fix: pass over a profile revision that won't parse

One unparsable commit in a profile's history stopped the walk and
marked every older prose citation unknown, which refused the recale and
bump for good (lindbergh-loader, df7346b3). The revision is now passed
over unless the pins differ on its two sides while the field's
citations changed across it.

scripts/profile_sync.py

@@ -3178,13 +3178,34 @@ def realign_prose(
     # Fields whose writing pin lies past a revision that cannot be read:
     # where their text came in is unknown, so they are not moved.
     unknown: dict[str, str] = {}
+    # A revision that does not parse is passed over unless the pins differ
+    # on its two sides while the field's citations changed across it: then
+    # the text may have come in there under either pin. Otherwise nothing
+    # it could hold changes the reading. lindbergh-loader
+    # committed invalid YAML once (df7346b3), fixed the same day, and every
+    # note older than that commit stayed blocked for good.
+    gap: str | None = None
     for sha in history:
         if not any(alive.values()):
             break
         past = _git_file_at(path, sha)
         if past is None:
-            unknown = {field: sha for field, live in alive.items() if live}
-            break
+            gap = gap or sha
+            continue
+        if gap:
+            newer = revisions[-1][1] if revisions else document
+            moved = any(
+                newer.get(pin) != past.get(pin)
+                for pin in ("source_commit", "upstream_commit")
+            )
+            values = _scalar_values(past)
+            for field, live in alive.items():
+                if live and moved and _scalar_citations(values.get(field)) != (
+                    _scalar_citations(current_values.get(field))
+                ):
+                    unknown[field] = gap
+                    alive[field] = False
+            gap = None
         revisions.append((sha, past))
         values = _scalar_values(past)
         for field in fields:
@@ -3196,6 +3217,10 @@ def realign_prose(
                 intro[field] = sha
             else:
                 alive[field] = False
+    if gap:
+        # The oldest revisions do not parse: where the text came in is past
+        # what can be read.
+        unknown.update({field: gap for field, live in alive.items() if live})
 
     messages: list[str] = []
     jobs: list[tuple[Citation, dict]] = []

tests/test_profile_sync.py

@@ -3845,6 +3845,46 @@ class TestRealignProse(unittest.TestCase):
         self.assertTrue(all(m.startswith("read again: notes:") for m in messages))
         self.assertIn("(a.c:10)", self.path.read_text())
 
+    def _repo_with_a_broken_revision(self, broken_pin: str):
+        self._git("init", "-q")
+        self._git("config", "user.email", "t@t")
+        self._git("config", "user.name", "t")
+        written = PROSE_SAMPLE.replace('"pin"', '"oldpin"')
+        self._commit(written, "profile")
+        broken = PROSE_SAMPLE.replace('"pin"', f'"{broken_pin}"')
+        self._commit(broken + "files: [\n", "invalid yaml")
+        self._commit(written, "repair")
+        self.path.write_text(PROSE_SAMPLE, encoding="utf-8")
+        self.files[("oldpin", "a.c")] = ["x"] * 9 + ["subject"]
+        self.files[("pin", "a.c")] = ["x"] * 13 + ["subject"]
+        self.files[("oldpin", "b.c")] = list("pqrstuvwx")
+        self.files[("pin", "b.c")] = list("pqrstuvwx")
+
+    def test_a_revision_that_does_not_parse_is_passed_over(self):
+        """lindbergh-loader committed invalid YAML once, repaired the same
+        day under the same pin. Every older note was reported unreadable
+        and the recale stayed refused for good."""
+        self._repo_with_a_broken_revision("oldpin")
+        messages = profile_sync.realign_prose(self.path, self.tmp.name, dry_run=True)
+        self.assertFalse(any("unreadable" in m for m in messages), messages)
+        self.assertTrue(any("would recale" in m for m in messages), messages)
+
+    def test_a_revision_that_does_not_parse_still_blocks_when_it_can_matter(self):
+        """The pins differ across the gap and the note's citations changed:
+        the text may have come in there under either pin."""
+        self._git("init", "-q")
+        self._git("config", "user.email", "t@t")
+        self._git("config", "user.name", "t")
+        self._commit(
+            PROSE_SAMPLE.replace('"pin"', '"older"').replace("a.c:10", "a.c:3"),
+            "profile",
+        )
+        self._commit(PROSE_SAMPLE + "files: [\n", "invalid yaml")
+        self._commit(PROSE_SAMPLE.replace('"pin"', '"oldpin"'), "repair")
+        self.path.write_text(PROSE_SAMPLE, encoding="utf-8")
+        messages = profile_sync.realign_prose(self.path, self.tmp.name, dry_run=True)
+        self.assertTrue(any("unreadable" in m for m in messages), messages)
+
     def test_scalar_written_at_the_current_pin_is_left_alone(self):
         self._git("init", "-q")
         self._git("config", "user.email", "t@t")