Re: 2.6.23.git build error

[Sam Ravnborg <sam@ravnborg.org>]

On Sat, Oct 13, 2007 at 09:35:09AM +0200, Mike Galbraith wrote:
> On Sat, 2007-10-13 at 09:16 +0200, Sam Ravnborg wrote:
> 
> > Pulled 2.6.23 tree and did a build
> > Added latest -linus on top of it.
> > Did a new build (x86_64).
> > 
> > 
> > As expected it failed due to wrong asm symlink but nothing
> > like the error Mike posted.
> > So I'm a bit puzzeled here.
> > 
> > I anyone can explain how to reproduce it I will chase it
> > but otherwise it will be left as a "consequence of
> > the x86 merge".
> 
> What I did was to pull, then diff against my working tree (/me=SCM
> weenie;) for review as usual, and then apply that diff to working tree
> and make oldconfig as usual.  That left me with the same failure this
> morning, despite the file being verified present in the git tree.  I
> then used git-archive --format=tar | (cd work_tree;tar -xvf -).  That
> spewed chunks when I tried to build it even though diff said the trees
> were identical.  After make mrproper and restoring my saved .config, all
> was peachy.
> 
> Methinks dontdiff bit me.

After a quick look:
[sam@neptun x86.git]$ find -name '*.i'
./include/asm-x86/alternative-asm_64.i
./include/asm-x86/alternative-asm_32.i
./include/asm-x86/frame.i
./include/asm-x86/alternative-asm.i
./include/asm-um/frame.i
./include/asm-um/alternative-asm.i
[sam@neptun x86.git]$ grep '*.i' Documentation/dontdiff 
*.i

But I cannot see how this could result in the bug you
reported..
Anyway at least one dontdiff buglet - or we should preferably
drop the .i files in the kernel src.

	Sam