Re: [GIT] Security subsystem changes for 2.6.33

[Eric Paris <eparis@redhat.com>]

On Mon, 2010-02-15 at 14:46 +1000, Dave Airlie wrote: 
> > Eric Paris (5):
> 
> >      SELinux: add .gitignore files for dynamic classes
> 
> This isn't such a good idea I've discovered after wasting half my day
> wondering why 2.6.33-rc8 oops in selinux
> code when vbetool runs.
> 
> The problem (not sure what we or git can do, hence Linus cc'ed) is:
> 
> a) git tree contains 2.6.32 + with old non-generated versions of
> flash.h and av_permissions.h
> b) git pull 2.6.33-rc8, nothing removes these old files, git status
> doesn't show them as they are now in .gitignore
> c) build 2.6.33, it picks up the include files from the old location
> not from the generated ones in the new tree.
> d) bad juju.
> 
> I'm also doing out of tree builds btw.

I'm not sure I understand why the old ones didn't get cleaned up.  They
did for me.

$ git checkout -b testing v2.6.32
$ find security/selinux/ -name flask\*
security/selinux/include/flask.h
[here I made a commit on MAINTAINERS just to make sure pull did
something]
$ git pull linus master
$ find security/selinux/ -name flask\*
[nothing]
$

I don't have those two old header files laying around.  I understand why
your old ones are still there.

I build selinux which generates the new files (in slightly new places)

$ make security/selinux/
$ find security/selinux/ -name flask\*
security/selinux/flask.h

They moved from
security/selinux/include
  to
security/selinux

I have a security/selinux/.gitignore which lists

$ cat security/selinux/.gitignore 
av_permissions.h
flask.h

Which I believe means they match the old locations as well.  Is there
some better way to ignore the new autogenereated files such that they
won't match include/flask.h?  I don't understand why the old files
weren't cleaned up?  Am I doing something wrong in my testing?  Should I
be doing something better in the code?

-Eric