▋
docs/testing/2026-10-11-vuh-2149-packages/README.md

VUH-2149 packages test pruning

Tracks: https://linear.app/vuhlp/issue/VUH-2149

Review and every deletion/merge reason: ADR 0221 inventory. Baseline: fetched origin/main 990a774a4cceaba7a5b3f6d2939d5529a70b8041.

Slice 1 touches only Discord presence-core and bridge tests. Focused Vitest before: 271/271 pass; after: 262/262 pass. Voice-session remains 215/215, including both native disconnect cases after the table merge. Retained bridge and user-session shutdown consumers are included in both runs. No eval or live Discord operation.

All commands use clankie heavy --. The focused tests name voice-floor, voice-session, bridge commands, bridge Vox ownership and user-session shutdown; the deleted helper file is included only in the baseline.

  • Typecheck: pnpm exec turbo run typecheck --only --concurrency=1 --filter=@clankie/discord-presence-core --filter=@clankie/discord-bridge.
  • Dead code: pnpm exec knip --workspace packages/discord-presence-core --workspace apps/discord-bridge --include files,dependencies,unlisted,unresolved,exports,types,enumMembers,namespaceMembers.
  • Lint and format: explicit changed test files; package lint also covers both touched packages.

An initial table merge failed TypeScript because destructuring separated correlated event discriminants. Passing the intact row fixes it; no cast, assertion or production change. The failed log is retained locally beside the final runs.

Raw check logs, JSON reports and the complete before/after static inventory are in this worktree's .local/vuh-2149-packages/. Final slice 1 exits: focused tests 0, both package typechecks 0, scoped Knip 0, changed-file lint 0, package lint 0 and format 0. These are focused gates, not a full landing gate.

Final slice

Based on current main after bf7fe53ec: settings persona loses a repeated authority sentence plus incidental introductory copy; addresses-character loses ordinary/ empty-input helper duplication already covered by real text ingress. Per-test reasons and all assigned-area inventory counts are in the review doc.

Final slice: 2 → 2 test files; 134 → 123 lines; 14 → 12 runtime cases. Focused before/after also includes retained text-ingress and model-provider cases: 120 → 118 passing cases. The config-path helper was considered but retained as the only direct check of its documented absent-path return. Its file is unchanged.

Combined cuts: 6 → 5 touched test files, 129 lines and 11 runtime cases removed. Complete static assigned-path inventory: 187 → 186 files, 46,862 → 46,733 lines, 1,610 → 1,598 test declarations. The extra declaration reduction is the merged voice-session table: both original runtime inputs remain. Candidate review is recorded, not an exhaustive claim about every unchanged test body. Protocol, credentials, native identity, filesystem/grant fences, recorded regressions, integration/golden cases and eval inputs are retained.

Final gates use clankie heavy --:

  • Typecheck: pnpm exec turbo run typecheck --only --concurrency=1 --filter=@clankie/settings --filter=@clankie/discord-presence-core.
  • Knip: the same issue types above, scoped to settings and discord-presence-core.
  • Lint: both touched package directories; format-check names the two changed tests, the complete shared review log, this README and the evidence manifest.
  • Focused Vitest: persona, addresses-character, text-ingress and the retained model-provider file, with JSON reports. Slice 1 evidence remains valid for its unchanged files and retained consumers.

Formatting incident

CI failed solely because the newly appended review tables were not formatted. A local revert was prepared and withdrawn before any push when the lead directed a format-only repair. bf7fe53ec formats the complete shared review log; CI passed. The first slice's tested cuts remained on main. The final slice checks the whole shared Markdown file. Three pushes total: first cuts, formatting repair, final cuts.

The archived checks.json contains logs (including the initial failed typecheck and doc-format check) and runtime reports; the complete inventories and before/ after Vitest reports are separate entries in evidence.json. No full gate, eval, release, live provider call or native service restart was run.