Replace re.sub by the faster str.translate.

anntzer · anntzer · commit cf40c9fcc3f7 · 2022-06-17T12:48:43.000+02:00
diff --git a/lib/matplotlib/backends/backend_pdf.py b/lib/matplotlib/backends/backend_pdf.py
@@ -13,7 +13,6 @@
 import logging
 import math
 import os
-import re
 import string
 import struct
 import sys
@@ -119,25 +118,6 @@ def _fill(strings, linelen=75):
     result.append(b' '.join(strings[lasti:]))
     return b'\n'.join(result)
 
-# PDF strings are supposed to be able to include any eight-bit data,
-# except that unbalanced parens and backslashes must be escaped by a
-# backslash. However, sf bug #2708559 shows that the carriage return
-# character may get read as a newline; these characters correspond to
-# \gamma and \Omega in TeX's math font encoding. Escaping them fixes
-# the bug.
-_string_escape_regex = re.compile(br'([\\()\r\n])')
-
-
-def _string_escape(match):
-    m = match.group(0)
-    if m in br'\()':
-        return b'\\' + m
-    elif m == b'\n':
-        return br'\n'
-    elif m == b'\r':
-        return br'\r'
-    assert False
-
 
 def _create_pdf_info_dict(backend, metadata):
     """
@@ -250,6 +230,15 @@ def _datetime_to_pdf(d):
     return r
 
 
+# PDF strings are supposed to be able to include any eight-bit data, except
+# that unbalanced parens and backslashes must be escaped by a backslash.
+# However, sf bug #2708559 shows that the carriage return character may get
+# read as a newline; these characters correspond to \gamma and \Omega in TeX's
+# math font encoding. Escaping them fixes the bug.
+_str_escapes = str.maketrans({
+    '\\': '\\\\', '(': '\\(', ')': '\\)', '\n': '\\n', '\r': '\\r'})
+
+
 def pdfRepr(obj):
     """Map Python objects to PDF syntax."""
 
@@ -289,8 +278,12 @@ def pdfRepr(obj):
     # escaped. Actually balanced parens are allowed, but it is
     # simpler to escape them all. TODO: cut long strings into lines;
     # I believe there is some maximum line length in PDF.
+    # Despite the extra decode/encode, translate is faster than regex.
     elif isinstance(obj, bytes):
-        return b'(' + _string_escape_regex.sub(_string_escape, obj) + b')'
+        return (
+            b'(' +
+            obj.decode('latin-1').translate(_str_escapes).encode('latin-1')
+            + b')')
 
     # Dictionaries. The keys must be PDF names, so if we find strings
     # there, we make Name objects from them. The values may be
