enable GitHub Actions

[alexlamsl]

No description provided.

[alexlamsl]

@kzc got this working on my local fork - all Linux+Windows tests in under 40 minutes is quite an improvement from the current Travis+AppVeyor combination.

Also, continuous fuzzing 👻
https://github.com/alexlamsl/UglifyJS2/actions?workflow=Fuzzing

[kzc]

Very cool. Much better than 4 or 5 hours.

I guess AppVeyer didn't dedicate many Windows servers to open source builds. Microsoft has a vested interest in making Windows builds faster.

[alexlamsl]

I've also tried out Windows build on Travis CI, but having five concurrent jobs are still too limiting to run the full suite of tests within reasonable time period.

Thanks you for the pointers (and GitHub for their generosity)!

[kzc]

The 5 hour test runs looked painful so when I happened to see that some projects ditched the other CI systems for this I thought it might be useful.

Microsoft acquiring GitHub has some benefits resource-wise. Just hope it's not a bait and switch. It's probably in their corporate interest to help open source software.

[alexlamsl]

So far I've experienced a bunch of network errors just as we did on Travis CI for benchmark.js and jetstream.js, which we resolved through usage of cache directive.

GitHub Actions have plans to implement the same feature:
https://github.community/t5/GitHub-Actions/Caching-files-between-GitHub-Action-executions/m-p/30974/highlight/true#M630

Failing that, it would mean bunching up the tmp folder and use the upload-artifact & download-artifact Actions.

[alexlamsl]

The ongoing intermittent outages on nodejs.org reveals an issue with GitHub Actions - when it fails a single sub-job it takes at least an hour before the whole CI job will be reported as failure (even though it will automatically cancel all other sub-jobs).

And until that happens, there is no way to re-run said CI job (the cancel button won't speed up the process either).

[kzc]

Interesting. Sounds like something github can fix easily enough.