Iteration without an iterator about hashbrown HOT 7 OPEN

Unfortunately we can't really guarantee that elements aren't shuffled around on removal. However have you had a look at the RawTable API that can be enabled with the raw feature? You might be able to build something on top of that.

from hashbrown.

Thanks @Amanieu . Does the RawTable provide the no-shuffle-on-remove guarantee then?

The other functionality needed is iteration (well, one step at a time) starting from an existing T known to be in the table. I can't see a way of doing that efficiently as is - it would still need a way of creating a RawIter<T> from a given starting position (e.g. from a Bucket<T> obtained from find(hash, eq)).

from hashbrown.

RawTable does guarantee no-shuffle-on-remove.

You can get a Bucket<T> from an index by calling the bucket function. You wouldn't use RawIter at all for your style of iteration.

from hashbrown.

I think I'm still missing something (or have not explained things very well) - I still need to use some of the machinery from RawIter to step from an item (I start with a T) to the next one.

I want to implement something like this function:

impl MyTable<T:Hash> {
    fn next(&self, item: &T) -> Option<&T> {
        // find the "next" entry in the table following the passed in item, if any
        ...
    }

Given a T I can get a Bucket<T> from the table via find. I don't see how to get an index for the current item (except via pointer subtraction), but definitely can't see a way (I'm looking at https://docs.rs/hashbrown/0.6.3/hashbrown/raw/struct.RawTable.html) of finding the next occupied bucket after that.

Thanks for bearing with me!

from hashbrown.

Oh, it seems that the bucket_index function is not marked pub. If you send a PR changing that I'll accept it and push a new version.

from hashbrown.

Thanks, I've opened the PR. Now that I've been looking at other private methods, it looks like I'll probably also need RawTable::ctrl() and raw::is_full() to be pub as well, and then I can just loop over an index looking for occupied buckets. I can add those to the PR (or open another) if that sounds right.

from hashbrown.

This would be useful to me as well. I have the requirement that the iterator cannot borrow from the map itself, it must borrow from a &RefCell<HashMap>, so an index based iteration API would be useful.

from hashbrown.