chore(quality): 启动代码质量 autoresearch 会话

- .auto/ 会话文件(prompt/measure/checks/ideas)
- pnpm-workspace.yaml 认可 @parcel/watcher 与 @swc/core 构建脚本,
  修复 pnpm 11 下 make code-check 无法运行的问题
This commit is contained in:
ryan
2026-08-16 17:10:28 +08:00
parent 5a8722ff07
commit 305d609d0d
5 changed files with 179 additions and 0 deletions
+15
View File
@@ -0,0 +1,15 @@
#!/bin/bash
# Correctness gate: must pass after every edit. Fails fast on real breakage.
set -euo pipefail
cd "$(dirname "$0")/.."
echo "==> go vet ./..."
go vet ./... 2>&1 | tail -20
echo "==> go build ./..."
go build ./... 2>&1 | tail -20
echo "==> golangci-lint run (repo config)"
golangci-lint run 2>&1 | tail -20
echo "OK: checks passed"
+10
View File
@@ -0,0 +1,10 @@
# Ideas backlog (代码质量)
- 后端 204 个 *_test.go 被 repo golangci 配置 `tests: false` 跳过;可以用
testifylint/usetesting/paralleltest 等对测试代码做一次质量扫描(当前基准不含测试,属于后续深化方向)。
- biome check 发现 76 处(基本全是格式漂移);repo 只跑 format 不跑 check。
可作为独立的一次性格式化提交处理,不进基准(避免纯格式噪声污染 metric)。
- repo `make code-check` 还有 `make format`(goimports + biome)契约:迭代中保持已改文件格式化干净。
- 观察 go vet 已启用但 `unusedresult`/`copylocks` 等子检查默认关闭;若需要可考虑仅作发现用,不进基准。
- errorlint 修复中有 4 处 `err != context.Canceled`,若 runner 实际从不 wrap,
用 errors.Is 是更稳妥的最佳实践,且语义不变 —— 已验证三处 cmd 入口同构。
+42
View File
@@ -0,0 +1,42 @@
#!/bin/bash
# Benchmark: total code-quality issues across backend + frontend (lower is better).
# Fixed linter set — see .auto/prompt.md. Never tune this file to game counts.
set -euo pipefail
cd "$(dirname "$0")/.."
start=$(date +%s)
# ---------- Backend: golangci-lint, repo config + fixed best-practice extras ----------
EXTRA_LINTERS="errorlint,errname,nilnil,forcetypeassert,copyloopvar,intrange,mirror,perfsprint,prealloc,usestdlibvars,modernize,sloglint,canonicalheader,nosprintfhostport,recvcheck,wastedassign"
golang_out=$(golangci-lint run --enable="$EXTRA_LINTERS" 2>&1 || true)
golang_total=0
while IFS= read -r line; do
if [[ "$line" =~ ^\*\ ([a-zA-Z0-9_]+):\ ([0-9]+)$ ]]; then
name="${BASH_REMATCH[1]}"
n="${BASH_REMATCH[2]}"
golang_total=$((golang_total + n))
echo "METRIC golint_${name}=$n"
fi
done <<< "$golang_out"
echo "METRIC golint_total=$golang_total"
# ---------- Frontend: eslint (repo gate) ----------
cd frontend
eslint_out=$(pnpm exec eslint . --max-warnings 0 2>&1 || true)
eslint_problems=0; eslint_errors=0; eslint_warnings=0
if [[ "$eslint_out" =~ ([0-9]+)\ problems? ]]; then eslint_problems="${BASH_REMATCH[1]}"; fi
if [[ "$eslint_out" =~ \(([0-9]+)\ errors?, ]]; then eslint_errors="${BASH_REMATCH[1]}"; fi
if [[ "$eslint_out" =~ ,\ ([0-9]+)\ warnings? ]]; then eslint_warnings="${BASH_REMATCH[1]}"; fi
echo "METRIC eslint_problems=$eslint_problems"
echo "METRIC eslint_errors=$eslint_errors"
echo "METRIC eslint_warnings=$eslint_warnings"
# ---------- Frontend: tsc (repo gate) ----------
tsc_out=$(pnpm exec tsc --noEmit --jsx preserve 2>&1 || true)
tsc_errors=$(grep -cE "error TS" <<< "$tsc_out" || true)
echo "METRIC tsc_errors=$tsc_errors"
end=$(date +%s)
total=$((golang_total + eslint_problems + tsc_errors))
echo "METRIC total_issues=$total"
echo "METRIC measure_s=$((end - start))"
+110
View File
@@ -0,0 +1,110 @@
# Autoresearch: 前后端代码质量符合最佳代码实践
## Objective
Improve backend (Go) and frontend (Next.js/TS) code quality so the codebase
conforms to best practices. NOT a performance task. Each experiment is a code
change that removes real, lint-diagnosed code-quality issues (dead assignments,
error-wrapping bugs, non-idiomatic loops, mixed receivers, unsafe error
comparisons, unnecessary string fmt, etc.) without changing behavior.
Genuine quality work only: fix code, never weaken the checks. Do NOT edit
`.golangci.yml`, eslint/biome config, or add `nolint`/`eslint-disable`
comments to reduce counts. Do NOT reformat code that isn't part of a fix
(no formatted-only churn).
## Metrics
- **Primary**: `total_issues` (unitless, lower is better) = backend golangci
issues (extended linter set below) + frontend eslint problems + tsc errors.
- **Secondary**: per-linter counts (`golint_modernize`, `golint_perfsprint`,
`golint_errorlint`, `golint_gosec`, `golint_canonicalheader`,
`golint_recvcheck`, `golint_wastedassign`, `golint_usestdlibvars`,
`golint_intrange`, `golint_forcetypeassert`, `golint_nilnil`,
`golint_prealloc`, `golint_errname`, `golint_sloglint`,
`golint_copyloopvar`, `golint_mirror`, `golint_nosprintfhostport`),
`eslint_problems`, `eslint_errors`, `eslint_warnings`, `tsc_errors`,
`measure_s` (benchmark wall time).
## How to Run
`./.auto/measure.sh` — outputs `METRIC name=value` lines. Parsed by
run_experiment automatically.
Correctness gate: `./.auto/checks.sh` runs `go vet ./...`, `go build ./...`,
and the repo's own `golangci-lint run` (repo config, tests excluded) — all
must pass. Note: `go test ./...` is NOT in checks.sh — several tests fail on
main today for environmental reasons (no local redis; flaky frpc process
tests). Don't "fix" those unless cheap and clearly unrelated to redis/flaky.
## Benchmark Definition (fixed — never change mid-session)
Backend: `golangci-lint run --enable=errorlint,errname,nilnil,forcetypeassert,
copyloopvar,intrange,mirror,perfsprint,prealloc,usestdlibvars,modernize,
sloglint,canonicalheader,nosprintfhostport,recvcheck,wastedassign`
(repo `.golangci.yml` linters stay active too; `tests: false` as configured).
Frontend: `pnpm exec eslint . --max-warnings 0` (repo gate) +
`pnpm exec tsc --noEmit --jsx preserve` (repo gate).
Do not add/remove linters or change settings to make the number go down.
## Files in Scope
Backend (Go): `cmd/`, `internal/`, `pkg/`. Anything lint-flagged in the
extended set above. Note: module name in go.mod is `github.com/Rain-kl/Wavelet`.
Frontend (TS/React): `frontend/app/`, `frontend/components/`, `frontend/lib/`,
`frontend/contexts/`, `frontend/hooks/`, `frontend/types/`, frontend scripts.
Infra: `frontend/pnpm-workspace.yaml` — approved @parcel/watcher + @swc/core
builds (fixes `make code-check` under pnpm 11; ERR_PNPM_IGNORED_BUILDS
otherwise). Already committed in setup.
## Off Limits
- `.golangci.yml`, `eslint.config.mjs`, `biome.json` — never touch to reduce counts.
- No `//nolint` / `eslint-disable` comments to silence checks.
- No reformat-only commits (biome/gofmt churn without a fix).
- No behavior changes: refactors must compile (checks.sh gate) and keep tests
semantics identical. Re-run checks.sh after every edit.
- `frontend/node_modules`, `frontend/bun.lock` (untracked, not ours).
- Do not run `go test` suites that need redis/network to declare success.
## Constraints
- Backend conventions (AGENTS.md): apps → repository → model layering;
`pkg/util/` must not import Gin/GORM/sessions; no `db.DB` in model;
response.Abort* for API errors; Chinese docs for content changes
(code-quality fixes are not content changes — no doc sync needed unless
behavior/UX changes; changelog only for user-visible changes, typically
none here).
- Frontend: run `pnpm exec biome format --write` only on files you edit
(repo `make format` uses biome); keep component placement rules.
- `golangci-lint --fix` is allowed and preferred for safe fixes
(modernize/intrange/perfsprint/usestdlibvars/canonicalheader/mirror/
copyloopvar/sloglint/errname) — review the resulting diff before keeping.
For no-fix linters (errorlint wrapping, wastedassign, recvcheck, nilnil,
prealloc, forcetypeassert) edit by hand.
## Workflow per iteration
1. Read current measure output: which categories remain, where.
2. Pick ONE category (or a coherent set of similar fixes), locate files, fix
by hand or with golangci-lint --fix scoped to that category.
3. `./.auto/measure.sh` → if total dropped → `./.auto/checks.sh` → log keep.
If flat/worse → discard or adjust.
## What's Been Tried
- Setup commit `ee6974d` (autoresearch/code-quality-2026-08-16): branch,
.auto/ session files, frontend/pnpm-workspace.yaml build approvals.
- Baseline (before any code fix): total_issues = 108
(golangci 107 = modernize 37, perfsprint 18, errorlint 12, canonicalheader 8,
recvcheck 7, wastedassign 7, usestdlibvars 3, intrange 3, forcetypeassert 3,
nilnil 3, prealloc 3, errname 1, gosec 2; eslint 1 warning
[react-hooks/exhaustive-deps in
app/(main)/pages/detail/components/pages-source-card.tsx:275]; tsc 0).
- Environment notes: golangci-lint 2.12.2 warm cache ~3s; eslint cold ~27s
(ignore stderr pnpm noise); go vet+go build ~15-30s after edits. Some
go test failures on main are pre-existing (redis down, flaky frpc) — not ours.
+2
View File
@@ -1,4 +1,6 @@
allowBuilds:
'@parcel/watcher': true
'@swc/core': true
core-js: true
sharp: true
unrs-resolver: true