It turns out that `syntaxAnalyzer.UnitParser()` in global now also
needs to be synchronized. The alternative is adding `synchronizeNames = true`
in the global constructor, but that already proved unreliable in the
case of #3743 (see comment https://github.com/sbt/sbt/issues/3170#issuecomment-355218833)
A thread blocking on System.in.read() cannot be interrupted, so check
System.in.available before blocking. This is how it used to work. It
requires https://github.com/sbt/io/pull/149 or else a cpu will be
pegged by the EventMonitor user input thread spinning on
System.in.available.
https://github.com/sbt/sbt/pull/4143 kept failing, and while I was debugging,
I noticed that the target that it was downloading scala-compiler was in
/tmp/... for the remote instance.
This moves the fakeboot to a full path, which should make the build
less susceptible to failure due to random network issues.
When source generators write into the unmanaged source directory, bad
things can happen. Continuous builds will loop indefinitely and
compiling will fail because the generated sources get added to the
source list twice, causing the incremental compiler to complain about
compiling classes it has already seen. My two-pronged solution is to
de-duplicate the sources task and to filter out managed source files in
watch sources. The drawback to the latter is that it causes the source
generation task to be executed twice per compile.
Fixes#3482 take 3
There are two bugs related REPL and JLine.
1. JLine getting disabled (up arrow not working) on the second run of `console` task.
2. Typed characters not showing up even on the `console` task.
The first issue was fixed by #4054 which added `t.init`. When I noticed the second problem, I fixed it in #4082 (adds `t.restore`) but undid the first fix. This attempts to fix both the issues.