feat(theta): add theta union

[ZENOTME]

Summary

• add stateful ThetaUnion and builder
• export ThetaUnion from the theta module
• add union coverage for exact, estimation, compact, seed mismatch, ordering, reset, and edge states

Tests

• cargo test -p datasketches --features theta

[ariesdevil]

seems no need check with entries.len() > 1, sort_unstable() no-op on single element array.

[ariesdevil]

Rest LGTM.

[tisonkun]

Generally LGTM. Some follow-up suggestions inline, and:

I may prefer to name result/result_with_order to to_sketch as we do for the CpcSketch and HllSketch. Also, we may not need to have two methods but only one with a ordered param.

This is a breaking change since we already use the result_with_order API in ThetaIntersection, but it should be worthy to do the refactor.

[tisonkun]

Later we may review all seed_hash != seed_hash checks and add # Errors document.

[tisonkun]

Now that #146 is merged. We may merge this as well.

Is there any logical conflict that should be resolved beforehand?

[ZENOTME]

Now that #146 is merged. We may merge this as well.

Is there any logical conflict that should be resolved beforehand?

Yes. Let me covert it as a generic version.

[tisonkun]

Thanks for your updates!

I'll go through all these changes and perhaps push a few commits to handle visibility issues.

Hopefully in days, but may be delay to next weekend.

[ariesdevil]

BTW: As @tisonkun mentioned on X, pub impl(crate) trait Foo {} just landed in nightly version; maybe we can use this feature?

[tisonkun]

This crate is a library and we should stay on stable toolchain. And we have an MSRV of 1.86.

[ariesdevil]

Make sense.

[tisonkun]

Pushed a few commits to handle part of the structure and visibility issues. I'll handle for:

• hash_table.rs
• sketch_view.rs
• union.rs

... under thetacommon as well and call for a merge.

[tisonkun]

There is one more pending issue: the two traits we made public in thetacommon are looking a bit strange right now. If you want to make them sealed or handle them differently, we probably need to refactor the tests in union.rs.

It feels very awkward to implement sealed traits for a struct inside the test module itself. We should take a step back and look at the overall scope of these two traits.