Merge Review build step is redundant?
Jima
jima at beer.tclug.org
Sun Feb 11 17:49:14 UTC 2007
On Fri, 9 Feb 2007, Fernando Nasser wrote:
> When reviewing a new package one needs to assert if the submitted SRPM really
> builds, so there is a step for the reviwer that consists in actually trying
> to build the package.
>
> But for the packages of the "Merge Review" type, that are already built in
> Core, isn't this step unnecessary?
Yes and no. While Matt Domsch's builds are very helpful as a baseline,
I've found (accidentally) that they're not quite perfect.
First, he doesn't do ppc builds. I honestly haven't encountered any
ppc-specific bugs (I've only done four Core reviews, though), and most
reviewers aren't going to have ppc builders to test them against, so I'm
not sure this is a particularly serious lack. Oh well.
Second: Unless I'm mistaken, he only built published SRPMs. (Anyone is
welcome to correct me if I'm wrong.) While this may seem like nit-picking
(it does to me, and I know better), it matters if a maintainer made bad
adaptations to a package after the last release was published. I have no
idea how likely this is, but it did happen in one of my four reviews; I
discovered a poorly implemented patch that made an inadvertant change that
caused the build to fail. (Oops!) It was a completely understandable
mishap, but I'm not sure anyone had picked up on it yet. I only did
because I did a CVS checkout, ran `make srpm`, and threw that against my
buildsys. Lucky catch.
I don't think anyone is advocating doing rebuilds against the CVS copy;
the only reason I did is because I got tired of leaching a bunch of files
from Matt's stash. :-)
I do strongly suspect that for 90+% of cases, Matt's builds are probably
dead-on. As usual, he's doing very helpful work. Thanks, Matt!
Jima
More information about the Fedora-maintainers
mailing list