fix(lint): exclude [ from frag and alias classes too (linear scan)

Follow-up to the target-class fix (e03e322): the frag class #[^\]|]* and
alias class [^\]]+ still allowed [, so the same O(n^2) backtracking
survived on long ]-free runs seeded with [[a#... or [[a|... (a bare
[[a#*10000 hung ~15s per pass, and lint scans each file twice).

Extend the same [-exclusion to both inner classes:
  frag   #[^\]|]*  ->  #[^\[\]|]*
  alias  [^\]]+    ->  [^\[\]]+

Every pathological shape is now linear (20k -> <20ms), and matching on
valid Obsidian syntax is byte-for-byte unchanged (headings, block refs,
aliases, note/attachment embeds, same-page fragments). Adds regression
tests for the frag/alias runs alongside the existing [-run one.

Addresses review follow-up on #180.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: Aldominguez12

openkb/lint.py

@@ -24,10 +24,13 @@
 # same-page fragment links [[#Heading]], and embeds ![[file]]. Named groups
 # keep the pieces separable so Obsidian fragment/embed syntax survives
 # validation and rewriting instead of being demoted to plain text.
-# `[` is excluded from the target class: Obsidian forbids it in note names,
-# and allowing it makes scans quadratic on long runs of unmatched `[`.
+# `[` is excluded from every inner class (target, frag, and alias): Obsidian
+# forbids it in note names, and allowing it in any class lets that class run
+# past a following `[[` that starts a new unclosed link, so a long `]`-free
+# region seeded with `[[a#`… or `[[a|`… backtracks quadratically — not just a
+# bare run of `[`. Bounding all three keeps the scan linear.
 _WIKILINK_RE = re.compile(
-    r"(?P<embed>!)?\[\[(?P<target>[^\[\]|#]*)(?P<frag>#[^\]|]*)?(?:\|(?P<alias>[^\]]+))?\]\]"
+    r"(?P<embed>!)?\[\[(?P<target>[^\[\]|#]*)(?P<frag>#[^\[\]|]*)?(?:\|(?P<alias>[^\[\]]+))?\]\]"
 )
 
 # Extension-like suffix on a wikilink target: 1-8 alphanumeric chars with at

tests/test_lint.py

@@ -701,6 +701,30 @@ def test_unmatched_bracket_run_scans_in_linear_time(self):
         assert out == payload
         assert ghosts == []
 
+    def test_unmatched_fragment_run_scans_in_linear_time(self):
+        # Regression: excluding "[" from the *target* class alone was not
+        # enough — the frag class `#[^\]|]*` still swallowed following `[[`s,
+        # so a long `]`-free run of `[[a#`… stayed quadratic (~15s at 10k).
+        # Excluding "[" from the frag class too keeps it linear.
+        payload = "[[a#" * 20_000
+        start = time.perf_counter()
+        out, ghosts = strip_ghost_wikilinks(payload, set())
+        elapsed = time.perf_counter() - start
+        assert elapsed < 1.0
+        assert out == payload
+        assert ghosts == []
+
+    def test_unmatched_alias_run_scans_in_linear_time(self):
+        # Same quadratic class via the alias group `[^\]]+` on a `[[a|`… run;
+        # excluding "[" from the alias class bounds it.
+        payload = "[[a|" * 20_000
+        start = time.perf_counter()
+        out, ghosts = strip_ghost_wikilinks(payload, set())
+        elapsed = time.perf_counter() - start
+        assert elapsed < 1.0
+        assert out == payload
+        assert ghosts == []
+
     # --- find_broken_links ------------------------------------------------
 
     def test_heading_link_to_existing_page_not_broken(self, tmp_path):