bpo-32030: Rework memory allocators (python#4625)

* Fix _PyMem_SetupAllocators("debug"): always restore allocators to
  the defaults, rather than only caling _PyMem_SetupDebugHooks().
* Add _PyMem_SetDefaultAllocator() helper to set the "default"
  allocator.
* Add _PyMem_GetAllocatorsName(): get the name of the allocators
* main() now uses debug hooks on memory allocators if Py_DEBUG is
  defined, rather than calling directly malloc()
* Document default memory allocators in C API documentation
* _Py_InitializeCore() now fails with a fatal user error if
  PYTHONMALLOC value is an unknown memory allocator, instead of
  failing with a fatal internal error.
* Add new tests on the PYTHONMALLOC environment variable
* Add support.with_pymalloc()
* Add the _testcapi.WITH_PYMALLOC constant and expose it as
   support.with_pymalloc().
* sysconfig.get_config_var('WITH_PYMALLOC') doesn't work on Windows, so
   replace it with support.with_pymalloc().
* pythoninfo: add _testcapi collector for pymem

Doc/c-api/memory.rst

@@ -100,9 +100,10 @@ The following function sets are wrappers to the system allocator. These
 functions are thread-safe, the :term:`GIL <global interpreter lock>` does not
 need to be held.
 
-The default raw memory block allocator uses the following functions:
-:c:func:`malloc`, :c:func:`calloc`, :c:func:`realloc` and :c:func:`free`; call
-``malloc(1)`` (or ``calloc(1, 1)``) when requesting zero bytes.
+The :ref:`default raw memory allocator <default-memory-allocators>` uses
+the following functions: :c:func:`malloc`, :c:func:`calloc`, :c:func:`realloc`
+and :c:func:`free`; call ``malloc(1)`` (or ``calloc(1, 1)``) when requesting
+zero bytes.
 
 .. versionadded:: 3.4
 
@@ -165,7 +166,8 @@ The following function sets, modeled after the ANSI C standard, but specifying
 behavior when requesting zero bytes, are available for allocating and releasing
 memory from the Python heap.
 
-By default, these functions use :ref:`pymalloc memory allocator <pymalloc>`.
+The :ref:`default memory allocator <default-memory-allocators>` uses the
+:ref:`pymalloc memory allocator <pymalloc>`.
 
 .. warning::
 
@@ -270,7 +272,8 @@ The following function sets, modeled after the ANSI C standard, but specifying
 behavior when requesting zero bytes, are available for allocating and releasing
 memory from the Python heap.
 
-By default, these functions use :ref:`pymalloc memory allocator <pymalloc>`.
+The :ref:`default object allocator <default-memory-allocators>` uses the
+:ref:`pymalloc memory allocator <pymalloc>`.
 
 .. warning::
 
@@ -326,6 +329,31 @@ By default, these functions use :ref:`pymalloc memory allocator <pymalloc>`.
    If *p* is *NULL*, no operation is performed.
 
 
+.. _default-memory-allocators:
+
+Default Memory Allocators
+=========================
+
+Default memory allocators:
+
+===============================  ====================  ==================  =====================  ====================
+Configuration                    Name                  PyMem_RawMalloc     PyMem_Malloc           PyObject_Malloc
+===============================  ====================  ==================  =====================  ====================
+Release build                    ``"pymalloc"``        ``malloc``          ``pymalloc``           ``pymalloc``
+Debug build                      ``"pymalloc_debug"``  ``malloc`` + debug  ``pymalloc`` + debug   ``pymalloc`` + debug
+Release build, without pymalloc  ``"malloc"``          ``malloc``          ``malloc``             ``malloc``
+Release build, without pymalloc  ``"malloc_debug"``    ``malloc`` + debug  ``malloc`` + debug     ``malloc`` + debug
+===============================  ====================  ==================  =====================  ====================
+
+Legend:
+
+* Name: value for :envvar:`PYTHONMALLOC` environment variable
+* ``malloc``: system allocators from the standard C library, C functions:
+  :c:func:`malloc`, :c:func:`calloc`, :c:func:`realloc` and :c:func:`free`
+* ``pymalloc``: :ref:`pymalloc memory allocator <pymalloc>`
+* "+ debug": with debug hooks installed by :c:func:`PyMem_SetupDebugHooks`
+
+
 Customize Memory Allocators
 ===========================
 
@@ -431,7 +459,8 @@ Customize Memory Allocators
    displayed if :mod:`tracemalloc` is tracing Python memory allocations and the
    memory block was traced.
 
-   These hooks are installed by default if Python is compiled in debug
+   These hooks are :ref:`installed by default <default-memory-allocators>` if
+   Python is compiled in debug
    mode. The :envvar:`PYTHONMALLOC` environment variable can be used to install
    debug hooks on a Python compiled in release mode.
 
@@ -453,9 +482,9 @@ to 512 bytes) with a short lifetime. It uses memory mappings called "arenas"
 with a fixed size of 256 KiB. It falls back to :c:func:`PyMem_RawMalloc` and
 :c:func:`PyMem_RawRealloc` for allocations larger than 512 bytes.
 
-*pymalloc* is the default allocator of the :c:data:`PYMEM_DOMAIN_MEM` (ex:
-:c:func:`PyMem_Malloc`) and :c:data:`PYMEM_DOMAIN_OBJ` (ex:
-:c:func:`PyObject_Malloc`) domains.
+*pymalloc* is the :ref:`default allocator <default-memory-allocators>` of the
+:c:data:`PYMEM_DOMAIN_MEM` (ex: :c:func:`PyMem_Malloc`) and
+:c:data:`PYMEM_DOMAIN_OBJ` (ex: :c:func:`PyObject_Malloc`) domains.
 
 The arena allocator uses the following functions:
 

Doc/using/cmdline.rst

@@ -687,6 +687,8 @@ conflict.
 
    Set the family of memory allocators used by Python:
 
+   * ``default``: use the :ref:`default memory allocators
+     <default-memory-allocators>`.
    * ``malloc``: use the :c:func:`malloc` function of the C library
      for all domains (:c:data:`PYMEM_DOMAIN_RAW`, :c:data:`PYMEM_DOMAIN_MEM`,
      :c:data:`PYMEM_DOMAIN_OBJ`).
@@ -696,20 +698,17 @@ conflict.
 
    Install debug hooks:
 
-   * ``debug``: install debug hooks on top of the default memory allocator
+   * ``debug``: install debug hooks on top of the :ref:`default memory
+     allocators <default-memory-allocators>`.
    * ``malloc_debug``: same as ``malloc`` but also install debug hooks
    * ``pymalloc_debug``: same as ``pymalloc`` but also install debug hooks
 
-   When Python is compiled in release mode, the default is ``pymalloc``. When
-   compiled in debug mode, the default is ``pymalloc_debug`` and the debug hooks
-   are used automatically.
+   See the :ref:`default memory allocators <default-memory-allocators>` and the
+   :c:func:`PyMem_SetupDebugHooks` function (install debug hooks on Python
+   memory allocators).
 
-   If Python is configured without ``pymalloc`` support, ``pymalloc`` and
-   ``pymalloc_debug`` are not available, the default is ``malloc`` in release
-   mode and ``malloc_debug`` in debug mode.
-
-   See the :c:func:`PyMem_SetupDebugHooks` function for debug hooks on Python
-   memory allocators.
+   .. versionchanged:: 3.7
+      Added the ``"default"`` allocator.
 
    .. versionadded:: 3.6
 

Include/pymem.h

@@ -21,6 +21,9 @@ PyAPI_FUNC(void) PyMem_RawFree(void *ptr);
    allocators. */
 PyAPI_FUNC(int) _PyMem_SetupAllocators(const char *opt);
 
+/* Try to get the allocators name set by _PyMem_SetupAllocators(). */
+PyAPI_FUNC(const char*) _PyMem_GetAllocatorsName(void);
+
 #ifdef WITH_PYMALLOC
 PyAPI_FUNC(int) _PyMem_PymallocEnabled(void);
 #endif
@@ -230,7 +233,12 @@ PyAPI_FUNC(void) PyMem_SetupDebugHooks(void);
 #endif
 
 #ifdef Py_BUILD_CORE
-PyAPI_FUNC(void) _PyMem_GetDefaultRawAllocator(PyMemAllocatorEx *alloc);
+/* Set the memory allocator of the specified domain to the default.
+   Save the old allocator into *old_alloc if it's non-NULL.
+   Return on success, or return -1 if the domain is unknown. */
+PyAPI_FUNC(int) _PyMem_SetDefaultAllocator(
+    PyMemAllocatorDomain domain,
+    PyMemAllocatorEx *old_alloc);
 #endif
 
 #ifdef __cplusplus

Lib/test/pythoninfo.py

@@ -56,6 +56,14 @@ def copy_attributes(info_add, obj, name_fmt, attributes, *, formatter=None):
         info_add(name, value)
 
 
+def copy_attr(info_add, name, mod, attr_name):
+    try:
+        value = getattr(mod, attr_name)
+    except AttributeError:
+        return
+    info_add(name, value)
+
+
 def call_func(info_add, name, mod, func_name, *, formatter=None):
     try:
         func = getattr(mod, func_name)
@@ -168,11 +176,10 @@ def format_attr(attr, value):
     call_func(info_add, 'os.gid', os, 'getgid')
     call_func(info_add, 'os.uname', os, 'uname')
 
-    if hasattr(os, 'getgroups'):
-        groups = os.getgroups()
-        groups = map(str, groups)
-        groups = ', '.join(groups)
-        info_add("os.groups", groups)
+    def format_groups(groups):
+        return ', '.join(map(str, groups))
+
+    call_func(info_add, 'os.groups', os, 'getgroups', formatter=format_groups)
 
     if hasattr(os, 'getlogin'):
         try:
@@ -184,11 +191,7 @@ def format_attr(attr, value):
         else:
             info_add("os.login", login)
 
-    if hasattr(os, 'cpu_count'):
-        cpu_count = os.cpu_count()
-        if cpu_count:
-            info_add('os.cpu_count', cpu_count)
-
+    call_func(info_add, 'os.cpu_count', os, 'cpu_count')
     call_func(info_add, 'os.loadavg', os, 'getloadavg')
 
     # Get environment variables: filter to list
@@ -219,7 +222,9 @@ def format_attr(attr, value):
     )
     for name, value in os.environ.items():
         uname = name.upper()
-        if (uname in ENV_VARS or uname.startswith(("PYTHON", "LC_"))
+        if (uname in ENV_VARS
+           # Copy PYTHON* and LC_* variables
+           or uname.startswith(("PYTHON", "LC_"))
            # Visual Studio: VS140COMNTOOLS
            or (uname.startswith("VS") and uname.endswith("COMNTOOLS"))):
             info_add('os.environ[%s]' % name, value)
@@ -313,12 +318,10 @@ def collect_time(info_add):
     )
     copy_attributes(info_add, time, 'time.%s', attributes)
 
-    if not hasattr(time, 'get_clock_info'):
-        return
-
-    for clock in ('time', 'perf_counter'):
-        tinfo = time.get_clock_info(clock)
-        info_add('time.%s' % clock, tinfo)
+    if hasattr(time, 'get_clock_info'):
+        for clock in ('time', 'perf_counter'):
+            tinfo = time.get_clock_info(clock)
+            info_add('time.%s' % clock, tinfo)
 
 
 def collect_sysconfig(info_add):
@@ -331,14 +334,14 @@ def collect_sysconfig(info_add):
         'CCSHARED',
         'CFLAGS',
         'CFLAGSFORSHARED',
-        'PY_LDFLAGS',
         'CONFIG_ARGS',
         'HOST_GNU_TYPE',
         'MACHDEP',
         'MULTIARCH',
         'OPT',
         'PY_CFLAGS',
         'PY_CFLAGS_NODIST',
+        'PY_LDFLAGS',
         'Py_DEBUG',
         'Py_ENABLE_SHARED',
         'SHELL',
@@ -422,6 +425,16 @@ def collect_decimal(info_add):
     copy_attributes(info_add, _decimal, '_decimal.%s', attributes)
 
 
+def collect_testcapi(info_add):
+    try:
+        import _testcapi
+    except ImportError:
+        return
+
+    call_func(info_add, 'pymem.allocator', _testcapi, 'pymem_getallocatorsname')
+    copy_attr(info_add, 'pymem.with_pymalloc', _testcapi, 'WITH_PYMALLOC')
+
+
 def collect_info(info):
     error = False
     info_add = info.add
@@ -444,6 +457,7 @@ def collect_info(info):
         collect_zlib,
         collect_expat,
         collect_decimal,
+        collect_testcapi,
     ):
         try:
             collect_func(info_add)

Lib/test/support/__init__.py

@@ -2848,3 +2848,8 @@ def save(self):
     def restore(self):
         for signum, handler in self.handlers.items():
             self.signal.signal(signum, handler)
+
+
+def with_pymalloc():
+    import _testcapi
+    return _testcapi.WITH_PYMALLOC

Lib/test/test_capi.py

@@ -654,8 +654,7 @@ class PyMemMallocDebugTests(PyMemDebugTests):
     PYTHONMALLOC = 'malloc_debug'
 
 
-@unittest.skipUnless(sysconfig.get_config_var('WITH_PYMALLOC') == 1,
-                     'need pymalloc')
+@unittest.skipUnless(support.with_pymalloc(), 'need pymalloc')
 class PyMemPymallocDebugTests(PyMemDebugTests):
     PYTHONMALLOC = 'pymalloc_debug'
 

Lib/test/test_cmd_line.py

@@ -5,6 +5,7 @@
 import os
 import subprocess
 import sys
+import sysconfig
 import tempfile
 import unittest
 from test import support
@@ -559,10 +560,14 @@ def test_xdev(self):
         except ImportError:
             pass
         else:
-            code = "import _testcapi; _testcapi.pymem_api_misuse()"
+            code = "import _testcapi; print(_testcapi.pymem_getallocatorsname())"
             with support.SuppressCrashReport():
                 out = self.run_xdev("-c", code, check_exitcode=False)
-            self.assertIn("Debug memory block at address p=", out)
+            if support.with_pymalloc():
+                alloc_name = "pymalloc_debug"
+            else:
+                alloc_name = "malloc_debug"
+            self.assertEqual(out, alloc_name)
 
         try:
             import faulthandler
@@ -573,6 +578,49 @@ def test_xdev(self):
             out = self.run_xdev("-c", code)
             self.assertEqual(out, "True")
 
+    def check_pythonmalloc(self, env_var, name):
+        code = 'import _testcapi; print(_testcapi.pymem_getallocatorsname())'
+        env = dict(os.environ)
+        if env_var is not None:
+            env['PYTHONMALLOC'] = env_var
+        else:
+            env.pop('PYTHONMALLOC', None)
+        args = (sys.executable, '-c', code)
+        proc = subprocess.run(args,
+                              stdout=subprocess.PIPE,
+                              stderr=subprocess.STDOUT,
+                              universal_newlines=True,
+                              env=env)
+        self.assertEqual(proc.stdout.rstrip(), name)
+        self.assertEqual(proc.returncode, 0)
+
+    def test_pythonmalloc(self):
+        # Test the PYTHONMALLOC environment variable
+        pydebug = hasattr(sys, "gettotalrefcount")
+        pymalloc = support.with_pymalloc()
+        if pymalloc:
+            default_name = 'pymalloc_debug' if pydebug else 'pymalloc'
+            default_name_debug = 'pymalloc_debug'
+        else:
+            default_name = 'malloc_debug' if pydebug else 'malloc'
+            default_name_debug = 'malloc_debug'
+
+        tests = [
+            (None, default_name),
+            ('debug', default_name_debug),
+            ('malloc', 'malloc'),
+            ('malloc_debug', 'malloc_debug'),
+        ]
+        if pymalloc:
+            tests.extend((
+                ('pymalloc', 'pymalloc'),
+                ('pymalloc_debug', 'pymalloc_debug'),
+            ))
+
+        for env_var, name in tests:
+            with self.subTest(env_var=env_var, name=name):
+                self.check_pythonmalloc(env_var, name)
+
 
 class IgnoreEnvironmentTest(unittest.TestCase):
 

Lib/test/test_sys.py

@@ -753,8 +753,15 @@ def test_debugmallocstats(self):
     @unittest.skipUnless(hasattr(sys, "getallocatedblocks"),
                          "sys.getallocatedblocks unavailable on this build")
     def test_getallocatedblocks(self):
+        try:
+            import _testcapi
+        except ImportError:
+            with_pymalloc = support.with_pymalloc()
+        else:
+            alloc_name = _testcapi.pymem_getallocatorsname()
+            with_pymalloc = (alloc_name in ('pymalloc', 'pymalloc_debug'))
+
         # Some sanity checks
-        with_pymalloc = sysconfig.get_config_var('WITH_PYMALLOC')
         a = sys.getallocatedblocks()
         self.assertIs(type(a), int)
         if with_pymalloc: