Re: [PATCH 1/6] rbtree: add rb_find_add_cached() to rbtree.h

[Qu Wenruo]

在 2024/12/17 08:52, Peter Zijlstra 写道:
> On Tue, Dec 17, 2024 at 08:43:26AM +1030, Qu Wenruo wrote:
> > 在 2024/12/13 19:36, Peter Zijlstra 写道:
> > > On Thu, Dec 12, 2024 at 10:46:18AM -0600, Roger L. Beckermeyer III wrote:
> > > > Adds rb_find_add_cached() as a helper function for use with
> > > > red-black trees. Used in btrfs to reduce boilerplate code.
> > > >
> > > > Suggested-by: Josef Bacik <josef@xxxxxxxxxxxxxx>
> > > > Signed-off-by: Roger L. Beckermeyer III <beckerlee3@xxxxxxxxx>
> > >
> > > Acked-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> >
> > I guess it's fine to merge this change through btrfs tree?
>
> Yeah, I think so. I don't think there's anything else pending for this
> file -- its not touched much.
>
> > Just curious about the existing cmp() and less() functions, as they only
> > accept the exist node as const.
> >
> > I'm wondering if this is intentional to allow the less/cmp() functions
> > to modify the new node if needed.
> > As I normally assume such cmp()/less() should never touch any node nor
> > its entries.
>
> Oh yeah, they probably should not. I think it's just because the
> callchain as a whole does not have const on the new node (for obvious
> raisins), and I failed to put it on for the comparators.
>
> You could add it (and fix up the whole tree) and see if anything comes
> apart.
Thanks for confirming this.

I'll make the cmp() for the new helper to accept all const parameter, 
and give a try to do a tree-wide cleanup to make existing cmp/less() to 
accept all const parameters. (pretty sure a lot of things will fall 
apart though).

Thanks,
Qu