pythongh-86179: Implement realpath() on Windows for getpath.py calculations (pythonGH-113033)

Author: zooba

Lib/test/test_venv.py

@@ -47,13 +47,18 @@ def check_output(cmd, encoding=None):
     p = subprocess.Popen(cmd,
         stdout=subprocess.PIPE,
         stderr=subprocess.PIPE,
-        encoding=encoding)
+        env={**os.environ, "PYTHONHOME": ""})
     out, err = p.communicate()
     if p.returncode:
         if verbose and err:
-            print(err.decode('utf-8', 'backslashreplace'))
+            print(err.decode(encoding or 'utf-8', 'backslashreplace'))
         raise subprocess.CalledProcessError(
             p.returncode, cmd, out, err)
+    if encoding:
+        return (
+            out.decode(encoding, 'backslashreplace'),
+            err.decode(encoding, 'backslashreplace'),
+        )
     return out, err
 
 class BaseTest(unittest.TestCase):
@@ -273,8 +278,18 @@ def test_sysconfig(self):
             ('get_config_h_filename()', sysconfig.get_config_h_filename())):
             with self.subTest(call):
                 cmd[2] = 'import sysconfig; print(sysconfig.%s)' % call
-                out, err = check_output(cmd)
-                self.assertEqual(out.strip(), expected.encode(), err)
+                out, err = check_output(cmd, encoding='utf-8')
+                self.assertEqual(out.strip(), expected, err)
+        for attr, expected in (
+            ('executable', self.envpy()),
+            # Usually compare to sys.executable, but if we're running in our own
+            # venv then we really need to compare to our base executable
+            ('_base_executable', sys._base_executable),
+        ):
+            with self.subTest(attr):
+                cmd[2] = f'import sys; print(sys.{attr})'
+                out, err = check_output(cmd, encoding='utf-8')
+                self.assertEqual(out.strip(), expected, err)
 
     @requireVenvCreate
     @unittest.skipUnless(can_symlink(), 'Needs symlinks')
@@ -296,8 +311,18 @@ def test_sysconfig_symlinks(self):
             ('get_config_h_filename()', sysconfig.get_config_h_filename())):
             with self.subTest(call):
                 cmd[2] = 'import sysconfig; print(sysconfig.%s)' % call
-                out, err = check_output(cmd)
-                self.assertEqual(out.strip(), expected.encode(), err)
+                out, err = check_output(cmd, encoding='utf-8')
+                self.assertEqual(out.strip(), expected, err)
+        for attr, expected in (
+            ('executable', self.envpy()),
+            # Usually compare to sys.executable, but if we're running in our own
+            # venv then we really need to compare to our base executable
+            ('_base_executable', sys._base_executable),
+        ):
+            with self.subTest(attr):
+                cmd[2] = f'import sys; print(sys.{attr})'
+                out, err = check_output(cmd, encoding='utf-8')
+                self.assertEqual(out.strip(), expected, err)
 
     if sys.platform == 'win32':
         ENV_SUBDIRS = (

Misc/NEWS.d/next/Windows/2023-12-12-20-58-09.gh-issue-86179.YYSk_6.rst

@@ -0,0 +1 @@
+Fixes path calculations when launching Python on Windows through a symlink.

Modules/getpath.c

@@ -499,6 +499,54 @@ getpath_realpath(PyObject *Py_UNUSED(self) , PyObject *args)
     PyMem_Free((void *)path);
     PyMem_Free((void *)narrow);
     return r;
+#elif defined(MS_WINDOWS)
+    HANDLE hFile;
+    wchar_t resolved[MAXPATHLEN+1];
+    int len = 0, err;
+    Py_ssize_t pathlen;
+    PyObject *result;
+
+    wchar_t *path = PyUnicode_AsWideCharString(pathobj, &pathlen);
+    if (!path) {
+        return NULL;
+    }
+    if (wcslen(path) != pathlen) {
+        PyErr_SetString(PyExc_ValueError, "path contains embedded nulls");
+        return NULL;
+    }
+
+    Py_BEGIN_ALLOW_THREADS
+    hFile = CreateFileW(path, 0, 0, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
+    if (hFile != INVALID_HANDLE_VALUE) {
+        len = GetFinalPathNameByHandleW(hFile, resolved, MAXPATHLEN, VOLUME_NAME_DOS);
+        err = len ? 0 : GetLastError();
+        CloseHandle(hFile);
+    } else {
+        err = GetLastError();
+    }
+    Py_END_ALLOW_THREADS
+
+    if (err) {
+        PyErr_SetFromWindowsErr(err);
+        result = NULL;
+    } else if (len <= MAXPATHLEN) {
+        const wchar_t *p = resolved;
+        if (0 == wcsncmp(p, L"\\\\?\\", 4)) {
+            if (GetFileAttributesW(&p[4]) != INVALID_FILE_ATTRIBUTES) {
+                p += 4;
+                len -= 4;
+            }
+        }
+        if (CompareStringOrdinal(path, (int)pathlen, p, len, TRUE) == CSTR_EQUAL) {
+            result = Py_NewRef(pathobj);
+        } else {
+            result = PyUnicode_FromWideChar(p, len);
+        }
+    } else {
+        result = Py_NewRef(pathobj);
+    }
+    PyMem_Free(path);
+    return result;
 #endif
 
     return Py_NewRef(pathobj);