Override Console.out and Console.err the REPL

[allanrenucci]

This make it possible to test command such as println in the REPL
scripted tests.

[allanrenucci]

I am not 100% this is the right solution. I guess hijacking Console.in and Console.out when the JVM is shared can lead to surprising behaviors (like when running inside sbt).

[felixmulder]

Very good to see some progress on this!

I agree that it's tricky. Since System.out and System.in are final MUTABLE in Java - maybe that influences this implementation?

I reckon Console.in is bound to System.in somewhere along the chain.

It looks good to me in that it solves the issue - but how resilient would it be if there is logging to stdout for instance? Eg logback and friends.

Maybe that's hard to test. Right now I don't have bandwidth to do a bug bash, but I've raised my concerns and I'm happy to see this go in since it improves the state of things! :)

[allanrenucci]

Console.out is initialised to System.out. Then if System.out is later on mutated, it will not impact Console.out. However Console.out is effectively mutated by Console.withOut(out) { ... }. The old value is restored while returning but the mutation can be seen by another thread.

It looks good to me in that it solves the issue - but how resilient would it be if there is logging to stdout for instance? Eg logback and friends.

This implementation is not mutating System.out, so anything that you run within the REPL that uses the Java API will output to System.out. We could also set System.out and System.err in withRedirectedOutput, but unless there is a use case, I would be conservative and not play with 🔥

[felixmulder]

Sounds reasonable, we did something similar in Vulpix FYI.

But yeah, LGTM :)