Merge pull request #2249 from geek-alpha/config-utf8-bom

fix: skip a UTF-8 BOM when reading config files

Author: Byron

git/config.py

@@ -550,9 +550,14 @@ def parse_value(value: str) -> str:
 
         while True:
             # We assume to read binary!
-            line = fp.readline().decode(defenc)
-            if not line:
+            raw_line = fp.readline()
+            if not raw_line:
                 break
+            if lineno == 0 and raw_line.startswith(b"\xef\xbb\xbf"):
+                # A UTF-8 BOM is not part of the content. git skips it, so a
+                # config file written by a Windows editor still parses.
+                raw_line = raw_line[3:]
+            line = raw_line.decode(defenc)
             lineno = lineno + 1
             # Comment or blank line?
             if line.strip() == "" or self.re_comment.match(line):

test/test_config.py

@@ -354,6 +354,17 @@ def test_comment_backslash_does_not_continue_value(self, rw_dir):
         with GitConfigParser(config_path) as config:
             self.assertEqual(config.get_value("a", "x"), "two")
 
+    def test_utf8_bom_is_skipped_like_git(self):
+        """git skips a UTF-8 BOM at the start of a config file, so one written by
+        a Windows editor still parses. Expectations are what
+        `git config -f <file> --list` prints on git 2.47.3."""
+        content = b"\xef\xbb\xbf[core]\n\tbare = true\n"
+        config_file = io.BytesIO(content)
+        config_file.name = "bom.config"
+        config = GitConfigParser(config_file)
+        config.read()
+        self.assertIs(config.get_value("core", "bare"), True)
+
     def test_config_value_with_trailing_new_line(self):
         config_content = b'[section-header]\nkey:"value\n"'
         config_file = io.BytesIO(config_content)