ponder-lab / Hybridize-Functions-Refactoring / #3790

89%

main: 89%

Build Type

Pull #1158

github

Committed by khatchad

Commit Message

Read a dataset component as a container, and advance pins to Ariadne's master

A dataset element's structured component, such as the inputs Keras's train step hands call when fit is given from_tensor_slices(((x, mask), y)), is a tuple at run time that no allocation stands for, so no instance key the parameter points to was a container, and the parameter got no nested specification. The engine now reads such a component through ContainerElements.ofValue (ponder-lab/ML#1087). A component that is a tuple of known length, or may be one, is a container, and the existing rules apply to it: a nested specification only when it's complete with an agreeing length. A component known to be one tensor is not a container. The pair reaching call, and the function call passes it to, now reduce member by member, which admits the pair; before, the inputs were typed as one tensor of several shapes, and that specification rejected it.

The merged pins from main record today's behavior on the released engine, each with the value a pending engine change gives. This branch builds against the snapshot that carries those changes, so each moves to the value its TODO names.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NTnpjPjJj2aZy1ZkUD1eAn

Pull Request Pull Request #1158: Ask the engine for a container's length, element types and None elements

Coverage Stats

72 of 72 new or added lines in 2 files covered. (100.0%)

7 existing lines in 2 files now uncovered.

5447 of 6116 relevant lines covered (89.06%)

0.89 hits per line