diff options
author | 2014-07-29 20:44:26 -0400 | |
---|---|---|
committer | 2014-07-29 20:44:26 -0400 | |
commit | ed1eed137515a3b478d3c83cb3b8548303b229fa (patch) | |
tree | f41ca4b06e31149ff572090f91b17b94253d9e1f | |
parent | 4333edf3fd5c5a255d01293f91b0d729a160bf34 (diff) |
devblog
-rw-r--r-- | doc/devblog/day_207__at_last.mdwn | 34 |
1 files changed, 34 insertions, 0 deletions
diff --git a/doc/devblog/day_207__at_last.mdwn b/doc/devblog/day_207__at_last.mdwn new file mode 100644 index 000000000..936ac98f0 --- /dev/null +++ b/doc/devblog/day_207__at_last.mdwn @@ -0,0 +1,34 @@ +It took 9 hours, but I finally got to make [[!commit c0dc134cded6078bb2e5fa2d4420b9cc09a292f7]], +which both removes 35 lines of code, and adds chunking support to all +external special remotes! + +The groundwork for that commit involved taking the type scheme I sketched +out yesterday, completely failing to make it work with such high-ranked +types, and falling back to a simpler set of types that both I and GHC seem +better at getting our heads around. + +Then I also had more fun with types, when it turned out I needed to +run encryption in the Annex monad. So I had to go convert several parts of +the utility libraries to use MonadIO and exception lifting. Yurk. + +The final and most fun stumbling block caused git-annex to crash when +retriving a file from an external special remote that had neither +encryption not chunking. Amusingly it was because I had not put in an +optimation (namely, just renaming the file that was retrieved in this case, +rather than unnecessarily reading it in and writing it back out). It's +not often that a lack of an optimisation causes code to crash! + +So, fun day, great result, and it should now be very simple to convert +the bup, ddar, gcrypt, glacier, rsync, S3, and WebDAV special remotes +to the new system. Fingers crossed. + +But first, I will probably take half a day or so and write a +`git annex testremote` that can be run in a repository and does live +testing of a special remote including uploading and downloading files. +There are quite a lot of cases to test now, and it seems best to get +that in place before I start changing a lot of remotes without a way to +test everything. + +---- + +Today's work was sponsored by Daniel Callahan. |