Zinc doesn't invalidate when it should

[benjyw]

See https://github.com/benjyw/pants/blob/benjy/invalidation_bug_repro/BUG_REPRO.md for details and a branch that reproduces this.

[benjyw]

Giving this a bump. I don't remember if we think this is a Pants issue or a Zinc issue. If the former, I can take a look myself.

[stuhood]

@benjyw : One thing to check would be whether it still fails in #4729, which upgrades us to much newer zinc.

[benjyw]

That change isn't in master though, right?

[stuhood]

It is not, no. I hope to get some more time to work on landing it soon, only relatively minor issues remain (related to requiring that zinc run in JDK8+, and forking if an older JVM is requested for java compilation).

[benjyw]

I notice that 33e98e1 upgraded us to 1.0.0-X20 though. Isn't that newer than -X16?

[benjyw]

Wait, we're at 1.0.3 in master. I'm too confused. However it does appear that this does not repro in current master. I tried to sync back past 33e98e1 but failed to build the native engine. Sigh.

[benjyw]

Anyway, this no longer repros, possibly thanks to 33e98e1. Closing.

[stuhood]

33e98e1 updates the JVM portion of zinc, but doesn't use it in pants. That's what #4729 will do.

[benjyw]

Hmm, well then perhaps 3e5f680 is the fix.

[benjyw]

TODO: For peace of mind, find the commit that fixed this before closing this issue.

Doing so requires a solution for building older native engine versions. See #4909.