This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCC 3.0 Release Criteria
> But you might face the same problem with any distribution:
> Distribution makers release updates and bugfixes and AFAIK Debian
> doesn't have a static distribution but still changes. Therefore if
> you say distribution X, you still have to say e.g: The original
> distribution without any updates and without any self compiled
> packages.
I think this is essentially the plan. The criteria document lists
"RedHat Linux 6.2" and "Debian 2.1", which, I believe, identifies a
specific collection of packages.
> And if you do this, you can say directly any linux system with glibc
> 2.1.3, Linux 2.2.14 headers etc.
No. From a portability point of view, yes, this is probably
sufficient. From a release criteria point of view, just listing those
two is not: For example, the exact version of make may result in
failures in one case, and passes in another (when building). Likewise,
the exact shell could make a difference - especially when distributors
chose not to use bash as /bin/sh. I'm sure that when I think longer, I
can find more things that might have impact.
Of course, in real life, none of these differences should matter, or
gcc would be a poor software. However, *as a release criterion*, they
do matter, because there is the potential risk that you cannot release
the software, if somebody finds a strange Linux installation where it
does not work properly, and nobody can reproduce it.
> Btw. for gcc 2.95 we just tested on bootstrapping on linux.
Yes. Release criteria tighten from release to release, as we gain
experience with that aspect of engineering as well. IMO, this is part
of the evolution from a "cathedral" project to a "bazaar" project:
More people get to know how the processes work, instead of just seeing
the outcome. I'm sure GCC releases always have been tested on a
specific distribution, at least since Linux is shipped in
distributions - it just was not documented what specific distributions
were used for testing.
Of course, you don't have to have all the distributions, and you can
still test GCC. Also, I'm pretty sure that a well-analyzed serious
failure of a release candidate on SuSE 7.9 would prevent gcc from
being released, even though that distribution is not listed in the
criteria document.
> P.S. Martin, my name is Jaeger - without an Umlaut.
Oops, sorry. I'm always assuming that people have the same problem as
I do with their names...
Regards,
Martin