Typically git-annex merge is fast, but it could still be sped up.

git-annex merge runs git-hash-object once per file that needs to be merged. Elsewhere in git-annex, git-hash-object is used in a faster mode, reading files from disk via --stdin-paths. But here, the data is not in raw files on disk, and I doubt writing them is the best approach. Instead, I'd like a way to stream multiple objects into git using stdin. Sometime, should look at either extending git-hash-object to support that, or possibly look at using git-fast-import instead.

Well, I went with git hash-object --stdin-paths despite it needing to read from temp files. --Joey

git-annex merge also runs git show once per file that needs to be merged. This could be reduced to a single call to git-cat-file --batch, There is already a Git.CatFile library that can do this easily. --Joey

This is now done, part above remains todo. --Joey

Merging used to use memory proportional to the size of the diff. It now streams data, running in constant space. This probably sped it up a lot, as there's much less allocation and GC action. --Joey