Re: [RFC] What are the goals for the architecture of an in-kernel IR system?

["David Härdeman" <david@hardeman.nu>]

On Sun, Mar 28, 2010 at 09:51:17PM -0300, Mauro Carvalho Chehab wrote:
> 
> I spoke too soon... removing the index causes a problem at the read ioctl: there's no way
> to retrieve just the non-sparsed values.
> 
> There's one solution that would allow both read/write and compat to work nicely,
> but the API would become somewhat asymmetrical:
> 
> At get (EVIOCGKEYCODEBIG):
> 	use index/len as input and keycode/scancode as output;
> 
> At set (EVIOCSKEYCODEBIG):
> 	use scancode/keycode/len as input (and, optionally, index as output).
> 

This was exactly the approach I had in mind when I suggested using 
indexes.

> Having it asymmetrical doesn't sound good, but, on the other hand, 
> using index for
> the set function also doesn't seem good, as the driver may reorder the entries after
> setting, for example to work with a binary tree or with hashes.

I don't think the assymetry is really a problem. As I see it, there are 
basically two user cases:

1) Userspace wants scancode X to generate keypress Y
   (In which case userspace doesn't care one iota what the index is)

2) Userspace wants to get the current keytable from the kernel
   (In which case a loop with an index from 0 to n is appropriate)

and, possibly:

3) Userspace wants to know what keycode (if any) scancode X generates
   (In which case approach 2 will work just as well, but this usecase
    seems a bit contrived anyway...)

-- 
David Härdeman