From 76395585bff9686da744e36631616262ba4840d4 Mon Sep 17 00:00:00 2001
From: devdavidejesus <davidejesus.log@proton.me>
Date: Fri, 23 Oct 2026 21:36:27 -0300
Subject: [PATCH] gh-160556: Fix quadratic time in csv.Sniffer with long runs
 of spaces

Sniffer._split_dormant() tried a match at every space of a run of
spaces when the space was a candidate delimiter, and rescanned the
rest of the run each time.  A match starting at a space preceded by a
space is never the leftmost one, so skip those positions with a
lookbehind.  The results do not change.

The regression was introduced in gh-83273 (25fea6d64e1).
---
 Lib/csv.py                                                 | 7 +++++--
 Lib/test/test_csv.py                                       | 7 +++++++
 .../Library/2026-10-24-00-36-27.gh-issue-160556.YTzSZI.rst | 2 ++
 3 files changed, 14 insertions(+), 2 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Library/2026-10-24-00-36-27.gh-issue-160556.YTzSZI.rst

diff --git a/Lib/csv.py b/Lib/csv.py
index b406d82efffd9f3..18dfeb1db230b7f 100644
--- a/Lib/csv.py
+++ b/Lib/csv.py
@@ -437,9 +437,12 @@ def _split_dormant(self, part, quotechar, combos):
         while remaining:
             # Include only the delimiters not found yet, so that the
             # search skips over the found ones; the compiled patterns
-            # come from the re cache.
+            # come from the re cache.  The lookbehind skips matches
+            # starting inside a run of spaces: an equivalent match
+            # starts at the beginning of the run, and rescanning the
+            # run from every space in it would take quadratic time.
             cls = re.escape(''.join(sorted(remaining)))
-            m = re.compile(fr'(?:^|([\r\n{cls}]))'
+            m = re.compile(fr'(?:^|([\r\n{cls}])(?<!  ))'
                            fr' *{quotechar}').search(part, pos)
             if m is None:
                 break
diff --git a/Lib/test/test_csv.py b/Lib/test/test_csv.py
index 36fa7e3572a5ac2..b1d8c0b00f11f11 100644
--- a/Lib/test/test_csv.py
+++ b/Lib/test/test_csv.py
@@ -1631,6 +1631,13 @@ def test_sniff_single_column_quoted(self):
         self.assertRaisesRegex(csv.Error, "Could not determine delimiter",
                                sniffer.sniff, sample)
 
+    def test_sniff_long_run_of_spaces(self):
+        # gh-160556: a long run of spaces not followed by a quote character
+        # used to take quadratic time.
+        sniffer = csv.Sniffer()
+        sample = ('a,b' + ' ' * 100_000 + 'c"\n') * 4
+        self.assertEqual(sniffer.sniff(sample).delimiter, ',')
+
     def test_sniff_non_ascii_delimiter(self):
         # gh-111820: an explicitly requested delimiter can be non-ASCII.
         sniffer = csv.Sniffer()
diff --git a/Misc/NEWS.d/next/Library/2026-10-24-00-36-27.gh-issue-160556.YTzSZI.rst b/Misc/NEWS.d/next/Library/2026-10-24-00-36-27.gh-issue-160556.YTzSZI.rst
new file mode 100644
index 000000000000000..be312452673984f
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2026-10-24-00-36-27.gh-issue-160556.YTzSZI.rst
@@ -0,0 +1,2 @@
+Fix quadratic time in :meth:`csv.Sniffer.sniff` for samples with long runs
+of spaces.  The regression was introduced in 3.16.0a1.