[llvm-commits] [llvm-gcc-4.2] r54260 - /llvm-gcc-4.2/trunk/gcc/c-common.c

[Mon Ping Wang]

Hi Duncan,

I completely agree with you. I view the checkin as a temporary workaround.
The main reason why I checked it in was that I was concerned that my
breaking the build is preventing other people from get any work done on
llvm-gcc.  I'm planning to try to figure out why the compiler is doing
this weekend and try to fix it. But from your email, it sounds like this
bug doesn't badly disrupt people's work flow. If that is the case, I don't
have any problem of ripping out my hack.

  -- Mon Ping


On Fri, August 1, 2008 12:52 am, Duncan Sands wrote:
> Hi,
>
>> Removed an unnecessary condition check that seems to cause an
>> optimization
>> error that prevent building after r54240
>
> I think this is completely the wrong approach.  If I
> understand right, this code is being miscompiled.  The
> correct solution is to fix the compiler, not to tweak
> the code!  If it was being miscompiled by someone else's
> compiler, then I'd be alright with tweaking the code.
> But it's being miscompiled by our own compiler!
> What's more, I was told on irc [*] that this only happens
> (in stage3) if you bootstrap, not if you do a normal
> build.  This suggests that llvm-gcc has miscompiled
> itself, and the miscompiled llvm-gcc is then miscompiling
> this code.  This is badness, and must be fixed.
>
> So please revert your changes.
>
> Thanks,
>
> Duncan.
>