Subject: Re: use .d for kernel
To: Hubert Feyrer <feyrer@cs.stevens.edu>
From: Jonathan Stone <jonathan@Pescadero.dsg.stanford.edu>
List: tech-kern
Date: 11/07/2005 21:45:46
overflowtext="",overflowoffset=0

In message <Pine.LNX.4.61.0511080519010.21188@m24s24.vlinux.de>Hubert Feyrer writes
>
>On Tue, 8 Nov 2005, YAMAMOTO Takashi wrote:
>> the attached patch is to use .d files for kernel "make depend".
>> can anyone familiar with make review?
>
>Can you please explain how this is different from how it's done today, and 
>what the benefits of the new or drawbacks of the old version are?

Hmm. In addition to circa 1100 .o files and 400-odd other files
(mostly config-generated .h files for options), a kernel build
directory will, with the patch, have another 1100-odd .d files with
source dependencies.

To me, that sounds like one significant drawback with the "new"
approach.

But then I've been wondering for some time wondering what it'd take to
move the config-generated headers into a subdirectory, say ./opt.
(OTOH, the thrash of changing all those #include "foo_opt.h"
directives to #include "opt/foo_opt.h", for 400 or 500 distinct option
headers, and the consequent pain in backporting any overlapping
changes, strikes me as a very good argument to the contrary).