Re: [PATCH v5 03/78] xarray: Add the xa_lock to the radix_tree_root

[Matthew Wilcox]

On Tue, Jan 02, 2018 at 10:01:55AM -0800, Darrick J. Wong wrote:
> On Tue, Dec 26, 2017 at 07:58:15PM -0800, Matthew Wilcox wrote:
> >         spin_lock_irqsave(&mapping->pages, flags);
> >         __delete_from_page_cache(page, NULL);
> >         spin_unlock_irqrestore(&mapping->pages, flags);
> >
> > More details here: https://9p.io/sys/doc/compiler.html
> 
> I read the link, and I understand (from section 3.3) that replacing
> foo.bar.baz.goo with foo.goo is less typing, but otoh the first time I
> read your example above I thought "we're passing (an array of pages |
> something that doesn't have the word 'lock' in the name) to
> spin_lock_irqsave? wtf?"

I can see that being a bit jarring initially.  If you think about what
object-oriented languages were offering in the nineties, this is basically
C++ multiple-inheritance / Java interfaces.  So when I read the above
example, I think "lock the mapping pages, delete from page cache, unlock
the mapping pages", and I don't have a wtf moment.  It's just simpler to
read than "lock the mapping pages lock", and less redundant.

> I suppose it does force me to go dig into whatever mapping->pages is to
> figure out that there's an unnamed spinlock_t and that the compiler can
> insert the appropriate pointer arithmetic, but now my brain trips over
> 'pages' being at the end of the selector for parameter 1 which slows
> down my review reading...
> 
> OTOH I guess it /did/ motivate me to click the link, so well played,
> sir. :)

Now if only I can trick you into giving your ACK on patch 1,
"xfs: Rename xa_ elements to ail_"