From dca72d38d6a8887534fba947afa527eb8c18eaa7 Mon Sep 17 00:00:00 2001
From: Serhiy Storchaka <storchaka@gmail.com>
Date: Mon, 31 Aug 2026 17:00:44 +0300
Subject: [PATCH] gh-156713: Use the filesystem encoding in nturl2path

urllib.request.pathname2url() and url2pathname() use the filesystem encoding
and error handler since gh-85168, but nturl2path, which implements them on
Windows before 3.14, was left unchanged.  Paths containing surrogate
characters raised UnicodeEncodeError.
---
 Lib/nturl2path.py                             | 14 ++++++++++---
 Lib/test/test_nturl2path.py                   | 21 +++++++++++++++++++
 ...-08-31-17-10-00.gh-issue-156713.Nt2URL.rst |  4 ++++
 3 files changed, 36 insertions(+), 3 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Library/2026-08-31-17-10-00.gh-issue-156713.Nt2URL.rst

diff --git a/Lib/nturl2path.py b/Lib/nturl2path.py
index 57c7858dff0b818..47c49ca2020c5c8 100644
--- a/Lib/nturl2path.py
+++ b/Lib/nturl2path.py
@@ -22,7 +22,10 @@ def url2pathname(url):
     #   ///C:/foo/bar/spam.foo
     # become
     #   C:\foo\bar\spam.foo
+    import sys
     import urllib.parse
+    encoding = sys.getfilesystemencoding()
+    errors = sys.getfilesystemencodeerrors()
     if url[:3] == '///':
         # URL has an empty authority section, so the path begins on the third
         # character.
@@ -40,7 +43,8 @@ def url2pathname(url):
         if url[1:2] == '|':
             # Older URLs use a pipe after a drive letter
             url = url[:1] + ':' + url[2:]
-    return urllib.parse.unquote(url.replace('/', '\\'))
+    return urllib.parse.unquote(url.replace('/', '\\'),
+                                encoding=encoding, errors=errors)
 
 def pathname2url(p):
     """OS-specific conversion from a file system path to a relative URL
@@ -50,7 +54,10 @@ def pathname2url(p):
     # becomes
     #   ///C:/foo/bar/spam.foo
     import ntpath
+    import sys
     import urllib.parse
+    encoding = sys.getfilesystemencoding()
+    errors = sys.getfilesystemencodeerrors()
     # First, clean up some special forms. We are going to sacrifice
     # the additional information anyway
     p = p.replace('\\', '/')
@@ -65,10 +72,11 @@ def pathname2url(p):
             # an authority section with a zero-length authority, and a path
             # section starting with a single slash.
             drive = f'///{drive}'
-        drive = urllib.parse.quote(drive, safe='/:')
+        drive = urllib.parse.quote(drive, encoding=encoding, errors=errors,
+                                   safe='/:')
     elif root:
         # Add explicitly empty authority to path beginning with one slash.
         root = f'//{root}'
 
-    tail = urllib.parse.quote(tail)
+    tail = urllib.parse.quote(tail, encoding=encoding, errors=errors)
     return drive + root + tail
diff --git a/Lib/test/test_nturl2path.py b/Lib/test/test_nturl2path.py
index a6a3422a0f75b2e..b4532137968d6e3 100644
--- a/Lib/test/test_nturl2path.py
+++ b/Lib/test/test_nturl2path.py
@@ -1,4 +1,6 @@
+import sys
 import unittest
+import urllib.parse
 
 from test.support import warnings_helper
 
@@ -58,6 +60,15 @@ def test_pathname2url(self):
         for url in urls:
             self.assertEqual(fn(nturl2path.url2pathname(url)), url)
 
+    def test_pathname2url_surrogates(self):
+        # gh-156713: the filesystem encoding and error handler are used,
+        # so that paths containing surrogate characters can be converted.
+        encoding = sys.getfilesystemencoding()
+        errors = sys.getfilesystemencodeerrors()
+        tail = urllib.parse.quote('a\udcff', encoding=encoding, errors=errors)
+        self.assertEqual(nturl2path.pathname2url('C:\\a\udcff'),
+                         '///C:/' + tail)
+
     def test_url2pathname(self):
         fn = nturl2path.url2pathname
         self.assertEqual(fn('/'), '\\')
@@ -103,5 +114,15 @@ def test_url2pathname(self):
             self.assertEqual(fn(nturl2path.pathname2url(path)), path)
 
 
+    def test_url2pathname_surrogates(self):
+        # gh-156713: the filesystem encoding and error handler are used, so
+        # that URLs containing percent-encoded surrogates can be converted.
+        encoding = sys.getfilesystemencoding()
+        errors = sys.getfilesystemencodeerrors()
+        url = urllib.parse.quote('a\udcff', encoding=encoding, errors=errors)
+        self.assertEqual(nturl2path.url2pathname('///C:/' + url),
+                         'C:\\a\udcff')
+
+
 if __name__ == '__main__':
     unittest.main()
diff --git a/Misc/NEWS.d/next/Library/2026-08-31-17-10-00.gh-issue-156713.Nt2URL.rst b/Misc/NEWS.d/next/Library/2026-08-31-17-10-00.gh-issue-156713.Nt2URL.rst
new file mode 100644
index 000000000000000..1d21fc70fd9256f
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2026-08-31-17-10-00.gh-issue-156713.Nt2URL.rst
@@ -0,0 +1,4 @@
+Fix :func:`!nturl2path.pathname2url` and :func:`!nturl2path.url2pathname`:
+the filesystem encoding and error handler are now used for percent-encoding
+and decoding, as in :mod:`urllib.request`.  Previously paths containing
+surrogate characters raised :exc:`UnicodeEncodeError`.