At Mon, 29 May 2006 18:34:06 +0100,
James Courtier-Dutton wrote:
>
> Takashi Iwai wrote:
> >
> > I agree to use numid.
> > My argument is that numid can belong to the common header block.
> >
> > The code you find in control.c is anyway a kind of object oriented.
> > It's a message to a control element.
> > But in this protocol, you get numid at very last, even depending on
> > the type. Thus the code looks like
> >
> > type = get_type();
> > len = get_length();
> > switch (type) {
> > case DB_SCALE:
> > numid = get_value();
> > kctl = get_kcontrol(numid);
> > kctl->info2(kctl, ptr);
> > ...
> > }
> >
> > When you add more different types,
> >
> > type = get_type();
> > len = get_length();
> > switch (type) {
> > case DB_SCALE:
> > numid = get_value();
> > kctl = get_kcontrol(numid);
> > kctl->info2(kctl, ptr);
> > ...
> > case ANOTHER_DB_SCALE:
> > numid = get_value();
> > kctl = get_kcontrol(numid);
> > kctl->info2(kctl, ptr);
> > ...
> > case YET_ANOTHER_DATA:
> > numid = get_value();
> > kctl = get_kcontrol(numid);
> > kctl->info2(kctl, ptr);
> > ...
> >
> > }
> >
> > If the numid is in the common block, we can simply extract numid
> > first, then evaluate the type and value later, eventually call the
> > callback at the single place.
> >
> >
> > Takashi
>
> My point is that there will only be one DB_SCALE case in the request
> handler in control.c in the kernel. It simply calls info2(). The driver
> then returns a TLV. The returned TLV will have all the variations on
> db_scales. control.c won't care what the actual TLV value means. It will
> simply return it to user space, and let alsa-lib work out what it is.
> Another call the snd_ctl_misc might not even need a numid. It might
> instead be requesting some other information, like card name or some
> other future bit of info needed. My point is this is extensible, without
> breaking kernel/alsa-lib binary compatibility.
Sure, it's extensible. But my point is that the complexity of this
encapsulation.
For example, in order to get a numid, you have to extract three
levels down. The version + length, the type + length, again type +
length, then finally you reach to numid value. Each extraction
requires a function call to convert from a byte-stream to an integer
and a check of the length. I feel it's a bit too much abstraction.
We've learned that too much abstraction is rather more evil -- see the
implementation of alsa-lib and alsa-kernel API...
Takashi
_______________________________________________
Alsa-devel mailing list
Alsa-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/alsa-devel
