From 818ea5242eb03b2ef61af8d7b1b3071e44f16125 Mon Sep 17 00:00:00 2001
From: Victor Stinner <vstinner@python.org>
Date: Thu, 20 Jun 2024 14:44:00 +0200
Subject: [PATCH 1/3] gh-119182: Optimize PyUnicode_FromFormat()

Use strchr() and ucs1lib_find_max_char() to optimize the code path
formatting sub-strings between '%' formats.
---
 Objects/unicodeobject.c | 38 +++++++++++++++++++-------------------
 1 file changed, 19 insertions(+), 19 deletions(-)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 1f8c89dd12a528..6f1dadfebe4219 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -2891,31 +2891,31 @@ unicode_from_format(_PyUnicodeWriter *writer, const char *format, va_list vargs)
                 goto fail;
         }
         else {
-            const char *p;
+            const char *p = strchr(f, '%');
             Py_ssize_t len;
-
-            p = f;
-            do
-            {
-                if ((unsigned char)*p > 127) {
-                    PyErr_Format(PyExc_ValueError,
-                        "PyUnicode_FromFormatV() expects an ASCII-encoded format "
-                        "string, got a non-ASCII byte: 0x%02x",
-                        (unsigned char)*p);
-                    goto fail;
-                }
-                p++;
+            if (p != NULL) {
+                len = p - f;
             }
-            while (*p != '\0' && *p != '%');
-            len = p - f;
-
-            if (*p == '\0')
+            else {
+                len = strlen(f);
                 writer->overallocate = 0;
+            }
 
-            if (_PyUnicodeWriter_WriteASCIIString(writer, f, len) < 0)
+            int is_ascii = (ucs1lib_find_max_char((Py_UCS1*)f, (Py_UCS1*)f + len) < 128);
+            if (!is_ascii) {
+                Py_ssize_t i;
+                for (i=0; i < len && (unsigned char)f[i] <= 127; i++);
+                PyErr_Format(PyExc_ValueError,
+                    "PyUnicode_FromFormatV() expects an ASCII-encoded format "
+                    "string, got a non-ASCII byte: 0x%02x",
+                    (unsigned char)f[i]);
                 goto fail;
+            }
 
-            f = p;
+            if (_PyUnicodeWriter_WriteASCIIString(writer, f, len) < 0) {
+                goto fail;
+            }
+            f += len;
         }
     }
     va_end(vargs2);

From 57d71526a1c9c0ce5d6c5139b042be09cbe69156 Mon Sep 17 00:00:00 2001
From: Victor Stinner <vstinner@python.org>
Date: Thu, 20 Jun 2024 16:17:27 +0200
Subject: [PATCH 2/3] Run ucs1lib_find_max_char() only once

---
 Objects/unicodeobject.c | 32 +++++++++++++++-----------------
 1 file changed, 15 insertions(+), 17 deletions(-)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 6f1dadfebe4219..37362db8142161 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -2875,16 +2875,26 @@ unicode_fromformat_arg(_PyUnicodeWriter *writer,
 static int
 unicode_from_format(_PyUnicodeWriter *writer, const char *format, va_list vargs)
 {
-    writer->min_length += strlen(format) + 100;
+    Py_ssize_t len = strlen(format);
+    writer->min_length += len + 100;
     writer->overallocate = 1;
 
-    va_list vargs2;
-    const char *f;
-
     // Copy varags to be able to pass a reference to a subfunction.
+    va_list vargs2;
     va_copy(vargs2, vargs);
 
-    for (f = format; *f; ) {
+    int is_ascii = (ucs1lib_find_max_char((Py_UCS1*)format, (Py_UCS1*)format + len) < 128);
+    if (!is_ascii) {
+        Py_ssize_t i;
+        for (i=0; i < len && (unsigned char)format[i] <= 127; i++);
+        PyErr_Format(PyExc_ValueError,
+            "PyUnicode_FromFormatV() expects an ASCII-encoded format "
+            "string, got a non-ASCII byte: 0x%02x",
+            (unsigned char)format[i]);
+        goto fail;
+    }
+
+    for (const char *f = format; *f; ) {
         if (*f == '%') {
             f = unicode_fromformat_arg(writer, f, &vargs2);
             if (f == NULL)
@@ -2892,7 +2902,6 @@ unicode_from_format(_PyUnicodeWriter *writer, const char *format, va_list vargs)
         }
         else {
             const char *p = strchr(f, '%');
-            Py_ssize_t len;
             if (p != NULL) {
                 len = p - f;
             }
@@ -2901,17 +2910,6 @@ unicode_from_format(_PyUnicodeWriter *writer, const char *format, va_list vargs)
                 writer->overallocate = 0;
             }
 
-            int is_ascii = (ucs1lib_find_max_char((Py_UCS1*)f, (Py_UCS1*)f + len) < 128);
-            if (!is_ascii) {
-                Py_ssize_t i;
-                for (i=0; i < len && (unsigned char)f[i] <= 127; i++);
-                PyErr_Format(PyExc_ValueError,
-                    "PyUnicode_FromFormatV() expects an ASCII-encoded format "
-                    "string, got a non-ASCII byte: 0x%02x",
-                    (unsigned char)f[i]);
-                goto fail;
-            }
-
             if (_PyUnicodeWriter_WriteASCIIString(writer, f, len) < 0) {
                 goto fail;
             }

From 614e132c188717ac563a3fabc2d9fd40ebd35292 Mon Sep 17 00:00:00 2001
From: Victor Stinner <vstinner@python.org>
Date: Thu, 20 Jun 2024 20:40:55 +0200
Subject: [PATCH 3/3] Add comment

---
 Objects/unicodeobject.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
index 37362db8142161..e6feed47fbb2bf 100644
--- a/Objects/unicodeobject.c
+++ b/Objects/unicodeobject.c
@@ -2883,6 +2883,8 @@ unicode_from_format(_PyUnicodeWriter *writer, const char *format, va_list vargs)
     va_list vargs2;
     va_copy(vargs2, vargs);
 
+    // _PyUnicodeWriter_WriteASCIIString() below requires the format string
+    // to be encoded to ASCII.
     int is_ascii = (ucs1lib_find_max_char((Py_UCS1*)format, (Py_UCS1*)format + len) < 128);
     if (!is_ascii) {
         Py_ssize_t i;