Re: [PATCH v4] staging: greybus: Convert uart.c from IDR to XArray

From: Matthew Wilcox <willy@infradead.org>
To: Johan Hovold <johan@kernel.org>
Cc: "Fabio M. De Francesco" <fmdefrancesco@gmail.com>,
	Alex Elder <elder@kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	greybus-dev@lists.linaro.org, linux-staging@lists.linux.dev,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v4] staging: greybus: Convert uart.c from IDR to XArray
Date: Mon, 30 Aug 2021 14:31:35 +0100	[thread overview]
Message-ID: <YSzdtzH1GdZqt66G@casper.infradead.org> (raw)
In-Reply-To: <YSzQAd0Rg5CF/eLe@hovoldconsulting.com>

On Mon, Aug 30, 2021 at 02:33:05PM +0200, Johan Hovold wrote:
> On Mon, Aug 30, 2021 at 01:16:07PM +0100, Matthew Wilcox wrote:
> > On Mon, Aug 30, 2021 at 01:52:48PM +0200, Johan Hovold wrote:
> > > Whether the API is better is debatable. As I said, almost no drivers use
> > > the new XArray interface, and perhaps partly because the new interface
> > > isn't as intuitive as has been claimed (e.g. xa_load() instead of
> > > ida_find()). And IDR/IDA isn't marked/documented as deprecated as far as
> > > I know.
> > 
> > I can't just slap a 'deprecated' attribute on it.  That'll cause a
> > storm of warnings.  What would you suggest I do to warn people that
> > this interface is deprecated and I would like to remove it?
> 
> I'd at least expect a suggestion in the IDR documentation to consider
> using XArray instead.

Fair enough.

+The IDR interface is deprecated; please use the `XArray`_ instead.

Just running that through htmldocs to make sure I've got the syntax
right, and then I'll commit it.

> > Why do you think that idr_find() is more intuitive than xa_load()?
> > The 'find' verb means that you search for something.  But it doesn't
> > search for anything; it just returns the pointer at that index.
> > 'find' should return the next non-NULL pointer at-or-above a given
> > index.
> 
> We're looking up a minor number which may or may not exist. "Find" (or
> "lookup" or "search") seems to describe this much better than "load"
> (even if that may better reflect the implementation of XArray).

It's not the _implementation_ that it fits, it's the _idiom_.
The implementation is a lookup in a trie.  The idiom of the XArray
is that it's a sparse array, and so it's a load.

> And no, I would not expect a find implementation to return the next
> entry if the requested entry does not exist (and neither does idr_find()
> or radix_tree_lookup()).

Oh dear.  You've been corrupted by the bad naming of the IDR functions
;-(