Re: gnumeric and compiler attitude

[jacob berkman <jacob ximian com>]

On Mon, 2002-06-17 at 22:52, Malcolm Tredinnick wrote:
> On Mon, Jun 17, 2002 at 09:48:02PM -0400, Jody Goldberg wrote:
> > > Can someone please explain to me why gnumeric has put a block in
> > > configure about gcc 2.96 (ie default compiler on RH/Mandrake) 
> > > 
> > > I fail to see why I should install a seperate compiler for one package 
> > 
> > Feel free to remove the check if you are willing to get incorrect
> > answers for some of the statistical functions.  There are so many
> > patch levels of 2.96 floating around and there are no neat simple
> > tests for the bogus results that we are left with a blanket no 2.96
> > restriction for now.
> 
> Are there known test cases that give the bad results? I hunted around the
> gnumeric mailing list for an explanation of this when the check originally went
> in and found nothing.

(outside of gnumeric, http://primates.ximian.com/~jacob/test.c gives
different results when built with -O2 or -O0 on rhl 7.2 or 7.3.  either
7.0 or 7.1 worked correctly.  yes, this is filed with red hat.  iirc the
reply was that it worked with latest gcc 3.x.  which i don't blame them
for - i certainly wouldn't want to debug this)

 - jacob