I am excited about the planned addition of diff feature because I find overwriting a whole file (working) with a new one that may have wrong codes, a bit frustrating.
It will be nice to only update existing file with the new codes and also modify the parts that need to be changed ONLY.
I am not a developer even though Iv worked closely with them before and know one or two things about coding and debugging, I wish there was a way I could contribute in a way to this project.
Yeah, I actually did such work before for my other side project
There is also open source aider that is pretty good at that GitHub - Aider-AI/aider: aider is AI pair programming in your terminal
I think I or we will bring it over.
But there are also issues, different models would react differently to it.
Like Aider has a leaderboard
You can see a column of edit format, diff or whole.
And you can see that ollama/qwen2.5-coder:32b and o1-mini perform better in Aider evals with whole rewrites then with diffs.
There was proposal before to add “prompt” library and we may need to enable/disable diff mode and change prompts based on model.
In that sense I would first invest in that. Infrastructure for customizable per model prompts or something.
That’s great.
For Aider, I have been wanting to try it out, I even have it as an extension already on my vscode.
When we’re talking about rewrites here, is it expected that models like Qwen2.5 do entire codebase rewrites (especially in oTToDev)?
Or is it just expected that it rewrites specific files it’s working on?
I’ve found that running Qwen2.5 32B with a context of 128k (131072) in oTToDev, it rewrites the entire project for any ask. Wondering if this is what’s part of the diff roadmap, or if this is unexpected behaviour.
Certainly looking forward to the future of oTToDev! It’s amazing what’s been achieved already.