Re: [PATCH 0/3] Couple of sysfs patches

[Greg KH <greg@kroah.com>]

On Wed, Jun 09, 2004 at 05:54:23PM -0500, Dmitry Torokhov wrote:
> On Wednesday 09 June 2004 05:45 pm, Greg KH wrote:
> > On Wed, Jun 09, 2004 at 05:32:28PM -0500, Dmitry Torokhov wrote:
> > > Actually, I myself want someting else -
> > > 
> > > int platform_device_register_simple(struct platform_device **ppdev,
> > > 				    const char *name, int id)
> > > 
> > > It will allocate platform device, set name and id and release function to
> > > platform_device_simple_release which in turn will be hidden from outside
> > > world. Since the function does allocation for user is should prevent the
> > > abuse you were concerned about.
> > 
> > Ok, that sounds good.  I'll take patches for that kind of interface.
> > 
> > But have the function return the pointer, like the class_simple
> > functions work.  Not the ** like you just specified.
> 
> I want to do both allocation + registration in one shot and I knowing
> the error code may be important to users.

That's fine to do.  Again, look at how the class_simple_create()
function works.  If an error happens, convert it to ERR_PTR() and return
that.  The caller can check it with IS_ERR() and friends.

> Why do you oppose having double pointers in interface?

It's messy, and with the ERR_PTR() macros, not needed :)

thanks,

greg k-h