unresolved import for some `#[path = ".."]

[qryxip]

• OS: Linux (Arch LInux)
• Toolchain: stable-x86_64-unknown-linux-gnu (1.42.0)
• rust-analyzer: 779555c

unresolved import errors show up for:

1. File paths whose extensions are not rs
2. Absolute file paths
3. Outside of the workspace (or simply workspace_root or somewhere?)

These files do exist.

$ tree -I target "$PWD"
/home/ryo/src/local/a
├── Cargo.lock
├── Cargo.toml
├── empty-file
├── empty-file.rs
└── src
    └── main.rs

1 directory, 7 files
$ touch ./src/main.rs && cargo clippy
    Checking a v0.1.0 (/home/ryo/src/local/a)
    Finished dev [unoptimized + debuginfo] target(s) in 0.47s
$ # OK

[flodiebold]

It seems unlikely that we'll ever support these. The set of files we look at has to be limited by design, and ".rs files in the workspace or dependencies" is a pretty reasonable set, I think. The only solution I could maybe imagine is having a setting for extra files.

[hhamana]

I ran into a similar issue. I'll post it here.

Defining local dependencies outside of the workspace in cargo.toml as such also has weird behaviour.

mydep = { path = "../mydep/" }

In this case, it does not output any error or warning, but also does not compute types/suggestions/tree/docs.
Function definition can be linked to similarly-named function in the impl of unrelated structs. (ie, mydep::MyDepStruct::new() is linking to crate::CrateStruct::new()

[jonas-schievink]

The problem here is that we aren't just limiting ourselves to looking at files in the workspace, but in the package. That means that no files in parents of lib.rs can be included, which breaks core::arch (#5982).

[mitinarseny]

I run into same issue. Is there any progress?

[ilyvion]

It seems unlikely that we'll ever support these. The set of files we look at has to be limited by design, and ".rs files in the workspace or dependencies" is a pretty reasonable set, I think. The only solution I could maybe imagine is having a setting for extra files.

Couldn't rust-analyzer be "automagically" informed of said extra files by what these #[path] attributes have been set to? It's not like you need to load "every file outside the workspace/package" but the ones brought in by #[path] kind of seem like obvious "extra" files to bring in. It's not a particularly pleasant dev experience to have this and the lack of code-completion that it comes with:

Especially given that this is probably 99% of use-cases for #[path] to begin with. In my case, I'm including files that are normally individual examples binary files into a bigger "combined bin" project for what I think is a very good reason (a combined WASM build of all the examples for publication on the project's github.io page) and I can't really think of another clean way to achieve this without any code/file duplication or hassle.

[bmulder-innoseis]

Basically this. In my particular project, it basically makes rust-analyzer as well as VSCode useless for working with Rust. My project structure looks as follows:

• my top level workspace has 3 members: 2 binary crates and one library crate that both binary crates make use of.
• Then, there is a library folder in the top level which is a git submodule that is used in different projects as well as this one. Some of the files from there are used in the library crate, some in the binary crates directly. They are being picked and placed through lib.rs using the path attribute. I cannot make them into a crate because they are just utilities shared by the team and might link to dependencies that are unavailable in some cases. I want full control over which source files I include and which not.

Now, since my submodule folder is used in both my binary crates and my library crate, it needs to be above them in the directory hierarchy. But my workspace members' cargo.toml are one directory below. Rust-analyzer does not allow this project structure, either forcing me to include this submodule three times as a subdirectory of all 3 crates, or copying my shared source files over many times.
At this point it forces me to either buy Rust Rover or just code in vim or emacs, since VSCode does not add much besides annoying error squiggles.

[bmulder-innoseis]

Currently I worked around it by adding a symlink to the submodule inside each crate that depends on it. In general, symlinks can be used as a workaround here from the directory that contains the lib.rs to the external directory to achieve the same thing as the #[path = "../../foo"] statement.