Next edit suggestion manually

[towry]

For what this feature is about, see https://githubnext.com/projects/copilot-next-edit-suggestions/

I don't want to implement the real-time function, I want to be able to trigger a "next edit suggestion" manually. I think the current plugin can achieve this.

1. In Vim, based on the user's recent edits, let the LLM guess the user's editing intention and list multiple ones.
2. Then let the user choose or directly select the first one.
3. Then, based on the intention selected by the user, provide the "next edit suggestion" using the inline assist function of the current plugin.

What do you think?

[towry]

Usage: suggest edit change based on pasted code.

This code is copy & pasted from another file, i need to change the formdata to formdataMx because in current file, the name is formdataMx.

So i do:

1. gv to select the pasted code.
2. :'<,'>CodeCompanion /buffer based on the pasted selection, refactor the pasted code so that it use variables in current file (ignore my bad English).
3. ga to accept the change (the result is perfect).

[olimorris]

Are you thinking that code on the clipboard could be shared with the LLM when you write an inline prompt? That's a really cool idea.

I'm in the middle of rewriting the Inline Assistant so it will be much more resilient so happy to take any suggestions to improve it.

[olimorris]

Regarding next edit suggestions, my position hasn't changed from #352

[towry]

"manually" means use current codecompanion api to compose a workflow to trigger (by command or keymap) inline assist to achieve the next edit suggestion, not auto trigger while you edit.

I am just proposing an idea here.

If it is a command, like CodeCompanionNext, it could be like:

1. get recent edit changes to compose a prompt to gather a list of user intentions from llm.
2. let the user choose the correct intention, like refactor variable name foo to bar in current file.
3. use the user's chosen intention to compose another prompt and feed it to the inline assist.

The user experience would be improved by only trigger a command instead of selecting code, write a prompt ...

[olimorris]

My overarching thought is, does this really improve a developers workflow? I'm not entirely sure it adds a great deal. Why doesn't the developer know what to do next? Or why wouldn't they prompt the LLM for feedback in the chat buffer and then tag the @editor tool to make the changes for them?

Something like "refactor variable name foo to bar in current file" is unlikely to be an action an LLM is going to suggest unless you specifically include it in the system prompt. At which point, you're better off writing a prompt in the prompt library.

I don't think the UX improvement versus complexity is favourable on this.