Description
Currently, we assists and fixes produced poorly formatted code: it uses exactly the same whitespace as specified during building of an edit, and it ignores indentation completely.
We need to figure out an API which will give us well-formatted edits.
I don't know how to handle this in a nice way, but I think something like this would work. Say, we want to handle add impl
for a nested struct:
fn foo() {
struct S {
field: u32
}
}
The assist initially produces some impl, and inserts it into the file "randomly". Note that insertion into the file is cheap, because trees are immutable. This will give us the following temporary file:
fn foo() {
struct S {
field: u32
}impl S {}
}
We than find a freshly inserted node and reformat only this node, whihc gives use
fn foo() {
struct S {
field: u32
}
impl S {
}
}
We using this reformatted node, we compute the final edit.
The code and tests for add impl
example live in this file: https://github.com/rust-analyzer/rust-analyzer/blob/17aaece6b39c2fb525be0eccce4626fc622e8236/crates/ra_assists/src/add_impl.rs
There's a nascent formatting crate here: https://github.com/rust-analyzer/rust-analyzer/blob/17aaece6b39c2fb525be0eccce4626fc622e8236/crates/ra_fmt/src/lib.rs
I think, as a first step, we should only handle auto-indent.
Here's a Zulip stream to discuss the issue: https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Fwg-rls-2.2E0/topic/rust-analyzer.23923.3A.20producing.20nice.20code.20for.20refactorings