[c++,committed] Revert patch for PR/c++2294

[Bernardo Innocenti]

Eric Botcazou wrote:
>>Yeah, "hum" is the right word here. Do all those regressions show up with
>>ONLY my patch applied to mainline?
> 
> I just reverted Bernardo's reverting patch and ran the C++ testsuite: the 
> hundreds of failures didn't show up.  Weird.  I was away this week-end so 
> everything was automated; on Saturday morning hundreds of C++ failures, on 
> Sunday morning they were gone.
>
> Sorry for the noise.

Concurrent development sucks :-)

Anyway, I had a phone talk with Giovanni. He was pretty sure his
patch was fine just before I committed it.

I've pre-processed the Java source causing the problem and tried
to reduce it to a managable test case.

The error appears on a rather simple class which is wrapped inside
two namespaces. I was sure the failure lied in there, but the
error disappeared when I got rid of the Java interface stuff.

My guess is that the error is being triggered by some GCC
extension used to glue C++ with Java classes.

I've Cc'd the java list too; perhaps the GCJ people have better
insight on this topic.

-- 
  // Bernardo Innocenti - Develer S.r.l., R&D dept.
\X/  http://www.develer.com/