feat(chat): filter namespace messages from history if it exists in metadata VSCODE-611

[gagik]

Scrub the namespace messages from the history when we have identified and are building the messages to send to the ai. We already include the namespace in the query assistant message. We might want to move these to be additions to the user's message instead (in both query and schema).

Description

Checklist

• New tests and/or benchmarks are included
• Documentation is changed or added
• I have signed the MongoDB Contributor License Agreement (https://www.mongodb.com/legal/contributor-agreement)

Motivation and Context

• Bugfix
• New feature
• Dependency update
• Misc

Open Questions

Dependents

Types of changes

• Backport Needed
• Patch (non-breaking change which fixes an issue)
• Minor (non-breaking change which adds functionality)
• Major (fix or feature that would cause existing functionality to change)

[Anemy]

Nice, left a couple code quality suggestions.

[Anemy]

This function is starting to get a bit long and hard to follow all of the things happening in it.
It already has the // eslint-disable-next-line complexity which is usually an indicator that we should break it into a few functions, even if it comes with a slightly hit to performance (like running through the messages multiple times).
Should we do that now? Break this function into multiple parts where each is doing a certain thing? That'll also make it more easily unit testable.

[alenakhineika]

More like a question about this area in code and maybe a bit of a request :) if you will refactor this code, could we split the history into something like getUserHistoryMessages and getAssistantHistoryMessages. When testing this functionality I found it difficult to print the history content because it is already wrapped into the vscode.LanguageModelChatMessage format when it is returned here: https://github.com/mongodb-js/vscode/blob/main/src/participant/prompts/promptBase.ts#L130

Maybe, it could be something like:

const messages = [
  vscode.LanguageModelChatMessage.Assistant(this.getAssistantPrompt(args)),
  vscode.LanguageModelChatMessage.Assistant(this.getAssistantHistoryMessages()),
  vscode.LanguageModelChatMessage.User(this.getUserHistoryMessages()),
  vscode.LanguageModelChatMessage.User(prompt),
];

Or something like that. The idea here is that we have some unformatted string value to print to see what message we send to the model.

[Anemy]

The ordering of the messages here is important, although the last message will always be the user prompt and the first is the assistant prompt, the user history and assistant history aren't sequential.
We do currently log information about the messages we are sending to the model:

vscode/src/participant/participant.ts

Line 143 in d5a9345

messages: modelInput.messages.map(

@alenakhineika we can add something there that will log them in their entirety, and not just the metadata, if an environment variable is set. How does that sound?

[gagik]

I am also dealing changing getHistoryMessages in VSCODE-632 so I think it'll be best to deal with some refactoring for it in the PR there.

[gagik]

For the sake of keeping this simple to review, going to merge and do any greater potential refactoring work in the PR for VSCODE-632.