On Fri, Apr 26, 2002 at 12:17:44PM -0400, Matt Zimmerman wrote: > On Fri, Apr 26, 2002 at 09:07:01AM -0500, Steve Langasek wrote: > > On Fri, Apr 26, 2002 at 09:48:29AM -0400, Matt Zimmerman wrote: > > > A testing + testing-updates system would give the worst of testing and > > > unstable: possibly broken packages that lag behind unstable. > > All software is possibly broken; whether we call it stable or testing or > > unstable doesn't matter, except to adjust the odds a little bit. > But those odds are exactly what matter. We want to minimize the chance of > broken software being released. Providing a backdoor into the release > without the sanity-checking of unstable seems like more work for the release > manager and more potential for bad bugs in the release. > > If we can't trust maintainers to use enough care when uploading to > > woody-proposed-updates that the odds still come out in our favor, then > > you're right that testing is a bad idea. I don't think it's a foregone > > conclusion that maintainers can't be trusted to use woody-proposed-updates > > properly. I suppose we'll never know for sure if no one ever implements > > it, though. > Maintainers are a motley group. Some could be trusted to make minimal > changes, and test them thoroughly, before making an upload to > testing-proposed-updates, and others could not. All of the updates would > certainly need to be reviewed by hand if they were allowed. What about a woody(testing)-proposed-updates with entrance guidelines only slightly more lax than potato(stable)-proposed-updates? Uploads to testing could be allowed in with only cursory examination if they 1) fix RC bugs, and 2) contain few or no additional changes. Granted, that does involve more work on the release manager's part; but dammit, that should be something the RM can delegate if he doesn't have time for it himself. I can't believe that the implementation of testing was unilateral on aj's part, or that there's no one else both willing and able to help him with such duties. If there are really developers who think testing was a good idea, then /some/ compromise should be possible here. Steve Langasek postmodern programmer
Attachment:
pgpyhEvTonK4g.pgp
Description: PGP signature