[connortsui20]

Summary

Array serde lived in two places: serialize and deserialize on VTable, and the ArrayPlugin trait that the session registry actually calls, bridged by a blanket impl<V: VTable> ArrayPlugin for V. Because of the blanket impl, every vtable could be registered as a plugin, so every encoding had to carry serde methods even when it had no wire form. About 15 vtables (Filter, Slice, ScalarFn, Shared, and others) carried Ok(None) and bail! stubs only to satisfy the trait. Vtables whose wire form lives in a separate plugin (BitPacked, FoR, DecimalByteParts, Decimal) could still be registered directly and silently lose serde, which the old vtable_serde_requires_plugin tests guarded at runtime. Removing the blanket impl turns that mistake into a compile error and leaves one serde path.

Changes

Each encoding with a wire form implements ArrayPlugin on its existing vtable type rather than on a new FooPlugin struct. This keeps the number of types the same and leaves every register(Primitive) call site unchanged. The cost is that X.id() is ambiguous where both traits are imported, so a few call sites now spell VTable::id(&X).

VTable does not get ArrayPlugin as a supertrait. That would force in-memory-only encodings to keep the stubs this PR removes, and it does not fit encodings whose wire form belongs to a separate plugin.

Single-ID plugins no longer check serialized_id, which the blanket impl used to do. The ArrayPlugin::deserialize docs now state that callers must pass an ID from read_ids, which the registry already guarantees. read_ids and write_ids default to the plugin's own ID, and the second commit removes three overrides that matched the default.

PythonVTable keeps an explicit stub impl, so vortex.registry.register still works. PiecewiseSequence is no longer registered in the default session, since it never had a wire form, and serializing it now fails at the registry lookup with a different message. The three tests that registered a bare vtable and expected no serde are removed, because that no longer compiles. zoned::writer::tests::zone_maps_hold_no_variant_state::case_2_variant_in_struct fails in the full vortex-layout suite on this branch and on its base commit alike, so it is unrelated.

API Changes

VTable::serialize and VTable::deserialize are removed, so third-party encodings must implement ArrayPlugin to be serializable. Vtables no longer implement ArrayPlugin automatically, so ArraySession::register no longer accepts in-memory-only vtables such as Filter, Slice, ScalarFn, PiecewiseSequence, BitPacked, or FoR. ArrayPlugin::read_ids and ArrayPlugin::write_ids now have default implementations.

---

Generated by Claude Code

[codspeed]

Merging this PR will regress 1 benchmark

1 warning

⚠️ Walltime results may be unreliable

Some benchmarks used the Walltime instrument outside CodSpeed Macro Runners, which can make the results inconsistent.

Learn more about Macro Runners

See affected benchmarks

⚡ 2 improved benchmarks
❌ 1 regressed benchmark
✅ 1914 untouched benchmarks
⏩ 47 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	CPU Simulation	take_filter_primitive_small_uncached_random_mask_random_indices[4096, 10]	26.6 µs	29.8 µs	-10.5%
⚡	CPU Simulation	missing_stats[16]	91.8 µs	74.5 µs	+23.25%
⚡	CPU Simulation	max_f64	317.1 µs	283.8 µs	+11.76%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ct/upbeat-ride-if0rb0 (26e275a) with develop (5852636)}

Footnotes

1. 47 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[mhk197]

Superseded by #11285