gstack's PR gate now runs every paid check when its import scan fails

50 minutes ago

gstack 1.91.128.0 to 1.91.130.0: the PR gate runs every paid check when its import scan fails, each PR plan previews a tighter exercised-path case map in shadow mode, shebang scripts get scanned, and the free-suite duration seeds refresh weekly. 1.91.128.0 notes: https://github.com/garrytan/gstack/blob/main/CHANGELOG.md#1911280---2026-10-18 1.91.129.0 notes: https://github.com/garrytan/gstack/blob/main/CHANGELOG.md#1911290---2026-10-18 1.91.130.0 notes: https://github.com/garrytan/gstack/blob/main/CHANGELOG.md#1911300---2026-10-18 Testing internals: https://github.com/garrytan/gstack/blob/main/docs/TESTING_INTERNALS.md

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

  1. 0:00Fallback
  2. 0:12Shadow map
  3. 0:27Fixes
  4. 0:41Your CI
Show transcript

Fallback

gstack logo

gstack's PR gate now runs every paid check when its import scan fails

Scan error, beforeno consumers
Scan error, nowfull gate
Label on the planimport scan failed + the file to fix
FallbackShadow mapFixesYour CI
garrytan/gstack · CHANGELOG.md 1.91.129.0

A failed import scan in gstack's pull request gate used to count as no consumers, so nothing ran. Now it restores the full gate, and labels the file to fix.

Shadow map

Every PR plan now previews a tighter case map, without changing what runs

The map counts only what a case executes: its touchfiles, its test's imports, its skill's shell, the bins its trials invoked
Shadow mode: one line per plan with the cases it would drop and add, and selection stays the same
Replayed over ten days of red PR runs, it kept all 47 failed gate cases
FallbackShadow mapFixesYour CI
garrytan/gstack · CHANGELOG.md 1.91.130.0

Each plan now carries a shadow line: the checks a tighter map would run, counting only what each case executes. It changes nothing yet. Replayed over ten days of failed runs, it kept every failing case.

Fixes

Three smaller fixes

Shebang scripts get scanned: extensionless bun and node scripts count as code, so 41 files reach the paid cases that use them1.91.129.0
Duration seeds refresh weekly: one bot PR, through normal review, when either seed's file set changes1.91.128.0
bun run test:health leads with cost per merged PR1.91.129.0
FallbackShadow mapFixesYour CI
garrytan/gstack · CHANGELOG.md 1.91.128.0 · 1.91.129.0

Extensionless bun and node scripts now get scanned, so forty-one files reach the checks that use them. The free suite's timing seeds refresh themselves weekly. And test health leads with cost per merged PR.

Your CI

Does your own CI run everything when it can't tell what a change touches?

docs/TESTING_INTERNALS.mdexercised-path map, PR paid-lane fallback
FallbackShadow mapFixesYour CI
garrytan/gstack · CHANGELOG.md 1.91.129.0 · 1.91.130.0

The testing internals doc in the gstack repo covers the shadow map and the fallback rules. So, does your own CI run everything when it can't tell what a change touches?