Seems that the recent release of git 2.13.0 contained a reversion that broke git-annex sync in an adjusted branch. After bisecting git, producing a minimal test case, and reporting that to the git developers, I was able to work around it in git-annex. The workaround is normally not expensive, but could be when a repository has thousands of unpacked refs. So I hope this will get fixed in git and I can remove the workaround.

I think I will hurry up the next git-annex release somewhat to get the workaround out. It's not a super bad bug, but it does make the test suite fail and I've already had 3 people report the problem.

Seems it would be good to have an integration test that runs git-annex's test suite against new commits to git. Ævar Arnfjörð Bjarmason has stepped up to add that to git's test suite.


Also, I dealt with some fallout from removing MissingH; a exponential speed blowup in a directory traversal function.

Getting back to the ssh password prompting with -J I was working on last week, dealt with the ssh prompt interfering with the regional display manager. The fix is not perfect, but good enough; before ssh prompts (and only if it prompts), git-annex temporarily clears the regional display. Then the display gets redrawn under the ssh output. That needed some changes to concurrent-output (which I did over the weekend), so will only be done when it's built with a new enough version. A better approach would be to save and restore the cursor position, but the ansi-terminal library does not yet support that.