The range of COMMIT..COMMIT does not count the starting commit, so
single commit PRs are completely unchecked, and multi commit PRs don't
have their first commit checked.
This reverts commit fd2997c2db.
Core tests are no longer online, so no longer are flaky nor tie up many
minutes of test runner time. So re-enable for all jobs.
They're still slightly flaky even with insanely long timeouts, and even when not they waste a huge amount of our runner time sitting and waiting for the DHT to disconnect and reconnect.
Still run on our fedora job to get almost all of the benefit, plus AddressSanitizer is enabled there for better testing synergy.
Keep disabled for jobs that produce artifacts to not affect user performance.
Keep disabled for other distros that don't produce artifacts as well since other versions of Qt fail themselves in e.g. TestSmileyPack's QGuiApplication usage.
Don't enable thread sanitizer yet due to many Qt internal errors.
This should still give us a significant increase in qTox coverage, even if we have low platform coverage.
Hasn't been run since the travis days.
Now managed through a deploy key rather than an access token to limit scope to
a single repo. Private key is stored in qTox/qTox as a repository secret, and
the public portion is saved to qTox/doxygen as a deploy key.
Matches Windows and AppImage build script behaviour.
Irrelevant for CI runs, but since the build dir is inside the user's mounted
qTox dir, this stops manual runs from deleting a possibly existing directory,
and allows the user to chose the build dir.
Requires removing of the check for committed generated cmake files, but I can't
imagine those making it past review anyway.