[FATAL] All input boxes are broken, not just ollama or lmstuido

[System233]

Which version of the app are you using?

v3.3.17

Which API Provider are you using?

OpenAI Compatible

Which Model are you using?

any

What happened?

All input boxes will lose data when the input speed is too fast, and the model configurations will overwrite each other when switching between profiles.
I see that #931 only modifies the code related to ollama and lmstudio. Are you sure these modifications can fix all input boxes?
In fact, there are problems with the configuration input boxes of all providers, and the configuration profile is completely unusable.

Steps to reproduce

The following two operations can reproduce the fault

• Quickly type in multiple inputs/checkboxes.
• Switching between multiple profiles.

Relevant API REQUEST output

Additional context

• Considering that this issue is so widespread, the settings menu may need to be reworked.
• This issue is somewhat similar to "Match precision" option cannot be adjusted #257. I am not sure whether the wrong value displayed in the settings will become the actual value of the configuration.

2025-02-12.14-02-38-1.mp4

2025-02-12.14-02-38-2.mp4

[mrubens]

Looking at this.

[mrubens]

Do you have a screencast of how you're reproducing this? Can you make sure you're on 3.3.18 and have the latest code? I'd really appreciate it - I've heard people complaining about this but have not been able to repro the OpenAI-compatible one myself.

[System233]

I updated to v3.3.18 and can still reproduce the problem in any provider. I will upload a video later.

[mrubens]

I updated to v3.3.18 and can still reproduce the problem in any provider. I will upload a video later.

Thanks! What operating system are you on btw?

[System233]

I updated to v3.3.18 and can still reproduce the problem in any provider. I will upload a video later.

Thanks! What operating system are you on btw?

I am using Windows 11 10.0.26100.2894 and VS Code 1.97.0 which I just installed yesterday.
I can reproduce this issue on the previous system and other computers.

[System233]

Check this out. I split the video into two because the file is too large.

2025-02-12.14-02-38-1.mp4

2025-02-12.14-02-38-2.mp4

[mrubens]

Thank you! Will take a look tonight / tomorrow morning.

[System233]

The OpenAI Compatible configuration in the video uses an invalid URL. If the URL and KEY are valid, the model ID input box will not work properly.

[mrubens]

The OpenAI Compatible configuration in the video uses an invalid URL. If the URL and KEY are valid, the model ID input box will not work properly.

Can you tell me more about this too?

[System233]

The OpenAI Compatible configuration in the video uses an invalid URL. If the URL and KEY are valid, the model ID input box will not work properly.

Can you tell me more about this too?

Because there is a function to get the model list in the OpenAI Compatible API, when the URL and KEY are valid, the MODEL ID input box will be repeatedly overwritten by the search results, even if enter it at a very slow speed.

It may be related to this https://developer.mozilla.org/en-US/docs/Web/API/Element/compositionstart_event. In short, may need to reduce the update frequency of the settings menu and update or save the settings when there is no user input for a period of time.

Another related issue:

When the API is returning content/ROO is editing code, the settings menu will also be repeatedly rolled back/overwritten, and no settings can be modified.

[aman-4321]

Hey I spent 10-20 minutes reproducing the issue just as you told , writing very fast and switching very fast between options but was not able to reproduce this same issue.

I am on arch linux and the all input boxes works fine for me

[System233]

Hey I spent 10-20 minutes reproducing the issue just as you told , writing very fast and switching very fast between options but was not able to reproduce this same issue.

I am on arch linux and the all input boxes works fine for me

After a brief code review, I believe the high latency of useExtensionState->handleInputChange->postMessage on Windows has amplified this issue (the unordered nature of Async tasks), but this doesn't mean the issue doesn't exist.
A similar situation occurred in the SanjulaGanepola/github-local-actions extension, where the VSCode backend was running on a remote Linux machine.

[System233]

Next I want to make the following changes:
x. Load all profiles at once on the settings page.
x. Separate the "Done" button into "Save" and "Cancel" buttons.

1. Automatically save when switching profiles, and prevent switching when the profile is invalid.