On Sun, 02 Mar 2014, David Bremner <david@tethera.net> wrote: > Jani Nikula <jani@nikula.org> writes: >> >> I've dropped most of the content in patches 7 and 10 due to their >> size. The patches (and the whole series) are available in the >> boolean-folder-and-path-v2 branch at >> git://gitorious.org/jani/notmuch.git. Web interface at >> https://gitorious.org/jani/notmuch/commits/0b3dd2d1cc6c413ea07ea326883ac448499c0e79. >> > > Hi Jani; > > I put a some proposed changes at > > http://pivot.cs.unb.ca/git?p=notmuch.git;a=shortlog;h=refs/heads/boolean-folder-and-path-v2 > > After playing with it a bit, I decided to > > - leave the source for the corpuses in the main tree. As we > discussed, we don't expect them to change much, and they are > text files. > > - store the binary database blobs externally. We could make the > same argument about infrequent changes here, except that even > the tiniest change will in practice require a whole new copy be > stored. > > - compute a checksum file > - test: add machinery to fetch and publish test databases > - test: commit database-v1.tar.xz checksum, ignore actual database > > - this is an independant bug fix that should probably be pushed anyway: > > - test: don't use $(dir) in recipes. > > - I also made some unrelated changes including switching to xz > compression. I intentionally didn't squash my patches into yours > so you see and decide > > - convert to XZ compression > - convert test to XZ > > Originally I thought metadata in the database tarball is > needed. It turns out not. I think it's harmless, but feel free > to drop it. > > - add some metadata to the database tarball I'm fine with all of the above. > - I struggled to find a way that would allow people to manage the > tarballs in a nice way while not requiring everyone to > install/learn new tools. My conclusion is in > > - test: add support for git annex managing test databases > > We still need to think about what to do for people installing from > tarballs. As it stands they will need to either have network access or > skip the relevant tests. Probably at least one followup patch is needed > here. The alternative would be to include the database tarballs in the > source; this will pretty quickly double the size of the source tarball. I think we should use the prereq mechanism in the tests to check if a required test database is in place, skipping the test if not. This should be fairly straightforward to do, and does not care *how* the test database is to be fetched. (I'm undecided whether the test should verify the checksum, or the part that fetches the database. Minor detail.) I know git annex in principle, but don't have any practical experience with it. I don't really have any ideas for that. BR, Jani.