Messages in this thread | | | Date | Wed, 2 Sep 2009 17:05:06 -0700 | From | Greg KH <> | Subject | Re: [PATCH] pci: separate out pci_add_dynid() |
| |
On Thu, Sep 03, 2009 at 07:43:28AM +0900, Tejun Heo wrote: > Ummm... another thought. > > Tejun Heo wrote: > > It's one of the gripes I've had with the driver core interface. > > Returning the object which had its reference bumped up is nice to do > > do_something(get_driver(driver)); but it should have been made clear > > that the return value always equals the input parameter. IIRC, in the > > early days of the current driver model and kref, there were intentions > > to make kref somehow more capable and handle the last reference > > problem from within it so that if (get_driver(driver)) test is > > actually meaningful but that turned out to be not too feasible, so we > > ended up with unnecessary if () around driver model get functions. I > > try to remove them whenever an affected piece of code is being > > modified but there still are plenty left. I'll update comment on > > driver core get functions. > > Greg, I think I suggested this before. What do you think about just > making those get functions return void?
I have no objection to that at all.
> The current interface combined with widespread if() around them is > quite misleading.
I agree.
> Most device/driver structures are inherited and extended when actually > in use, so the driver or its subsystem code usually can't do much with > the return value other than checking for bogus error condition anyway.
I totally agree.
thanks,
greg k-h
| |