Ben Reser ([EMAIL PROTECTED]) wrote: > On Thu, Nov 14, 2002 at 11:23:00AM +0100, Luca Olivetti wrote: > > > I don't agree: I suppose that mandrakesoft has an automated build > > procedure, a warning will probably go unnoticed, while a package > > that fails to build will surely get a maintainer attention. The > > result will be many less broken packages (I hope). > > The developer should be doing a build on their own machine *BEFORE* > they upload it to any system that is going to do an automated build. > And ports should only be building packages that have already been > built and tested on i586.
OTOH I have seen quite a few RPMS that were build only on the packagers machine. That person rpm now depends on libraries that exist on his machine the don't exist in mdk-cooker. ie Don't upload binaries built on your machine and use the mdk build-hosts to make them. > Unfortunately, so many times people upload packages without bother to > even test them. Yes mistakes will get made. But I have to wonder how > packages with syntax errors in the perl scripts get uploaded.... Only > thing I can see is that someone didn't bother to test the package > first. Most Perl makefiles support the test target. When I make a Perl-SRPM I turn on that test target just to make sure everything is OK. > Ultimately this small change will not solve the problem of untested > RPMS that are screwy. It will still be a problem, until developers > take responsibility for their packages. Difficult situation. But indeed it is better to wait at least to the next day when you release a package. Don't rush it out. After a few releases I learned to recognize the feel of the rush and that I had to sit on my hands and wait until the next morning after the coffee to double-check everything. I also make packages for OpenBSD and the knowledge of that build-system is very useful for making RPMS, vice versa as well btw. //Han -- http://www.xs4all.nl/~hanb/software