🐛 fix(tsl-syntax-reference): close verified syntax gaps

skills/tsl-syntax-reference/scripts/lookup.py

@@ -41,6 +41,7 @@ STRUCTURAL_METADATA_RE = re.compile(
     r"<!--\s*(?:section-id|quickstart-rule)\s*:.*?-->",
     re.DOTALL | re.IGNORECASE,
 )
+HTML_COMMENT_RE = re.compile(r"<!--.*?-->", re.DOTALL)
 IDENTITY_PREFIX = "代码块身份："
 BLOCK_DESCRIPTION_PREFIX = "代码块说明："
 ALLOWED_IDENTITIES = {
@@ -156,7 +157,6 @@ CHINESE_QUERY_PARTICLES = ("的", "是", "吗", "呢", "吧")
 ASCII_FILTER_STOP_TOKENS = {
     "debug",
     "please",
-    "program",
     "tinysoft",
     "tsl",
     "tsf",
@@ -480,7 +480,7 @@ def load_sections(references_dir: Path = DEFAULT_REFERENCES_DIR) -> list[Section
                 raise ReferenceStructureError(f"重复 section ID：{base_id}")
             seen_ids.add(base_id)
             tags = _section_tags(local_body)
-            searchable_body = STRUCTURAL_METADATA_RE.sub(" ", local_body)
+            searchable_body = HTML_COMMENT_RE.sub(" ", local_body)
             searchable_text = normalize(
                 "\n".join((page.stem, page_title, *heading_path, *tags, searchable_body))
             )
@@ -1122,9 +1122,9 @@ def _text_contains_exact_query(text: str, query: str) -> bool:
 
 def _intent_score(section: Section, query: str) -> int:
     aliases = PAGE_INTENT_ALIASES.get(section.page.name, ())
-    return PAGE_INTENT_SCORE * sum(
+    return PAGE_INTENT_SCORE if any(
         _query_contains_phrase(query, alias) for alias in aliases
-    )
+    ) else 0
 
 
 def _has_chinese_context(section: Section, query: str) -> bool:
@@ -1161,7 +1161,9 @@ def _tag_matched_tokens(tags: tuple[str, ...], query_token_set: set[str]) -> int
 def _code_text(body: str) -> str:
     inline = INLINE_CODE_RE.findall(body)
     fenced = FENCED_CODE_RE.findall(body)
-    return normalize("\n".join((*inline, *fenced)))
+    # 标识符信号只来自 ASCII 代码术语。中文散文会走标题、tag 和正文得分；
+    # 若把围栏里的“下标数组”等输出标签也当标识符，中文查询会被样例值劫持。
+    return normalize("\n".join(ASCII_TOKEN_RE.findall("\n".join((*inline, *fenced)))))
 
 
 def _score_section(section: Section, query: str, mode: str) -> ScoreBreakdown:
@@ -1169,8 +1171,7 @@ def _score_section(section: Section, query: str, mode: str) -> ScoreBreakdown:
     tokens = query_tokens(query)
     heading_text = normalize("\n".join(section.heading_path))
     page_title_text = normalize(section.page_title)
-    body_without_metadata = STRUCTURAL_METADATA_RE.sub(" ", section.local_body)
-    body_text = normalize(SECTION_TAG_RE.sub(" ", body_without_metadata))
+    body_text = normalize(HTML_COMMENT_RE.sub(" ", section.local_body))
     tag_text = normalize("\n".join(section.tags))
     term_text = _code_text(section.local_body)
     expanded_only_tokens = _synonym_tokens(query) - _base_query_tokens(query)
@@ -1356,6 +1357,7 @@ def query_sections(
         )
     ranked.sort(
         key=lambda match: (
+            match.weak,
             -match.score,
             *(-value for value in match.priority),
             match.section.page.as_posix(),
@@ -1400,10 +1402,9 @@ def _safe_json_string(value: str) -> str:
 
 
 def _plain_text_summary(body: str, limit: int = 180) -> str:
-    # 标签是检索元数据，不是事实正文；不能泄进候选摘要。
-    without_metadata = STRUCTURAL_METADATA_RE.sub(" ", body)
-    without_tags = SECTION_TAG_RE.sub(" ", without_metadata)
-    without_fences = FENCED_CODE_RE.sub(" ", without_tags)
+    # HTML 注释都是维护元数据，不是事实正文；不能泄进候选摘要。
+    without_comments = HTML_COMMENT_RE.sub(" ", body)
+    without_fences = FENCED_CODE_RE.sub(" ", without_comments)
     without_links = re.sub(
         r"!?\[([^\]]*)\]\([^)]+\)", lambda match: match.group(1), without_fences
     )
@@ -1464,7 +1465,7 @@ def render_candidates(result: QueryResult) -> str:
 
 
 def render_section(section: Section) -> str:
-    body = STRUCTURAL_METADATA_RE.sub("", section.body)
+    body = HTML_COMMENT_RE.sub("", section.body)
     body = re.sub(r"\n{3,}", "\n\n", body).rstrip()
     lines = [
         "# TSL Syntax Section",