Re: [PATCH] do not overwrite untracked symlinks

From: Clemens Buchacher <drizzd@aon.at>
Date: Feb 22, 2011, 19:26 UTC
Message-ID: <20110222192632.GB4881@localhost>
In-Reply-To: <7v62scftjl.fsf@alter.siamese.dyndns.org>

On Mon, Feb 21, 2011 at 10:54:38PM -0800, Junio C Hamano wrote:

Show 6 quoted lines

> 
> Yes, I agree with that reading.  I'd squash in this on top, first making
> sure that "do not overwrite untracked symlinks" (which is the title of
> this patch) won't be broken, and then the original test that wanted to
> make sure that checking out a/b/c/d when a/b is pointing to an unrelated
> part of the tree does not nuke whatever is pointed at that symbolic link.

Ok, perfect.

> My unhappiness with the undescriptive "check_leading_path()" still
> remains, though...

Ok. If I can't figure anything out this weekend, I'll rename it. Promise.

It's just that I'm even more unhappy about the actual code rather than the function name. And each time I look at it I end up trying to fix it, only to find out that it affects so many parts of the code that don't make any sense.

For example, we already have a function check_path() in entry.c, which does something very similar and is called by checkout_entry(), which in turn does something similar to verify_absent(). In fact, I suspect that both functions are called twice by the same code path around unpack_trees, for the same paths and for the same reason.

I strongly feel that we should separate the merge process into two steps.

 - First, do everything in the index, ignoring the work tree.

 - Second, checkout the index to the work tree while making sure no
   changes or untracked files get overwritten.

I know I have talked about this before, and I don't know if I will ever find the time to implement such a major change. But I can't get myself to feel good about fixing the function name, but not the function.

Clemens