Re: [RFC] Build in clone

From: Daniel Barkalow <barkalow@iabervon.org>
Date: Feb 26, 2008, 17:36 UTC
Message-ID: <alpine.LNX.1.00.0802261128360.19024@iabervon.org>
In-Reply-To: <200802260321.14038.johan@herland.net>

On Tue, 26 Feb 2008, Johan Herland wrote:

Show 5 quoted lines

> Other than the failing tests, it seems to work fairly well. I've been
> playing around with it for a few minutes, and on a test repo I have with
> 1001 branches and 10000 tags, it cuts down the runtime of a local git-clone
> from 25 seconds to ~1.5 seconds. (simply by eliminating the overhead of
> invoking git-update-ref for every single ref) :)

Good to hear. A certain amount of the point is performance, and I've only got relatively simple repositories on Linux to test with, where everything is too fast to tell anyway.

Show 5 quoted lines

> I've tried to test this by diffing a cloned repo against an equivalent
> clone done by the old script. Below I pasted in a few immediate fixes I
> found. With these fixes, the only remaining diff between the clones is
> that refs/remotes/origin/HEAD used to be a symbolic ref (with no reflog),
> but is now a "regular" ref (with reflog).

I think that's just a call to the wrong function (and a lack of very very explicit documentation).

> The fixes are, in order of importance:
> - Call git_config(git_default_config) in order to properly set up
>   user.name and user.email for reflogs (This BREAKS test #9 in
>   t1020-subdirectory.sh. Have yet to figure out why)

I should have read email last night; I could have identified a bunch of the odd errors for you, but you've figured most of them out by now.

I need to look into the config system further; things should be configured such that the local config is in the new directory and the global config is unchanged. If no environment variable is set and pwd is a certain way, git_config_set() will write to the wrong file.

> - Fix "clone from $repo" reflog messages (using strbufs; something tells
>   me more of this code would benefit from using strbufs)

Most likely. I think Kristian wrote most of this before strbuf existed or something of the sort.

> - Høgsberg's name should be in UTF-8 (not sure if this will survive this
>   mail)

It is for me; I bet it didn't survive the mail from me to you. (That is, your mailer got my UTF-8 message and converted it to Latin-1 on export, and so applying it gave a Latin-1 ø in your tree, which you changed to UTF-8, and then your mailer again thought your patch was changing one Latin-1 character to two, and sent the patch like that.)

> - The two whitespace errors you mentioned
> 
> I'm sorry that my patch below sucks from a style POV. Feel free to ignore.
> Will redo when it's not in the middle of the night.

I haven't gotten to the level of style in this code myself, so no worries there. :) And whitespace noise is welcome at this point, since I'm still incorporating changes and evaluating the resulting state, instead of evaluating the changes themselves.

	-Daniel
*This .sig left intentionally blank*