> Hmm, it does not core dump every time. The errors vary from time to
> time, too. Sometimes the compiler gets SIGSEGV, sometimes it just
> dies with an "internal compiler error" and I get told to "submit a
> full bug report".

This sounds very much like the cache corruption bug.  It amounts to
something randomly corrupting little blocks of memory, probably one
cache line in size.  What effect this has depends on what the program
is using that memory for - as I mentioned, upthread, once I saw it
strike in a way that just corrupted part of a picture.

What's the hardware?  I didn't see that in your message.

One characteristic of it is that it's not reproducible in detail; if
the compiler falls over on a given file this build, it will fail at a
different place next run, for example, whereas real bugs will generally
fail in exactly the same way on the same file every run.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B