From mboxrd@z Thu Jan  1 00:00:00 1970
From: hjl@lucon.org (H.J. Lu)
To: khan@xraylith.wisc.edu (Mumit Khan)
Cc: egcs@cygnus.com, jason@cygnus.com
Subject: Re: (971023) Addendum: i386-linux-gnulibc1 g++ test results
Date: Thu, 23 Oct 1997 23:12:00 -0000
Message-id: <m0xOcyV-0004edC@ocean.lucon.org>
References: <199710240427.XAA27044@eh_pc11.xraylith.wisc.edu>
X-SW-Source: 1997-10/msg01003.html

> 
> I had written earlier regarding 39 unexpected failures for g++:
> 
> > Platform: i386-linux-gnulibc1 (libc 5.3.12)
> > Configure args: --enable-shared
> > 
> > g++ note:
> > Note the EH failures. The problem that I had reported for 971016
> > still exists; basically -fexceptions don't work, but -fsjlj-exceptions 
> > do work. Interesting that HJ Lu reported that he doesn't see the EH
> > failures using 5.4.(something) libc.
> > 
> > fyi, the same EH problem and work-around on i386-cygwin32, so I'm
> > inclined to believe it has nothing to do with the libc version.
> 
> I just re-ran the test cases substituting -fexceptions with 
> -fsjlj-exceptions, and here're the results:

I think this patch

Sat Oct 18 09:49:46 1997  Jason Merrill  <jason@yorick.cygnus.com>

       * libgcc2.c (__throw): Don't copy the return address.
       * dwarf2out.c (expand_builtin_dwarf_reg_size): Ignore return address.

breaks the EH on x86. Using _eh.o from egcs-971013 seems to fix the EH
bug on x86.


H.J.