Recently, Github Copilot makes a big release, and I heard a lot of people saying they have came back to Github Copilot. Today I make a Github Copilot subscription renew, and tried a lot coding, I don’t find anything new to convince myself to switch back to Github Copilot. Here are my some thought about Github Copilot and Cursor.
After the initial release of Github Copilot, It really help me a lot in my daily coding. But there are always something I didn’t get from Github Copilot.
1、Code auto completion does not work in the middle of codes. For example:
If I move the cursor in the front of display, and try to type some other style, the code auto completion will not work there. Only I type something in the end of the line, or switch to the new line, the code auto completion will work again. This really sucks. And for a developer, we don’t always proceed the code further, at least more than half the time, we are changing the codes.
2、The code auto completion speed matters a lot. If I always need to wait sometime to get some code suggestion, I wasted a lot of time.
As to the chat and model, which seems to be considered the most important thing for the AI coding tool, honestly, I don’t heavily rely on this.
For faster code auto completion and middle-line code auto completion, I almost used all well known AI code assistant tools, and finally found SuperMaven and Cursor.
SuperMaven really have the unbelievable speed for the code auto completion and can do the middle-line code auto completion. Unfortunately, after their announcement of funding raising, they seemed have stoped the plugin development and switched to their own editor.
Compared to SuperMaven, the Cursor seems satisfied my needs perfectly:
1、The speed for the code auto completion is faster than Github Copilot
2、Cursor can do the middle-line code auto completion
3、The cursor tab and prediction is really magic while I do the code refactor, which is lack in Github Copilot and SuperMaven(it seems have this, but not so good)
As to the chat model, I really can’t find any differences among Github Copilot, SuperMaven and Cursor. Even they don’t have the chat model, we can still use any available models in the website.
So for me, the Cursor is still far ahead of other AI coding assistance tool.
Why do so many people said the Github Copilot will catch up Cursor and beat up the Cursor? Do I miss something?
For the code auto completion, all the tools currently seems not good enough:
It seems that all the tools make the code auto completion based on a short context, maybe codes surrounding the edit point, which is not enough.
The SuperMaven’s code auto completion seem better, because they have a large code context window. It seems that they will consider the files you are opening, and give the code prediction.
I don’t know what context the Cursor will provide when it gives the code prediction.
Cursor provides the index functionality, but these indexed codes are not provided for the code completion but for the compose or chat.
The perfect code auto completion for me is:
The tools can index all the project code base, and give code prediction based on these indexes. For example, If I defined some constants in a file named constants.js, when I want to use these constants in other files, the code auto completion should provide constants in the contants.js, rather than give some unrelated prediction.