ponder-lab / Hybridize-Functions-Refactoring / #3884

90%

main: 90%

push

github

Release the engine's per-builder tensor caches when the analysis caches are cleared (#1191)

The engine clears its per-builder tensor caches at the end of its own
analysis, but the dataset-component lookups made afterward while inferring
parameters write fresh entries keyed on the builder. Their values strongly
reach the weakly held builder, so the entries never expire, and every
analysis's builder, call graph, and class hierarchy stayed alive.

The processor's clearCaches() now calls TensorGenerator#clearCaches on each
builder it holds before forgetting them, which also covers the evaluator's
per-project clearing. The test harness keeps the processors each test
creates and clears their caches after the test.


Claude-Session: https://claude.ai/code/session_01NTnpjPjJj2aZy1ZkUD1eAn

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

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

5857 of 6543 relevant lines covered (89.52%)

0.9 hits per line