Re: [PATCH] Detect renames in diff family.

From: Linus Torvalds <torvalds@osdl.org>
Date: May 19, 2005, 16:19 UTC
Message-ID: <Pine.LNX.4.58.0505190901340.2322@ppc970.osdl.org>
In-Reply-To: <7vu0kz1p6k.fsf@assigned-by-dhcp.cox.net>

On Thu, 19 May 2005, Junio C Hamano wrote:

Show 6 quoted lines

> 
> Special request for Linus is to check if I did not screw up the
> various calls into the diff core from diff-tree.  Essentially
> the idea is to start one patchset session with diff_setup() and
> close it with diff_flush() before you start another patchset
> session.

It all looks ok from a quick setup, and with this I can now do

	git-whatchanged -M

in the kernel, and searching for renames I find:

	diff --git a/arch/um/kernel/sys_call_table.c b/arch/um/sys-x86_64/sys_call_table.c
	rename old arch/um/kernel/sys_call_table.c
	rename new arch/um/sys-x86_64/sys_call_table.c
	--- a/arch/um/kernel/sys_call_table.c
	+++ b/arch/um/sys-x86_64/sys_call_table.c
	@@ -1,4 +1,4 @@
	-/* 
	+/*
	  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
	  * Copyright 2003 PathScale, Inc.
	  * Licensed under the GPL
	@@ -14,6 +14,12 @@
	 #include "sysdep/syscalls.h"
	 #include "kern_util.h"
	 
	+#ifdef CONFIG_NFSD
	....

which looks quite correct.

I notice that you left some debugging output in there ("**score **" stuff), and I'll remove it, but it's merged and pushed out and passed my trivial tests.

[ rambling mode on: ]

One thing that struck me is that there is nothing wrong with having the same old file marked twice for a rename, or considering new files to be copies of old files. So if we ever allow that, then "rename" may be the wrong name for this, since the logic certainly allows the old file to still exist (or be removed and show up multiple times in a new guise).

In other words, let's say that we create a new architecture or a new filesystem, and we have tons of _new_ files, but not a lot of removed files. It would literally be very cool to see that the new files are based on contents of old files, and that it would thus potentially be very interesting to see a diff like

	diff --git a/arch/i386/kernel/irq.c b/arch/x86-64/kernel/irq.c
	based-on old arch/i386/kernel/irq.c
	creates new arch/x86-64/kernel/irq.c
	--- arch/i386/kernel/irq.c
	+++ arch/x86_64/kernel/irq.c
	@@ -1,205 +1,31 @@
	 /*
	- *     linux/arch/i386/kernel/irq.c
	+ *     linux/arch/x86_64/kernel/irq.c
	  *
	  *     Copyright (C) 1992, 1998 Linus Torvalds, Ingo Molnar
	  *
	...

(the above is a made-up example, but it's at least _half-way_ valid).

I'm not suggesting you actually do this, if only because it's quite expensive: it means that any newly added file would have to be compared with _all_ files in the previous archive, which is just too damn expensive. But I'd like people to kind of keep this in mind as a possibility, because maybe wasting CPU time in a big way might actually be acceptable in some cases, and having a separate flag to enable this kind of thing might be interesting, no?

		Linus