(971023) Addendum: i386-linux-gnulibc1 g++ test results

[Robert Lipe]

Jason Merrill wrote:
> I believe that the problem on linux is that you need a recent gas snapshot;
> 2.8.1 fills a .align in .eh_frame with NOPs instead of zeroes.

FWIW, I'm seeing test failures similar (but not identical) to those 
on Linux and I'm not using GAS.

> I guess we should go back to calculating the size of the .eh_frame section
> directly instead of using label subtraction, because of this and the
> assemblers that can't handle it.

There definitely seem to be a number of assemblers that are croaking 
on this.   Several targets on egcs and gcc2 encountered this.

RJL