From dcdd5d45c3d01c4d52fcf3e226d5078fc0ac2908 Mon Sep 17 00:00:00 2001
From: 1ndahous3 <9205230+1ndahous3@users.noreply.github.com>
Date: Sat, 10 Oct 2026 13:35:48 +0300
Subject: [PATCH] Avoid copying strings when replace finds no match (#9029)

Assisted-by: Codex:gpt-6
---
 crates/vm/src/builtins/str.rs       | 32 +++++++++++++++++++----------
 extra_tests/snippets/builtin_str.py | 17 +++++++++++++++
 2 files changed, 38 insertions(+), 11 deletions(-)

diff --git a/crates/vm/src/builtins/str.rs b/crates/vm/src/builtins/str.rs
index 7d339ad1b..5e3c10f77 100644
--- a/crates/vm/src/builtins/str.rs
+++ b/crates/vm/src/builtins/str.rs
@@ -1245,19 +1245,29 @@ impl Py<PyStr> {
         }
 
         let s = zelf.as_wtf8();
-        let replaced = if count < 0 {
-            s.replace(old.as_wtf8(), new.as_wtf8())
+        let limit = if count < 0 {
+            usize::MAX
         } else {
-            let s_is_empty = s.is_empty();
-            let old_is_empty = old.is_empty();
-
-            if s_is_empty && !old_is_empty {
-                s.to_owned()
-            } else if s_is_empty && old_is_empty {
-                new.as_wtf8().to_owned()
-            } else {
-                s.replacen(old.as_wtf8(), new.as_wtf8(), count as usize)
+            count as usize
+        };
+        let replaced = if old.is_empty() {
+            s.replacen(old.as_wtf8(), new.as_wtf8(), limit)
+        } else {
+            let mut matches = s.find_iter(old.as_wtf8()).take(limit);
+            let Some(first) = matches.next() else {
+                drop(matches);
+                return PyStr::result_unchanged(zelf, vm);
+            };
+            let mut result = Vec::with_capacity(s.len());
+            let mut last = 0;
+            for start in core::iter::once(first).chain(matches) {
+                result.extend_from_slice(&s.as_bytes()[last..start]);
+                result.extend_from_slice(new.as_bytes());
+                last = start + old.byte_len();
             }
+            result.extend_from_slice(&s.as_bytes()[last..]);
+            // SAFETY: A nonempty WTF-8 needle matches only at code point boundaries.
+            unsafe { Wtf8Buf::from_bytes_unchecked(result) }
         };
         vm.ctx.new_str(replaced)
     }
diff --git a/extra_tests/snippets/builtin_str.py b/extra_tests/snippets/builtin_str.py
index a8da6ab96..9ab9538c5 100644
--- a/extra_tests/snippets/builtin_str.py
+++ b/extra_tests/snippets/builtin_str.py
@@ -1098,6 +1098,23 @@ def test_replace_unchanged():
 test_replace_unchanged()
 
 
+def test_replace_missing_pattern():
+    for text in ("ordinary text", "\u00e1\u03b2\U0001f600", "a\ud800b\udfff"):
+        for count in (-1, 1, 10):
+            assert text.replace("\r", "replacement", count) is text
+            result = MyString(text).replace("\r", "replacement", count)
+            assert type(result) is str
+            assert result == text
+
+    # Continue the same search after the first match, including lone surrogates.
+    assert "ababa".replace("aba", "x") == "xba"
+    assert "\ud800x\ud800".replace("\ud800", "\udfff", 1) == "\udfffx\ud800"
+    assert "\ud800x\ud800".replace("\ud800", "\udfff") == "\udfffx\udfff"
+
+
+test_replace_missing_pattern()
+
+
 def test_expandtabs_zero_tabsize():
     # With no width to advance to, the tabs come out and nothing else moves.
     # A tab that followed a character used to ask for a run of usize::MAX