David Gwynne wrote: > i recently proposed replacing a hash with an rb tree somewhere in > the network stack, but it was pointed out that rb trees are big. > > in hindsight i think the other person was talking about the size > of an RB_ENTRY inside each thing you're tracking, but it made me > look at the code size of rb trees again. it turns out on amd64 its > about 2.5k of code per type of rb tree. a type being each RB_ENTRY > inside a particular struct. ie, if a struct has two RB_ENTRYs in > it, then it generates two chunks of code, one for each of them.
I love everything about this, but didn't actually look much at the diff or try it out.
