git tree browsing in redmine

Hi,

I'm trying to find some ideas for a better way to implement browsing of a git repository within the redmine application.

My top level directory in my git repository has 300 files. Redmine wants to display each file name, its most recent revision, and the comment for that revision. Currently it does a git log -1 on each file to find the latest revision. Is there a quicker way of doing this for a whole directory? git log runs from .2 seconds to 3 seconds depending on the file on my server.

Here is the open issue at redmine.org which has more details: http://www.redmine.org/issues/show/1435

A fix for this would help a whole bunch of redmine / git users including myself.

Burt Culver wrote:

Not yet, but if you clone libgit2 and start helping out on that, things like this will become fairly trivial to do in a single pass instead of, as redmine does it today, using one call to "git log" for every file in the directory (which is just insane).

You should be aware that git is fully snapshot based though, while svn is file-revision based (a bunch of file-revisions make up a "changeset"), so it's fundamentally easy to get the file-revisions for a particular changeset in svn and fundamentally hard to get it in git.

That's because the command used by redmine ("git log <cruft> -1 -- $path") stops as soon as it finds the first commit that touches the path. Some paths haven't been touched in a long time, so more revisions have to be loaded, parsed, analyzed and discarded before the correct one is found.

I have no idea what redmine is, but one solution for them would be to run "git log --pretty=fuller -p $branch" and then parse the diffs to get the latest commit that touched all files (in one go). This would be slightly slower for small repos (where it shouldn't matter anyway), but time should increase linearly with the number of commits in the repository. It's almost certainly required to be able to turn this feature off, btw, as it won't work at all for super-huge repositories (think kde or OO.org).

You should also make sure to have your repository properly packed. Run "git gc" on it every once in a while.

"Burt Culver" <burt@fishpond.co.nz> writes:

First, while it might be common and cheap view (information) for Subversion, it is not the case for Git. Git is whole project snapshot based, not per-file snapshot based, or patch based. Take a look at one of git web interfaces: gitweb, or cgit, or ViewGit to see how repository browsing is implemented there: it is log-like view, then tree, not tree, then history of files. Note also that for example history of two files is more than union of histories of individual files (history simplification, including merge simplification).

Second, I tried to implement such view (which I named 'tree blame' view, because it looks a bit like blame/annotate, but for directories/trees) for gitweb. You can view my attempts in 'gitweb/tree_blame' branch of my git fork:

  http://repo.or.cz/w/git/jnareb-git.git?a=shortlog;h=refs/heads/gitweb/tree_blame

But I think the correct solution, if you decide that you absolutely need so unnatural (and I think not that useful) view, would be to implement such view in git core, allowing 'git blame' to work for trees (directories). You would probably need to implement also some kind of '--porcelain' output for script/Redmine.

Or as suggested add your work to libgit2 library, and make use of it in Redmine...