fix(core): map DataFusion SchemaError to InvalidInput

[jja725]

Summary

Schema errors from DataFusion (e.g. a filter referencing an unknown column, a projection naming a missing field) are user-input failures, not internal lance failures. Today they're mapped to Error::Schema, which makes downstream consumers that pattern-match on Error::InvalidInput for user-facing error classification report them as "internal."

This PR moves DataFusionError::SchemaError into the same bucket as SQL/Plan/Configuration, all of which already route to Error::InvalidInput. Behaviorally these are the same kind of error — bad input from the caller — and they should classify the same way.

Concrete impact

The C/C++ FFI binding (lance-format/lance-c) maps Error::Schema to its Internal error code (catch-all), so a caller doing:

lance_dataset_delete(ds, "no_such_column = 1", NULL);

sees LANCE_ERR_INTERNAL instead of LANCE_ERR_INVALID_ARGUMENT. Any binding doing the same kind of variant-based mapping (Python, Java, …) hits the same problem. lance-c #31 currently asserts the wrong code in its tests as a workaround — once this lands, that PR can flip its assertions to InvalidArgument and the wart goes away.

The path that produces this is straightforward: Scanner::filter parses the SQL via Planner::parse_filter → filter.to_field(&df_schema)? → DataFusion returns DataFusionError::SchemaError for unknown fields → today, Error::Schema { .. }. With this fix → Error::InvalidInput { .. }, with the column name preserved in the boxed source.

Risk

Any code that pattern-matches on Error::Schema to handle DataFusion-derived schema errors will no longer match.

A grep across rust/lance, rust/lance-index, and rust/lance-datafusion finds two such matches, both in lance-index/src/scalar/inverted/index.rs (lines 1862, 1878). Both are fed by lance-io's internal Error::schema(...) calls (not by the DataFusion conversion), and continue to behave unchanged.

External consumers (Python bindings, Java bindings, lance-c) that classify Error::Schema as "internal" today will start classifying these errors as "invalid input" — which is the desired direction.

Test plan

Adds a unit test (test_datafusion_schema_error_is_invalid_input) that constructs a DataFusionError::SchemaError(SchemaError::FieldNotFound { ... }) directly and asserts:

1. The converted Error is InvalidInput.
2. The offending column name ("missing_col") is preserved in the displayed error message.

$ cargo test -p lance-core --features datafusion test_datafusion_schema_error_is_invalid_input
test error::test::test_datafusion_schema_error_is_invalid_input ... ok
test result: ok. 1 passed; 0 failed

cargo fmt -p lance-core --check clean.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[codecov]

Codecov Report

❌ Patch coverage is 92.30769% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
rust/lance-core/src/error.rs	92.30%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Xuanwo]

Thank you for this change!