Crash in C1, using indy

[Christian Thalinger]

On Apr 12, 2012, at 8:06 AM, Charles Oliver Nutter wrote:

> FWIW, the user has offered up his contact info, in case you want to
> poke him directly: whitequark at whitequark.org

Thanks.  I'll contact him directly and see what I can find out.

-- Chris

> 
> On Thu, Apr 12, 2012 at 10:00 AM, Charles Oliver Nutter
> <headius at headius.com> wrote:
>> A JRuby user was refactoring some code to avoid the perf drop off from
>> non-inlined invokedynamic call sites and ran into a crasher:
>> 
>> u3: http://pastie.org/3774422
>> u4: http://pastie.org/3774417
>> 
>> This occurs on the OS X u4 preview and on a Linux u3 build.
>> Unfortunately the app is both complex and private so I'm not able to
>> provide a reproduction right now.
>> 
>> Disabling TieredCompilation makes the problem go away (only uses C2).
>> Running with -client fails in GVN instead:
>> 
>> u3: http://pastie.org/3774468
>> u4: http://pastie.org/3774461
>> 
>> The bytecode of the method that crashes is at the following URL. The
>> MethodHandles backing up the indy call sites can't easily be shown,
>> but Christian T. knows basically how we structure them.
>> 
>> http://kaunan.whitequark.org/files/javafail2.tbz2
>> 
>> PrintCompilation is included in this output:
>> 
>> http://kaunan.whitequark.org/files/javafail3.tbz2
>> 
>> Each fo the tarballs includes core + hs_err, etc, for consistency.
>> 
>> I can ask the user to run anything you suggest, but it seems like an
>> invokedynamic crasher in C1.
>> 
>> - Charlie