Re: how to represent sequence of brightnesses in /sys (was Re: [PATCH] Add the LED burst trigger)

[Greg KH <greg@kroah.com>]

On Sun, Dec 29, 2013 at 12:50:46AM +0100, Pavel Machek wrote:
> On Sat 2013-12-28 13:50:42, Greg KH wrote:
> > On Sat, Dec 28, 2013 at 10:25:23PM +0100, Pavel Machek wrote:
> > > Hi!
> > > 
> > > > >> Sysfs is meant to be human-readable/writable, so please use plain ASCII
> > > > >> numbers in strings instead.
> > > > >
> > > > > Actually, sysfs is meant to be one value per file, and it is
> > > > 
> > > > Ideally, yes.
> > > > 
> > > > > understood that data that are "natively blob" are just passed as
> > > > > blob. (I believe this qualifies).
> > > > 
> > > > But it doesn't buy us much here, does it? It will make e.g. shell scripts
> > > > needlessly complicated.
> > > 
> > > echo -ne '\012' is not that bad, and parsing array of integers from
> > > kernel will be an ugly piece of code.
> > 
> > Ick, no.  What are you trying to do here?  Have the kernel intrepret a
> > sequence of bytes to flash an LED?  I thought we have frameworks in
> > userspace already to handle this type of thing.  Please don't invent new
> > ways of doing stuff...
> 
> Idea would be "sequence of brigtnesses" (one file) and "delay between
> changes" (second file).

Ick.

> Reason to do it in kernel is that some machines actually have
> "coprocessor" on i2c that can do it while main CPU is suspended. (For
> more reasons, see beggining of thread).

Ick ick.

> Binary attribute with array of bytes should be acceptable, rights?

Not at all.

> (IOW write(..., buf, size) )
> 
> Ascii array of decimal integers -- no so, right?
> 
> (IOW printf("%d %d ..", buf[0], buf[1]) )

Use an ioctl with a structure to get things correct as a character
device.  As odds are, you aren't going to be able to create a "generic"
format for all of this for all types of devices that support such a
"co-processor".

greg k-h