Hey @pwilkin Thank you for taking the time to solve this issue.

It seems like some unit tests are failing, can you take a look?

Let me know if you have any questions!

Yeah, sorry, I did those unit tests on an old version, then saw they were deleted, didn't reallize they just moved to a new place :> will commit.

BTW: any idea why the webView tests are duplicated? The ClineProvider and webViewMessageHandler tests for the requestRouterModels are virtually the same.

@daniel-lxs I think at this point I'm pretty confident this was the wrong route to take :)

Instead, I think the proper approach to solve this is:

• store info about whether a model's state was updated when loaded somehere (/api/providers/fetchers/lmstudio.ts maybe? basically we need to distinguish if we loaded from LLMInstanceInfo vs LLMInfo)
• in useSelectedModel.ts, on the getSelectedModel call, force a live reload of the model data if we don't have its live data yet (that means basically doing:

const client = new LMStudioClient({ baseUrl: lmsUrl })
const model = await client.llm.model(apiConfiguration.lmStudioModelId)
const modelInfo = await model.getModelInfo()

and then updating the cache + returning the new modelInfo processed via parseLMStudioModel.

This way, we don't wait until the first request is made to get the data and we don't need to do all that hacking + refreshing of the state.

The only thing I'm afraid about is a case when getSelectedModel is called in a weird UI state, since the operation will be blocking and loading a model in LM Studio can take a looooong time (on slow setups + big models I'd say even 2-3 minutes). The alternative is doing this on upsertProviderProfile in ClineProvider.ts when provider is LM Studio - but that means the user won't have a current view of the model's context size when selecting models in the LM Studio configurator.

Let me know what you think!

@daniel-lxs Added proper fix in #6183, closing this one.