On 05/09/2008, Joachim Breitner <[EMAIL PROTECTED]> wrote: > Hi, > > Am Freitag, den 05.09.2008, 12:33 +0200 schrieb Michal Suchanek: > > > > When you upload a chain of depending packages (e.g. xmonad, > > > xmonad-contrib), using the delayed queue with a delay of two days for > > > the second should be fine. This is not really elegant, but somewhat > > > nicer than versioned Build-Depends IMHO. > > > > Why would there need to be a delay? How does this scale for longer > > dependency chains? > > > It scales easily to chains up to 8 packages, as there is a delayed queue > for 0 to 15 days :-) > > > > Are the buildds really that braindead that they cannot order the > > packages either by upload order or by dependency order? > > > It's not braindead: For most packages, e.g. C packages, it does not > matter if a depending version is not the very latest, as the ABI does > not change as easily as it does for Haskell libraries. >
I suspect that build ordering is not needed for ghc only but in other cases it would be probably exceptional case rather than the standard requirement. > > > Even if they are, and build the packages in the wrong order they > > should eventually crawl through the lot, right? > > > They will create a package that was built against a previous ABI. This > will be uninstallable once the new version of the dependency is built, > thus requiring a new built of the depending library. > > This is why we currently have strict source dependencies, and why I'm > advocating the use of >= source dependencies. Actually, the issue of > Source Dependencies (==, >=, or no versions) and binary Dependencies (== > or iface hashes) is relatively independent. > With >= dependencies that force building against the latest packages backporting might be difficult. But then maybe just backporting the whole ghc suite as a whole might be needed anyway if such thing is ever desired. Ideally buildd would do the right thing and build dependencies would be real library version requirements but apparently it is not the way it is. Whatever works for the package maintainers should be OK. Thanks Michal _______________________________________________ debian-haskell mailing list [email protected] http://urchin.earth.li/mailman/listinfo/debian-haskell
