Re: GSoC 2010

From: Scott Chacon <schacon@gmail.com>
Date: Feb 12, 2010, 19:47 UTC
Message-ID: <d411cc4a1002121147i67317ec8k69c7274ba49d6e38@mail.gmail.com>
In-Reply-To: <fabb9a1e1002121130uf16efeeyea6f763be18d27d@mail.gmail.com>

Hey,

On Fri, Feb 12, 2010 at 11:30 AM, Sverre Rabbelier <srabbelier@gmail.com> wrote:

Show 11 quoted lines

> Heya,
>
> On Fri, Feb 12, 2010 at 20:16, Daniel Barkalow <barkalow@iabervon.org> wrote:
>> I think that 'export' will always have this effect, because the
>> export/reimport roundtrip is practically never going to be exact with a
>> remote system that isn't actually git (and the state that you see the
>> remote as having after a successful push should match what you see if you
>> were to clone again).
>
> Not so, currently round-trips to Mercurial using hg-git result in
> identical commits.

In most cases, this is true, but unfortunately not every VCS is as denormalized data-wise as Git is. Hg records rename info and a manifest of changed files in the changeset explicitly, whereas Git figures them out from the linked trees. This means that if some plugin does not write the file listing or rename info properly, it's difficult to recreate perfectly. It also doesn't help that the method that writes that changed-file list from contents in memory operates subtly differently than the one that writes it from files in the working directory. This caused days of headaches for me trying to keep bidirectional losslessness.

Possibly not terribly helpful information, but interesting to me nonetheless.

Scott