I think I found it: Re: 2.0.15 SIGBUS in mod_autoindex?

[Dale Ghent <da...@elemental.org>]

bt showed that the httpd child was dying in apr_vformatter()

I recompiled JUST apr_strings.c without -O3 (with the rest of the oblect
fles compiled with -O3) and ran the binary... no failure!

This pretty much confirmed that the SIGBUSes that I'm seeing are being
caused by a optimized apr_vformatter(). I then recompiled just
apr_strings.c with -O3 and -g for debugging info, and it died again... and
I got the following out of the stack trace:

apr_vformatter (flush_func=0xc7b00 <snprintf_flush>, vbuff=0xfed04350, 
    fmt=0xea52c "fM", ap=0xfed054a4) at apr_snprintf.c:953
953                     fp_num = va_arg(ap, double);

So, my hunch is that line is the one that doesnt like optimization of any
sort. Though, what might be wrong with it, and what to do about it, is way
beyond my C comprehension level at this point.

Possible short-term solution: Have autoconf ignore OPTIM if it is to set
on Solaris 8 machines. A bummer, yes.

/dale

Re: I think I found it: Re: 2.0.15 SIGBUS in mod_autoindex?

[Dale Ghent <da...@elemental.org>]

On Tue, 27 Mar 2001, Dale Ghent wrote:

| Possible short-term solution: Have autoconf ignore OPTIM if it is to set
| on Solaris 8 machines. A bummer, yes.

Also, I'm seeing this problem with -O2 as well. This is unfortunate
because -O2 is generally considered a "safe" optimization level, I
suppose.

humm. Can anyone else duplicate this? My config is Solaris 8, kernel
108528-06, gcc 2.95.2, and the /usr/ccs suite of compiling tools.

/dale