bugbug
bugbug copied to clipboard
mozilla
Platform for Machine Learning projects on Software Engineering
Corrected 'are adding lines is the first' to 'are adding lines as the first'. ```suggestion # means that we are adding lines as the first ``` _Originally posted by @Copilot...
Implement a hierarchical subagent system where the main code review agent can **delegate** function/class analysis to a specialized subagent that can recursively invoke itself to analyze dependencies. This should remove...
E.g. many users don’t want to see comments that are review tips, some other users want to see them. This will help considerably reduce noise.
See [Cheat Sheet: Mastering Temperature and Top_p in ChatGPT API](https://community.openai.com/t/cheat-sheet-mastering-temperature-and-top-p-in-chatgpt-api/172683) For example, make comment generation more creative, but keep filtering phase strict. ea7cd01f20217521db07846ea161a1b5b4a3e698 was the first step for this. This...
An ensemble to generate more comments, with a filtering step to only output actually unique ones. For example, GPT-4o + Mistral. Or, GPT-4o with the patch in the default order...
As suggested by @asutherland: > for stuff like [this manual expansion of nsIWidget's subclasses that says searchfox doesn't have the functionality](https://github.com/mozilla/bugbug/blob/c98b7078911f442a99c22407bcd4ffc9beaa59e8/bugbug/code_search/searchfox_data.py#L199-L206), searchfox does have that data. [inheritance-diagram:'nsIWidget' depth:4](https://searchfox.org/mozilla-central/query/default?q=inheritance-diagram%3A%27nsIWidget%27%20depth%3A4) is a...
@marco-c: > For severity classes, we could define them as either numbers, or words (e.g. “very important”, “important”, “not important”)
> Could you check if the number of comments asking to verify something exists are more or less the same before/after the change (by using a LLM, even if not...
> We should maybe add one _Originally posted by @marco-c in https://github.com/mozilla/bugbug/pull/5005#discussion_r2080509990_ AND > Here we should reuse the old generated explanations when there are some _Originally posted by @marco-c...
Similar to #5002 Currently, the `suggestions_feedback` collection in the vector database does not have the creation data. Thus, we cannot filter based on the date. We need to include that...