Ben Lippmeier wrote: > The patch history doesn't provide a working build at every point > anyway.
Not every point, but from what I've seen the build breakage is rather rare. > People push *bundles* of patches, and the build needs to work > after the bundle, but it doesn't need to work after every patch. No, but the bundles with a single git tree are usually pretty easy to detect by their date stamp in the git commit history. If you plot the commit date vs time of patches in a project with a single committer who only commits in the master branch you will get a function that increases monotonically. For ghc, with multiple commiters commiting bundles, the patches within a bundle have a monotonically increasing date stamp and the start of a bundle is marked by commit N+1 having a date earlier than commit N. Obviously this is somewhat simplistic and doesn't consider overlapping patch bundles, but to a certain extent the git history contains info to be able to guess which commits should build. The real difficulty with the ghc sources is that fact that over a dozen different trees, with independant commit histories are needed to retrieve a working revision. Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/ _______________________________________________ Cvs-ghc mailing list [email protected] http://www.haskell.org/mailman/listinfo/cvs-ghc
