Re: [PATCH v2] VERIFY_OCTAL_PERMISSIONS: Move to <linux/sysfs.h> where it belongs

[Stephen Rothwell <sfr@canb.auug.org.au>]

[-- Attachment #1: Type: text/plain, Size: 2535 bytes --]

Hi George,

On 15 Dec 2014 19:14:53 -0500 "George Spelvin" <linux@horizon.com> wrote:
>
> Stephen Rothwell <sfr@canb.auug.org.au> wrote:
> > Please do *not* mix changes up like this.  Split this out into a
> > separate patch, please (1 logical change per patch).
> 
> Um... I thought I was doing that.  More particularly, the task of
> untangling header file dependencies eseemed sufficiently cohesive
> that it could be considered one logical change.

Well, given the subject of the commit, I expected a simple change that
just did the move (and any immediately associated include changes).
You then said "Some other extraneous header files pruned while I was at
it" and that part I would expect to be in a separate patch.

> It was one round of thinking and analysis about what declarations the
> affected files depend on.

Which is separate from what VERIFY_OCTAL_PERMISSIONS() requires.

> Although syntactically possible, given the small size of the change (I
> deleted a total of 5 #includes, 2 from moduleparam.h and 3 from sysfs.h),
> it didn't seem worth breaking it up further.
> 
> > And testing only
> > on x86_64 is not "sure" when talking about header file pruning (but at
> > least you did the "all" configs).
> 
> Well, the first round was reading and understanding the headers; the
> compile was just to make sure.

Understood, it was more a "actually changing architectures was more
likely to show breakage than building lots of stuff".  I guess I see
more breakage from pruning includes than most people since I build for
multiple architectures more than most.

> The files I was messing with (moduleparam.h and sysfs.h) don't have a
> lot of architecture-specificness within them.  The main danger is that
> they're used in a zillion places and some caller might depend on the
> over-inclusion.

The problem is not the direct includes and direct architecture
depenedencies, but the fact that lost of stuff ends up include asm/
include files at some point in the chain and that affects the set of
files implicitly included.  X86 seems to implicitly include more than
some other architectures.

> If my haste made me judge wrong, I apologize.  Was I very wrong, or
> just a bit over the line?

Probably just a bit over the line.  The advantage of the split would be
that when it hits Andrew's tree and then breaks linux-next (for me), I
can pick on a smaller patch to get rid of/correct.

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]