From: Junio C Hamano <junkio@cox.net>
Date: Thu, 08 Dec 2005 18:17:23 GMT
Subject: Re: Wine + GIT
Message-ID: <7v4q5jjvws.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20051208092945.GA4921@c165.ib.student.liu.se>

Fredrik Kuivinen <freku045@student.liu.se> writes:

> However, in some other cases such as rename/rename, file/directory and
> add/add conflicts "a~A" and "a~B" will be created in the working
> directory. The recursive strategy will (or, is supposed to) tell you
> about such things, though.

I think one reason for this was because read-tree -m was the
only way to create higher stage entries in the index, and after
detecting renames, the recursive merge strategy did not have a
good way to insert these as different stages in the index.

I extended update-index --index-info a little bit to let people
insert higher stage entries in the index, so we could if we
wanted to.

Here are the differences I know about between resolve and
recursive other than ~A ~B:

 - in add/add case, merge-one-file tries to be cuter and leaves
   conflict-marked merge result.

 - content conflict in a renamed path gets its index entry
   collapsed, unlike other content conflicts.