Re: Release update: base and standard frozen
Matthew Palmer <mpalmer@debian.org> writes:
> On Mon, Aug 09, 2004 at 02:04:42AM +0200, Adeodato Sim? wrote:
>> * Matthew Palmer [Mon, 09 Aug 2004 09:51:52 +1000]:
>>
>> > There'd hopefully be something better than that available, because
>> > disallowing uploads to sid just seems Wrong. I only know what it looks like
>> > from the developer's end, but a testing migration test that produced
>> > something in the excuses like "Uploaded after 2004-08-15: not a candidate
>> > for testing migration" or something of that sort.
>>
>> $ grep-excuses gnutls11 | grep freeze
>> Package is in freeze; use testing-proposed-updates for changes
>
> Well that looks interesting; I wonder what it's basis for making those
> decisions is, though. If it's just a list of "these packages won't be
> transitioned", that (a) won't really work for a whole-archive freeze, and
> (b) doesn't solve the problem of buildd backlog.
Two things would be nice:
1. testing candidate is every package which source was uploaded X day
before the freeze (X being the sarge delay of the urgency). If buildds
take long to build a package it could enter sarge way past the freeze
time.
2. wanna-builds sorts sources that are testing candidates before
packages that are not. This is to prevent uploads post freeze to
starve sources uploaded before freeze.
A list of testing candidates should be posted somewhere with the
wanna-build state for each arch to show where special attention is
required to get the freeze hard frozen. Scripts to collect the w-b
info per package are available already.
> What I'm suggesting is a decision on testing transition that is based on the
> date that the version of the package entered unstable. While that won't
> necessarily solve all our problems, it will at least solve the problem of
> buildd backlog possibly causing packages to miss the freeze (although that's
> arguably the maintainer's fault for not uploading early enough). See
> libapache-mod-auth-mysql on ARM for reasons why it's not always the
> maintainer's fault, though...
The wanna-build queue is not a fifo. In fact under load it quite often
starves packages from being build for a long time (month). You can't
realy blame a maintainer for uploading a package only 2 month before a
freeze.
> - Matt
MfG
Goswin
Reply to: