Repository navigation
Commit 1461c35
committed
fix: skip a UTF-8 BOM when reading config files
git ignores a UTF-8 BOM at the start of a config file, so a file written by a
Windows editor still parses. GitConfigParser raised MissingSectionHeaderError
instead, because the BOM was decoded into the first line, which then no longer
matched a section header.
Evidence on git 2.47.3: `git config -f bom.cfg --list` prints core.bare=true
for a file starting with the three BOM bytes, while GitPython raised
MissingSectionHeaderError. With this change both read the same values, also
when the BOM file is pulled in through include.path.
The new test fails without the change.1 parent 303c48f commit 1461c35
2 files changed
Lines changed: 18 additions & 2 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
@@ -550,9 +550,14 @@ def parse_value(value: str) -> str: | |||
| 550 | 550 | ||
| 551 | 551 | while True: | |
| 552 | 552 | # We assume to read binary! | |
| 553 | - line = fp.readline().decode(defenc) | ||
| 554 | - if not line: | ||
| 553 | + raw_line = fp.readline() | ||
| 554 | + if not raw_line: | ||
| 555 | 555 | break | |
| 556 | + if lineno == 0 and raw_line.startswith(b"\xef\xbb\xbf"): | ||
| 557 | + # A UTF-8 BOM is not part of the content. git skips it, so a | ||
| 558 | + # config file written by a Windows editor still parses. | ||
| 559 | + raw_line = raw_line[3:] | ||
| 560 | + line = raw_line.decode(defenc) | ||
| 556 | 561 | lineno = lineno + 1 | |
| 557 | 562 | # Comment or blank line? | |
| 558 | 563 | if line.strip() == "" or self.re_comment.match(line): | |
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
@@ -354,6 +354,17 @@ def test_comment_backslash_does_not_continue_value(self, rw_dir): | |||
| 354 | 354 | with GitConfigParser(config_path) as config: | |
| 355 | 355 | self.assertEqual(config.get_value("a", "x"), "two") | |
| 356 | 356 | ||
| 357 | + def test_utf8_bom_is_skipped_like_git(self): | ||
| 358 | + """git skips a UTF-8 BOM at the start of a config file, so one written by | ||
| 359 | + a Windows editor still parses. Expectations are what | ||
| 360 | + `git config -f <file> --list` prints on git 2.47.3.""" | ||
| 361 | + content = b"\xef\xbb\xbf[core]\n\tbare = true\n" | ||
| 362 | + config_file = io.BytesIO(content) | ||
| 363 | + config_file.name = "bom.config" | ||
| 364 | + config = GitConfigParser(config_file) | ||
| 365 | + config.read() | ||
| 366 | + self.assertIs(config.get_value("core", "bare"), True) | ||
| 367 | + | ||
| 357 | 368 | def test_config_value_with_trailing_new_line(self): | |
| 358 | 369 | config_content = b'[section-header]\nkey:"value\n"' | |
| 359 | 370 | config_file = io.BytesIO(config_content) | |
0 commit comments