fix(todo): only count real comment markers around TODO keywords (#1128)

The comment filter accepted `#`, `--`, `;` and `%` anywhere before the keyword, so `foo(); TODO`, `if (a % 2) TODO`, `i--; TODO`, URLs and strings produced items. Inline openers (// /* <!-- # -- {- (*) now count only at the line start or after whitespace; `*` `;` `%` and docstring quotes only at the line start. A TODO anywhere inside a comment still counts, as in IntelliJ. Lines the backend truncated keep the lenient rule.

Co-authored-by: qtvbwfn <[email protected]>

src/features/todo/model.ts

@@ -71,10 +71,20 @@ export function buildTodoQuery(s: TodoSettings, scope: SearchScope = { kind: "pr
   };
 }
 
-const COMMENT_MARKER = /(\/\/|\/\*|^\s*\*|#|--|<!--|;|%|\{-|\(\*|"""|''')/;
+// Comment openers that count anywhere on the line, at its start or after whitespace: // /* <!-- # -- {- (*.
+// The whitespace rule drops `//` in URLs, `#` in strings, `f(*)` and `a * b`. Left over without knowing
+// the language: `"x # y"` in a string and a prefix `--y`.
+// `*` `;` `%` `"""` `'''` are code characters too, so they count only at the line start (LINE_START_MARKER).
+// IDEA's default pattern matches a TODO anywhere inside a comment, so the keyword need not lead it.
+const INLINE_MARKER = /(?:^|\s)(?:\/\/|\/\*|<!--|#|--|\{-|\(\*)/;
+const LINE_START_MARKER = /^\s*(?:\*|;|%|"""|''')/;
+// A long line the backend truncated (ide-search line.rs, `textOffset > 0`) doesn't start at the real line
+// start, so neither rule can apply: any marker counts, a false positive rather than a missed TODO.
+const LENIENT_MARKER = /(\/\/|\/\*|<!--|#|--|\{-|\(\*|\*|;|%|"""|''')/;
 
-export function looksLikeComment(prefix: string): boolean {
-  return COMMENT_MARKER.test(prefix);
+export function looksLikeComment(prefix: string, truncated = false): boolean {
+  if (truncated) return LENIENT_MARKER.test(prefix);
+  return INLINE_MARKER.test(prefix) || LINE_START_MARKER.test(prefix);
 }
 
 function cleanText(text: string): string {
@@ -96,7 +106,7 @@ export function itemsFromMatches(fm: FileMatches, s: TodoSettings): TodoItem[] {
       const matched = lm.text.slice(start, end);
       const p = patternFor(s, matched);
       if (!p) continue;
-      if (s.commentsOnly && !looksLikeComment(lm.text.slice(0, start))) continue;
+      if (s.commentsOnly && !looksLikeComment(lm.text.slice(0, start), lm.textOffset > 0)) continue;
       out.push({
         path: fm.path,
         line: lm.line,
@@ -125,7 +135,7 @@ export function scanText(path: string, text: string, s: TodoSettings): TodoItem[
     }
     const p = patternFor(s, m[0]);
     if (!p) continue;
-    if (s.commentsOnly && !looksLikeComment(text.slice(lineStart, idx))) continue;
+    if (s.commentsOnly && !looksLikeComment(text.slice(lineStart, idx), false)) continue;
     out.push({ path, line, col: idx - lineStart, keyword: p.keyword.trim(), text: cleanText(m[0]) });
   }
   return out;

src/features/todo/todo.test.ts

@@ -12,6 +12,7 @@ import {
   DEFAULT_TODO_SETTINGS,
   groupTodos,
   itemsFromMatches,
+  looksLikeComment,
   mergeLive,
   scanText,
   todoRegexSource,
@@ -22,6 +23,52 @@ import { todoDecorations } from "./highlight";
 
 const S = DEFAULT_TODO_SETTINGS;
 
+describe("comment markers", () => {
+  const keep = [
+    "// TODO",
+    "/* TODO */",
+    " * TODO", // Javadoc continuation
+    "/** TODO: document this */", // doc comment led by the keyword
+    "# TODO", // line-start Python/Shell
+    "foo() # TODO", // trailing Python/Shell comment
+    "SELECT 1 -- TODO", // trailing SQL comment
+    "<!-- TODO -->",
+    "  ; TODO", // line-start ini comment
+    "  % TODO", // line-start LaTeX comment
+    '""" TODO', // Python docstring
+    "''' TODO",
+    "x = 1; // TODO", // marker after whitespace, not just line start
+    "x (* TODO *)", // OCaml trailing comment
+    "x {- TODO -}", // Haskell trailing comment
+    // IDEA's default pattern matches anywhere inside a comment, so mentions count too:
+    "/** 业务类别：WORK_NOTICE 工作通知，TODO 待办。 */", // documents an enum value
+    "// fixed the TODO bug",
+    "/* see the TODO list */",
+  ];
+  const drop = [
+    "foo(); TODO", // statement separator
+    "if (a % 2) TODO", // modulo
+    "i--; TODO", // postfix decrement
+    'let s = "a#b"; TODO', // # inside a string, no preceding space
+    "x = a--b; TODO", // -- with no preceding space
+    "see http://example.com TODO", // // inside a URL
+    '"//"; TODO', // // inside a string
+    "f(*) TODO", // OCaml-style wildcard
+    "a * b; TODO", // infix multiplication
+  ];
+
+  it.each(keep)("keeps %s", (line) => expect(looksLikeComment(`${line.slice(0, line.indexOf("TODO"))}`)).toBe(true));
+  it.each(drop)("drops %s", (line) => expect(looksLikeComment(`${line.slice(0, line.indexOf("TODO"))}`)).toBe(false));
+
+  it("falls back to the lenient rule on truncated lines", () => {
+    // A 600-char shell line truncated by the backend: the window starts
+    // mid-line, so the leading rule cannot apply — any marker counts.
+    expect(looksLikeComment("very long code # remainder ", true)).toBe(true);
+    expect(looksLikeComment("very long code % remainder ", true)).toBe(true);
+    expect(looksLikeComment("i-- ", true)).toBe(true);
+  });
+});
+
 describe("query", () => {
   it("builds the keyword regex from enabled patterns", () => {
     expect(todoRegexSource(S)).toBe("\\b(TODO|FIXME|HACK|XXX)\\b.*");