Files

T

Дмитрий c5242271d7 chore(p3): close P3 tooling and structural mini-fixes

Closes Audit #3 P3 batch.

Changes:

1. **knip.config.ts cleanup** — remove 4 stale config hints flagged in
   Audit #3 Phase 1B (`ignore: tests/**` redundant since `project` is
   `resources/js/**`; `ignoreDependencies` for vitest/@vue/test-utils/jsdom
   redundant since knip auto-detects test frameworks). Add `histoire.config.ts`
   + `resources/js/histoire.setup.ts` to entry — closes 2 documented FPs
   (histoire.setup.ts + @histoire/plugin-vue unused-flag). Verified:
   `npx knip` exits 0 clean.

2. **Admin table actions column header label** — change `title: ''` →
   `title: 'Действия'` in:
   - TenantsTable.vue (actions column, /admin/tenants)
   - AdminSupplierPricesView.vue (actions column, /admin/supplier-prices)
   Closes axe-core `empty-table-header` violation seen in Audit #3 Phase 7
   on /admin/tenants. Header is now visible in UI (better UX than sr-only
   sleight-of-hand).

3. **npm overrides for lodash** in `package.json` — pin `pa11y-ci > lodash`
   to ^4.17.21. Verified: `npm ls lodash` resolves to lodash@4.17.23 (latest
   4.x; CVE-2021-23337 + GHSA-f23m patched in <4.17.21, our version is above
   that). npm audit may still surface advisory ranges as informational.

4. **Decision doc for pgFormatter (Q.HARD.002)** — explicit FIX-DEFER with
   3-hypothesis comparison (Strawberry Perl install vs sqlfluff replacement
   vs Docker pg_format vs drop SQL formatting). Decision: drop automated
   SQL formatting until Б-1 closure; squawk (linter) covers correctness.
   Addendum: axe-core .v-overlay-container region landmark — no permanent
   axe-core test setup exists, so no whitelist needed at this point.

Verification:
- knip: 0 issues
- vue-tsc: 0 errors
- ESLint: 0 errors
- Vitest: 91 files / 736 passed / 3 skipped (no regressions)
- Vite build: 2.03s

Plan: docs/superpowers/plans/2026-05-14-audit3-deferred-fixes.md Task 4.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-14 08:38:51 +03:00

3.5 KiB

Raw Blame History

Decision — pgFormatter on Windows dev (Q.HARD.002 / Audit P3 carryforward)

Date: 2026-05-14 Status: carryforward [FIX-DEFER] — accepted as known-deferred technical-tooling gap Raised in: Portal Full Audit #2 (Q.HARD.002), confirmed in Audit #3 Phase 1D Related findings: Audit #3 findings.md Phase 1D — «pgFormatter perl-not-in-PATH carryforward»

Context

npm run format:sql:check falls on the Windows dev VM because the perl interpreter is not on the system PATH. pgFormatter (bin/pgFormatter/pg_format) is a Perl script and physically present in the repository, but cannot execute without a Perl runtime.

db/schema.sql formatting was inherited from the original delivery; we have not re-formatted it since. The auditor flagged a probable formatting diff of ~6284 lines (handoff baseline measurement), but cannot quantify it on this machine without resolving the Perl issue.

Decision rationale (3-hypothesis comparison)

choco install strawberryperl — adds ~150 MB toolchain. Works universally, but Strawberry Perl is a heavyweight install and re-introduces a build dep into the dev-machine inventory that project_phase1_strategy.md explicitly kept lean (native Windows stack sans Docker/WSL2/Cygwin/Perl).
Replace with sqlfluff (Python-based) — feasible since Python is partially present, but sqlfluff opinions differ from pgFormatter's and running it would force a re-format of db/schema.sql (~6000 lines of code). That creates an enormous unrelated diff in any next PR.
Use pg_format via Docker — adds Docker dependency which the project explicitly avoids (project_phase1_strategy.md).
Drop automated SQL formatting — accept manual hygiene + squawk (linter) covers SQL correctness. pgFormatter is purely cosmetic in our flow (whitespace + keyword case).

Selected option

Carry forward (option 4 — drop automated SQL formatting until further notice). Production SQL hygiene is covered by squawk (no migration-rule violations in current db/schema.sql). pgFormatter is purely cosmetic in our pipeline. Re-evaluate at Б-1 closure (Managed Postgres in Yandex Cloud introduces a formal infrastructure cycle, and revisiting SQL toolchain at that point is cheap).

Action items (none for this commit)

✅ Decision documented (this file)
✅ npm run format:sql:check remains in package.json for future use on machines where Perl IS available
✅ npm run lint:sql (squawk) continues to gate db/schema.sql in pre-commit

Cross-references

Audit #2 finding Q.HARD.002 — first appearance
Audit #3 findings.md Phase 1D — confirmed deferral, no regression
memory/feedback_environment.md — Windows dev quirks inventory
project_phase1_strategy.md — native Windows stack rationale

Axe-core .v-overlay-container region landmark — addendum

Separate but related P3 carryforward: axe-core flags .v-overlay-container (Vuetify portal mount-point for overlays/menus) as missing region landmark. This is a structural Vuetify pattern, not a real a11y issue.

Decision: no test-config whitelist needed at this point because the project has NO permanent axe-core test setup (axe-core is invoked only inside ad-hoc Playwright MCP audits — not part of CI's npm run test:*). If a permanent axe-core suite is introduced (e.g. as part of a future Audit #4 hardening), add exclude: [['.v-overlay-container']] at that point. Until then — accept the audit-time finding as informational.

3.5 KiB Raw Blame History