🐛 repro: external resolve advances action-orchestrated loop (#1188)

[taras]

Reproduction for #1188. These tests are expected to fail — they document the bug and should go green once the underlying fix lands.

What it shows

When shutdown is triggered externally via scope.run(() => resolve(...)) from inside an action-orchestrated loop, the loop advances into the next iteration (round-2-start) after the shutdown trigger, instead of stopping once the current iteration finishes.

Two tests are added to test/scoped.test.ts:

1. does not advance loop after external resolve with nested scoped + race — mirrors the original report (action wrapping a loop, each round runs scoped(scoped(race(...)))). Observed events:

   round-1-start
   round-1-result:false
   round-failed-trigger-shutdown
   round-2-start        ← leaks
   

2. does not advance loop after external resolve in minimal action baseline — strips out scoped() and race() entirely; just action + a for loop + sleep + external resolve. Still leaks:

   round-1-start
   round-failed-trigger-shutdown
   round-2-start        ← leaks
   

Why the baseline matters

The minimal baseline has no scoped and no race, yet it still advances to round-2-start. That isolates the root cause to the action + external-resolve path — i.e. resolving an action from outside (via scope.run) does not unwind the operation that is currently suspended awaiting that action before the loop takes its next step.

This is the broader category #1188 refers to: #1185 was the narrower scoped + race manifestation (fixed in #1186 / 49fe66f0), but the same "external resolve doesn't halt the in-flight iteration" problem reproduces with a plain action loop.

Run

deno test --allow-run --allow-read --allow-env --allow-ffi test/scoped.test.ts

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/effection@1189

commit: d1e5aa6

[codspeed-hq]

Merging this PR will not alter performance

✅ 6 untouched benchmarks

---

_{Comparing repro/1188-action-loop-external-resolve (d1e5aa6) with v4 (49fe66f)}

[taras]

this is not a problem. It just shows that we can interrupt sync loop which is not expected to be possible