Stu Hood

I'll take a look at fixing this today.

It looks like this one might go fairly deep into a rabbit hole. It's very, very easy to reproduce in a zinc test: I've pushed a repro here: https://github.com/sbt/zinc/compare/1.x...stuhood:stuhood/double-dollar ....

...but @yufangong 's linked PR definitely looks relevant.

I was able to work around this by using a different namespacing character in this generated code, so I think we're unblocked.

Correct: pants builds are not currently bitwise identical. zinc is one piece of that problem, but there are plenty of others. We're moving toward an execution model that will allow...

> An idea that @smarter gave me last weekend is that we could test how good incremental compilation is now by creating a script that compiles from scratch every commit...

> protobuf deserialization logic To be clear, I'm not blaming the deserialization logic: since most of the objects created there are short lived, they should get cleaned up quite easily....

> I agree this is a superior solution than String.intern, which can have unexpected results if the JVM string table gets too full. Are you going to submit a patch...

Hm, so: it looks like maybe `mapMiniSetup` is only applied in `TextAnalysisFormat`. But the protobuf analysis format actually does something a bit better: it applies `mapOutputDir` to the output dir....

I was totally wrong about `MiniSetup` not being portable. Have updated the title/description.