fix: preserve non-empty theta v2 exact sketches

[tisonkun]

Summary

Fix legacy Theta serialization version 2 exact images being deserialized as logically empty when they contain retained entries.

The v2 exact branch now derives logical emptiness from num_entries == 0 instead of setting it unconditionally.

Closes #186.

Regression coverage

The new serialization compatibility tests construct valid v2 exact images using the current Theta family ID and default seed hash, then verify:

• a non-empty image retains all entries,
• is_empty() is false,
• exact estimate and bounds equal the retained count,
• entry iteration is unchanged,
• a current-format round trip preserves the state, and
• a zero-entry v2 exact image remains empty.

Relationship to other implementations

This restores agreement with the current C++ and Go legacy readers. Both treat a v2 exact image (preamble_longs == 2) as empty only when num_entries == 0:

• C++: https://github.com/apache/datasketches-cpp/blob/c22888581964fc490feee835766cc8c3adb722e0/theta/include/theta_sketch_impl.hpp#L605-L631
• Go: https://github.com/apache/datasketches-go/blob/c558cc2d64f9a307c196a7adb0067eadd1b776f7/theta/decoder.go#L295-L343

Go retains compatibility fixtures identified as Java-generated v2 images. Current Java reads compact versions 3 and 4 rather than v2 directly, so the active legacy-reader behavior is represented by C++ and Go:

• Go Java-v2 fixture test: https://github.com/apache/datasketches-go/blob/c558cc2d64f9a307c196a7adb0067eadd1b776f7/theta/sketch_serialization_test.go#L210-L228
• Current Java version dispatch: https://github.com/apache/datasketches-java/blob/4067ffefabbcd03944dc3617ff2948ab760f3b75/src/main/java/org/apache/datasketches/theta/CompactThetaSketch.java#L97-L115

The fix changes no serialization bytes emitted by Rust; it only corrects the state reconstructed from legacy input.

Validation

• cargo x prepare-testdata
• cargo x check
• cargo x test
• cargo x lint

[Copilot]

Pull request overview

Fixes legacy Theta compact serialization v2 (“exact”, preamble_longs == 2) being deserialized as logically empty even when it contains retained entries, aligning Rust behavior with the C++ and Go legacy readers and preventing downstream set-operation short-circuiting.

Changes:

• Fix v2 exact deserialization to set empty based on num_entries == 0 instead of unconditionally true.
• Add regression tests that construct valid v2 exact images and assert non-empty behavior, estimate/bounds correctness, iteration stability, and current-format round-trip preservation.
• Add a regression test ensuring zero-entry v2 exact images remain logically empty.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
datasketches/src/thetafamily/theta/sketch.rs	Corrects v2 exact logical-emptiness reconstruction by tying empty to num_entries.
datasketches/tests/serde_tests/theta.rs	Adds compatibility tests and a helper to synthesize v2 exact images for regression coverage.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ZENOTME]

LGTM. Maybe we refine test later.

[ZENOTME]

I'm not sure there are better way to provide a v2 serialize snapshot instead of construct by hand here. 🤔

[tisonkun]

v2 serialize snapshot

Maybe. But existing C++/Java/Go impls do not provide one. We may provide one when implementing #176 , and if you'd like to, you can go to other impls to ask them to provide one and update https://github.com/apache/datasketches-tck.