Re: Last mile to 1.0?

barkalow@iabervon.org· Jul 30, 2005, 05:40 UTC · lore

I've been missing for a couple of weeks due to my server dying and needing to be replaced; I think stuff is mostly back in order now. (But I'm not yet resubscribed, and I suspect only dumb mailers are currently willing to talk to my mail server, for some reason I don't yet understand; I'm sort of reading the archives, at least.)

I still have the patch to make git-http-pull download packs, and I should be able to get it to read the objects/info/packs file without too much trouble.

I was also in progress on a series to rearrange the code in pull.c such that it passes the objects it needs through a list such that it can request them in parallel. I'd gotten as far as making it blast out requests for objects over git-ssh-pull (and then break because it doesn't expect to be able to accidentally read part of the next object while reading the current object); it shouldn't be too hard to make the HTTP version use this to parallelize requests as well.

I expect I'll get to work on this on Sunday, in any case.

 	-Daniel
*This .sig left intentionally blank*

Junio C Hamano· Jul 30, 2005, 08:07 UTC · re: barkalow@iabervon.org · lore

barkalow@iabervon.org writes:

> I still have the patch to make git-http-pull download packs, and I
> should be able to get it to read the objects/info/packs file without
> too much trouble.

Another thing that may help you gain more parallelism in the initial set of requests is the rev-cache file. You can find out the ancestry information upfront by reading it, without waiting for the commits you asked to arrive.