Re: [patch 0/4] mm: de-skew page refcount

[Nick Piggin <npiggin@suse.de>]

On Thu, Jan 19, 2006 at 09:27:14AM -0800, Linus Torvalds wrote:
> 
> 
> On Thu, 19 Jan 2006, Nick Piggin wrote:
> > 
> > Hmm... this is what the de-skew patch _did_ (although it was wrapped
> > in a function called get_page_unless_zero), in fact the main aim was
> > to prevent this twiddling and the de-skewing was just a nice side effect
> > (I guess the patch title is misleading).
> > 
> > So I'm confused...
> 
> The thing I minded was the _other_ changes, namely the de-skewing itself. 
> It seemed totally unnecessary to what you claimed was the point of the 
> patch.
> 
> So I objected to the patch on the grounds that it did what you claimed 
> badly. All the _optimization_ was totally independent of that de-skewing, 
> and the de-skewing was a potential un-optimization.
> 

No longer confused...

> But if you do the optimizations as one independent set of patches, and 
> _then_ do the counter thing as a "simplify logic" patch, I don't see that 
> as a problem.
> 
> Side note: I may be crazy, but for me when merging, one of the biggest 
> things is "does this pass my 'makes sense' detector". I look less at the 
> end result, than I actually look at the _change_. See?
> 
> That's why two separate patches that do the same thing as one combined 
> patch may make sense, even if the _combined_ one does not (it could go the 
> other way too, obviously).
> 

I agree, and the patches really are cleaner this way too, so again,
thanks for the input on them.

I'll resend soonish (with a trimmed cc list).

Nick