To make git-annex faster when it's dealing with a lot of urls, I decided to make it use the http-conduit library for all url access by default. That way, http pipelining will speed up repeated requests to the same web servers. This is kind of a follow-up to the recent elimination of rsync.

Some users rely on some annex.web-options or a .netrc file to configure how git-annex downloads urls. To keep that supported, when annex.web-options is set, git-annex will use curl. To use a .netrc file, curl needs an option, so you would configure:

git config annex.web-options --netrc

I get the feeling that nobody has implemented resuming interrupted downloads of files using http-conduit before, because it was unexpectedly kind of hard and http-types lacks support for some of the necessary range-related HTTP stuff.

Today's work was supported by the NSF-funded DataLad project.

---

Stewart V. Wright announced recastex, a program that publishes podcasts and other files from by git-annex to your phone.