fix: a finished all-run is not a run to resume

`/pygienium-all` loaded the previous run-state and skipped every check
whose entry was terminal, so a second invocation dispatched nothing at
all. The one check that did run was whichever an interrupted run had
left non-terminal — registry order made that `comments`, which read as
"comments ran, everything after it was skipped".

Resumability belongs to unfinished work: continue only when the stored
run is not complete (`in_progress`, `partial`, `failed`), so completed
checks are carried forward cheaply and the interrupted one picks up
where it stopped. A completed run means the next invocation is a new
scan, which is what the command name promises; `--fresh` and
`/pygienium-resume` keep their explicit jobs.

Settle a terminal check's phases on the way past too: an interrupted run
can leave a phase `in_progress` behind a check that reads `complete`,
which made the state claim both finished and still running.

Tests: reverted src/ fails "a completed run is not resumed", "settles a
phase an interrupted run left in_progress", and the integration test "an
unfinished run-state is continued, not restarted"; all pass with the
fix. Suite 164 pass, 1 pre-existing worktree-only failure (MISC-46).
bun run typecheck clean; port-to-omp regenerates and typechecks.

MISC-47

Co-Authored-By: nw-flash-flex <noreply@neuralwatt.com>
This commit is contained in:
2026-10-20 23:29:01 -04:00
parent f01d4c2264
commit 8990a94321
4 changed files with 148 additions and 33 deletions

View File

@@ -22,7 +22,7 @@ import {
type AgentRunner,
} from "../src/agent-runner.js";
import { handleAllCommand, type PygieniumCtx } from "../src/commands.js";
import { loadRunState } from "../src/run-state.js";
import { loadRunState, markCheckStatus, markRunStatus, saveRunState } from "../src/run-state.js";
import { canonicalChecksRoot } from "../src/export.js";
/** Stub check whose fake-runner task writes artifacts + echoes a line. */
@@ -147,16 +147,27 @@ describe("/pygienium-all end-to-end (task 14)", () => {
expect(out.join("\n")).toContain("no checks registered");
});
it("is resumable: a second call reuses the existing run-state", async () => {
it("is resumable: an unfinished run-state is continued, not restarted", async () => {
registerCheck(fakeCheck("alpha"));
// A completed run is finished work: the next call starts a new run-state.
await captureStdout(() => handleAllCommand("--fix", stubCtx(cwd)));
const first = await loadRunState(cwd);
const firstStarted = first!.startedAt;
const completed = await loadRunState(cwd);
expect(completed!.status).toBe("complete");
await captureStdout(() => handleAllCommand("--fix", stubCtx(cwd)));
const fresh = await loadRunState(cwd);
expect(fresh!.startedAt).toBeGreaterThan(completed!.startedAt);
// An unfinished one is continued: same run, same startedAt, and the
// completed check is not dispatched again.
const interrupted = await loadRunState(cwd);
markRunStatus(interrupted!, "in_progress");
markCheckStatus(interrupted!, "alpha", "in_progress");
await saveRunState(interrupted!);
const before = (await loadRunState(cwd))!.startedAt;
// Second run reloads the existing run-state (same startedAt).
await captureStdout(() => handleAllCommand("--fix", stubCtx(cwd)));
const second = await loadRunState(cwd);
expect(second!.startedAt).toBe(firstStarted);
expect(second!.status).toBe("complete");
const resumed = await loadRunState(cwd);
expect(resumed!.startedAt).toBe(before);
expect(resumed!.status).toBe("complete");
});
});