AI Feature: Predictive debugging #6

tichmangono commented 2025年12月14日 15:21:56 +01:00

Predictive debugging (as you write) - before executing code , analyze it against current runtime to predict issues, some IDE's do this but its surface level .. with your access to all those state values and docs, can do more and display warnings in real time before execution. And AI can enhance this process

Predictive debugging (as you write) - before executing code , analyze it against current runtime to predict issues, some IDE's do this but its surface level .. with your access to all those state values and docs, can do more and display warnings in real time before execution. And AI can enhance this process

sczi commented 2025年12月15日 04:01:15 +01:00

Interesting, I haven't seen this. Do you have links to what other IDE's are doing with this? I searched predictive debugging and found this but it looks like something different.

I'm wary of it happening before executing code, as I want instant feedback and LLMs take a bit to respond. Or of it happening as I write, as what I'm writing is unfinished, any issues identified at that point could just be distracting noise of some issue that will be fixed when I finish writing. But I think maybe a way to do this would be to have it after I evaluate the code, start to analyze in the background and report back some seconds later if it finds an issue like some edge case I might not have handled?

Interesting, I haven't seen this. Do you have links to what other IDE's are doing with this? I searched predictive debugging and found [this](https://blog.jetbrains.com/dotnet/2023/07/27/introducing-predictive-debugging-a-game-changing-look-into-the-future/) but it looks like something different. I'm wary of it happening before executing code, as I want instant feedback and LLMs take a bit to respond. Or of it happening as I write, as what I'm writing is unfinished, any issues identified at that point could just be distracting noise of some issue that will be fixed when I finish writing. But I think maybe a way to do this would be to have it after I evaluate the code, start to analyze in the background and report back some seconds later if it finds an issue like some edge case I might not have handled?

tichmangono commented 2025年12月16日 01:34:40 +01:00

Point taken here on the delay this would introduce. Probably good to assume the user is correct until you find an error. When I said some IDEs do this, I was more thinking about live linting, and type safety check like typescript and pydantic... overall, probably not worth pursuing since it pulls against the whole idea of instant feedback and autoreload

Point taken here on the delay this would introduce. Probably good to assume the user is correct until you find an error. When I said some IDEs do this, I was more thinking about live linting, and type safety check like typescript and pydantic... overall, probably not worth pursuing since it pulls against the whole idea of instant feedback and autoreload