Re: Event API changes - EVIOCGID

[Brad H. <bh...@bi...>]

On Sun, 28 Jul 2002 18:22, Vojtech Pavlik wrote:
> On Sun, Jul 28, 2002 at 05:45:37PM +1000, Brad Hards wrote:
<snip>
> > I am not happy about the change from uint16_t to __u16, which you appear
> > to have made before sending this to Linus.
> >
> > That is a broken change - there is a standard type, and you've changed
> > it to a non-standard type. This is confusing to userspace programmers,
> > and I cannot provide a satisfactory explaination for this in
> > documentation.
> >
> > Please change it back.
>
> Well, I know this has been discussed back and forth.
And I was right last time too :-)

> __u16 is a kernel type and is defined if you #include <linux/input.h>.
> uint16_t isn't.
__u16 is no more a kernel type than uint16_t.
It is a one line fix: include <linux/types.h> instead of <asm/types.h>
Which you probably should be doing anyway, since there is no
reason to rely on any assembler types in <linux/input.h>

> __u* is used extensively in the input API anyway, so you'd have to
> explain it to userspace programmers nevertheless. So I prefer keeping
> the input.h include use just one type of explicit sized types.
So do I, and it had better be a standard type.

Note that the input API does *NOT* use __u* extensively. In fact
if you take out the force feedback stuff (which Johannes already
agreed to change:), this is the *only* _u* usage in any part of the 
input API.

> Sure, we can change them all to uint*_t, but then do it all at once and
> provide a satisfactory explanation for it. ;)
I am doing it all. Johannes agreed to the change, and I did the only
other required entry. If Johannes agrees, I'll do the trivial changes
for force-feedback.
The reason why I am not doing it all at once is to provide patches
that do one API change at a time. Or, depending on how you look
at it, I did the only change all-at-once, and you reverted it :)

Brad

-- 
http://conf.linux.org.au. 22-25Jan2003. Perth, Australia. Birds in Black.