2011/8/20 Frédéric Delanoy frederic.delanoy@gmail.com:
I thought the Date: field was set by the client, so server races shouldn't matter.
Sorry, missed that you use that and not the order you receive it.
That's not always reliable: say you commit locally patches [1-2/3] on day D and patch [3/3] on day D+1 P3 - baz - D+1 P2 - bar - D P1 - foo - D
For whatever reason, before submission, you decide patch [3/3] should be the first one, so you use "git rebase" do to that: P3 - bar - D P2 - foo - D P1 - baz - D+1
and you git-send the mails.
git-send's messages' timestamps use current time, not commit time: http://git.kernel.org/?p=git/git.git;a=blob;f=git-send-email.perl#l98 And it is careful to increment the time after each message! So I think my usage is compatible with git-send.
So, you should always use the numbering specified by the author IMHO
I wish it were so easy. It is very difficult to recognize patch series without relying on them being sent in order. I've tried: take all unprocessed messages divide them by sender further divide them into groups based on length of patch series in subject line sort each group by patch number from subject line but that breaks down if the developer sends two patch series with the same length, which happens very often, e.g. when somebody retries a patch series. Maybe one could further subdivide messages by retry number, but that sounds hard and fragile. - Dan