[llvm-commits] Your commit 150662

[Duncan Sands]

Hi Peter,

>>>>> I've tried to reproduce this on my Mac by getting binaries for gcc 4.2 and building in as close a setup as I can to the buildbot but I can't reproduce it.
>>>>
>>>> I can reproduce this using gcc 4.3 on a 32 bit linux machine.  I was unable
>>>> to reproduce it using different versions of gcc or on a 64 bit system.  I
>>>> will try to work out what the cause of the problem is and find a workaround.
>>> Thanks a lot Duncan.  I appreciate it.
>>
>> I committed a workaround in revision 151358.
> Nice!  Thanks a lot of fixing that.  I'm surprised it was StringRef that had the problem and not SmallVector.

I can only imagine that the SmallVector operations just after the miscompiled
statement (the check for an empty StringRef - folded to false) somehow helped
confuse gcc.  It's pretty obscure.

Ciao, Duncan.