community: add include_comment_forest to RedditSearch

[hemati]

• Description:
  Introduces a new include_comment_forest boolean parameter in the RedditSearchSchema and RedditSearchRun tool. When set to True, this fetches and recursively parses the entire nested comment tree using _parse_comment_forest, and formats it using _format_comment_forest. Updated run and results methods to support displaying the full comment hierarchy.

• Issue:
  (No specific issue to reference.)

• Dependencies:
  None.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Feb 20, 2025 8:24am

[ccurme]

Do you think it makes more sense to initialize the tool with include_comment_forest, instead of requiring the model to populate True/False?

Note that as-is, this PR would require all models using this tool to populate the new parameter. Some models will also start filling out True, will pull a huge comment history, and cause problems for maintainers. IMO this should be an opt-in feature.

[hemati]

Thanks for the thoughtful feedback!

My original reason for having include_comment_forest as a parameter in each request was to let the model/agent decide on a per-call basis whether it needs the full comment tree. For example:

1. Selective Deep-Dive
   • The agent might first retrieve only post metadata. If it finds something important, it can dynamically request the full comment forest (include_comment_forest=True) on a second call.
2. Contextual User Requests
   • If a user specifically asks, “What are people saying about X in the comments?” the agent can enable include_comment_forest=True only when needed, instead of always pulling comments.
3. Adaptive Workflows
   • The agent can decide on the fly based on resource limits or user instructions, rather than one static setting for all calls.

However, I understand your concern that this could lead to unintended large data pulls if an LLM or user call sets include_comment_forest=True by default.

Proposed Compromise:

1. Keep include_comment_forest in the request schema, defaulted to False.
2. Add a constructor parameter, for example allow_comment_forest, which defaults to False.
3. If allow_comment_forest is False, the tool will ignore any True value for include_comment_forest in the call, effectively preventing deep comment pulls.
4. If someone needs dynamic control, they explicitly set allow_comment_forest=True so the agent can opt in on a per-call basis.

This way:

• Maintain flexibility: Developers who want dynamic deep dives can enable them at runtime.
• Protect performance: If maintainers want to prevent accidental large data usage, they leave allow_comment_forest=False.

Would that address both the need for runtime flexibility and your concern about possibly large or unintended data pulls?

[ccurme]

Thanks for the update. This could work but it would be confusing for a model to specify include_comment_forest with no effect. You could dynamically select the schema. But I think ideally we get this out of langchain-community so we can test and version changes like this properly (there also appear to be no tests on this tool, which makes reviewing new features difficult).

I believe you contributed the Discord integration, would you be interested in doing something similar for Reddit?

[hemati]

Thanks for the update. This could work but it would be confusing for a model to specify include_comment_forest with no effect. You could dynamically select the schema. But I think ideally we get this out of langchain-community so we can test and version changes like this properly (there also appear to be no tests on this tool, which makes reviewing new features difficult).

I believe you contributed the Discord integration, would you be interested in doing something similar for Reddit?

Thanks for the feedback! Yes, I did contribute the Discord integration. I’d be happy to work on a similar, fully-tested and separately versioned Reddit integration if that’s the direction you prefer.

Do you want to close this PR?

[ccurme]

Awesome, yes. Will close for now. Thank you!