At Tue, 27 Oct 2009 16:39:31 -0700, Phil Nelson wrote: > > On Tuesday 27 October 2009 4:11:55 pm Marko Schütz wrote: > > > We just about manage three builds a day on TNF hardware at present. > > > > Hmmm? Not more? That strikes me as very few. Are those `./build.sh > > release`s from a fresh checkout? > > Actually, that is quite good. The builds include a full ./build.sh for > every architecture on every build. So a build of the -current branch does > a total of 57 ./build.sh commands. So assuming a build of a 4- branch, a > 5- branch and -current, that would be around 166 ./build.sh builds run and > these are full release builds that build .iso images for many architectures. > That is a *lot* of work. Until some time in 2009, the TNF cluster was getting > less than one tag totally built. Improvements in the TNF cluster has achieved > this speed. > > > > We tend to have a bit more than three commits per day. (On a good day, by > > > a factor of 20). > > [..] > > > > You seem to imply that for every commit we'd need to do one of those > > (heavyweight) builds you mention above. Is this really necessary? I'm > > not convinced that it is. > > While it may not require a complete release build, to test it properly so > one could be sure it would build correctly as above, one would need to > test all architecture builds. This would imply doing builds of all the > needed cross-tools to test a small changes. Given that we do a complete > build of every currently active branch at least every other day and we > do a complete build of -current every day, I see no reason to have any > kind of incremental build added to the job mix. Is that setup using ccache? Best regards, Marko
Attachment:
pgp5nPF9JCHf5.pgp
Description: PGP signature