codecov results for fork PRs are delivered before all coverage is submitted, claim massive loss of coverage

[exarkun]

If a PR is submitted from a GitHub? fork of tahoe-lafs then codecov makes its report comment very early - before most builds have finished - and the report is ugly. It often claims something like a 30% loss of coverage spread across the whole project.

It seems like codecov is generating a report from the coverage data uploaded by the *first* job to finish. It typically updates its comment later when more data comes in but the first report is obnoxious, misleading noise (it will often be emailed out in addition to showing up on the PR page).

Also, the codecov status item is not always updated. For example, right now ​https://github.com/tahoe-lafs/tahoe-lafs/pull/939/checks?check_run_id=1571210288 shows 63% coverage even though the report on codecov itself ​https://codecov.io/gh/tahoe-lafs/tahoe-lafs/pull/939 says 92% coverage. These pages are totally out of my control, of course, so who knows what they'll look like after any amount of time passes.

[exarkun]

​https://github.com/tahoe-lafs/tahoe-lafs/pull/941

[GitHub <noreply@…>]

In 6dc2928/trunk:

Merge pull request #941 from LeastAuthority?/3568.better-codecov-reporting.wait_for_ci

Try the wait_for_ci option

Fixes: ticket:3568

[exarkun]

This option has not fixed the problem. eg ​https://github.com/tahoe-lafs/tahoe-lafs/pull/948 had a bogus codecov report submitted to it long before most CI finished.