Re: [IR-RFC PATCH v4 0/6] In-kernel IR support using evdev

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Thu, Nov 26, 2009 at 10:34:55PM -0500, Jon Smirl wrote:
> On Thu, Nov 26, 2009 at 10:12 PM, Jarod Wilson <jarod@wilsonet.com> wrote:
> > This part... Not so wild about. The common thought I'm seeing from people is
> > that we should be using setkeycode to load keymaps. I mean, sure, I suppose
> > this could be abstracted away so the user never sees it, but it seems to be
> > reinventing a way to set up key mapping when setkeycode already exists, and
> > is used by a number of existing IR devices in the v4l/dvb subsystem (as well
> > as misc things like the ati rf remotes, iirc). Is there some distinct
> > advantage to going this route vs. setkeycode that I'm missing?
> 
> The configfs scheme and keymaps offer the same abilities. One is an
> ancient binary protocol and the other one uses Unix standard commands
> like mkdir and echo to build the map. You need special commands -
> setkeycodes, getkeycodes, showkey, loadkeys, xmodmap, dump-keys to use
> a keymap.  I've been using Linux forever and I can't remember how
> these commands work.

Nor you really should - it all is mostly being used transparently for
the end-user. I mean udev loading your laptop-specific keymap is not
using loadkeys but specially written utility that issues EVIOCSKEYCODE
directly.

> 
> Keymaps are a binary protocol written by Risto Kankkunen in 1993.
> Configfs was added by Oracle about two years ago but it has not been
> used for mapping purposes.

Nor it is even enabled by default... Do we want to make in mandatory on
all consumer systems out there?

> 
> It's another discussion, but if IR goes the configfs route I'd
> consider writing a patch to switch keymaps/keycodes onto the configfs
> model. It is a huge advantage to get rid of these pointless special
> purpose commands that nobody knows how to use. I'd keep the legacy
> IOCTLs working and redirect the data structure to a configfs one
> instead of the existing structure.

What is the memory footprint for configfs solution though? I would hate
to see the cost of user-modifiable keymap explode tenfold so that we can
rely less (not even get rid of, since it is published userspace API/ABI)
on setkeycodes and related ioctls.

> 
> The same idea is behind getting rid of IOCTLs and using sysfs. Normal
> Unix commands can manipulate sysfs. IOCTLs have problems with strace,
> endianess and the size of int (32/64b).

The size of long you mean, right? Besides, now that we know better we
should simply use explicitely-sized fields in ioctl structures.

-- 
Dmitry