So I thought I’d just make a thread to let the owners know the code auto fixing is getting significantly worse as time/updates go on. RIght now it’s so bad that it’s just adding the new code fix directly below the existing code. Before this it was near enough perfect and was one of my favourite features, but right now it’s gotten itself into a unusable state.
2 Likes
Appreciate the report. When you say code autofixing, what feature do you mean? The Command-dot “Fix with AI” or something else? A screenshot could help. Also, are you using GPT-4?
Hey, thanks for the prompt response. Specifically I’m referring to the option you get on a generated code snipped in the chat window with a play icon that when clicked, fixes/adds to your code for you. Also as for a screenshot, I’ll be sure to grab one the next time it happens, has only been a noticeable issue for the last 2-3 days.
Just happened again, had a code fix presented to me, selected for it to be applied to my code and it ran through the code and then add the code fix to the bottom of the code instead of replacing the code. Seems to be mostly doing this in PHP as I can’t recall it doing it many times in python.
Also another glitch I regularly notice is when I choose for a code fix to be applied, it will scan through the code and not fix anything and then when I go to fix it manually it will show up with the code fix when I start typing on the same line and the line will suddenly become green. Almost as if the code fix is behind the code and when I start typing on that line it becomes present, that glitch has been around for a week or so now, that happens for both python and PHP for me. Also yes I’m using GPT-4 only.
@truell20 This is still a ongoing issue for me, sometimes the cmd+k method refuses to add a fix. Or, when I use the “Apply to current file” function, it adds the fix to the end of my code, therefore duplicating the start/my whole code over again. This seems to only happen for me in php also to add, not seen these issues in python.
Yes Cursor also re-adds entire code sections below existing code and like… re-write the entire file with duplicate. Often the file gets past 250 lines and then the file to too large for it to edit to fix its own mess. Really hope Cursor can edit larger than 250 line files soon. It’s getting painful.
Also getting quite a few instances of adding existing code at the end of the file.
Also code snippets in chat in the format:
@@ -155,10 +147,2 @@
with - before lines to be removed, and + before lines to be added.
As in, the chat is showing what I assume is the internal formatting before the change in the actual file.
Yes Cursor also re-adds entire code sections below existing code and like… re-write the entire file with duplicate. Often the file gets past 250 lines and then the file to too large for it to edit to fix its own mess. Really hope Cursor can edit larger than 250 line files soon. It’s getting painful.
Also getting quite a few instances of adding existing code at the end of the file.
Also code snippets in chat in the format:
@@ -155,10 +147,2 @@
with - before lines to be removed, and + before lines to be added.
These are both things I’m suffering with first hand, hoping further updates will improve the situation. I seen the recent updated format of code suggestions like this “@@ -155,10 +147,2 @@” and assumed it was a new fix from the cursor side, but then every time it appears it has never fixed the code correctly once so I assume that’s coming from OpenAI’s side. Hoping Cursor devs can take a closer look to see what’s happening here as it’s becoming more and more apparent and more and more of a headache.
1 Like
I’m thinking of 2 specific forum thread/Github issues are in order, because 2 two issues here can easily get lost from the subject title, and not much confirmation about the 2 issues being acknowledged by the Cursor team…
Both issues are absolutely acknowledged! This is on the model side and we are heavily tuning things to make this not happen. We are actually invested in this!
Can either of you all give us a couple of code files where this happens and we will add it to our eval so it doesnt happen again.
1 Like
That’s great! Glad to know it’s being looked into, I took a short break of coding but will be back to it today and will post any code examples I come across of it doing either of the issues mentioned:)
Glad it’s being looked at. Will look to see if I can find any recent examples.
Haven’t seen it as much the last few days. So that’s good.
The 250 line limit for the notebook (interpreter mode) - is that something on your to-do list as well?