tower120/hi_sparse_array

WIP

hi_sparse_array's Issues

Try to get rid of level1_mask, level2_mask, data_mask and use only _meta

Basically try to use LevelMasksIter instead of LevelMasks.
Benchmark contains/data_mask for reduce specifically - since that could potentially bring overhead.

`Fold` can work with non-Cloneable iterator

Fold can work with non-Cloneable iterator that produce values or references.

We already store references to Iterator Items in fold's state. We can store Iterator Items in Fold instead. This will make Fold big though.

Iterator now can iterate empty blocks in non-EXACT_HIERARCHY

Compile-time customization of levels count

That could be done through [Level;N], using PassThrough, using some kind of Tuple<impl Level>.
Also some specialization of LevelMasksIter required since get_data... accept last hierarchy level block meta, as input.
All of this requires better meta-programming support from the Rust side. Namely - specialization.

I can't come up with compile-time configurable depth level in current Rust state, without making lib
unreadable. But this is probably doable with macros, or just by re-implementing Array N times, for each
level count.

`Apply` could work with different type DataBlocks

Data can be stored continiouglsy for fast unordered iteration

Rust 1.79 : LevelMasks could return Borrow

This rust-lang/rust#122055 should appear in Rust 1.79
We could try to return Borrow in LevelMasks instead of value.
This should allow to remove Clone constraint from BitBlock.

UPDATE. It turns out that we need more robust GAT support from Rust as well. for<'a> does not work in places where we need it.

SparseHierarchy::get can return Empty DataBlock, if index outside range

SparseHierarchy::get can return Empty DataBlock, if index outside range, instead of Option None.

To do this we probably need to add SparseHierarchy::empty_data_block(&self) -> Self::DataBlock<'_>.
This will make SparseHierarchy interface more complex, and benefits are currently unclear.