Re: [PATCH 2/3 v3] Use an external program to implement fetching with curl

From: Johannes Schindelin <johannes.schindelin@gmx.de>
Date: Aug 5, 2009, 20:10 UTC
Message-ID: <alpine.DEB.1.00.0908052156360.8306@pacific.mpi-cbg.de>
In-Reply-To: <fabb9a1e0908051115u145b2b0bi2973bcf41d5ef68f@mail.gmail.com>

Hi,

On Wed, 5 Aug 2009, Sverre Rabbelier wrote:

Show 16 quoted lines

> On Wed, Aug 5, 2009 at 08:45, Daniel Barkalow<barkalow@iabervon.org> 
> wrote:
> > Someday, I want to have a diff output format that makes these things 
> > clear somehow. I think it would be not-too-hard to get the diff code 
> > to determine that an addition matches or almost matches a deletion (or 
> > some unchanged code), and provide library access to this information, 
> > but representing it to humans (and getting patch to still work) is 
> > hard.
> 
> I started on this a while ago (as part of some post-GSoC git-stats 
> work), but I had a hard time finding a good rule to determine whether an 
> added hunk is similar enough to a deleted hunk elsewhere. Perhaps a 
> variant of Levenshtein can be used to determine how different two hunks 
> are; I tried diffing the two hunks and then looking at the ration 
> between the size of the diff and the size of the original hunk, but as 
> said that didn't really work out.

I think that there are two complications:

- how to present it in a format that helps the human to understand, yet is 
  well-defined enough to be used as a machine-readable edit script (I 
  really do not think that it can be well-designed if it only fulfills one 
  of the two purposes).  I could imagine something like this:

	diff --git a/remote-curl.c b/remote-curl.c
	partial copy from transport.c +394,130
		#ifndef NO_CURL
		static int curl_transport_push(struct transport *transport, int refspec_nr, const char **refspec, int flags)
		{
		[...]
		}

		#endif
	@@ -1,4 +1,6
	+#include "cache.h"
	+#include "transport.h"
	+#include "refs.h"
	+
	 #ifndef NO_CURL
	 static int curl_transport_push(struct transport *transport, int refspec_nr, const char **refspec, int flags)

- how to determine efficiently where to spend a lot of time to determine 
  what could be similar enough.

For example, the diff would still look pretty unreadable if you determined that there was a code move which involved a reindentation, so I am not at all sure if it is worth to try hard to detect that that was a move after all.

Ciao, Dscho