplus

Lib/ctypes/test/test_values.py

@@ -56,35 +56,37 @@ class PythonValuesTestCase(unittest.TestCase):
                         ("size", c_int)]
         FrozenTable = POINTER(struct_frozen)
 
-        ft = FrozenTable.in_dll(pythonapi, "PyImport_FrozenModules")
-        # ft is a pointer to the struct_frozen entries:
         modules = []
-        for entry in ft:
+        for group in ["Bootstrap", "Stdlib", "Test"]:
-            # This is dangerous. We *can* iterate over a pointer, but
+            ft = FrozenTable.in_dll(pythonapi, f"_PyImport_Frozen{group}")
-            # the loop will not terminate (maybe with an access
+            # ft is a pointer to the struct_frozen entries:
-            # violation;-) because the pointer instance has no size.
+            for entry in ft:
-            if entry.name is None:
+                # This is dangerous. We *can* iterate over a pointer, but
-                break
+                # the loop will not terminate (maybe with an access
-            modname = entry.name.decode("ascii")
+                # violation;-) because the pointer instance has no size.
-            modules.append(modname)
+                if entry.name is None:
-            with self.subTest(modname):
+                    break
-                # Do a sanity check on entry.size and entry.code.
+                modname = entry.name.decode("ascii")
-                self.assertGreater(abs(entry.size), 10)
+                modules.append(modname)
-                self.assertTrue([entry.code[i] for i in range(abs(entry.size))])
+                with self.subTest(modname):
-                # Check the module's package-ness.
+                    # Do a sanity check on entry.size and entry.code.
-                with import_helper.frozen_modules():
+                    self.assertGreater(abs(entry.size), 10)
-                    spec = importlib.util.find_spec(modname)
+                    self.assertTrue([entry.code[i] for i in range(abs(entry.size))])
-                if entry.size < 0:
+                    # Check the module's package-ness.
-                    # It's a package.
+                    with import_helper.frozen_modules():
-                    self.assertIsNotNone(spec.submodule_search_locations)
+                        spec = importlib.util.find_spec(modname)
-                else:
+                    if entry.size < 0:
-                    self.assertIsNone(spec.submodule_search_locations)
+                        # It's a package.
+                        self.assertIsNotNone(spec.submodule_search_locations)
+                    else:
+                        self.assertIsNone(spec.submodule_search_locations)
 
         with import_helper.frozen_modules():
             expected = _imp._frozen_module_names()
         self.maxDiff = None
-        self.assertEqual(modules, expected, "PyImport_FrozenModules example "
+        self.assertEqual(modules, expected,
-            "in Doc/library/ctypes.rst may be out of date")
+                         "_PyImport_FrozenBootstrap example "
+                         "in Doc/library/ctypes.rst may be out of date")
 
         from ctypes import _pointer_type_cache
         del _pointer_type_cache[struct_frozen]

Misc/NEWS.d/next/C API/2021-10-06-15-54-40.bpo-45395.yVhdAl.rst

@@ -0,0 +1,4 @@
+Custom frozen modules (the array set to ``PyImport_FrozenModules``) are now
+treated as additions, rather than replacing all the default frozen modules.
+Frozen stdlib modules can still be disabled by setting the "code" field of
+the custom array entry to NULL.

Programs/_freeze_module.c

@@ -23,13 +23,16 @@
    of frozen modules instead, left deliberately blank so as to avoid
    unintentional import of a stale version of _frozen_importlib. */
 
-static const struct _frozen _PyImport_FrozenModules[] = {
+static const struct _frozen no_modules[] = {
     {0, 0, 0} /* sentinel */
 };
 static const struct _module_alias aliases[] = {
     {0, 0} /* sentinel */
 };
 
+const struct _frozen *_PyImport_FrozenBootstrap;
+const struct _frozen *_PyImport_FrozenStdlib;
+const struct _frozen *_PyImport_FrozenTest;
 const struct _frozen *PyImport_FrozenModules;
 const struct _module_alias *_PyImport_FrozenAliases;
 
@@ -188,7 +191,10 @@ main(int argc, char *argv[])
 {
     const char *name, *inpath, *outpath;
 
-    PyImport_FrozenModules = _PyImport_FrozenModules;
+    _PyImport_FrozenBootstrap = no_modules;
+    _PyImport_FrozenStdlib = no_modules;
+    _PyImport_FrozenTest = no_modules;
+    PyImport_FrozenModules = NULL;
     _PyImport_FrozenAliases = aliases;
 
     if (argc != 4) {

Programs/_testembed.c

@@ -8,6 +8,7 @@
 #include <Python.h>
 #include "pycore_initconfig.h"    // _PyConfig_InitCompatConfig()
 #include "pycore_runtime.h"       // _PyRuntime
+#include "pycore_import.h"        // _PyImport_FrozenBootstrap
 #include <Python.h>
 #include <inttypes.h>
 #include <stdio.h>
@@ -1804,30 +1805,10 @@ static int test_unicode_id_init(void)
 
 static int test_frozenmain(void)
 {
-    // Get "_frozen_importlib" and "_frozen_importlib_external"
-    // from PyImport_FrozenModules
-    const struct _frozen *importlib = NULL, *importlib_external = NULL;
-    for (const struct _frozen *mod = PyImport_FrozenModules; mod->name != NULL; mod++) {
-        if (strcmp(mod->name, "_frozen_importlib") == 0) {
-            importlib = mod;
-        }
-        else if (strcmp(mod->name, "_frozen_importlib_external") == 0) {
-            importlib_external = mod;
-        }
-    }
-    if (importlib == NULL || importlib_external == NULL) {
-        error("cannot find frozen importlib and importlib_external");
-        return 1;
-    }
-
     static struct _frozen frozen_modules[4] = {
-        {0, 0, 0},  // importlib
-        {0, 0, 0},  // importlib_external
         {"__main__", M_test_frozenmain, sizeof(M_test_frozenmain)},
         {0, 0, 0}   // sentinel
     };
-    frozen_modules[0] = *importlib;
-    frozen_modules[1] = *importlib_external;
 
     char* argv[] = {
         "./argv0",
@@ -1846,7 +1827,12 @@ static int test_frozenmain(void)
 static int list_frozen(void)
 {
     const struct _frozen *p;
-    for (p = PyImport_FrozenModules; ; p++) {
+    for (p = _PyImport_FrozenBootstrap; ; p++) {
+        if (p->name == NULL)
+            break;
+        printf("%s\n", p->name);
+    }
+    for (p = _PyImport_FrozenStdlib; ; p++) {
         if (p->name == NULL)
             break;
         printf("%s\n", p->name);

Python/frozen.c

@@ -63,14 +63,15 @@
 
 /* Note that a negative size indicates a package. */
 
-static const struct _frozen _PyImport_FrozenModules[] = {
+static const struct _frozen bootstrap_modules[] = {
-    /* import system */
     {"_frozen_importlib", _Py_M__importlib__bootstrap,
         (int)sizeof(_Py_M__importlib__bootstrap)},
     {"_frozen_importlib_external", _Py_M__importlib__bootstrap_external,
         (int)sizeof(_Py_M__importlib__bootstrap_external)},
     {"zipimport", _Py_M__zipimport, (int)sizeof(_Py_M__zipimport)},
-
+    {0, 0, 0} /* bootstrap sentinel */
+};
+static const struct _frozen stdlib_modules[] = {
     /* stdlib - startup, without site (python -S) */
     {"abc", _Py_M__abc, (int)sizeof(_Py_M__abc)},
     {"codecs", _Py_M__codecs, (int)sizeof(_Py_M__codecs)},
@@ -87,8 +88,9 @@ static const struct _frozen _PyImport_FrozenModules[] = {
     {"os", _Py_M__os, (int)sizeof(_Py_M__os)},
     {"site", _Py_M__site, (int)sizeof(_Py_M__site)},
     {"stat", _Py_M__stat, (int)sizeof(_Py_M__stat)},
-
+    {0, 0, 0} /* stdlib sentinel */
-    /* Test module */
+};
+static const struct _frozen test_modules[] = {
     {"__hello__", _Py_M____hello__, (int)sizeof(_Py_M____hello__)},
     {"__hello_alias__", _Py_M____hello__, (int)sizeof(_Py_M____hello__)},
     {"__phello_alias__", _Py_M____hello__, -(int)sizeof(_Py_M____hello__)},
@@ -103,8 +105,11 @@ static const struct _frozen _PyImport_FrozenModules[] = {
     {"__phello__.spam", _Py_M____phello___spam,
         (int)sizeof(_Py_M____phello___spam)},
     {"__hello_only__", _Py_M__frozen_only, (int)sizeof(_Py_M__frozen_only)},
-    {0, 0, 0} /* modules sentinel */
+    {0, 0, 0} /* test sentinel */
 };
+const struct _frozen *_PyImport_FrozenBootstrap = bootstrap_modules;
+const struct _frozen *_PyImport_FrozenStdlib = stdlib_modules;
+const struct _frozen *_PyImport_FrozenTest = test_modules;
 
 static const struct _module_alias aliases[] = {
     {"_frozen_importlib", "importlib._bootstrap"},
@@ -124,4 +129,4 @@ const struct _module_alias *_PyImport_FrozenAliases = aliases;
 /* Embedding apps may change this pointer to point to their favorite
    collection of frozen modules: */
 
-const struct _frozen *PyImport_FrozenModules = _PyImport_FrozenModules;
+const struct _frozen *PyImport_FrozenModules = NULL;

Python/import.c

@@ -1070,27 +1070,6 @@ resolve_module_alias(const char *name, const struct _module_alias *aliases,
 
 /* Frozen modules */
 
-static bool
-is_essential_frozen_module(const char *name)
-{
-    /* These modules are necessary to bootstrap the import system. */
-    if (strcmp(name, "_frozen_importlib") == 0) {
-        return true;
-    }
-    if (strcmp(name, "_frozen_importlib_external") == 0) {
-        return true;
-    }
-    if (strcmp(name, "zipimport") == 0) {
-        return true;
-    }
-    /* This doesn't otherwise have anywhere to find the module.
-       See frozenmain.c. */
-    if (strcmp(name, "__main__") == 0) {
-        return true;
-    }
-    return false;
-}
-
 static bool
 use_frozen(void)
 {
@@ -1115,26 +1094,76 @@ list_frozen_module_names()
         return NULL;
     }
     bool enabled = use_frozen();
-    for (const struct _frozen *p = PyImport_FrozenModules; ; p++) {
+    const struct _frozen *p;
+#define ADD_MODULE(name) \
+    do { \
+        PyObject *nameobj = PyUnicode_FromString(name); \
+        if (nameobj == NULL) { \
+            goto error; \
+        } \
+        int res = PyList_Append(names, nameobj); \
+        Py_DECREF(nameobj); \
+        if (res != 0) { \
+            goto error; \
+        } \
+    } while(0)
+    // We always use the bootstrap modules.
+    for (p = _PyImport_FrozenBootstrap; ; p++) {
         if (p->name == NULL) {
             break;
         }
-        if (!enabled && !is_essential_frozen_module(p->name)) {
+        ADD_MODULE(p->name);
-            continue;
+    }
+    // Frozen stdlib modules may be disabled.
+    for (p = _PyImport_FrozenStdlib; ; p++) {
+        if (p->name == NULL) {
+            break;
         }
-        PyObject *name = PyUnicode_FromString(p->name);
+        if (enabled) {
-        if (name == NULL) {
+            ADD_MODULE(p->name);
-            Py_DECREF(names);
-            return NULL;
         }
-        int res = PyList_Append(names, name);
+    }
-        Py_DECREF(name);
+    for (p = _PyImport_FrozenTest; ; p++) {
-        if (res != 0) {
+        if (p->name == NULL) {
-            Py_DECREF(names);
+            break;
-            return NULL;
+        }
+        if (enabled) {
+            ADD_MODULE(p->name);
+        }
+    }
+#undef ADD_MODULE
+    // Add any custom modules.
+    if (PyImport_FrozenModules != NULL) {
+        for (p = PyImport_FrozenModules; ; p++) {
+            if (p->name == NULL) {
+                break;
+            }
+            PyObject *nameobj = PyUnicode_FromString(p->name);
+            if (nameobj == NULL) {
+                goto error;
+            }
+            int found = PySequence_Contains(names, nameobj);
+            if (found < 0) {
+                Py_DECREF(nameobj);
+                goto error;
+            }
+            else if (found) {
+                Py_DECREF(nameobj);
+            }
+            else {
+                int res = PyList_Append(names, nameobj);
+                Py_DECREF(nameobj);
+                if (res != 0) {
+                    goto error;
+                }
+            }
         }
     }
     return names;
+
+error:
+    Py_DECREF(names);
+    return NULL;
 }
 
 typedef enum {
@@ -1180,6 +1209,54 @@ set_frozen_error(frozen_status status, PyObject *modname)
     }
 }
 
+static const struct _frozen *
+look_up_frozen(const char *name)
+{
+    const struct _frozen *p;
+    // We always use the bootstrap modules.
+    for (p = _PyImport_FrozenBootstrap; ; p++) {
+        if (p->name == NULL) {
+            // We hit the end-of-list sentinel value.
+            break;
+        }
+        if (strcmp(name, p->name) == 0) {
+            return p;
+        }
+    }
+    // Prefer custom modules, if any.  Frozen stdlib modules can be
+    // disabled here by setting "code" to NULL in the array entry.
+    if (PyImport_FrozenModules != NULL) {
+        for (p = PyImport_FrozenModules; ; p++) {
+            if (p->name == NULL) {
+                break;
+            }
+            if (strcmp(name, p->name) == 0) {
+                return p;
+            }
+        }
+    }
+    // Frozen stdlib modules may be disabled.
+    if (use_frozen()) {
+        for (p = _PyImport_FrozenStdlib; ; p++) {
+            if (p->name == NULL) {
+                break;
+            }
+            if (strcmp(name, p->name) == 0) {
+                return p;
+            }
+        }
+        for (p = _PyImport_FrozenTest; ; p++) {
+            if (p->name == NULL) {
+                break;
+            }
+            if (strcmp(name, p->name) == 0) {
+                return p;
+            }
+        }
+    }
+    return NULL;
+}
+
 struct frozen_info {
     PyObject *nameobj;
     const char *data;
@@ -1209,19 +1286,9 @@ find_frozen(PyObject *nameobj, struct frozen_info *info)
         return FROZEN_BAD_NAME;
     }
 
-    if (!use_frozen() && !is_essential_frozen_module(name)) {
+    const struct _frozen *p = look_up_frozen(name);
-        return FROZEN_DISABLED;
+    if (p == NULL) {
-    }
+        return FROZEN_NOT_FOUND;
-
-    const struct _frozen *p;
-    for (p = PyImport_FrozenModules; ; p++) {
-        if (p->name == NULL) {
-            // We hit the end-of-list sentinel value.
-            return FROZEN_NOT_FOUND;
-        }
-        if (strcmp(name, p->name) == 0) {
-            break;
-        }
     }
     if (info != NULL) {
         info->nameobj = nameobj;  // borrowed

Tools/freeze/freeze.py

@@ -367,12 +367,6 @@ def main():
         else:
             mf.load_file(mod)
 
-    # Alias "importlib._bootstrap" to "_frozen_importlib" so that the
-    # import machinery can bootstrap.  Do the same for
-    # importlib._bootstrap_external.
-    mf.modules["_frozen_importlib"] = mf.modules["importlib._bootstrap"]
-    mf.modules["_frozen_importlib_external"] = mf.modules["importlib._bootstrap_external"]
-
     # Add the main script as either __main__, or the actual module name.
     if python_entry_is_main:
         mf.run_script(scriptfile)

Tools/scripts/freeze_modules.py

@@ -60,6 +60,7 @@ PCBUILD_FILTERS = os.path.join(ROOT_DIR, 'PCbuild', '_freeze_module.vcxproj.filt
 OS_PATH = 'ntpath' if os.name == 'nt' else 'posixpath'
 
 # These are modules that get frozen.
+TESTS_SECTION = 'Test module'
 FROZEN = [
     # See parse_frozen_spec() for the format.
     # In cases where the frozenid is duplicated, the first one is re-used.
@@ -94,7 +95,7 @@ FROZEN = [
         'site',
         'stat',
         ]),
-    ('Test module', [
+    (TESTS_SECTION, [
         '__hello__',
         '__hello__ : __hello_alias__',
         '__hello__ : <__phello_alias__>',
@@ -103,7 +104,7 @@ FROZEN = [
         f'frozen_only : __hello_only__ = {FROZEN_ONLY}',
         ]),
 ]
-ESSENTIAL = {
+BOOTSTRAP = {
     'importlib._bootstrap',
     'importlib._bootstrap_external',
     'zipimport',
@@ -527,16 +528,24 @@ def regen_frozen(modules):
         header = relpath_for_posix_display(src.frozenfile, parentdir)
         headerlines.append(f'#include "{header}"')
 
-    deflines = []
+    bootstraplines = []
+    stdliblines = []
+    testlines = []
     aliaslines = []
     indent = '    '
     lastsection = None
     for mod in modules:
-        if mod.section != lastsection:
+        if mod.frozenid in BOOTSTRAP:
-            if lastsection is not None:
+            lines = bootstraplines
-                deflines.append('')
+        elif mod.section == TESTS_SECTION:
-            deflines.append(f'/* {mod.section} */')
+            lines = testlines
-        lastsection = mod.section
+        else:
+            lines = stdliblines
+            if mod.section != lastsection:
+                if lastsection is not None:
+                    lines.append('')
+                lines.append(f'/* {mod.section} */')
+            lastsection = mod.section
 
         symbol = mod.symbol
         pkg = '-' if mod.ispkg else ''
@@ -544,11 +553,11 @@ def regen_frozen(modules):
                 ) % (mod.name, symbol, pkg, symbol)
         # TODO: Consider not folding lines
         if len(line) < 80:
-            deflines.append(line)
+            lines.append(line)
         else:
             line1, _, line2 = line.rpartition(' ')
-            deflines.append(line1)
+            lines.append(line1)
-            deflines.append(indent + line2)
+            lines.append(indent + line2)
 
         if mod.isalias:
             if not mod.orig:
@@ -559,11 +568,13 @@ def regen_frozen(modules):
                 entry = '{"%s", "%s"},' % (mod.name, mod.orig)
             aliaslines.append(indent + entry)
 
-    if not deflines[0]:
+    for lines in (bootstraplines, stdliblines, testlines):
-        del deflines[0]
+        # TODO: Is this necessary any more?
-    for i, line in enumerate(deflines):
+        if not lines[0]:
-        if line:
+            del lines[0]
-            deflines[i] = indent + line
+        for i, line in enumerate(lines):
+            if line:
+                lines[i] = indent + line
 
     print(f'# Updating {os.path.relpath(FROZEN_FILE)}')
     with updating_file_with_tmpfile(FROZEN_FILE) as (infile, outfile):
@@ -579,9 +590,23 @@ def regen_frozen(modules):
         )
         lines = replace_block(
             lines,
-            "static const struct _frozen _PyImport_FrozenModules[] =",
+            "static const struct _frozen bootstrap_modules[] =",
-            "/* modules sentinel */",
+            "/* bootstrap sentinel */",
-            deflines,
+            bootstraplines,
+            FROZEN_FILE,
+        )
+        lines = replace_block(
+            lines,
+            "static const struct _frozen stdlib_modules[] =",
+            "/* stdlib sentinel */",
+            stdliblines,
+            FROZEN_FILE,
+        )
+        lines = replace_block(
+            lines,
+            "static const struct _frozen test_modules[] =",
+            "/* test sentinel */",
+            testlines,
             FROZEN_FILE,
         )
         lines = replace_block(