Re: [PATCH 0/2] Custom low-level merge driver support.

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Apr 18, 2007, 20:35 UTC
Message-ID: <alpine.LFD.0.98.0704181330050.2828@woody.linux-foundation.org>
In-Reply-To: <7vk5w9fnjx.fsf@assigned-by-dhcp.cox.net>

On Wed, 18 Apr 2007, Junio C Hamano wrote:

Show 9 quoted lines

> 
> Yes, but in the case I had trouble with, I know what I want to
> see happen when we DO hit file-level conflict on RelNotes
> symlink.  The ancestor says Documentation/RelNotes-1.5.0.txt,
> the maint branch says Documentation/RelNotes-1.5.0.1.txt, while
> the current branch says Documentation/RelNotes-1.5.1.txt.  The
> logic in merge_file() simply says "we would punt on file-level
> conflicts for symlinks" without giving a chance for low-level
> drivers to interfere.

Ahh, ok. So it really is a file-level conflict, it's just that our traditional merger didn't handle them at all, so we said "nobody can do it". Fair enough. That does sound like a misfeature, although I would also claim that expecting merge strategies to handle symlinks is likely to fail horribly.

So maybe each strategy could have "sub-strategies" for other file types.

Ie something like

	[merge "ours"]
		name = pick our own version
		driver = /bin/true
		symlinks = /bin/true

ie we'd use tyhe "driver" name for regular files, and the "symlinks" name for symlinks and if no "symlinks" entry exists, we error it out as a conflict?

		Linus