building with GCC-4.2 on amd64

[Kelly O'Hair]

This one?

    http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34304

Thanks.

---

Handling these changes that are 'workarounds' for something have always
been a bit of a pain. Connecting the workaround in the impacted software
with the actual bug that is. I've seen too many places where the workaround
in the makefiles just looks like an arbitrary makefile change.

-kto

Matthias Klose wrote:
> Please note that I didn't label this message as "[patch]". It's basically the
> first time that I get this working on amd64 with a FSF GCC (didn't check
> versions earlier than 4.1). This is now http://gcc.gnu.org/bugzilla/PR34304
> 
>   Matthias
> 
> Kelly O'Hair schrieb:
>> This kind of change concerns me a little.
>> Are we punishing the jdk performance of everybody because of a problem
>> building
>> with GCC 4.2?
>>
>> I know this is always a difficult situation, effectively we have a
>> workaround
>> for an unknown problem. Whether it's a bug in GCC 4.2 or a bug in the
>> jdk source
>> due to some code constructs that violate some C or C++ compiler
>> assumptions.
>>
>> My tendency is to file a bug against something, GCC or jdk, and then in the
>> makefiles, refer to this bug number, explaining that the opt level had
>> to be
>> downgraded for this reason. That gives people like ma a fighting chance
>> when
>> I decide to try and crank up the opt level 2 or 3 years from now. ;^)
>>
>> Do other people have different ways of handling this?
>>
>> In the hotspot Makefiles, they usually try and isolate the particular
>> source file
>> that triggers the failure, and then only downgrade the opt level for
>> that one
>> source file.
>>
>> -kto
>>
>> Matthias Klose wrote:
>>> When building with GCC 4.2 from the 4.2 branch, all the demos in
>>> demos/jfc and
>>> the gcjwebplugin crash. Lowering the optimization for the jdk to -O2
>>> results in
>>> working demos and plugin. The build with GCC 4.1 does fail for me on
>>> amd64, GCC
>>> 4.3 from trunk works fine with -O2 and -O3. Appended is a patch to
>>> lower the
>>> optimization.
>>>
>>>   Matthias
>>>
>