Re: Revisiting large binary files issue.

From: Linus Torvalds <torvalds@osdl.org>
Date: Jul 11, 2006, 17:09 UTC
Message-ID: <Pine.LNX.4.64.0607111004360.5623@g5.osdl.org>
In-Reply-To: <20060711145527.GA32468@hpsvcnb.fc.hp.com>

On Tue, 11 Jul 2006, Carl Baldwin wrote:

Show 5 quoted lines

>
> I'd like to get my hands dirty and see for myself where the issue lies.
> I hope to have some time next week to devote to this.  Is it reasonable
> to hope for a solution that is at least a lot lighter weight than the
> current status quo?

Ok, I decided to see how nasty an object database change would be.

It doesn't look too bad. The following three patches implement what looks like a workable model.

NOTE NOTE NOTE! It makes the new "binary headers" the default, and that will mean that unless you have applied the first two patches, any repository that has had objects added with the new git version WILL NOT BE READABLE BY AN OLDER GIT VERSION!

So I think that the first two patches can be added to the main git branch pretty much immediately (after people have tested this all a _bit_ more, of course), because the first two patches just add the capability to _read_ a mixed-format repository.

The third patch is the one that actually starts _writing_ new-format repo. It should be applied with extreme care, although it can basically be de-fanged by setting the default initial value of the "use_binary_headers" to 0, which will make it not write the binary headers by default.

I have _not_ verified that the actual object format is identical to the pack-file one, but it should be. It's simple enough.

The three patches will be sent as replies to this email.

		Linus