Re: git diff -B: splitting up complete renames

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Jan 5, 2008, 17:37 UTC
Message-ID: <alpine.LFD.1.00.0801050926180.2811@woody.linux-foundation.org>
In-Reply-To: <200801051818.40009.jnareb@gmail.com>

On Sat, 5 Jan 2008, Jakub Narebski wrote:

Show 8 quoted lines

> 
> Shouldn't the last block read for the modern git read:
> 
>   ------------------------------------------------
>   :100644 000000 bcd1234... 0123456... M99 file0
>   ------------------------------------------------
> 
> or is it only for '-B' togehter with '-M' or '-C'?

Without -M/C, git re-joins them, so yeah, either the documentation (or that re-joining logic) is wrong.

Show 5 quoted lines

> Second: do you have per chance good examples of broken (and perhaps 
> merged together) pair, such that it is affected by --diff-filter=B;
> with -B<num> or -B<num1>/<num2> parameters needed, if required. COPYING 
> and Makefile are one good example, but do you have literal files with 
> similar behavior?

Hmm. A kernel example that shows *something* is

	git show -B --raw -r acbbbe9f5ab52da90a8edec02ec973e7f44dae81

which shows

	:100644 100644 eb14b18... 1f2d6d5... M093       include/asm-x86/byteorder.h

even though that 093 is the *dissimilarity* index, so according to the documentation you quote you'd expect it to be

	:100644 000000 eb14b18... 0000000... D	include/asm-x86/byteorder.h
	:000000 100644 0000000... 1f2d6d5... A	include/asm-x86/byteorder.h

instead of a M. Whether it's the docs or the behavior that should be changed, I dunno.

Especially as using -B -M makes it all sane again, because then it shows it as the real rename:

	:100644 100644 435074a... 26b0dcd... M  include/asm-x86/Kbuild
	:100644 100644 a45470a... 1f2d6d5... R061       include/asm-x86/byteorder_32.h  include/asm-x86/byteorde
	:100644 000000 5e86c86... 0000000... D  include/asm-x86/byteorder_64.h

and it's all good.

My personal opinion is that -B doesn't make much sense without M (or C), but at least you can see an example of the breaking behaviour in the kernel for a real case.

			Linus