ponder-lab / Hybridize-Functions-Refactoring / #2549

82%

main: 83%

Pull #792

github

Cover the per-element bottom and the tuple branch

Driven by the patch-coverage report on #792, mapped to changed lines via a
local JaCoCo run: `reportElementDrop` was entirely uncovered (no fixture
bottomed at an element position), as was `Util.isSequenceType`'s tuple branch.

- `testInputSignatureContainerElementDtypeConflict`: two call sites pass
  singleton lists whose elements are `float32` and `int32`; the form is
  modeled, the reduction reaches the element position, and the dtype union
  bottoms there with `HETEROGENEOUS_DTYPE` and an element-citing diagnostic.
  Covers the element-blocking path and the first `reportElementDrop` arm.
- `testInputSignatureTupleParameter`: a singleton tuple reduces to the same
  nested rendering a list does. Probed first on TF 2.9.3: `input_signature`
  structure matching does not distinguish list from tuple (a list-shaped spec
  accepts a tuple argument and vice versa), so the `[...]` rendering is sound
  for tuple-valued parameters.

The remaining uncovered arms are the element-level `UNKNOWN_DTYPE` and
`HETEROGENEOUS_SPARSITY` mirrors (requiring an analysis-⊤ or sparse element
inside a container fixture) and `extractContainerElements`' defensive returns
(mixed container/non-container values, non-contiguous indices, and two
inconsistent-state guards).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K8tXuyy3YWhM7BzvmCxKAn

128 of 138 new or added lines in 5 files covered. (92.75%)

2323 of 2816 relevant lines covered (82.49%)

0.82 hits per line