run-llama / llama-agents / 35542187147

85%

push

github

Support unhashable workflows in serializer caching (#762)

`Runtime._serializer_cache` uses `WeakKeyDictionary`, which still hashes
workflow instances and rejects subclasses that set `__hash__ = None`.

The runtime now uses one `WeakIdentityKeyDictionary` for workflow
tracking and serializer caching. It keys entries by object identity,
verifies the live referent, and removes entries when the workflow is
collected. Direct coverage checks equal unhashable keys, replacement,
removal, and garbage collection. Workflow coverage includes the public
`workflow.run()` path and a real
`llama_index.core.agent.workflow.ReActAgent` run with its default
context.

Fixes #761

42 of 42 new or added lines in 1 file covered. (100.0%)

18308 of 21565 relevant lines covered (84.9%)

1.14 hits per line