Imported From: https://issues.scala-lang.org/browse/SI-5333?orig=1
Reporter: Marek Wiącek (marekw)
Assignee: @jrudolph
Affected Versions: 2.9.0-1, 2.10.0, 2.11.0-M8
See #5699
Attachments:

• Const.java (created on Dec 22, 2011 6:46:41 PM UTC, 61 bytes)
• IntAnnotation.java (created on Dec 22, 2011 6:46:41 PM UTC, 51 bytes)
• Test.scala (created on Dec 22, 2011 6:46:41 PM UTC, 44 bytes)

@soc said:
I can verify this with Scala version 2.10.0-20120511-235304-d55ea54589 (OpenJDK 64-Bit Server VM, Java 1.6.0_24), which contains the fix for #2764.

Looks like another issue related to the classfile parser like #5699.

@vigdorchik said:
I believe this one is different from #5699:
the former is about incorrect parsing of annotation definition, while this one is about the parser not parsing java field initializer, and thus confusing scala constant evaluator.

@lrytz said (edited on Oct 20, 2014 12:56:08 PM UTC):
Problem is: it's an intrusive change trying to fix this one, because we don't compile java files (just parse signatures). in Scalac, constants folding happens thanks to the type system, (1 + 1) has ConstantType(2). To make the following work, we need to push expressions through the typer:

public class A {
  public static final int N = 0;
  public static final int K = A.N + A.N; // needs to have ConstantType
}

Also, the way things are set up in Scalac, a compile-time constant (in Scala) needs to have an inferred type. In Java, there's always an explicit type. So we need to be able to decide somehow if the type (for K) should be inferred or not.

@adriaanm said:
A good initial implementation was done by @jrudolph in scala/scala#5487. A full implementation would move this logic into the type checker, but I think this could be tracked as a new issue when it comes up.