Using ggit_diff_blob_to_buffer(), the hunk callback always gets passed a pointer to the very first hunk in the diff, whatever the number of hunk may be for that diff.

This seems to be because libgit2-glib uses libgit2's hunk object pointer address as a key for its hunks cache hash table while libgit2 always passes the same pointer and simply memcopy() new hunk data in place at each run...

Most probably, libgit's hunk object pointer address should not be use as a key in that hash table. I can work on a fix, but I'd like to discuss what's best to use here as a unique key first.

The key should identify a particular hunk for a particular delta (ie. file). Thus it should be build with something unique from the delta and something unique from the hunk. For the hunk, the new_start (or old_start) integer should be unique within a given delta. The content string should also be. For the delta, the only thing that I can think of right now is the new_file (or old_file) path. Would using a combination of both of these be acceptable? Any better option?