[connortsui20]

Summary

Builds on #11375.

Exporting a versioned Vortex UUID to Arrow and importing it back currently drops the declared version, even when every value is unchanged. Nested extensions can also lose metadata or fail to import when Arrow fields disagree with their storage. The full dtype, including nested metadata and nullability, must survive this round trip.

The raw import and export traits leave metadata transport to each plugin. ArrowExtVTable puts that handling behind a typed registration adapter, so extension authors implement their native conversions while shared code preserves Vortex metadata.

Changes

The session checks that inferred fields reconstruct the complete source dtype and that successful plugin imports return the dtype inferred from their field. These checks apply to raw and typed plugins. Explicit export targets can still request conversions.

UUID needs a separate Vortex payload because canonical arrow.uuid metadata must be empty. Keeping that representation lets other Arrow tools recognize UUID columns. A field without the Vortex payload imports with an unspecified version, so ordinary Arrow producers remain supported.

Nested conversions need the full Arrow field because a DataType alone cannot carry extension identity. GeoArrow's required child names also belong to its layout contract. Metadata restoration uses inferred Vortex storage because it can differ from Arrow's physical storage.

Regression coverage retains complete dtype and array comparisons while removing redundant parameter products and assertions. Variant remains excluded pending #11359, and equality derivation remains separate in #11115.

API Changes

Extension plugins implement ArrowExtVTable, declare metadata_policy(), and register through ArrowSession::register_extension(vtable). The typed import callbacks are associated functions. Raw plugins remain available, but inferred fields and imported arrays must satisfy the same dtype checks.

Validation

• Formatting with the pinned nightly-2026-09-10.
• Commit hook: workspace Rust and TOML formatting.
• Pre-push hook: DCO checks and workspace Clippy with all targets, all features, and warnings denied.
• Runtime tests, doctests, and fuzz runs were not repeated for these refinements.

[codspeed]

Merging this PR will not alter performance

2 warnings

⚠️ 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 benchmarks are dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

✅ 1916 untouched benchmarks
⏩ 48 skipped benchmarks¹

---

_{Comparing ct/arrow-roundtrip-fixes (0607c3b) with develop (b367941)}

Footnotes

1. 48 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. ↩

[robert3005]

Note to self - these todos are stale

[robert3005]

The arrow parts looks good, we can make the runend fallback better