pythongh-99726: Improves correctness of stat results for Windows, and…

… uses faster API when available (pythonGH-102149)

This deprecates `st_ctime` fields on Windows, with the intent to change them to contain the correct value in 3.14. For now, they should keep returning the creation time as they always have.

Doc/library/os.rst

@@ -2858,6 +2858,12 @@ features:
       Added support for the :class:`~os.PathLike` interface.  Added support
       for :class:`bytes` paths on Windows.
 
+   .. versionchanged:: 3.12
+      The ``st_ctime`` attribute of a stat result is deprecated on Windows.
+      The file creation time is properly available as ``st_birthtime``, and
+      in the future ``st_ctime`` may be changed to return zero or the
+      metadata change time, if available.
+
 
 .. function:: stat(path, *, dir_fd=None, follow_symlinks=True)
 
@@ -2973,10 +2979,12 @@ features:
 
    .. attribute:: st_ctime
 
-      Platform dependent:
+      Time of most recent metadata change expressed in seconds.
 
-      * the time of most recent metadata change on Unix,
-      * the time of creation on Windows, expressed in seconds.
+      .. versionchanged:: 3.12
+         ``st_ctime`` is deprecated on Windows. Use ``st_birthtime`` for
+         the file creation time. In the future, ``st_ctime`` will contain
+         the time of the most recent metadata change, as for other platforms.
 
    .. attribute:: st_atime_ns
 
@@ -2989,29 +2997,48 @@ features:
 
    .. attribute:: st_ctime_ns
 
-      Platform dependent:
+      Time of most recent metadata change expressed in nanoseconds as an
+      integer.
+
+      .. versionchanged:: 3.12
+         ``st_ctime_ns`` is deprecated on Windows. Use ``st_birthtime_ns``
+         for the file creation time. In the future, ``st_ctime`` will contain
+         the time of the most recent metadata change, as for other platforms.
+
+   .. attribute:: st_birthtime
+
+      Time of file creation expressed in seconds. This attribute is not
+      always available, and may raise :exc:`AttributeError`.
+
+      .. versionchanged:: 3.12
+         ``st_birthtime`` is now available on Windows.
+
+   .. attribute:: st_birthtime_ns
 
-      * the time of most recent metadata change on Unix,
-      * the time of creation on Windows, expressed in nanoseconds as an
-        integer.
+      Time of file creation expressed in nanoseconds as an integer.
+      This attribute is not always available, and may raise
+      :exc:`AttributeError`.
+
+      .. versionadded:: 3.12
 
    .. note::
 
       The exact meaning and resolution of the :attr:`st_atime`,
-      :attr:`st_mtime`, and :attr:`st_ctime` attributes depend on the operating
-      system and the file system. For example, on Windows systems using the FAT
-      or FAT32 file systems, :attr:`st_mtime` has 2-second resolution, and
-      :attr:`st_atime` has only 1-day resolution.  See your operating system
-      documentation for details.
+      :attr:`st_mtime`, :attr:`st_ctime` and :attr:`st_birthtime` attributes
+      depend on the operating system and the file system. For example, on
+      Windows systems using the FAT32 file systems, :attr:`st_mtime` has
+      2-second resolution, and :attr:`st_atime` has only 1-day resolution.
+      See your operating system documentation for details.
 
       Similarly, although :attr:`st_atime_ns`, :attr:`st_mtime_ns`,
-      and :attr:`st_ctime_ns` are always expressed in nanoseconds, many
-      systems do not provide nanosecond precision.  On systems that do
-      provide nanosecond precision, the floating-point object used to
-      store :attr:`st_atime`, :attr:`st_mtime`, and :attr:`st_ctime`
-      cannot preserve all of it, and as such will be slightly inexact.
-      If you need the exact timestamps you should always use
-      :attr:`st_atime_ns`, :attr:`st_mtime_ns`, and :attr:`st_ctime_ns`.
+      :attr:`st_ctime_ns` and :attr:`st_birthtime_ns` are always expressed in
+      nanoseconds, many systems do not provide nanosecond precision.  On
+      systems that do provide nanosecond precision, the floating-point object
+      used to store :attr:`st_atime`, :attr:`st_mtime`, :attr:`st_ctime` and
+      :attr:`st_birthtime` cannot preserve all of it, and as such will be
+      slightly inexact. If you need the exact timestamps you should always use
+      :attr:`st_atime_ns`, :attr:`st_mtime_ns`, :attr:`st_ctime_ns` and
+      :attr:`st_birthtime_ns`.
 
    On some Unix systems (such as Linux), the following attributes may also be
    available:
@@ -3041,10 +3068,6 @@ features:
 
       File generation number.
 
-   .. attribute:: st_birthtime
-
-      Time of file creation.
-
    On Solaris and derivatives, the following attributes may also be
    available:
 
@@ -3117,6 +3140,25 @@ features:
       files as :const:`S_IFCHR`, :const:`S_IFIFO` or :const:`S_IFBLK`
       as appropriate.
 
+   .. versionchanged:: 3.12
+      On Windows, :attr:`st_ctime` is deprecated. Eventually, it will
+      contain the last metadata change time, for consistency with other
+      platforms, but for now still contains creation time.
+      Use :attr:`st_birthtime` for the creation time.
+
+   .. versionchanged:: 3.12
+      On Windows, :attr:`st_ino` may now be up to 128 bits, depending
+      on the file system. Previously it would not be above 64 bits, and
+      larger file identifiers would be arbitrarily packed.
+
+   .. versionchanged:: 3.12
+      On Windows, :attr:`st_rdev` no longer returns a value. Previously
+      it would contain the same as :attr:`st_dev`, which was incorrect.
+
+   .. versionadded:: 3.12
+      Added the :attr:`st_birthtime` member on Windows.
+
+
 .. function:: statvfs(path)
 
    Perform a :c:func:`statvfs` system call on the given path.  The return value is

Doc/whatsnew/3.12.rst

@@ -306,6 +306,16 @@ os
   functions on Windows for enumerating drives, volumes and mount points.
   (Contributed by Steve Dower in :gh:`102519`.)
 
+* :func:`os.stat` and :func:`os.lstat` are now more accurate on Windows.
+  The ``st_birthtime`` field will now be filled with the creation time
+  of the file, and ``st_ctime`` is deprecated but still contains the
+  creation time (but in the future will return the last metadata change,
+  for consistency with other platforms). ``st_dev`` may be up to 64 bits
+  and ``st_ino`` up to 128 bits depending on your file system, and
+  ``st_rdev`` is always set to zero rather than incorrect values.
+  Both functions may be significantly faster on newer releases of
+  Windows. (Contributed by Steve Dower in :gh:`99726`.)
+
 os.path
 -------
 
@@ -469,6 +479,12 @@ Deprecated
   warning at compile time. This field will be removed in Python 3.14.
   (Contributed by Ramvikrams and Kumar Aditya in :gh:`101193`. PEP by Ken Jin.)
 
+* The ``st_ctime`` fields return by :func:`os.stat` and :func:`os.lstat` on
+  Windows are deprecated. In a future release, they will contain the last
+  metadata change time, consistent with other platforms. For now, they still
+  contain the creation time, which is also available in the new ``st_birthtime``
+  field. (Contributed by Steve Dower in :gh:`99726`.)
+
 Pending Removal in Python 3.13
 ------------------------------
 

Include/internal/pycore_fileutils.h

@@ -66,7 +66,7 @@ PyAPI_FUNC(PyObject *) _Py_device_encoding(int);
 
 #ifdef MS_WINDOWS
 struct _Py_stat_struct {
-    unsigned long st_dev;
+    uint64_t st_dev;
     uint64_t st_ino;
     unsigned short st_mode;
     int st_nlink;
@@ -80,8 +80,11 @@ struct _Py_stat_struct {
     int st_mtime_nsec;
     time_t st_ctime;
     int st_ctime_nsec;
+    time_t st_birthtime;
+    int st_birthtime_nsec;
     unsigned long st_file_attributes;
     unsigned long st_reparse_tag;
+    uint64_t st_ino_high;
 };
 #else
 #  define _Py_stat_struct stat

Include/internal/pycore_fileutils_windows.h

@@ -0,0 +1,80 @@
+#ifndef Py_INTERNAL_FILEUTILS_WINDOWS_H
+#define Py_INTERNAL_FILEUTILS_WINDOWS_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "Py_BUILD_CORE must be defined to include this header"
+#endif
+
+#ifdef MS_WINDOWS
+
+#if !defined(NTDDI_WIN10_NI) || !(NTDDI_VERSION >= NTDDI_WIN10_NI)
+typedef struct _FILE_STAT_BASIC_INFORMATION {
+    LARGE_INTEGER FileId;
+    LARGE_INTEGER CreationTime;
+    LARGE_INTEGER LastAccessTime;
+    LARGE_INTEGER LastWriteTime;
+    LARGE_INTEGER ChangeTime;
+    LARGE_INTEGER AllocationSize;
+    LARGE_INTEGER EndOfFile;
+    ULONG FileAttributes;
+    ULONG ReparseTag;
+    ULONG NumberOfLinks;
+    ULONG DeviceType;
+    ULONG DeviceCharacteristics;
+    ULONG Reserved;
+    FILE_ID_128 FileId128;
+    LARGE_INTEGER VolumeSerialNumber;
+} FILE_STAT_BASIC_INFORMATION;
+
+typedef enum _FILE_INFO_BY_NAME_CLASS {
+    FileStatByNameInfo,
+    FileStatLxByNameInfo,
+    FileCaseSensitiveByNameInfo,
+    FileStatBasicByNameInfo,
+    MaximumFileInfoByNameClass
+} FILE_INFO_BY_NAME_CLASS;
+#endif
+
+typedef BOOL (WINAPI *PGetFileInformationByName)(
+    PCWSTR FileName,
+    FILE_INFO_BY_NAME_CLASS FileInformationClass,
+    PVOID FileInfoBuffer,
+    ULONG FileInfoBufferSize
+);
+
+static inline BOOL _Py_GetFileInformationByName(
+    PCWSTR FileName,
+    FILE_INFO_BY_NAME_CLASS FileInformationClass,
+    PVOID FileInfoBuffer,
+    ULONG FileInfoBufferSize
+) {
+    static PGetFileInformationByName GetFileInformationByName = NULL;
+    static int GetFileInformationByName_init = -1;
+
+    if (GetFileInformationByName_init < 0) {
+        HMODULE hMod = LoadLibraryW(L"api-ms-win-core-file-l2-1-4");
+        GetFileInformationByName_init = 0;
+        if (hMod) {
+            GetFileInformationByName = (PGetFileInformationByName)GetProcAddress(
+                hMod, "GetFileInformationByName");
+            if (GetFileInformationByName) {
+                GetFileInformationByName_init = 1;
+            } else {
+                FreeLibrary(hMod);
+            }
+        }
+    }
+
+    if (GetFileInformationByName_init <= 0) {
+        SetLastError(ERROR_NOT_SUPPORTED);
+        return FALSE;
+    }
+    return GetFileInformationByName(FileName, FileInformationClass, FileInfoBuffer, FileInfoBufferSize);
+}
+
+#endif
+
+#endif

Lib/test/test_os.py

@@ -556,6 +556,15 @@ def trunc(x): return x
             nanosecondy = getattr(result, name + "_ns") // 10000
             self.assertAlmostEqual(floaty, nanosecondy, delta=2)
 
+        # Ensure both birthtime and birthtime_ns roughly agree, if present
+        try:
+            floaty = int(result.st_birthtime * 100000)
+            nanosecondy = result.st_birthtime_ns // 10000
+        except AttributeError:
+            pass
+        else:
+            self.assertAlmostEqual(floaty, nanosecondy, delta=2)
+
         try:
             result[200]
             self.fail("No exception raised")
@@ -4234,7 +4243,8 @@ def assert_stat_equal(self, stat1, stat2, skip_fields):
             for attr in dir(stat1):
                 if not attr.startswith("st_"):
                     continue
-                if attr in ("st_dev", "st_ino", "st_nlink"):
+                if attr in ("st_dev", "st_ino", "st_nlink", "st_ctime",
+                            "st_ctime_ns"):
                     continue
                 self.assertEqual(getattr(stat1, attr),
                                  getattr(stat2, attr),

Misc/NEWS.d/next/Windows/2023-02-22-17-26-10.gh-issue-99726.76t957.rst

@@ -0,0 +1,2 @@
+Improves correctness of stat results for Windows, and uses faster API when
+available