Imported From: https://issues.scala-lang.org/browse/SI-9192?orig=1
Reporter: @sschaef
Affected Versions: 2.11.5, 2.11.6

@retronym said (edited on Feb 28, 2015 2:02:20 AM UTC):
Here's a smaller minimization:

object X {
  implicitly[com.nicta.scoobi.core.WireFormat[Nothing]]
}

-Xlog-implicits shows that we are cutting off lots of divergent nested implicit searches. We tread a fairly perilous path with divergent implicits, it seems in this case we are too lenient and should call the whole search off rather than continuing. But it is hard to be sure about what's going on without first creating a smaller set of implicits that demonstrates the essense of Scoobi's WireFormat and AvroSchema implicits.

I remember that these implicits were involved in another compilation speed problem I spent some time on: NICTA/scoobi#324

AFAICT, this behaviour isn't a regression from 2.11.5, so I'm demoting it as a blocker for 2.11.6.

@adriaanm said:
Agreed. Changing implicit behavior is probably a 2.12 thing (generally speaking) unless it's a regression.

Alex Cozzi (alecozzi) said:
Thanks for looking into it.
Not sure about the compiler itself, but least from the scala-ide I have the impression that it got worse with 2.11. I was able to edit scoobi programs with the scala ide based on 2.10. It did lock up once in a while, but was overall usable. When I updated to scala-ide based on 2.11 it would lock up almost immediately when editing a scoobi program.

Alex Cozzi (alecozzi) said:
If you change the build file as:
resolvers += "staging1" at "https://oss.sonatype.org/content/repositories/orgscala-lang-1184/"

scalaVersion in ThisBuild := "2.10.4"

libraryDependencies += "com.nicta" %% "scoobi" % "0.9.2"
it does not fail:
Java HotSpot(TM) 64-Bit Server VM warning: ignoring option MaxPermSize=256M; support was removed in 8.0
[info] Loading global plugins from /Users/acozzi/.sbt/0.13/plugins
[info] Loading project definition from /Users/acozzi/Documents/workspace-4.0/scoobi-blowup/project
[info] Set current project to scoobi-blowup (in build file:/Users/acozzi/Documents/workspace-4.0/scoobi-blowup/)
[info] Updating {file:/Users/acozzi/Documents/workspace-4.0/scoobi-blowup/}scoobi-blowup...
[info] Resolving org.fusesource.jansi#jansi;1.4 ...
[info] Done updating.
[info] Compiling 1 Scala source to /Users/acozzi/Documents/workspace-4.0/scoobi-blowup/target/scala-2.10/classes...
[error] /Users/acozzi/Documents/workspace-4.0/scoobi-blowup/x.scala:5: diverging implicit expansion for type com.nicta.scoobi.core.WireFormat[K]
[error] starting with method hlistHasWireFormat in trait WireFormatImplicits
[error] def filtered = lines.groupBy { x => ??? }
[error] ^
[error] one error found
[error] (compile:compile) Compilation failed
[error] Total time: 9 s, completed Mar 1, 2015 9:13:42 PM

but from 2.11.0 and above it does lock up.

@retronym said:
That's probably related to scala/scala#2428 and scala/scala#3669. This is a pretty shadowy area of the compiler, we seem to be oscillating between solutions which don't strike the right balance between recovering from diverging implicit expansions vs bailing out the search blows up.