middle-end/10472: [3.4 regression] ICE in instantiate_virtual_regs_lossage

[Wolfgang Bangerth]

> First I tried to determine under what circumstances and platforms the
> bug appears.  While the PR was opened against cygwin, it seems it
> should occur on any x86 target.

I'm sorry, I should have made that clearer: the reduced testcase ICEs on 
my plain vanilla x86-linux-gnu box with SuSE 8.0.


> (and why I'm not being chased with torches and pitchforks)

We write hate-mails these days -- mental torture is just as effective :-)


> BTW, IMHO the reduced testcase in the PR is slightly misleading.  It
> lead me at least to believe the bug had something to do with null
> pointers.

Again my fault. 


> In the mean time, we can disable the optmization on x86 so that cygwin
> people can resume bootstrap.  (Yes this is lame, sorry I'm not able to
> provide a proper fix.)

Alternatively, one could work around the bootstrap failure with something 
like the following instead of your small testcase (I'm surprised this 
works):
    void foo (char *s)
    {
      char * tmp = __builtin_stpcpy (s, "hi");
      f (tmp);
    }
That's just as lame, and fragile on top of that, but we don't have to wield 
the sledgehammer to stomp this bug, it would be simpler to add a testcase 
and we don't have to remember to re-enable the optimization once someone 
gets around to fixing the underlying problem. There can't be that many 
places in gcc where such a fix would be necessary, and Giovanni would 
surely be willing to find them on his box, right? :-)

W.

-------------------------------------------------------------------------
Wolfgang Bangerth              email:            bangerth@ices.utexas.edu
                               www: http://www.ices.utexas.edu/~bangerth/