Comparing rule evaluators#

parcel_runtime::differential compares the CEL interpreter and DataFusion on the same data and callers. This is the comparison used by parcel check.

`differential`#

pub async fn differential(
    c: &Compilation,
    batch: &RecordBatch,
    callers: &[Caller],
) -> parcel_runtime::plan::Result<DiffReport>;

c supplies compiled rules, batch supplies the rows to test, and callers supplies each context. The batch is conformed to the contract’s row schema. Schema conversion and execution failures return RuntimeError; differing evaluation results are recorded in the returned report.

For existing artifacts, batch, and caller values:

let report = parcel_runtime::differential::differential(
    &artifacts, &batch, &[caller],
).await?;
assert!(report.passed(), "{:?}", report.mismatches);

`DiffReport`#

Field	Type	Meaning
`rows`	`usize`	Rows tested.
`callers`	`usize`	Caller profiles tested.
`evaluations`	`usize`	Individual comparisons performed.
`both_errored`	`usize`	Comparisons where both evaluators rejected the input, such as division by zero.
`mismatches`	`Vec<Mismatch>`	Comparisons whose results differ.

impl DiffReport {
    pub fn passed(&self) -> bool;
}

Returns true when mismatches is empty. It can be true when both_errored is nonzero: matching errors count as agreement.

`Mismatch`#

Field	Type	Meaning
`rule`	`String`	Rule or calculated-field identifier.
`caller`	`String`	Caller label for this comparison.
`row`	`usize`	Zero-based row index in the tested batch.
`reference`	`String`	CEL interpreter result as text.
`datafusion`	`String`	DataFusion result as text.