GCC Status Report (2004-04-05)

[Richard Guenther]

Richard Guenther wrote:
> Mark Mitchell wrote:
> 
>> GCC 3.4
>> =======
>>
>> I believe that we are not at the point of diminishing marginal
>> returns.  There is still a code-gen problem on PA that Zack and/or I
>> may look at, but I think that we have gone about as far as we should
>> go for 3.4.0.
>>
>> Therefore, I plan to roll the prerelease tonight.
>>
>> If you have an outstanding patch which I have already approved for 
>> 3.4.0, please check it in now.  Other than that, the 3.4.0 branch is 
>> now fully frozen to any and all commits.
>>
>> The actual release will follow in a week or so, assuming all goes
>> well.
>>
>> Thanks to Eric and Jakub for working hard on the aliasing issues this
>> past week!  And to everyone else who fixed bugs!
> 
> 
> I just checked the 3.4 branch wrt POOMA again, and viola, got another 
> miscompilation compared to g++ (GCC) 3.4.0 20040331 (prerelease).
> src/Particles/tests/bctest3.cpp is miscompiled with -O2 and passes with 
> -O on ia32-linux.  It's again some sort of Heisenbug (commenting out 
> unrelated code before the failure makes the test pass), as were previous 
> ones that just disappeared and as are the ones filed against 3.3.  So I 
> suspect there is some weird bug somewhere at least both in 3.3 and 3.4.

It also miscompiles my main application, entering an endless loop during 
initialization, g++ (GCC) 3.4.0 20040323 (prerelease) was fine here. 
I'll try to followup this one with a PR as it _seems_ to be easier to 
track down...

Richard.