a73x/claudealong
Watch
1
0
Fork
You've already forked claudealong
 0
Code Issues 1 Pull requests Projects Releases Packages Wiki Activity Actions

Multi-line block comments don't fire markers #1

New issue
Open
opened 2026-04-29 15:59:59 +00:00 by a73x · 0 comments
a73x commented 2026-04-29 15:59:59 +00:00 (Migrated from git.a73x.sh)
Copy link

[claude 2026-04-29] v1 limitation surfaced during manual smoke and discussion.

What's broken

Markers in multi-line block comments are silently ignored. Examples:

/*
 * @claude refactor this function
 * to use the visitor pattern
 */

<!--
  @claude add aria-labels to
  every interactive element below
-->

The scanner only matches single-line comments and single-line block comments (open and close on the same line). See `internal/scanner/scanner.go`: the loop checks each line independently. Multi-line block-comment spans never get a chance to be evaluated as marker text.

Why it bites users

A user trying a multi-line ask sees no response and has no diagnostic — looks identical to "the watcher didn't notice my save."

Two ways to fix

A. State machine (recommended for v1.1)

Track `/* ... */` and `` open/close across lines per-language. When a span closes, evaluate its accumulated body for an `@claude` marker. Stamping inserts the UUID after the first `@claude` token within the span (single-line or multi-line).

  • ~50 lines added to `internal/scanner`, no new dependencies
  • Handles realistic usage (block comments not nested inside string literals)
  • Edge case: `@claude` inside a string that happens to look like a comment open/close → wrong fire. Cost is one false positive that Claude reads and ignores.

B. Tree-sitter (defer)

Per-language grammar gives proper "is this a comment?" AST nodes including multi-line spans.

  • ~30 MB binary growth from grammars
  • CGO (or wasm) — breaks pure-Go cross-compile, adds runtime cost
  • Per-scan parse cost is real on hot saves
  • Only earns its weight if we also want AST-aware behavior (stamp at end of function body, scope reads to the symbol the marker refers to, etc.) — not needed for "find @claude inside a comment"

Recommendation

Ship A as v1.1. Revisit B in a v2 that does something genuinely AST-aware.

Files involved

  • `internal/scanner/scanner.go` — `Scan()` line loop, `blockCommentDelims` table
  • `internal/scanner/scanner_test.go` — needs new tests for multi-line spans across the existing supported block-comment languages (CSS, HTML, XML, MD)
  • `internal/rewriter/rewriter.go` — `stampLine` already operates on a single line; for multi-line, the stamping target is the line where `@claude` appears within the span, so existing logic should work without changes. Worth verifying with a test.

Migrated from git-collab issue e65f975003a5cc8bad8c80651df46f7dd8875362.

Original author: a73x dev@a73x.sh.

[claude 2026-04-29] v1 limitation surfaced during manual smoke and discussion. ## What's broken Markers in multi-line block comments are silently ignored. Examples: ```js /* * @claude refactor this function * to use the visitor pattern */ ``` ```html <!-- @claude add aria-labels to every interactive element below --> ``` The scanner only matches single-line comments and single-line block comments (open and close on the same line). See \`internal/scanner/scanner.go\`: the loop checks each line independently. Multi-line block-comment spans never get a chance to be evaluated as marker text. ## Why it bites users A user trying a multi-line ask sees no response and has no diagnostic — looks identical to "the watcher didn't notice my save." ## Two ways to fix ### A. State machine (recommended for v1.1) Track \`/* ... */\` and \`<!-- ... -->\` open/close across lines per-language. When a span closes, evaluate its accumulated body for an \`@claude\` marker. Stamping inserts the UUID after the first \`@claude\` token within the span (single-line or multi-line). - ~50 lines added to \`internal/scanner\`, no new dependencies - Handles realistic usage (block comments not nested inside string literals) - Edge case: \`@claude\` inside a string that happens to look like a comment open/close → wrong fire. Cost is one false positive that Claude reads and ignores. ### B. Tree-sitter (defer) Per-language grammar gives proper "is this a comment?" AST nodes including multi-line spans. - ~30 MB binary growth from grammars - CGO (or wasm) — breaks pure-Go cross-compile, adds runtime cost - Per-scan parse cost is real on hot saves - Only earns its weight if we *also* want AST-aware behavior (stamp at end of function body, scope reads to the symbol the marker refers to, etc.) — not needed for "find @claude inside a comment" ## Recommendation Ship A as v1.1. Revisit B in a v2 that does something genuinely AST-aware. ## Files involved - \`internal/scanner/scanner.go\` — \`Scan()\` line loop, \`blockCommentDelims\` table - \`internal/scanner/scanner_test.go\` — needs new tests for multi-line spans across the existing supported block-comment languages (CSS, HTML, XML, MD) - \`internal/rewriter/rewriter.go\` — \`stampLine\` already operates on a single line; for multi-line, the stamping target is the line where \`@claude\` appears within the span, so existing logic should work without changes. Worth verifying with a test. --- Migrated from git-collab issue `e65f975003a5cc8bad8c80651df46f7dd8875362`. Original author: a73x <dev@a73x.sh>.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
No results found.
Labels
Clear labels   
backlog
No labels
backlog
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
a73x/claudealong#1
No description provided.