This bug has been present for a week, coinciding with what seems to be behind the scenes changes at Cursor to use fewer tokens.
If you have a code file with say 2000 lines of code and you ask Cursor to migrate specific functionality, comprising about 500 lines of code, from the original file to its own file, Cursor cannot currently do this.
It will write the migrated code correct to the new file, usually.
But when it comes to removing the code from the original file so it’s not duplicated, it cannot do it. Instead it will do one of four things:
- Fail outright with ‘no changes made’ or ‘error calling tool’
- Add a reference to the new file in the original but leave all the migrated code there.
- Add a reference to the new file and delete every other line of code in the original file.
- Attempt to recreate the entire original file from scratch, minus the migrated code, which ends up being a mess with huge chunks of the original code missing.
Even splitting the migration into 2 separate requests doesn’t work. The first request successfully writes the new code to its own file. Then in the next request we ask it to remove the migrated functionality from the original file. And here is the mess that occurs when you do that:
Have Cursor devs made any attempt to duplicate this issue? Many people experiencing it and yet it persists.