Details

Disposing a language model provider removes its models from LanguageModelsService without firing the model-change event. Consumers keep their previous model catalogs until another change occurs.

Change

Fire the vendor's model-change event after removing the provider, model groups and listener.

Before

When registering, discovering and disposing a 37-model provider over 37 cycles, CodexContinuationService keeps 37 model entries after disposal. The leaked objects are plain data records and do not appear in named-function-count3; the complete process-memory chart below provides context for the same scenario.

After

The 37 stale consumer entries are gone: exact model records stay at zero. Other heap growth remains; the process-memory totals below are context rather than an estimate of savings from these small objects.

Test Video

_{AI disclosure: Model: GPT-6.1-Sol. Worktime: 16 min}