Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 30 Apr 2020 11:26:48 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Michel Lespinasse <walken@google.com>
Cc:     linux-kernel@vger.kernel.org, dave@stgolabs.net, mingo@kernel.org,
        tglx@linutronix.de, oleg@redhat.com, irogers@google.com,
        juri.lelli@redhat.com, vincent.guittot@linaro.org
Subject: Re: [RFC][PATCH 1/7] rbtree: Add generic add and find helpers
Message-ID: <20200430092648.GB13616@hirez.programming.kicks-ass.net>
References: <20200429153258.563269446@infradead.org>
 <20200429153549.006661686@infradead.org>
 <20200430010405.GA237715@google.com>
 <20200430084617.GQ13592@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200430084617.GQ13592@hirez.programming.kicks-ass.net>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Apr 30, 2020 at 10:46:17AM +0200, Peter Zijlstra wrote:
> On Wed, Apr 29, 2020 at 06:04:05PM -0700, Michel Lespinasse wrote:

> > - Have you considered passing a cmp() function to rb_add() and
> >   rb_add_cached(), and having these test cmp() < 0 rather than less() ?
> >   I figure every user will need to have a cmp() function, so it'd be
> >   nicer if they didn't also need a less() function, if the generated
> >   code is similar (if you checked and rejected it because of bad code,
> >   please just say so).
> 
> I did consider it; in fact I my original helpers had that.
> 
> The reaosn I went with less() over cmp() is that the add() vs find()
> function signatures:
> 
>   bool (*less)(struct rb_node *, const struct rb_node *);
>   int (*cmp)(const void *, const struct rb_node *);
> 
> differ anyway in the left-hand argument; this is 'fixable' when you
> use an (on-stack) dummy object for find (as uprobes does), but that
> doesn't always work, esp. when the object is big. And given you need two
> functions anyway, I figured it was useful to name them differently.
> 
> If you've looked at the other patches a bit, you'll see I've implemented
> both functions as 'trivial' wrappers around a single compare function in
> many cases.

I just realized I'd already done all this at some point ;-) See
rbtree_latch.h. Clearly I failed to realize the full potential back when
I did that.