Support forward reference from annotation argument to member of the annotated class.

[scabug]

With Scalac 2.10, the use of any objects, class or method annotaded with the com.unboundid.util.ThreadSafetyLevel leads to java.lang.AssertionError.

For example, the following file Test.scala:

import com.unboundid.ldap.sdk.Entry

object Test extends App {
  
  val foo = new Entry("plop")
  
}

Leads to :

% scalac -classpath ./lib/unboundid-ldapsdk-2.3.1.jar src/Test.scala
warning: Caught: java.lang.AssertionError: assertion failed: 
     while compiling: src/Test.scala
        during phase: global=typer, atPhase=parser
     library version: version 2.10.0
    compiler version: version 2.10.0
  reconstructed args: -classpath ./lib/unboundid-ldapsdk-2.3.1.jar

  last tree to typer: Ident(Entry)
              symbol: <none> (flags: )
   symbol definition: <none>
       symbol owners: 
      context owners: value foo -> object Test -> package <empty>

== Enclosing template or block ==

Template( // val <local Test>: <notype> in object Test
  "App" // parents
  ValDef(
    private
    "_"
    <tpt>
    <empty>
  )
  // 2 statements
  DefDef( // def <init>: <?> in object Test
    <method>
    "<init>"
    []
    List(Nil)
    <tpt>
    Block(
      Apply(
        super."<init>"
        Nil
      )
      ()
    )
  )
  ValDef( // private[this] val foo: <?> in object Test
    private <local> <locked>
    "foo"
    <tpt>
    Apply(
      new Entry."<init>"
      "plop"
    )
  )
)

com.unboundid.util.ThreadSafetyLevel while parsing annotations in ./lib/unboundid-ldapsdk-2.3.1.jar(com/unboundid/util/ThreadSafetyLevel.class)
one warning found

ThreadSafety.java source here: http://ldap-sdk.svn.sourceforge.net/viewvc/ldap-sdk/trunk/src/com/unboundid/util/ThreadSafety.java?revision=410&view=markup

Link to unboundid-ldapsdk-2.3.1.jar: https://www.unboundid.com/products/ldap-sdk/files/unboundid-ldapsdk-2.3.1-se.zip

That seems to be linked to the following comment: https://github.com/scala/scala/blob/master/src/compiler/scala/tools/nsc/symtab/classfile/ClassfileParser.scala#L1033

// We want to be robust when annotations are unavailable, so the very least
// we can do is warn the user about the exception
// There was a reference to ticket 1135, but that is outdated: a reference to a class not on
// the classpath would *not* end up here. A class not found is signaled
// with a `FatalError` exception, handled above. Here you'd end up after a NPE (for example),
// and that should never be swallowed silently.

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-7014?orig=1
Reporter: Francois Armand (fanf)
Affected Versions: 2.10.0, 2.10.2
See #7551

[scabug]

Sebastian Meßmer (smessmer) said:
same bug here. It's only a warning, but it's a disturbing one with a lot of output.

[scabug]

Mariusz Sakowski (sakfa) said:
Same thing here, major issue for me because Play Framework 2.0 gets confused when displaying normal compilation errors (displays this warning instead)

[scabug]

Francois Armand (fanf) said:
Since Scala 2.10.2, that bug simply terminate the build, so basically, I can't use Scala 2.10.2 on my project. I would qualify that one as blocker now - I'm stuck with Scala 2.10.1.

[scabug]

@adriaanm said:
sorry to hear it -- looking into it
test files here: https://github.com/adriaanm/scala/tree/ticket-7014

[scabug]

@gkossakowski said:
Assigning to Adriaan who seemed to look into this. Feel free to unassign if you don't plan to work on this ticket.

[scabug]

Francois Armand (fanf) said (edited on Jul 16, 2013 11:47:37 AM UTC):
I just got eaten with the update to Scala-IDE 3.0.1 which uses Scala 2.10.2, and so is not usable on our project.
So, that bug is starting to be really annoying.

All in all, I prefered the past behaviour, with hundreds of assert error lines in the log. Our build output were crap, but actually could build the project...

[scabug]

Sonnenschein (sonnenschein) said:
I've also voted for this issue to be resolved.

For the time being, I decided to package my single layer dealing with LDAP into a separate JAR which I than use as an unmanaged library. Since the rest doesn't need a dependency to UnboundID-SDK, I can keep out the annoying noise from the rest.

Peter

[scabug]

Jun Yamog (jkyamog) said:
Same here. We are using unboundid in our play app. So eclipse is reporting SBT is crashing. We haven't upgraded our play project, so command line compiling is still working.

[scabug]

@adriaanm said:
Looks like RUNTIME retention annotations aren't supported: https://github.com/scala/scala/blob/2.10.x/src/compiler/scala/tools/nsc/symtab/classfile/ClassfileParser.scala#L911

Not sure what we can do for this in 2.10, as this is now a maintenance release. Happy to take PRs.
I'll still see if I can avoid the crash and continue with a dummy or something.

[scabug]

@adriaanm said:
I jumped to conclusions: the problem is the forward reference from the annotation argument to a member of the annotated class. We won't be able to fix this for 2.10.3, but I'll submit a PR that turns the crash into a warning. You should be able to compile normally, I assume.

[scabug]

@adriaanm said (edited on Aug 13, 2013 10:18:44 PM UTC):
Fix for crasher: scala/scala#2829
(Don't close on merging.)

[scabug]

@retronym said:
See also #7551 / https://github.com/retronym/scala/compare/ticket;7551

commit b545e410e072600b1dc77c6b8536eb3c22cd657a
Author: Jason Zaugg <jzaugg@gmail.com>
Date:   Tue Jun 4 09:32:36 2013 +0200

    WIP SI-7551 Ignore class file parser cycles in annotations

    But should we be parsing runtime visible annotations other than
    ScalaSignature at all?

[scabug]

Francois Armand (fanf) said:
I can confirm that the workaround allows to build our project again, and avoid displaying horrible debug information on the stdout.

Thanks!

[scabug]

Hendy Irawan (ceefour) said:
+1 for this