Strip CR from CRLF input in msgmerge-female

Working trees produced by Git's autocrlf=true on Windows ship .po/.pot
files with CRLF line endings. gettext msgmerge leaks the embedded \r
into string fields, producing mixed-ending output and spurious fuzzy
or duplicate entries from the polluted comparisons. Pre-canonicalize
PO and POT to LF before invoking gettext.

Author: burner1024

docs/changelog.md

@@ -1,5 +1,9 @@
 ## Changelog
 
+### 1.5.3
+
+- `msgmerge-female` now strips `\r` from CRLF line endings in PO and POT files before invoking gettext `msgmerge`. Working trees produced by Git's `autocrlf=true` on Windows would otherwise produce mixed-ending output and spurious fuzzy/duplicate entries.
+
 ### 1.5.2
 
 - Action: `poify` now always runs `msgmerge`, even when the POT file itself is unchanged. This lets the action clear stale PO/POT occurrence mismatches that can be reintroduced by translation-side PO updates.

msg2po/msgmerge.py

@@ -22,8 +22,26 @@
 from msg2po.po_utils import normalize_po, po_content_snapshot
 
 
+def _strip_cr(path: str) -> None:
+    """Strip CRs from CRLF line endings before invoking gettext.
+
+    On Windows checkouts under core.autocrlf=true, .po/.pot files arrive with
+    CRLF endings. gettext msgmerge then leaks the embedded \\r into string
+    fields (msgids, occurrence filenames, ...), producing mixed-ending output
+    and spurious fuzzy/duplicate entries from the polluted comparisons. The
+    output writer itself emits only LF, so feeding pre-canonicalized input
+    is sufficient to get clean output.
+    """
+    with open(path, "rb") as f:
+        data = f.read()
+    if b"\r\n" in data:
+        with open(path, "wb") as f:
+            f.write(data.replace(b"\r\n", b"\n"))
+
+
 def merge(po_path: str, pot_path: str):
     logger.info(po_path)
+    _strip_cr(po_path)
     exit_code = 0
     cmd = ["msgmerge", "--previous", "--no-wrap", "-U", "-q", "--backup=off", po_path, pot_path]
     res = subprocess.run(
@@ -67,6 +85,7 @@ def main():
 
     # single file
     if (args.PO is not None) and (args.POT is not None):
+        _strip_cr(args.POT)
         res = merge(args.PO, args.POT)
         if res != 0:
             logger.error(f"msgmerge failed for {args.PO}")
@@ -78,6 +97,10 @@ def main():
     po_files = find_files(po_dir, "po")
     pot_file = os.path.join(po_dir, CONFIG.src_lang + ".pot")
 
+    # Canonicalize POT once before the worker pool: each worker would
+    # otherwise race to rewrite the same shared file.
+    _strip_cr(pot_file)
+
     logger.info(f"Merging PO files in {po_dir} with {pot_file}")
     pool = Pool()
     try:

tests/test_metadata_regressions.py

@@ -64,6 +64,39 @@ def fake_run(*_args, **_kwargs):
         assert po[0].occurrences == [("game/one.msg", "100"), ("game/two.msg", "200")]
 
 
+class TestMsgmergeStripsCRLF:
+    def test_merge_canonicalizes_crlf_input_to_lf(self, tmp_path, monkeypatch):
+        """gettext msgmerge on Windows leaks \\r from CRLF input into msgids
+        and occurrence filenames; pre-canonicalize PO/POT before invoking it."""
+        po_path = tmp_path / "it.po"
+        pot_path = tmp_path / "en.pot"
+
+        po = polib.POFile()
+        po.metadata = {"Content-Type": "text/plain; charset=UTF-8"}
+        po.append(polib.POEntry(msgid="Hello", msgstr="ciao", occurrences=[("game/one.msg", "100")]))
+        po.save(po_path)
+        pot = polib.POFile()
+        pot.metadata = {"Content-Type": "text/plain; charset=UTF-8"}
+        pot.append(polib.POEntry(msgid="Hello", msgstr="", occurrences=[("game/one.msg", "100")]))
+        pot.save(pot_path)
+
+        for p in (po_path, pot_path):
+            data = p.read_bytes()
+            p.write_bytes(data.replace(b"\n", b"\r\n"))
+            assert b"\r\n" in p.read_bytes()
+
+        def fake_run(*_args, **_kwargs):
+            return type("Result", (), {"stdout": "", "stderr": "", "returncode": 0})()
+
+        monkeypatch.setattr("msg2po.msgmerge.subprocess.run", fake_run)
+
+        merge(str(po_path), str(pot_path))
+
+        assert b"\r\n" not in po_path.read_bytes()
+        # POT is canonicalized in main() outside merge() to avoid pool races,
+        # so the subprocess-mocked single-file call here only strips the PO.
+
+
 class TestDir2MsgstrOccurrenceOnlyNormalization:
     def test_dir2msgstr_saves_dedup_when_only_occurrences_change(self, tmp_path, monkeypatch):
         po_path = tmp_path / "it.po"