Add truffleruby-head to CI

[bjfish]

Please add truffleruby-head to travis CI so we can see the status of the latest TruffleRuby.

[fxn]

Awesome, looking forward to a green build! ❤️

[eregon]

Seems like truffleruby-head passed on this run:
https://travis-ci.com/github/fxn/zeitwerk/jobs/376296186
But failed here:
https://travis-ci.com/github/fxn/zeitwerk/jobs/376301423

So there looks like there is something transient.

[eregon]

That could be some unsynchronized modification of $LOADED_FEATURES, because the line failing is
https://github.com/oracle/truffleruby/blob/cc7f184c7c153b29f113c727a1695bec1af0c62a/src/main/ruby/truffleruby/core/truffle/feature_loader.rb#L145
which means $LOADED_FEATURES[i] is nil.

But I don't see any Thread in that test:

zeitwerk/test/lib/zeitwerk/test_reloading.rb

Line 156 in 1790e82

loader.eager_load

so it sounds unlikely.

So maybe the @loaded_features_index simply points to a removed entry from $LOADED_FEATURES?

@bjfish I'd suggest to try reproducing locally, and change
next if loaded_feature.size < feature.size
to
next if loaded_feature.nil? || loaded_feature.size < feature.size (+ a comment explaining why it can be nil)

[fxn]

In case it helps, Zeitwerk mutates $LOADED_FEATURES using Array#reject! here.

The motivation for that is that reloading is implemented as "remove constant + load its definition again". Since Module#autoload loads with require, we need to mutate $LOADED_FEATURES to workaround its idempotency.

[bjfish]

@fxn We found a bug related to the test failures and a fix has been pushed: oracle/truffleruby@985b21c. I expect future builds to pass for truffleruby-head.

[fxn]

@bjfish I am curious, why the emphasis on pop/shift? The API to mutate arrays has several methods, and you could combine them to edit the thing in place at arbitrary positions and preserve the size.

[eregon]

Other mutations will cause a copy of the the underlying storage (which is copy-on-write).
pop/shift do not need to modify the underlying storage, and so they just adjust the size or offset.

[eregon]

@fxn Assuming truffleruby-head passes the CI reliably now, do you think we could remove it from allow_failures?

[fxn]

@eregon Removed and green 🎉.