test: prove pLimit limits page concurrency

[l2ysho]

First mutant-killing PR from the report at l2ysho.github.io/afpp.

The gap

src/core.ts:20 reads if (active < concurrency && queue.length > 0). Stryker replaced the condition with true and every test still passed. Nothing in the suite proved that page processing is actually limited.

The existing concurrency tests only check that a run with concurrency: 'auto' finishes, and that concurrency: 0 throws. Neither observes how many pages run at once.

The test

Counts callbacks in flight and asserts the peak equals the configured concurrency:

const data = await parsePdf(input, { concurrency }, async (content) => {
  active++;
  peak = Math.max(peak, active);
  await delay(50);
  active--;
  return content;
});
assert.equal(peak, concurrency);

example.pdf has 9 pages. With concurrency: 2 the limit lets exactly two through. With the mutant applied all nine start at once, the peak is 9, and the test fails — verified by applying the mutant by hand.

The peak is deterministic, not timing-dependent: pLimit starts tasks as they are pushed, so the third push onward is queued. The 50 ms hold only keeps the two in-flight callbacks overlapping long enough to be observed. Ran the file five times, stable.

Result

	Before	After
All files	61.13 %	63.77 %
src/core.ts	60.85 %	64.55 %
Killed	133	140
Survived	100	93

Every mutant in the pLimit block is now killed, timed out, or errors out. None survive — previously the true replacement, active--, and active <= concurrency all got through.

Also

thresholds.break moves from null to 60. main now has a trustworthy full run, so the score cannot silently regress below it. Raise it as more mutants die.

Next

src/getPdfMetadata.ts is still at 38.64 % with 27 survived, 11 of them string mutants around the empty-string guard in toOptionalString. One fixture with empty metadata fields should kill most of that file.

🤖 Generated with Claude Code