Re: [PATCHv3 5/6] fast-export: Fix output order of D/F changes

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 7, 2010, 21:40 UTC
Message-ID: <7vy6dnrmvu.fsf@alter.siamese.dyndns.org>
In-Reply-To: <1278480034-22939-6-git-send-email-newren@gmail.com>

newren@gmail.com writes:

Show 11 quoted lines

> From: Elijah Newren <newren@gmail.com>
>
> The fast-import stream format requires incremental changes which take place
> immediately, meaning that for D->F conversions all files below the relevant
> directory must be deleted before the resulting file of the same name is
> created.  Reversing the order can result in fast-import silently deleting
> the file right after creating it, resulting in the file missing from the
> resulting repository.
>
> We correct the order by instructing the diff_tree machinery to compare
> entries using df_name_compare instead of base_name_compare.

I am not so sure about this one.

You can be walking two trees, one of which has "b-1" (blob), "b-2" (blob) and then "b" (tree), while the other one has "b" (blob), "b-1" (blob) and then "b-2" (blob). The patch tells the machinery that "b" (tree) sorts just like "b" (blob) only during comparison, but the actual data stream it is walking is sorted differently. Without some form of lookahead, can you reliably "correct the order"?