And... the PR doesn't trigger GitHub Actions (guess it has to be in the branch first). Here's how the workflow turned out on my fork: https://github.com/zmwangx/googler/actions/runs/114040792
There should be zero setup other than merging.
Note: to properly investigate failures we sometimes might need to get our hands on the raw HTML response instead of truncated strings from pytest, so ideally we need to package up the raw HTML responses as artifact. This is already possible right now with --debug (to make things easier we can set a special TMPDIR, then zip up the whole thing at the end). However, I'm saving this work for later if the need actually arises.
Testing overhaul
================
Rewrite test suite based on survey of past issues and regressions.
Use pytest instead of homegrown "framework". Use structured output to simplify
programmatic testing.
Migrate to GitHub Actions.
I was saying the lock app works for googler only because it was installed in the past, new installations don't work anymore. ("Grandfathered in" example: I'm grandfathered into *<*insert telecom*>*'s $*<*insert dollar amount*>* plan, which doesn't exist anymore if you sign up today.)
Btw, I caught this while thinking about rewriting the test suite to be not utterly useless (I surveyed issues within the last two years, and all regression PRs, apparently the CI test never caught anything...)
Will have a PR maybe tomorrow, or tonight if I'm motivated enough.
Fix extraction of matched keywords
==================================
At some point Google apparently switched to the em tag (at least in the version
I'm served).