ponder-lab / Hybridize-Functions-Refactoring / #4198

90%

main: 90%

Build Type

push

github

Committed by web-flow

Commit Message

Read only reached code for calls that key weights by name (#1281)

* Read only reached code for calls that key weights by name

The colliding-names shape also read every program method without a
call-graph node and let a name-keyed call there refuse every method.
That worked around the engine not modeling `fit`, `evaluate` and
`predict` calling the callbacks they're passed, and it let dead code
refuse everything: NLPGNN's never-instantiated `CheckpointSaver.save`
refused its GPT-2 `MLP.call` and `SegPosEmbedding.call`. Only reached
code is read now. A save in a Keras callback is missed until the engine
models those calls, which the callback test pins with a TODO, and a
save in a class the program never uses no longer refuses.

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

* Point the callback TODOs at wala/ML#1244

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

* Explain why the callback checkpoint test passes until wala/ML#1244

The callback in testMayBuildLayersCheckpointInCallback has no call-graph
node now, so the test passes whatever the callback saves. Its TODO says
so. The keysNames and getCalleeName targets are never null, and the
unused-class fixture names its class Exporter, not a keyed name.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Coverage Stats

4 of 4 new or added lines in 1 file covered. (100.0%)

1 existing line in 1 file now uncovered.

7075 of 7866 relevant lines covered (89.94%)

0.9 hits per line