Add Error Animation in Code Cell for Jupyter-AI Errors

[Darshan808]

References #1144

This PR introduces the error animation functionality to code cells for jupyter-ai. It aligns jupyter-ai's error reporting with JupyterLab's completer error animation feature.

• Added an error field in responses from jupyter-ai during error generation.

Testing

• Test is added to validate the error field emission.
• Currently seeking guidance on how to simulate error scenarios for testing.

Feedback on the testing approach is appreciated, particularly on methods to programmatically raise errors during tests.

Screencast

Simulated by raising custom exception from server side

ai-error.mp4

[krassowski]

I think you could override handle_stream_request in the Mock class to conditionally raise an exception (conditionally on arguments passed in to the mock?)

[andrewfulton9]

I think you could override handle_stream_request in the Mock class to conditionally raise an exception (conditionally on arguments passed in to the mock?)

I think you will also need to bump the version of jupyterlab/completer to 4.3.0 if the main branch of jupyter-ai isn't on that version yet. It wasn't when I looked at this issue.

[dlqqq]

@Darshan808 Welcome to Jupyter AI! I love seeing new contributors. Thank you for getting started on this feature request.

I've left a few suggestions below. In addition, I have a couple of questions:

1. Could you add a screenshot to this PR?
2. The original issue (Show error animation in code cell when error is triggered  #1144) shows an icon when the completer fails. How is the icon being added?

[dlqqq]

With this PR, it seems like whether an inline completion resulted in an error can be determined in 2 different ways:

1. An error occurred if only one InlineCompletionItem is returned and InlineCompletionItem.error exists.

2. An error occurred if InlineCompletionReply.error exists.

   • This attribute takes the type CompletionError, which is very similar to the InlineCompletionError model that you added. CompletionError is already defined here on line 49.

I suggest that we:

1. Have a single source of truth and drop either InlineCompletionItem.error or InlineCompletionReply.error. That way, there's only one way to determine if a completion reply has an error.

2. Drop either InlineCompletionError or CompletionError, as the models are very similar and can probably be unified.

[dlqqq]

Currently seeking guidance on how to simulate error scenarios for testing.

You can modify MockProvider to take an raise_exc param. If "raise_exc": True is set in lm_provider_params, then MockProvider should raise an exception in its _call() method. For example:

def _call(self, *args, **kwargs):
    if self.raise_exc:
        raise Exception("Test exception")
    else:
        return super()._call(*args, **kwargs)

In LangChain, LLM._call() is (usually) the method which is called when a non-streaming reply is requested. If this doesn't work, you can try following the above steps for the _acall() method instead.

[Darshan808]

Thank you for your feedback and suggestions, @dlqqq

Have a single source of truth and drop either InlineCompletionItem.error or InlineCompletionReply.error. That way, there's only one way to determine if a completion reply has an error.

Yes, this was redundant, so I dropped InlineCompletionItem.error since InlineCompletionReply.error already provides more context about the error.

The original issue (Show error animation in code cell when error is triggered #1144) shows an icon when the completer fails. How is the icon being added?

Whenever we return an item with the error field containing a message, JupyterLab's completer has implemented the necessary code to display that animation. So, I don't think we need to reimplement it here. Simply returning a completion with the error field does the job.

Could you add a screenshot to this PR?

Sure, I am adding a clip.

In LangChain, LLM._call() is (usually) the method which is called when a non-streaming reply is requested. If this doesn't work, you can try following the above steps for the _acall() method instead.

This worked perfectly.

[Darshan808]

A custom error can be raised here before sending the reply.
https://github.com/jupyterlab/jupyter-ai/blob/7c1c8c0ac2c78df260b965ad416c190c0108bd2c/packages/jupyter-ai/jupyter_ai/completions/handlers/default.py#L12C1-L16C26

[Darshan808]

The newly added test passes locally but fails on CI. Looking into it