On Fri, Jun 03, 2016 at 12:19:41PM +0100, Liviu Dudau wrote:
On Fri, Jun 03, 2016 at 11:36:33AM +0100, Russell King - ARM Linux wrote:
On Fri, Jun 03, 2016 at 10:40:48AM +0100, Liviu Dudau wrote:
On Fri, Jun 03, 2016 at 08:58:10AM +0100, Russell King wrote:
Convert DT component matching to use component_match_add_release().
Hi Russell,
Any reason for not keeping the component_match_add() calls in the drivers?
Sorry, I don't understand your comment.
As in: component_match_add() already exists as a macro that calls component_match_add_release(), but with a NULL release function. If it were to be changed to pass a default release_of() function then most of the drivers would not have to change, right? It is just one point of view and I was curious if there was a reason not to choose it, as it would have (probably) generated a smaller delta?
And what should the calls that don't pass a DT node be called? What happens to new users who aren't passing a DT node but use component_match_add() ?
What you're suggesting sound totally insane to me: you're making the change a flag-day: component_match_add() currently takes anything as the data pointer and users can pass anything that their compare function can handle - and changing that to a function which can only take a device_node. That means all non-DT users of that function need to change at the same time.
Flag days are really bad news in kernel development (or any distributed development project), and I won't generate a patch which causes a flag day to occur - especially not one which impacts a large number of users.