Files
board-viewer/.tasks/board-viewer-polish.md
vitya 469ff112cc feat(polish): close all 6 review nitpicks in one sweep
closes board-viewer-polish

- M1 src/parser.ts: comment explaining why **Status:** text field is intentionally
  ignored (emoji is canonical, prevents future-reader head-scratch)
- M2 src/render.ts + tests: clamp future ISO to '0m' (server clock drift safety)
  + inline comment + test
- M3 .tasks/board-viewer-gitea-reader.md: decisions-log clarifying orphan
  per-task files are out of scope by design (STATUS.md = index of record)
- M4 static/board.{js,css} + src/render.ts: filter rebuilt to query records
  (not DOM badge text); 5 status-filter pill buttons in header with active
  toggle state; smoke-verified blocked filter
- M5 deploy/Dockerfile.build: HEALTHCHECK (file fresher than 2×tick) and
  stderr redirect for failures with timestamp
- M6 src/reader.ts: parallelize getLatestCommitIso per-repo via Promise.all
  (preserves block order)

49 tests pass (was 43 pre-polish), typecheck clean, smoke against real Gitea
returns 81 records from 4 repos.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-22 15:46:47 +03:00

4.7 KiB
Raw Permalink Blame History

board-viewer-polish

Goal

Bundle of minor / nitpick findings from board-viewer-review checkpoint. Each — small, individually below the task-creation threshold; collectively worth doing in one sweep.

Items

M1. Document parser's **Status:** field neglect

  • Location: src/parser.ts (in parseOneBlock)
  • Problem: **Status:** text field is matched by FIELD_RE but never stored in StatusBlock (emoji is SoT). Future maintainer reads parser and wonders why.
  • Fix: comment // **Status:** text field intentionally ignored — emoji is canonical near where it's matched, OR delete fields['Status'] after parse to avoid the orphan field in local var. Choose the comment.

M2. formatAge future-timestamp behavior is undocumented

  • Location: src/render.ts:31 + static/board.js:49
  • Problem: Math.max(0, …) silently clamps future ISO to "0m". With server-clock drift на VDS vs Gitea это маскирует ошибку синхронизации.
  • Fix: add one test asserting formatAge(futureIso, now) === '0m' to lock behavior, and add inline comment "// clamps to '0m' on future iso — see test".

M3. gitea-reader acceptance "per-task без записи в STATUS.md (legacy)" is impossible by design

  • Location: .tasks/board-viewer-gitea-reader.md Decisions log
  • Problem: Acceptance line 42 promises tests for "per-task без записи в STATUS.md (legacy)" case, but the reader iterates STATUS.md blocks only — orphan per-task files are invisible by design (matches spec: STATUS.md is the index).
  • Fix: append a decisions-log entry to board-viewer-gitea-reader.md clarifying this is intentional; STATUS.md is the entry point, orphan per-task files are out of scope.

M4. Client filter doesn't match status in by-status mode; "multi-select по статусам" not implemented

  • Location: static/board.js:139-147 (applyFilter)
  • Problem: Filter reads .badge.status / .badge.project text from DOM. In by-status mode there's no status badge → typing "blocked" matches nothing. Также acceptance в board-viewer-html-render.md:27 обещает multi-select по статусам — не реализован.
  • Fix: rebuild applyFilter to query the underlying records array (status always available regardless of view mode). Decide on multi-select: either ship a status-pill filter row, or amend acceptance criteria. Recommendation: add multi-select pills below the filter input — simple <button data-status="…" class="status-filter"> toggle.

M5. Build container failures show only echo output, no Docker healthcheck

  • Location: deploy/Dockerfile.build:14
  • Problem: ... || echo 'render failed; will retry next tick' swallows non-zero exit. Operator must docker compose logs build to notice. No Docker healthcheck signals unhealthy.
  • Fix: emit failure to stderr (>&2 echo), and add a HEALTHCHECK to Dockerfile that checks [ -f /output/index.html ] AND [ $(date +%s) - $(stat -c %Y /output/index.html) -lt 600 ] (file exists AND fresher than 10 min). Compose web service can depends_on health.

M6. getLatestCommitIso called serially, N=80+ per tick

  • Location: src/reader.ts:36-37
  • Problem: for...of await serializes 80+ HTTP calls. Within budget for now but a Gitea hiccup serializes through all of them.
  • Fix: parallelize per-repo or globally with Promise.all(blocks.map(...)). Cap concurrency at ~10 if politeness is a concern.

Decisions log

  • 2026-05-22: bundle создан промоушеном из review-чекпойнта; severity варьируется minor → nitpick; bundled, чтобы не плодить 6 микро-тасок.

Completed steps

  • M1: comment в parser.ts объясняющий что **Status:** text field intentionally ignored (emoji canonical)
  • M2: тест clamps future timestamps to "0m" + inline comment в formatAge
  • M3: decisions-log entry в gitea-reader.md: orphan per-task без блока в STATUS.md — out of scope by design
  • M4: filter rebuilt to query underlying records (status always known); 5 status-filter pill buttons добавлены в header; click toggles active set; CSS .status-filter.active filled. Smoke-verified: click "blocked" → только 1 blocked задача видна.
  • M5: HEALTHCHECK в Dockerfile.build (index.html exists AND fresher than 2×tick); failures redirected to stderr с timestamp
  • M6: getLatestCommitIso parallelized via Promise.all per-repo (preserves block order)

Notes

Каждый item — TDD по тому, что есть testable. M1/M3 — comment/docs only ([skip-tdd: visual]). M2 — pure test. M4/M5/M6 — TDD logic.