Improve link matching to avoid many classes of bad edits

parse_cite: extend to skip {{cite}}/{{citation}}, {{short description}},
{{gli}}, {{defn}}, external links [https://...], italic text ''...'',
and bullet-point lines containing bare URLs (unformatted bibliography
entries). Uses brace-counting to handle nested templates correctly.

parse_links: yield [[Category:...]] links as 'category' tokens so they
are never modified.

add_link: handle three new boundary cases where the match spans an
existing [[link]]:
- match ends exactly at the link boundary: replace the whole thing with
  a single clean link (e.g. surface [[runoff (hydrology)|runoff]] →
  [[surface runoff]])
- match starts right after [[: absorb the stray [[ (e.g.
  [[anti-globalization]] movement → [[anti-globalization movement]])
- match starts partway inside a link: skip (would produce broken wikitext)
- match spans into but not through a link: use a piped prefix link
  (e.g. cross-platform [[interchange station]] →
  [[cross-platform interchange|cross-platform]] [[interchange station]])

Fallback search: mask [[Category:...]] spans with spaces so the pattern
cannot match inside them. Guard against matches that are part of a
longer named entity (title-case phrase followed by extra words then an
abbreviation in parentheses, e.g. "Anti-Globalization Movement of
Russia (AGMR)").

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

add_links/match.py

@@ -8,6 +8,12 @@ from .core import get_case_from_content, get_content_and_timestamp, get_revision
 from .util import is_title_case, lc_alpha
 
 re_link_in_text = re.compile(r"\[\[[^]]+?\]\]", re.I | re.S)
+re_category_link = re.compile(r"\[\[Category:[^\]]+\]\]", re.I)
+
+# Matches when extra words separate the match from a following (ABBREVIATION),
+# indicating the matched text is just part of a longer named entity.
+# e.g. "of Russia (AGMR)" matches; " (AGMR)" does not (no intervening words).
+re_named_entity_abbrev = re.compile(r"^(?:\s+[^\s(]+){1,6}\s*\([A-Z]{2,}\)")
 
 
 class LinkReplace(Exception):
@@ -72,14 +78,62 @@ re_cite = re.compile(
     re.I | re.S,
 )
 
+re_cite_template_start = re.compile(r"\{\{(?:cite|citation|short description|gli|defn)\b", re.I)
+re_external_link = re.compile(r"\[https?://[^\]]+\]")
+# Italic text (work titles in bibliographies). Handles apostrophes in content
+# (e.g. ''It's fine'') but requires at least one non-apostrophe character so
+# that ''' bold ''' is not consumed as italic.
+re_italic = re.compile(r"''[^']+(?:'[^']+)*''")
+# Bullet-point lines that contain a bare URL are unformatted bibliography entries.
+re_bullet_with_url = re.compile(r"^\*[^\n]*https?://[^\s\n]+[^\n]*", re.MULTILINE)
+
+
+def find_cite_template_spans(text: str) -> list[tuple[int, int]]:
+    """Find (start, end) spans of {{Cite ...}} templates, handling nested braces."""
+    spans: list[tuple[int, int]] = []
+    for m in re_cite_template_start.finditer(text):
+        start = m.start()
+        if any(s <= start < e for s, e in spans):
+            continue  # already inside a found span
+        depth = 0
+        i = start
+        while i < len(text):
+            if text[i : i + 2] == "{{":
+                depth += 1
+                i += 2
+            elif text[i : i + 2] == "}}":
+                depth -= 1
+                i += 2
+                if depth == 0:
+                    spans.append((start, i))
+                    break
+            else:
+                i += 1
+    return spans
+
 
 def parse_cite(text: str) -> typing.Iterator[tuple[str, str]]:
-    """Parse a citation template."""
+    """Parse citations yielding (type, chunk) tuples, skipping ref tags, cite templates, and external links."""
+    regions = [(m.start(), m.end()) for m in re_cite.finditer(text)]
+    regions.extend(find_cite_template_spans(text))
+    regions.extend((m.start(), m.end()) for m in re_external_link.finditer(text))
+    regions.extend((m.start(), m.end()) for m in re_italic.finditer(text))
+    regions.extend((m.start(), m.end()) for m in re_bullet_with_url.finditer(text))
+    regions.sort()
+
+    # Merge overlapping regions (e.g. a {{Cite}} that sits inside a <ref>)
+    merged: list[tuple[int, int]] = []
+    for start, end in regions:
+        if merged and start < merged[-1][1]:
+            merged[-1] = (merged[-1][0], max(merged[-1][1], end))
+        else:
+            merged.append((start, end))
+
     prev = 0
-    for m in re_cite.finditer(text):
-        yield ("text", text[prev : m.start()])
-        yield ("cite", m.group(0))
-        prev = m.end()
+    for start, end in merged:
+        yield ("text", text[prev:start])
+        yield ("cite", text[start:end])
+        prev = end
     yield ("text", text[prev:])
 
 
@@ -163,6 +217,8 @@ def parse_links(text: str) -> typing.Iterator[tuple[str, str]]:
             for prefix in ("file:", "image:")
         ):
             yield ("image", m.group(0))
+        elif m.group().lower().startswith("[[category:"):
+            yield ("category", m.group(0))
         else:
             yield ("link", m.group(0))
         prev = m.end()
@@ -186,12 +242,66 @@ def mk_link_matcher(q: str) -> typing.Callable[[str], re.Match[str] | None]:
 
 def add_link(m: re.Match[str], replacement: str, text: str) -> str:
     """Add link to text."""
-
     matched_text = m.group(0)
     if matched_text.startswith("[[") and matched_text.endswith("|"):
         return m.re.sub(lambda m: f"[[{replacement}|", text, count=1)
-    else:
-        return m.re.sub(lambda m: f"[[{replacement}]]", text, count=1)
+
+    inner_bracket = matched_text.find("[[")
+    if inner_bracket > 0:
+        prefix = matched_text[:inner_bracket].rstrip()
+        sep = matched_text[len(prefix) : inner_bracket]
+        suffix = matched_text[inner_bracket:]
+        link_dest = replacement.split("|")[0] if "|" in replacement else replacement
+        if text[m.end() : m.end() + 2] == "]]":
+            # The existing [[link]] is fully consumed by the match (its closing ]]
+            # immediately follows). Replace everything with a single clean link.
+            # e.g. "surface [[runoff (hydrology)|runoff]]" → "[[surface runoff]]"
+            return text[: m.start()] + f"[[{link_dest}]]" + text[m.end() + 2 :]
+        # The existing link is only partially consumed; keep it and link just the prefix.
+        # e.g. "cross-platform [[interchange station]]"
+        #   → "[[cross-platform interchange|cross-platform]] [[interchange station]]"
+        return text[: m.start()] + f"[[{link_dest}|{prefix}]]{sep}{suffix}" + text[m.end() :]
+
+    close_bracket = matched_text.find("]]")
+    if close_bracket > 0 and m.start() >= 2 and text[m.start() - 2 : m.start()] == "[[":
+        # Match started inside an existing [[link]] and spans beyond its closing ]].
+        # e.g. "[[anti-globalization]] movement" matched as "anti-globalization]] movement"
+        # → absorb the leading [[ and produce "[[anti-globalization movement]]"
+        return text[: m.start() - 2] + f"[[{replacement}]]" + text[m.end() :]
+
+    return m.re.sub(lambda m: f"[[{replacement}]]", text, count=1)
+
+
+def match_spans_existing_link(m: re.Match[str], text: str) -> bool:
+    """Return True if the match starts partway inside an existing [[link]].
+
+    The trans2 space pattern allows matching across ]] boundaries, which is
+    intentional for cases like [[anti-globalization]] movement (match starts
+    right after [[).  But when the match starts *inside* a link (e.g.
+    [[impervious surface]] runoff matching 'surface runoff'), absorbing the
+    brackets would produce broken wikitext — skip those matches instead.
+    """
+    close_bracket = m.group(0).find("]]")
+    if close_bracket <= 0:
+        return False
+    # close_bracket > 0: match spans ][. If it starts right after [[ that
+    # opening bracket is absorbed by add_link — not our problem here.
+    if m.start() >= 2 and text[m.start() - 2 : m.start()] == "[[":
+        return False
+    return True
+
+
+def is_part_of_named_entity(m: re.Match[str], text: str) -> bool:
+    """Return True if the match is a title-case fragment of a longer named entity.
+
+    Detects the pattern: matched text (title case) + extra words + (ABBREVIATION).
+    e.g. "Anti-Globalization Movement" followed by "of Russia (AGMR)" → True.
+    But "Anti-Globalization Movement" followed directly by "(AGM)" → False,
+    since the match itself is the full entity name.
+    """
+    if not m.group(0)[0].isupper():
+        return False
+    return bool(re_named_entity_abbrev.match(text[m.end() :]))
 
 
 def find_link_in_chunk(
@@ -252,8 +362,9 @@ def find_link_in_chunk(
     if not replacement:
         if bad_link_match:
             raise LinkReplace
-        m = search_for_link(content)
-        if m:
+        masked = re_category_link.sub(lambda c: " " * len(c.group(0)), content)
+        m = search_for_link(masked)
+        if m and not is_part_of_named_entity(m, content) and not match_spans_existing_link(m, content):
             found_text_to_link = m.group(0)
             replacement = match_found(m, q, linkto)
             new_content = add_link(m, replacement, content)
@@ -342,8 +453,9 @@ def find_link_and_section(q: str, content: str, linkto: str | None = None):
             if replacement:
                 text = new_text
             else:
-                m = search_for_link(text)
-                if m:
+                masked = re_category_link.sub(lambda c: " " * len(c.group(0)), text)
+                m = search_for_link(masked)
+                if m and not is_part_of_named_entity(m, text) and not match_spans_existing_link(m, text):
                     replacement = match_found(m, q, linkto)
                     text = add_link(m, replacement, text)
             new_content += text