Unrecognized insn internal error in egcs 1.1.2

[Joe Buck]

> >> OK, I'm pulling my hair out.  1.1.2 is giving me this weird error on
> >> Linux IA32 while compiling a C++ program.  Should I update to 2.95?
> > 
> > Since you only provide a code fragment, not a complete input, no one
> > can tell whether or not 2.95 fixes the bug you are seeing.
> 
> Unfortunately the complete input is rather huge.  I'll try to pare it down. 

Alternatively, you could build gcc 2.95 and try it.  Otherwise there's
no alternative than to send an example.

> Does all of this stuff spit out by the compiler (below) provide any
> clues?  Does anyone have any idea exactly what this output means?

The output means that, in the process of transforming your input, the
compiler has produced a bit of RTL code that cannot be turned into
any legal instruction.  This is not supposed to ever happen, and indicates
a compiler bug.  Generally speaking, people are never going to be able
to diagnose these just from a code fragment and an error message; such
things may depend on the entire state of the compiler just before it
reached this point.

> Am I getting i686 or i586 (or i486?) if I compiled egcs using its default
> compilation rules on a Pentium II machine?

Type "gcc -v" and the compiler will tell you.