ponder-lab / Hybridize-Functions-Refactoring / #3869

89%

main: 90%

Build Type

push

github

Committed by web-flow

Commit Message

Examine only a container's elements for primitives (#1183)

* Build against Ariadne's 0.52.111-SNAPSHOT

The snapshot carries the engine's container-elements API, which the container analysis is about to read. Tycho's wrap instructions come back so the timestamped snapshot wraps under its base version.

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

* Ask the engine for a container's length, element types and None elements

The container analysis decided a container's length, the tensor types at each position, and whether an element may be None by decoding the engine's object catalogs, its storage channels and its field keys. Reading the length off the catalog's constant indices gets it wrong for a list that appending, a comprehension or extend builds: such a list read as length 1 whatever it holds at run time, and the nested specification written for it rejected the call. It now reads all three from the engine's ContainerElements, which gives no length where the run-time length may differ, and the parameter then declines with TENSOR_CONTAINER_UNSUPPORTED. The nested-container check reads the engine's element keys as well.

Two pins move for a reason other than a corrected length, and each carries a TODO: the engine gives no length to a list that is the left operand of a +, as the front end lowers += to the same operation, nor to a tuple a library summary allocates.

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

* Ask the engine whether a container holds a tensor

Whether a value reaching a parameter is a container of tensors was answered by sweeping the tensor-type analysis for field keys of a container type. It now asks the engine for the container's element types, over every element key, and the container analysis reads no field key at all.

The two answers differ for a dict keyed by a ... (continued)

Coverage Stats

3 of 3 new or added lines in 1 file covered. (100.0%)

2 existing lines in 1 file now uncovered.

5766 of 6446 relevant lines covered (89.45%)

0.89 hits per line