Andreas Ericsson <ae@xxxxxx> wrote: > Shawn O. Pearce wrote: > >By saving the packfile we are transferring over the network to a > >local file we can remove the corresponding loose objects from the > >objects directory and immediately benefit from the packing work > >that was done to perform the network transport. This is a form > >of `git gc --auto` that happens automatically anytime the user > >performs a push. > > I'm sure this *could* be a good idea, but in my typical workflow I > push about twice a day, and usually not more than 5-8 commits at a > time. Since I'm a great believer in isolated changes, this usually > contains changes to one or two files at a time. Is it worth the > trouble saving 15-25 loose objects, creating two new packfiles / day? Yes, I know. Its a tuning thing. In my normal work on git-gui I am usually pushing under 20 objects when I publish to repo.or.cz. Since I also do that about once a week I probably wouldn't mind letting these packs be saved. But for other users that might be so small its not worth it. In my day-job I'm usually pushing about 30-40 objects when I push. I also believe in small isolated changes but often will do a long string of those and only push to the public location once I am reasonably happy with how that string turned out. 40 objects is almost worth packing if we also have a cheap pack file merge available (Junio's patch). I work with about 30 others at day-job through Git. Some of them are able to remember to run `git gc` every once in a while. Others don't. Those who don't also have a workflow where they are making a single commit and then pushing that commit as soon as it is complete. Usually that commit affects about 4-6 files one tree down, so its a total of 9 new objects. Packing 9 objects is most likely not worth saving. Yet these are the very users who also cannot remember to run `git gc`. :-\ > Otoh, I also rebase every once in a while, moving 50+ commits to some > other branch, and then it would most definitely be worth it. Yeah, its really only worthwhile if you are moving a reasonable number of objects. Maybe it should be enabled only if it at least meets a minimum threshold of objects such as the fetch limit (transfer.unpackLimit / fetch.unpackLimit) and also if the majority of what it is sending is currently loose (my patch has it at 50%). > >+ save_pack = 0; > >+ else if (progress) > >+ fprintf(stderr, "Also keeping saving packfile...\n"); > > keeping or saving? Pick one :) It was late and a proof of concept. I never meant for it to be applied in the state it is in. :) -- Shawn. - To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
