docs(coverage): c8 ignore hints actually work

[AriPerkkio]

When writing this part of the guide initially I thought c8's ignore hints would not work, as their regexp is very strict and does not allow additional content such as @preseve keywords.

The ignore hints don't work with @vitest/coverage-istanbul since esbuild strips the comments before code is instrumented. Istanbul uses ignore hints to check whether coverage counters should be placed in given code block.

With @vitest/coverage-c8 the coverage filtering is done after the coverage report has been collected. The exclusion is done by comparing results to the actual sources where original comments are in place. v8-to-istanbul used by c8 uses ignore hints to see whether uncovered lines should be marked as covered.

[AriPerkkio]

The c8 tests seem really unstable now on CI. Randomly failing on both ubuntu-latest, 16 and ubuntu-latest, 18. I'm unable to reproduce this locally.

I'll do some more testing on CI and try to find root cause.

[AriPerkkio]

After rebasing latest main this seems to be working fine. I think #2758 fixed the issue. Before #2758 the worker from earlier test run was interfering the next run, as coverage tests run two test runs in loop. This only happend on heavy CPU load. The #2758 doesn't still terminate workers early enough to stop writing unnecessary V8 reports but seems to help a lot.

With 0.28.3 I'm also seeing some really good performance and stability improvements. I've been testing the test/coverage-test C8 tests in 100x loop to reproduce these CI crashes locally.

async function runTests() {
+ for (const [round] of Array(100).fill(null).entries()) {
+   console.log(`Round #${1 + round}`)
    for (const threads of [true, false]) {
      for (const [directory, config] of configs) {
        await startVitest('test', [directory], {

100x rounds of pnpm test:c8 with version

• 0.28.3: 2m29.773s, does not crash with this PRs changes
• 0.28.2, 0.28.1, 0.28.0:
  • Without changes of this PR 3m34.741s
  • With this PR's changes crashes after 3-7 rounds
• 0.27.3:
  • Crashes even without the changes of this PR after 71 rounds. Even that takes 3m50.894s. So 100x rounds would be somewhere around 4m30s maybe?
  • With changes of this PR crashes after 3 rounds.