Re: [PATCH 2/2] Add keyword unexpansion support to convert.c

From: Rogan Dawes <lists@dawes.za.net>
Date: Apr 17, 2007, 20:27 UTC
Message-ID: <46252DAE.4020604@dawes.za.net>
In-Reply-To: <Pine.LNX.4.64.0704171121090.5473@woody.linux-foundation.org>

Linus Torvalds wrote:

Show 15 quoted lines

> 
> On Tue, 17 Apr 2007, Rogan Dawes wrote:
>> Well, if the only keyword we support is $BlobId:$, then if the tree/object
>> hasn't changed, then we still don't need to touch the object.
>>
>> Not so?
> 
> Correct. However, is that actually a useful expansion?
> 
> Most of the time, I'd expect people to want things like "last committer, 
> time, story of their life" etc.. I don't think the SHA1 ID's are pretty 
> enough that anybody would ever want to see them. But yes, they are 
> certainly stable.
> 
> 			Linus

Well, one example for wanting a keyword expansion option was where people modify the entire file, and just email it back to the maintainer. It surely helps to have the SHA1 of the original object when applying the changes.

You also stated in another email that doing keyword expansion prevents people from using non-git tools. I agree that you'd probably end up with diffs that may include the keyword (object id) being mailed to you if the submitter is not using git. But when a git maintainer applies those diffs using git-apply, the keyword unexpansion could still take place, making the diffs usable in practice.

None of what I said necessarily supports the view that it is a good idea from the perspective of trusting the results, of course.

Regards,

Rogan