Re: gnumeric and compiler attitude

[Jody Goldberg <jody gnome org>]

On Tue, Jun 18, 2002 at 09:12:51AM +0200, Thomas Vander Stichele wrote:
> > http://bugzilla.gnome.org/show_bug.cgi?id=68057
> 
> This one seems to say that it was NOT an error with gcc, but rather 
> gnumeric using the "wrong" header to pull stuff in.

It should not be getting the wrong header.
 
> > http://maths.newcastle.edu.au/~rking/R/devel/01c/0929.html
> 
> This one seems to say that gnumeric pulled in a 2.5 year old function from 
> R (whatever R is) that has been fixed in R already.

We did pull a function from R (a free replacement for S) 2 years
ago.  However, this is numeric code it does not change or rot
quickly.  There have not been updates.

> removing that check, I don't really like it when people say "oh, it's 
> redhat's gcc compiler who's to blame" when all that redhat's gcc has ever 
> done is show bugs or code with undefined behaviour that other gcc versions 
> let pass, but should have flagged as well.
We are not pointing fingers at redhat here.  Bug reports came in of
gnumeric producing incorrect numbers.  One of the developers managed
to replicate it, but mysteryously it worked for the rest of us.  He
eventually tracked it to a miscompilation due to different
interpretation of double in the 2.96 compiler in his mandrake
distro.  Tracking why a number is off by a bit is a pain in the
butt.  A serious drain of resources.  This miscompilation was
closely followed by a another, this time in the control-center,
which Jacob worked around.  This was not bizarre pseudo assembler
from the middle of the kernel it was 1 line of fairly simple code.
 	http://bugzilla.gnome.org/show_bug.cgi?id=77121

These sorts of issues take along time to track and I would rather
not.  If you don't mind getting the wrong number in the development
series of gnumeric remove the test.