git commit $some_unlocked_file seems like a reasonably common thing for someone to do, so it's surprising to find that it's a ?little bit broken, leaving the file staged in the index after (correctly) committing the annexed symlink.

This is caused by either a bug in git and/or by git-annex abusing the git post-commit hook to do something it shouldn't do, although it's not unique in using the post-commit hook this way. I'm talking this over with Junio, and the fix will depend on the result of that conversation. It might involve git-annex detecting this case and canceling the commit, asking the user to git annex add the file first. Or it might involve a new git hook, although I have not had good luck getting hooks added to git before.

---

Meanwhile, today I did some other bug fixing. Fixed the Internet Archive support for embedcreds=yes. Made git annex map work for remote repos in a directory with an implicit ".git" prefix. And fixed a strange problem where the repository repair code caused a git gc to run and then tripped over its pid file.

I seem to have enough fixes to make another release pretty soon. Especially since the current release of git-annex doesn't build with yesod 1.4.

Backlog: 94 messages