bpo-25996: Added support of file descriptors in os.scandir() on Unix. (#502)

os.fwalk() is sped up by 2 times by using os.scandir().

Doc/library/os.rst

@@ -2029,6 +2029,9 @@ features:
    attributes of each :class:`os.DirEntry` will be ``bytes``; in all other
    circumstances, they will be of type ``str``.
 
+   This function can also support :ref:`specifying a file descriptor
+   <path_fd>`; the file descriptor must refer to a directory.
+
    The :func:`scandir` iterator supports the :term:`context manager` protocol
    and has the following method:
 
@@ -2075,6 +2078,9 @@ features:
 
       The function accepts a :term:`path-like object`.
 
+   .. versionchanged:: 3.7
+      Added support for :ref:`file descriptors <path_fd>` on Unix.
+
 
 .. class:: DirEntry
 
@@ -2114,7 +2120,9 @@ features:
       The entry's full path name: equivalent to ``os.path.join(scandir_path,
       entry.name)`` where *scandir_path* is the :func:`scandir` *path*
       argument.  The path is only absolute if the :func:`scandir` *path*
-      argument was absolute.
+      argument was absolute.  If the :func:`scandir` *path*
+      argument was a :ref:`file descriptor <path_fd>`, the :attr:`path`
+      attribute is the same as the :attr:`name` attribute.
 
       The :attr:`path` attribute will be ``bytes`` if the :func:`scandir`
       *path* argument is of type ``bytes`` and ``str`` otherwise.  Use

Doc/whatsnew/3.7.rst

@@ -108,6 +108,9 @@ os
 Added support for :class:`bytes` paths in :func:`~os.fwalk`. (Contributed by
 Serhiy Storchaka in :issue:`28682`.)
 
+Added support for :ref:`file descriptors <path_fd>` in :func:`~os.scandir`
+on Unix.  (Contributed by Serhiy Storchaka in :issue:`25996`.)
+
 unittest.mock
 -------------
 
@@ -148,6 +151,10 @@ Optimizations
   :func:`~math.erfc` in the :mod:`math` module. (Contributed by Serhiy
   Storchaka in :issue:`26121`.)
 
+* The :func:`os.fwalk` function has been sped up by 2 times.  This was done
+  using the :func:`os.scandir` function.
+  (Contributed by Serhiy Storchaka in :issue:`25996`.)
+
 
 Build and C API Changes
 =======================

Lib/os.py

@@ -129,6 +129,7 @@ if _exists("_have_functions"):
     _add("HAVE_FCHMOD",     "chmod")
     _add("HAVE_FCHOWN",     "chown")
     _add("HAVE_FDOPENDIR",  "listdir")
+    _add("HAVE_FDOPENDIR",  "scandir")
     _add("HAVE_FEXECVE",    "execve")
     _set.add(stat) # fstat always works
     _add("HAVE_FTRUNCATE",  "truncate")
@@ -416,7 +417,7 @@ def walk(top, topdown=True, onerror=None, followlinks=False):
 
 __all__.append("walk")
 
-if {open, stat} <= supports_dir_fd and {listdir, stat} <= supports_fd:
+if {open, stat} <= supports_dir_fd and {scandir, stat} <= supports_fd:
 
     def fwalk(top=".", topdown=True, onerror=None, *, follow_symlinks=False, dir_fd=None):
         """Directory tree generator.
@@ -455,6 +456,7 @@ if {open, stat} <= supports_dir_fd and {listdir, stat} <= supports_fd:
             top = fspath(top)
         # Note: To guard against symlink races, we use the standard
         # lstat()/open()/fstat() trick.
+        if not follow_symlinks:
             orig_st = stat(top, follow_symlinks=False, dir_fd=dir_fd)
         topfd = open(top, O_RDONLY, dir_fd=dir_fd)
         try:
@@ -470,35 +472,41 @@ if {open, stat} <= supports_dir_fd and {listdir, stat} <= supports_fd:
         # necessary, it can be adapted to only require O(1) FDs, see issue
         # #13734.
 
-        names = listdir(topfd)
+        scandir_it = scandir(topfd)
+        dirs = []
+        nondirs = []
+        entries = None if topdown or follow_symlinks else []
+        for entry in scandir_it:
+            name = entry.name
             if isbytes:
-            names = map(fsencode, names)
-        dirs, nondirs = [], []
-        for name in names:
+                name = fsencode(name)
             try:
-                # Here, we don't use AT_SYMLINK_NOFOLLOW to be consistent with
-                # walk() which reports symlinks to directories as directories.
-                # We do however check for symlinks before recursing into
-                # a subdirectory.
-                if st.S_ISDIR(stat(name, dir_fd=topfd).st_mode):
+                if entry.is_dir():
                     dirs.append(name)
+                    if entries is not None:
+                        entries.append(entry)
                 else:
                     nondirs.append(name)
             except OSError:
                 try:
                     # Add dangling symlinks, ignore disappeared files
-                    if st.S_ISLNK(stat(name, dir_fd=topfd, follow_symlinks=False)
-                                .st_mode):
+                    if entry.is_symlink():
                         nondirs.append(name)
                 except OSError:
-                    continue
+                    pass
 
         if topdown:
             yield toppath, dirs, nondirs, topfd
 
-        for name in dirs:
+        for name in dirs if entries is None else zip(dirs, entries):
             try:
-                orig_st = stat(name, dir_fd=topfd, follow_symlinks=follow_symlinks)
+                if not follow_symlinks:
+                    if topdown:
+                        orig_st = stat(name, dir_fd=topfd, follow_symlinks=False)
+                    else:
+                        assert entries is not None
+                        name, entry = name
+                        orig_st = entry.stat(follow_symlinks=False)
                 dirfd = open(name, O_RDONLY, dir_fd=topfd)
             except OSError as err:
                 if onerror is not None:

Lib/test/test_os.py

@@ -3313,6 +3313,35 @@ class TestScandir(unittest.TestCase):
         self.assertEqual(entry.path,
                          os.fsencode(os.path.join(self.path, 'file.txt')))
 
+    @unittest.skipUnless(os.listdir in os.supports_fd,
+                         'fd support for listdir required for this test.')
+    def test_fd(self):
+        self.assertIn(os.scandir, os.supports_fd)
+        self.create_file('file.txt')
+        expected_names = ['file.txt']
+        if support.can_symlink():
+            os.symlink('file.txt', os.path.join(self.path, 'link'))
+            expected_names.append('link')
+
+        fd = os.open(self.path, os.O_RDONLY)
+        try:
+            with os.scandir(fd) as it:
+                entries = list(it)
+            names = [entry.name for entry in entries]
+            self.assertEqual(sorted(names), expected_names)
+            self.assertEqual(names, os.listdir(fd))
+            for entry in entries:
+                self.assertEqual(entry.path, entry.name)
+                self.assertEqual(os.fspath(entry), entry.name)
+                self.assertEqual(entry.is_symlink(), entry.name == 'link')
+                if os.stat in os.supports_dir_fd:
+                    st = os.stat(entry.name, dir_fd=fd)
+                    self.assertEqual(entry.stat(), st)
+                    st = os.stat(entry.name, dir_fd=fd, follow_symlinks=False)
+                    self.assertEqual(entry.stat(follow_symlinks=False), st)
+        finally:
+            os.close(fd)
+
     def test_empty_path(self):
         self.assertRaises(FileNotFoundError, os.scandir, '')
 
@@ -3328,7 +3357,7 @@ class TestScandir(unittest.TestCase):
         self.assertEqual(len(entries2), 0, entries2)
 
     def test_bad_path_type(self):
-        for obj in [1234, 1.234, {}, []]:
+        for obj in [1.234, {}, []]:
             self.assertRaises(TypeError, os.scandir, obj)
 
     def test_close(self):

Misc/NEWS

@@ -294,6 +294,9 @@ Extension Modules
 Library
 -------
 
+- bpo-25996: Added support of file descriptors in os.scandir() on Unix.
+  os.fwalk() is sped up by 2 times by using os.scandir().
+
 - bpo-28699: Fixed a bug in pools in multiprocessing.pool that raising an
   exception at the very first of an iterable may swallow the exception or
   make the program hang. Patch by Davin Potts and Xiang Zhang.

Modules/clinic/posixmodule.c.h

@@ -5926,7 +5926,7 @@ os_scandir(PyObject *module, PyObject **args, Py_ssize_t nargs, PyObject *kwname
     PyObject *return_value = NULL;
     static const char * const _keywords[] = {"path", NULL};
     static _PyArg_Parser _parser = {"|O&:scandir", _keywords, 0};
-    path_t path = PATH_T_INITIALIZE("scandir", "path", 1, 0);
+    path_t path = PATH_T_INITIALIZE("scandir", "path", 1, PATH_HAVE_FDOPENDIR);
 
     if (!_PyArg_ParseStackAndKeywords(args, nargs, kwnames, &_parser,
         path_converter, &path)) {
@@ -6493,4 +6493,4 @@ exit:
 #ifndef OS_GETRANDOM_METHODDEF
     #define OS_GETRANDOM_METHODDEF
 #endif /* !defined(OS_GETRANDOM_METHODDEF) */
-/*[clinic end generated code: output=5a0be969e3f71660 input=a9049054013a1b77]*/
+/*[clinic end generated code: output=5529857101c08b49 input=a9049054013a1b77]*/

Modules/posixmodule.c

@@ -11161,6 +11161,7 @@ typedef struct {
     unsigned char d_type;
 #endif
     ino_t d_ino;
+    int dir_fd;
 #endif
 } DirEntry;
 
@@ -11210,19 +11211,31 @@ DirEntry_fetch_stat(DirEntry *self, int follow_symlinks)
     PyObject *ub;
 
 #ifdef MS_WINDOWS
-    if (PyUnicode_FSDecoder(self->path, &ub)) {
+    if (!PyUnicode_FSDecoder(self->path, &ub))
+        return NULL;
     const wchar_t *path = PyUnicode_AsUnicode(ub);
 #else /* POSIX */
-    if (PyUnicode_FSConverter(self->path, &ub)) {
+    if (!PyUnicode_FSConverter(self->path, &ub))
+        return NULL;
     const char *path = PyBytes_AS_STRING(ub);
+    if (self->dir_fd != DEFAULT_DIR_FD) {
+#ifdef HAVE_FSTATAT
+        result = fstatat(self->dir_fd, path, &st,
+                         follow_symlinks ? 0 : AT_SYMLINK_NOFOLLOW);
+#else
+        PyErr_SetString(PyExc_NotImplementedError, "can't fetch stat");
+        return NULL;
+#endif /* HAVE_FSTATAT */
+    }
+    else
 #endif
+    {
         if (follow_symlinks)
             result = STAT(path, &st);
         else
             result = LSTAT(path, &st);
+    }
     Py_DECREF(ub);
-    } else
-        return NULL;
 
     if (result != 0)
         return path_object_error(self->path);
@@ -11633,20 +11646,36 @@ DirEntry_from_posix_info(path_t *path, const char *name, Py_ssize_t name_len,
     entry->stat = NULL;
     entry->lstat = NULL;
 
+    if (path->fd != -1) {
+        entry->dir_fd = path->fd;
+        joined_path = NULL;
+    }
+    else {
+        entry->dir_fd = DEFAULT_DIR_FD;
         joined_path = join_path_filename(path->narrow, name, name_len);
         if (!joined_path)
             goto error;
+    }
 
     if (!path->narrow || !PyBytes_Check(path->object)) {
         entry->name = PyUnicode_DecodeFSDefaultAndSize(name, name_len);
+        if (joined_path)
             entry->path = PyUnicode_DecodeFSDefault(joined_path);
     }
     else {
         entry->name = PyBytes_FromStringAndSize(name, name_len);
+        if (joined_path)
             entry->path = PyBytes_FromString(joined_path);
     }
     PyMem_Free(joined_path);
-    if (!entry->name || !entry->path)
+    if (!entry->name)
+        goto error;
+
+    if (path->fd != -1) {
+        entry->path = entry->name;
+        Py_INCREF(entry->path);
+    }
+    else if (!entry->path)
         goto error;
 
 #ifdef HAVE_DIRENT_D_TYPE
@@ -11674,6 +11703,9 @@ typedef struct {
 #else /* POSIX */
     DIR *dirp;
 #endif
+#ifdef HAVE_FDOPENDIR
+    int fd;
+#endif
 } ScandirIterator;
 
 #ifdef MS_WINDOWS
@@ -11758,6 +11790,10 @@ ScandirIterator_closedir(ScandirIterator *iterator)
 
     iterator->dirp = NULL;
     Py_BEGIN_ALLOW_THREADS
+#ifdef HAVE_FDOPENDIR
+    if (iterator->path.fd != -1)
+        rewinddir(dirp);
+#endif
     closedir(dirp);
     Py_END_ALLOW_THREADS
     return;
@@ -11933,7 +11969,7 @@ static PyTypeObject ScandirIteratorType = {
 /*[clinic input]
 os.scandir
 
-    path : path_t(nullable=True) = None
+    path : path_t(nullable=True, allow_fd='PATH_HAVE_FDOPENDIR') = None
 
 Return an iterator of DirEntry objects for given path.
 
@@ -11946,13 +11982,16 @@ If path is None, uses the path='.'.
 
 static PyObject *
 os_scandir_impl(PyObject *module, path_t *path)
-/*[clinic end generated code: output=6eb2668b675ca89e input=e62b08b3cd41f604]*/
+/*[clinic end generated code: output=6eb2668b675ca89e input=b139dc1c57f60846]*/
 {
     ScandirIterator *iterator;
 #ifdef MS_WINDOWS
     wchar_t *path_strW;
 #else
     const char *path_str;
+#ifdef HAVE_FDOPENDIR
+    int fd = -1;
+#endif
 #endif
 
     iterator = PyObject_New(ScandirIterator, &ScandirIteratorType);
@@ -11988,18 +12027,40 @@ os_scandir_impl(PyObject *module, path_t *path)
         goto error;
     }
 #else /* POSIX */
+    errno = 0;
+#ifdef HAVE_FDOPENDIR
+    if (path->fd != -1) {
+        /* closedir() closes the FD, so we duplicate it */
+        fd = _Py_dup(path->fd);
+        if (fd == -1)
+            goto error;
+
+        Py_BEGIN_ALLOW_THREADS
+        iterator->dirp = fdopendir(fd);
+        Py_END_ALLOW_THREADS
+    }
+    else
+#endif
+    {
         if (iterator->path.narrow)
             path_str = iterator->path.narrow;
         else
             path_str = ".";
 
-    errno = 0;
         Py_BEGIN_ALLOW_THREADS
         iterator->dirp = opendir(path_str);
         Py_END_ALLOW_THREADS
+    }
 
     if (!iterator->dirp) {
         path_error(&iterator->path);
+#ifdef HAVE_FDOPENDIR
+        if (fd != -1) {
+            Py_BEGIN_ALLOW_THREADS
+            close(fd);
+            Py_END_ALLOW_THREADS
+        }
+#endif
         goto error;
     }
 #endif