ponder-lab / Hybridize-Functions-Refactoring / #1407

71%

main: 72%

push

travis-pro

Split `Parameter.isTensorTyped` into `classifyAsTensor` + `isTensor` (#500)

* Split `Parameter.isTensorTyped` into `classifyAsTensor` + `isTensor`.

Closes #498.

`isTensorTyped` was named like a boolean query but mutated state (via `inferTensorTypes` populating the tensor-types cache). The combined classifier+query shape made the cache write a hidden side effect of a method whose name suggested pure read.

Refactor:

- Rename `Parameter.isTensorTyped` (both overloads) → `Parameter.classifyAsTensor`. Return type changes from `boolean` to `void`.
- Add `private Boolean isTensor;` field. The classifier sets it before each `return`.
- Add `public Boolean isTensor()` getter as the cache reader. Returns `null` if the classifier hasn't run.
- Update `Function.inferTensorParameters` to invoke `classifyAsTensor` then read `isTensor()`.

Backward-compatible behavior: `Function.inferTensorParameters` still sets `hasTensorParameter = TRUE` when any non-self parameter classifies as tensor-typed.

Two regression tests pin the classifier→query contract in both directions:

- `testParameterClassifyAsTensorContract` (forward): tensor parameter ⇒ `isTensor() == TRUE`; non-tensor parameter ⇒ `isTensor() == FALSE`; ∃ tensor parameter ⇒ `Function.getHasTensorParameter() == TRUE`.
- `testParameterClassifyAsTensorContractReverse` (reverse): `Function.getHasTensorParameter() == FALSE` ⇒ no non-self parameter has `isTensor() == TRUE`. Uses a function name (`f`) outside the speculative-context regex so the fallback doesn't mask the relationship.

`./mvnw verify` green: 473 tests, 0 failures, 0 errors, 11 skipped (+2 from baseline).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Pin more `Parameter`/`Function` classification invariants (#498).

Adds two more regression tests and tightens the existing forward test:

- `testSelfParameterClassification`: pins `self.isSelf() == TRUE ⇒ self.isTensor() == FALSE`. Self parameter must classify as non-tensor ... (continued)

6 of 8 new or added lines in 2 files covered. (75.0%)

1132 of 1586 relevant lines covered (71.37%)

0.71 hits per line