Re: Generic rbtree search & insert cores

[Peter Zijlstra <peterz@infradead.org>]

On Mon, 2012-04-30 at 06:11 -0500, Daniel Santos wrote:
> 
> So as long as our struct rbtree_relationship is a compile-time
> constant, the generated code will look pretty much (if not exactly)
> like that of the example code in rbtree.h.  Please let me know what
> you think.  I've tested this in a userspace program, but haven't fully
> stress tested it in kernelspace yet. 

Right, this ought to work.

I'm not sure the root_offset thing is worth it though, passing in the
rb_root pointer isn't much of a hassle, in fact I'd expect to pass rb
related pointers to a function called rbtree_insert().

Also, using a macro to create the rbtree_relationship thing would make
it easier. Something like:

	RB_RELATION(struct mouse, node, name, name_cmp);

I'd think you'd also want to provide an insertion variant that does the
leftmost thing, that's used in several places, and you'd also need one
for the augmented rb-tree.