Hashline: Edit by Content Hash #5474
AviseOffical
started this conversation in
General
Replies: 2 comments
|
this appears to be the biggest issue for the agent, it keeps having to rethink after failed edits resulting in a lot of unnecessary work and delay. I am running Pi with KAT Coder 2.5 Dev no quant on a 128gb Blackwell. |
0 replies
|
I also think this is indeed quite good, and I've already supported it in my own coding agent.
This implementation was the original approach of oh-my-pi, but now oh-my-pi's implementation of hashline has become very complex—it uses Lark to define a DSL. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
I saw this concept implemented in the oh-my-pi fork, and it seems like a massive quality-of-life improvement for the edit tool. The premise is:
“Perfect edits, fewer tokens. The model points at anchors instead of retyping the lines it wants to change, so whitespace battles and string-not-found loops just stop happening. Edit a stale file and the anchors diverge — we reject the patch before it corrupts anything.”
Because LLM string-matching and indentation failures are so common, is there any plan to bring hash-anchored editing upstream into pi's core file-editing tools? Or would the maintainers prefer this remains a third-party extension?
All reactions