For approximately 10+ days, Cursor has been affected by a bug that prohibits it from editing files properly. This issue manifests itself in preventing the edit_file tool from doing what the AI has instructed it to do to an existing file, causing it retry in different ways before eventually causing the AI to attempt to recreate the entire file from scratch.
Hey Cursor, this is costing you money. When Claude 3.7 unnecessarily recreates an entire 1200+ line file from scratch that’s a LOT of tokens.
Fix your bug please and save yourselves some money, and us a whole lot of aggravation.
I understand your frustration. Looking at the history, it seems the edit_file tool’s changes from commit 8d6cb4e either weren’t properly applied or were overwritten. This is concerning and suggests unreliability in the tool.
Let me explain what I see:
The commit 8d6cb4e claims to have removed all high-level methods, but clearly didn’t succeed
The high-level methods like create_stem, get_stem, etc. are still present
The edit_file tool appears to be unreliable for making large-scale changes
You’re right - I should not continue using the edit_file tool if it’s not reliably making changes. Instead, I suggest:
You could manually remove the high-level methods
Or you could use your IDE/editor to remove them
Or we could try a different approach like generating a patch file that you could apply
Would you like me to:
List exactly which methods need to be removed (without trying to edit the file myself)?
Help you identify the low-level methods that should remain?
Verify that the remaining low-level methods match the Java client exactly?
Please let me know how you’d like to proceed, and I’ll avoid using the unreliable edit_file tool.
Hey, we did have a bug that was causing the ‘No Changes Made’ status to come from file edits more often than they should have, and a fix should be out in the latest version already!
When the AI sees it has successfully edited the file, it shouldn’t then try it’s abstract methods to edit it.
@danperks I can confirm this has been much better since yesterday afternoon. Kudos. FWIW it would be nice to hear that you guys are working on bugs like this before they get fixed so many of us aren’t wondering if anyone cares and if a fix is ever coming. There were largely crickets for a while on this.
Hi @danperks this issue now manifests itself in “error calling tool” rather than “no changes made”. Seems like the same thing, just called something different now. See req id 9ad75725-734b-4ad5-86d5-1bd3b910adfb
