When ScalacWorker fails during macro evaluation, errors are not made available in build log

[thirtyseven]

When compiling a scala_library that requires a macro expansion, and there is a user error in that macro, the entire scalac process will crash, resulting in a Bazel error like this:

ERROR: /Users/tkaplan/<redacted>:3:14: scala @//<redacted> failed: Worker process did not return a WorkResponse:

---8<---8<--- Start of log snippet, file at /private/var/tmp/_bazel_tkaplan/fa1bc4e61b191b698b582e9d142cb714/bazel-workers/worker-6-Scalac.log ---8<---8<---
[... truncated ...]
cros$MacroExpander.apply(Macros.scala:609)
	at scala.tools.nsc.typechecker.Macros.standardMacroExpand(Macros.scala:784)
	at scala.tools.nsc.typechecker.Macros.standardMacroExpand$(Macros.scala:782)
	at scala.tools.nsc.Global$$anon$5.standardMacroExpand(Global.scala:482)
	at scala.tools.nsc.typechecker.AnalyzerPlugins$$anon$10.default(AnalyzerPlugins.scala:457)
	at scala.tools.nsc.typechecker.AnalyzerPlugins$$anon$10.default(AnalyzerPlugins.scala:454)
	at scala.tools.nsc.typechecker.AnalyzerPlugins.invoke(AnalyzerPlugins.scala:411)
	at scala.tools.nsc.typechecker.AnalyzerPlugins.pluginsMacroExpand(AnalyzerPlugins.scala:454)
	at scala.tools.nsc.typechecker.AnalyzerPlugins.pluginsMacroExpand$(AnalyzerPlugins.scala:454)
	at scala.tools.nsc.Global$$anon$5.pluginsMacroExpand(Global.scala:482)

There are several reasons why this isn't very user-friendly:

1. The actual reason for the error is in the truncated part of the stack trace (in this case, it was a NoSuchMethodError caused by mixing Scala 2.11 and 2.12 libs)
2. If the build is happening on CI, getting the worker-n-Scalac.log file can be difficult
3. The error won't be sent as part of the BES to any BES viewers like BuildBuddy

Maybe the ScalacWorker implementation should try to catch unchecked exceptions and determine if they are the result of macro expansion, rather than crashing.

Full worker log:
worker-6-Scalac.log

[johnynek]

if you have time to send a test that reproduces this issue it would be helpful.

I thought we used to handle nonfatal exceptions, but maybe I'm misremembering.

[thirtyseven]

I have a PR that fixes this for our case (#1489), it will take me a little longer to actually extract a repro that I can share.

[thirtyseven]

@johnynek Linked PR has an E2E test that reproduces the issue