Re: Problems found by piuparts
pe, 2006-02-24 kello 18:27 -0300, Gustavo Franco kirjoitti:
> What i thought in a first look to the Lars' list. I think that the
> best thing would include piuparts as a infrastructural test (oficially
> as a part of our archive code), or due to restrict admin time to do
> that, opt for something like piuparts.debian.org as we have
Do you mean that the archive should automatically reject uploads that
don't pass a piuparts check? I think that sounds like a nice idea, but
we're not yet in a situation where it is feasible. For example, the
reason piuparts testing fails may be due to a bug in an unrelated
package. This needs to be dealt with manually, and that requires quite a
bit of effort. For the time being, at least, I think it is better for
Debian to let uploads run smoothly, and do piuparts testing separately.
If we are to start doing checks on packages before accepting uploads, I
think it would be best to start with some subset of lintian and linda
errors. Lintian and linda are faster to run, anyway, and less risky, and
less likely to hang.
> I would be glad to help with a web interface to show the piuparts html
> results in a organized way
Something like piuparts-report.py?
Anyway, I think it is more productive if package maintainers run
piuparts themselves, and then people running piuparts centrally
reporting bugs. The lintian.debian.org experience seems to be that
having lots of info on the web is nice, but filing bug reports actually
gets things fixed.
As it happens, however, there is work going on in getting a centralized
machine to run piuparts testing, and that machine will be used to
publish logs and statistics. I haven't done that so far because the logs
are big, and I don't have gigabytes of disk space and bandwidth to spend
> Since Lars already did the check (for i386 only, i think),
I haven't tested the entire i386, either, only about half or so. The
rest is waiting for their dependencies to be fixed, or for something to
happen with regards to circular dependencies (which at the moment are
likely to make piuparts testing fail).
Boilerplate programming mean tools lack power.