Mike/pygienium
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

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

Browse Source 
`/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:
Michael Freno
2026-10-20 23:29:01 -04:00
parent f01d4c2264
commit 8990a94321
4 changed files with 148 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
tests/all-integration.test.ts
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");
});
});

102
tests/all.test.ts
View File

@@ -39,6 +39,9 @@ import {
loadRunState,
markCheckStatus,
applyPhaseStatus,
initRunState,
markRunStatus,
saveRunState,
PHASE_RECON,
PHASE_ANALYSIS,
PHASE_FIX,
@@ -229,41 +232,100 @@ describe("/pygienium-all orchestrator (task 12)", () => {
expect(summary).not.toContain("## beta");
});
it("skips terminal checks on re-run; --fresh re-runs them", async () => {
it("a completed run is not resumed: the next invocation runs every check again", async () => {
registerCheck(fakeCheck("alpha"));
registerCheck(fakeCheck("beta"));
// First run: both complete.
await captureStdout(() => handleAllCommand("--fix", stubCtx(cwd)));
expect(track.dispatched.length).toBe(4); // 2 checks × 2 phases
const firstAlphaFix = await readFile(
join(cwd, ".pygienium", "checks", "alpha", "changes.md"),
"utf8",
);
expect(track.dispatched).toEqual(["alpha", "alpha", "beta", "beta"]);
// Second run without --fresh: both already terminal → skipped.
// Second run, no flags: the stored run completed, so this is a new scan.
// Every check dispatches again — the regression this pins dispatched
// nothing but whichever check an interrupted run happened to leave open.
track.dispatched.length = 0;
const out = await captureStdout(() =>
handleAllCommand("--fix", stubCtx(cwd)),
);
expect(track.dispatched).toHaveLength(0);
expect(out.join("\n")).toContain("skipping");
const state2 = await loadRunState(cwd);
expect(state2?.status).toBe("complete");
expect(track.dispatched).toEqual(["alpha", "alpha", "beta", "beta"]);
expect(out.join("\n")).not.toContain("skipping");
const state = await loadRunState(cwd);
expect(state?.status).toBe("complete");
expect(state?.checks.alpha?.status).toBe("complete");
expect(state?.checks.beta?.status).toBe("complete");
// Third run with --fresh: both re-dispatched from scratch.
// --fresh remains the explicit reset and dispatches the same way.
track.dispatched.length = 0;
await captureStdout(() => handleAllCommand("--fix --fresh", stubCtx(cwd)));
expect(track.dispatched).toEqual(["alpha", "alpha", "beta", "beta"]);
const state3 = await loadRunState(cwd);
expect(state3?.status).toBe("complete");
// The fresh re-run overwrote alpha's changes.md (still valid content).
const alphaFix2 = await readFile(
join(cwd, ".pygienium", "checks", "alpha", "changes.md"),
"utf8",
});
it("resumes an unfinished run: terminal checks skip, the rest re-dispatch", async () => {
registerCheck(fakeCheck("alpha"));
registerCheck(fakeCheck("beta"));
// A run that stopped with alpha done and beta failed is unfinished work,
// so it continues instead of starting over.
const state = initRunState(cwd, [
{ name: "alpha", label: "alpha", fix: true },
{ name: "beta", label: "beta", fix: true },
]);
state.recon = {
complete: true,
path: join(cwd, ".pygienium", "recon.json"),
finishedAt: Date.now(),
};
markComplete(state, "alpha");
markCheckStatus(state, "beta", "failed", "provider reset mid-scan");
markRunStatus(state, "partial");
await saveRunState(state);
const out = await captureStdout(() =>
handleAllCommand("--fix", stubCtx(cwd)),
);
expect(alphaFix2).toContain("alpha changes");
void firstAlphaFix;
expect(track.dispatched).toEqual(["beta", "beta"]);
expect(out.join("\n")).toContain("skipping");
});
it("settles a phase an interrupted run left in_progress on a skipped check", async () => {
registerCheck(fakeCheck("alpha"));
registerCheck(fakeCheck("beta"));
// alpha finished, but a later interrupted run left its analysis phase open.
const state = initRunState(cwd, [
{ name: "alpha", label: "alpha", fix: false },
{ name: "beta", label: "beta", fix: false },
]);
state.recon = {
complete: true,
path: join(cwd, ".pygienium", "recon.json"),
finishedAt: Date.now(),
};
markComplete(state, "alpha");
// Directly, because applyPhaseStatus would also move the check itself back
// to in_progress — the corruption here is a phase left open behind a
// check that still reads complete.
const staleAnalysis = state.checks.alpha?.phases.find(
(p) => p.id === PHASE_ANALYSIS,
);
if (!staleAnalysis) throw new Error("alpha has no analysis phase");
staleAnalysis.status = "in_progress";
applyPhaseStatus(state, "beta", PHASE_ANALYSIS, "in_progress");
await saveRunState(state);
await captureStdout(() => handleAllCommand("", stubCtx(cwd)));
const after = await loadRunState(cwd);
// alpha was skipped as already complete: its stale phase settles to
// complete rather than leaving the check finished-but-running.
const alphaPhases = after?.checks.alpha?.phases ?? [];
expect(alphaPhases.every((p) => p.status !== "in_progress")).toBe(true);
expect(
alphaPhases.find((p) => p.id === PHASE_ANALYSIS)?.status,
).toBe("complete");
// beta never finished, so the run resumes it and its scan runs again
// (scan-only: this run passes no --fix).
expect(track.dispatched).toEqual(["beta"]);
});
it("resumes an interrupted run (re-dispatches non-terminal checks only)", async () => {