[17:37] <nacc> so I'm 99% sure this diff is very wrong: https://code.launchpad.net/~nacc/ubuntu/+source/apache2/+git/apache2/+merge/327021 ... I can clone those two repos down locally and diff the two refs and I get: http://paste.ubuntu.com/25040631/ . Is this an artifact of the object sharing?
[17:37] <nacc> cjwatson: --^?
[17:38] <cjwatson> Object sharing is an unlikely culprit
[17:38] <nacc> cjwatson: yeah it wasn't immediately clear why that would be, but it's the only thing i can think of :)
[17:38] <cjwatson> Exactly which diff syntax did you use?
[17:40] <cjwatson> The merge diff is actually done by "try a merge and then see what the diff in the merged index is"
[17:41] <nacc> cjwatson: locally? `git diff --stat pkg/debian/sid..nacc/lp1702582` where pkg is the ~usd-import-team remote and nacc is the ~nacc remote for the apache2 src packageds
[17:41] <nacc> ah
[17:41] <nacc> yeah we may be abusing the merge logical :)
[17:41] <nacc> *logic
[17:41] <nacc> cjwatson: it's odd, though, the act of this merge should be a FF
[17:41] <cjwatson> Since the whole point of it is to try to preview what would happen if you merged it
[17:41] <nacc> cjwatson: as debian/sid is an immediate ancestor of lp1702582 in my repo
[17:41] <cjwatson> If you've managed to confuse git's merge algorithms somehow ...
[17:41] <nacc> (the same hash)
[17:41] <nacc> cjwatson: let me see what happens locally
[17:42] <nacc> cjwatson: it's a FF locally
[17:45] <cjwatson> Yeah, I don't see anything obviously wrong and it's EOW for me.  You could dig into lp:turnip (turnip/api/store.py:get_merge_diff) to see if you see anything?
[17:45] <nacc> cjwatson: ack, thanks
[18:04] <nacc> cjwatson: interesting, pygit2 is generating that output, from the index. Debugging why (merge_base on the same commits shows that debian/sid is the merge-base, which implies a FF)
[18:04] <nacc> cjwatson: enjoy your w/e
[18:16] <nacc> rbasak: --^ this might be something for your list, as you come up to speed on launchpad
[18:16] <nacc> rbasak: i'm not sure why it worked in one case, but not in two