Follow-up to #332802. On the fetched /models deployment path, supportsUnifiedCompletions resolves asynchronously, so there is a brief cold-start window (typically the /models round-trip, ~100ms) where the separate github.copilot completions provider is not yet excluded and ghost text can appear before onModelListUpdated fires and re-registers with the excludes.
Fully closing the window would require persisting the last-resolved /models config and hydrating it synchronously at startup (ProxyModelsService currently fetches fresh over the network with no disk cache), so warm starts have the correct excludes immediately. This does not affect local/ExP model configuration, which already resolves synchronously.
Follow-up to #332802. On the fetched
/modelsdeployment path,supportsUnifiedCompletionsresolves asynchronously, so there is a brief cold-start window (typically the/modelsround-trip, ~100ms) where the separategithub.copilotcompletions provider is not yet excluded and ghost text can appear beforeonModelListUpdatedfires and re-registers with the excludes.Fully closing the window would require persisting the last-resolved
/modelsconfig and hydrating it synchronously at startup (ProxyModelsServicecurrently fetches fresh over the network with no disk cache), so warm starts have the correct excludes immediately. This does not affect local/ExP model configuration, which already resolves synchronously.