Configure completion to insert, not replace, the following unrelated text?

[barries]

When inserting before an existing word (to_path_buffer, below), is there a way to prevent rustanalyzer from deleting that word, since it is not relevant to / part of the completion?

Example before/after selecting the LSP's strip_prefix completion (^ is the cursor location):

What	Code
Before:	path.strip^to_path_buffer();
Result:	path.strip_prefix()^;
Desired:	path.strip_prefix()^to_path_buffer();

[ChayimFriedman2]

It doesn't remove for me in VSCode, so this is probably something with your editor.

Specifically, rust-analyzer sends the editor what range to delete. Could it be that your editor doesn't respect that?

[barries]

Thanks, good to know it works in VSCode. The LSP sending the range is why I thought it was an LSP issue.

The completion engine doesn't do this for locally-sourced snippets, which also makes it seems like an LSP issue.

However, who knows? I'm using nvim + blink.cmp. I'll look there-wards and see what I can find.

[zkwinkle]

Hello I have this exact same issue with nvim + blink.cmp, it's been annoying me for months but I haven't been able to figure out what's going on.
I found a similar issue in the blink.cmp repo (saghen/blink.cmp#2179) but they specifically say they're just using the LSP edit range, and not doing anything special on top of that... Which is why I looked up the issue in the context of rust-analyzer and arrived at this discussion.

Also I had this issue before using blink.cmp, it also happened with nvim-cmp.

[suguruwataru]

The range sent to delete is exactly why this happens. In the desired scenario OP describes, nothing needs to be deleted.

I've been debugging this on nvim + vim-lsp. This is not an editor problem, but indeed rust-analyzer behavior (I'm not surprised that rust-analyzer chose to go this way, as this is what JetBrains IDEA does).

Specifically, rust-analyzer puts a textEdit field in the CompletionItem it sends back to the client (or editor, if you will), which instructs the client to remove the word under cursor so that the completion replaces it instead of just inserting before it. In the scenerio described by OP, the textEdit sent from rust-analyzer indeed instructs the method name to be delete.

I can disable this in vim-lsp using g:lsp_text_edit_enabled. So perhaps you also have some setting in VS Code suppressing this.

As OP said, this isn't everyone's cup of tea. Definitely not mine. So having an option from rust-analyzer would be great.

[suguruwataru]

I filed an issue: #21365

[ZXY595]

Hi, guys, I found a way to solve this if you use nvim + blink.cmp.
Just insert this config to you blink.cmp config:

sources = {
	providers = {
		lsp = {
			transform_items = function(_, items)
				for _, item in ipairs(items) do
					if item.client_name == "rust-analyzer" and item.textEdit then
						item.insertText = item.textEdit.newText
						item.textEdit = nil
					end
				end
				return items
			end,
		}
	},
},

This works fine for me, although it's a bad idea to just map from textEdit to insertText :)

[suguruwataru]

Good job. This disallows us from taking advantage when textEdit is actually good tho. Like when completing the methods in a trait impl, with textEdit you get the whole method signature from the completion. Without textEdit you only get the method name.