Re: testing packages at build

[Steve Greenland <steveg@moregruel.net>]

On 09-Oct-03, 13:00 (CDT), Branden Robinson <branden@debian.org> wrote: 
> On Thu, Oct 09, 2003 at 08:24:43AM -0500, Steve Greenland wrote:
> > No. While they certainly do exist, >99% of the time, if code works at
> > -O0 but not at -O2, then the code is broken. 
>
> I find this difficult to swallow given my own experiences with XFree86
> when it first met GCC 3.3.

Well, that's a pretty extreme case - an ancient, giant codebase and new
compiler release with new optimization features. And if XFree86 really
still requires --traditional with the pre-processor, it's likely to
violate (or perhaps more accurately, exceed) the standard in a variety
of ways. (But maybe not, I've not really looked at it...)

And in any case, do you think that it would be a good idea to just
automatically recompile Xfree86 with '-O0' if the binaries failed tests?
(Rhetorical question, I presume.)

Over many years, with a large variety of languages, OSs, and compilers,
I found that most "optimization bugs" turned out to be coding errors[1].
GCC may, in fact, be more likely to have optimization bugs than, say,
the old DEC Fortran compiler. But when something breaks when the
optimizer is turned on, I look at the code first: it's certainly the way
to bet.

Steve

[1] Well, there were certainly a few compilers whose optimizer *was*
broken, in which case we soon learned to not turn it on except for
amusement value.

-- 
Steve Greenland
    The irony is that Bill Gates claims to be making a stable operating
    system and Linus Torvalds claims to be trying to take over the
    world.       -- seen on the net