setting rust-analyzer.cargo.extraEnv.key = null has no effect

[jyn514]

rust-analyzer version: rust-analyzer version: 0.3.2379-standalone (8365cf8 2025-04-13) [/home/jyn/.vscode/extensions/rust-lang.rust-analyzer-0.3.2379-linux-x64/server/rust-analyzer]

rustc version (eg. output of rustc -V): rustc 1.88.0-nightly (78f2104e3 2025-04-16)

editor or extension: VSCode 1.96.2 (fabdb6a30b49f79a7aba0f2ad9df9b399473380f), extension version 0.3.2379

relevant settings:

$ cat .vscode/settings.json
{
    "rust-analyzer.cargo.extraEnv": {
        "ENV": null
    }
}

repository link (if public, optional): N/A; happens on every project i've tried

code snippet to reproduce: N/A

if i run strace on rust-analyzer, i see this output:

execve("/home/jyn/.local/lib/cargo/bin/cargo", ["/home/jyn/.local/lib/cargo/bin/cargo", "check", "--quiet", "--workspace", "--message-format=json", "--manifest-path", "/home/jyn/src/example/Cargo.toml", "--all-targets", "--all-features", "--keep-going"], [/*...*/ "ENV=/home/jyn/.profile])

i expect ENV to be unset here, since i've explicitly unset it in the config file.

i tracked through a bunch of the RA codebase but couldn't figure out why this happens. the typescript extension passes it through unchanged, i see extraEnv: { ENV: null } in the typescript output. the "rust-analyzer Language Server" "output" panel shows Server process exited with code 0., which is odd because the LSP is working fine, goto-definition etc work fine.

i found that this config is parsed into cargo_extraEnv: FxHashMap<String, String> using get_field_json, and i would have expected that function to return an error given that serde_json::from_value::<String>(null) returns an error. but i don't see that anywhere in the logs.

[Veykril]

and i would have expected that function to return an error given

it likely does, and as such, the environment passed to cargo is unchanged (that is whatever is inherited). The config only adds to that environment. We should support setting env keys to null to explicitly unset them though, since overwriting it already possible.

[Veykril]

So,

rust-analyzer/crates/rust-analyzer/src/config.rs

Line 605 in a09a550

cargo_extraEnv: FxHashMap<String, String> = FxHashMap::default(),

and

rust-analyzer/crates/rust-analyzer/src/config.rs

Line 655 in a09a550

check_extraEnv | checkOnSave_extraEnv: FxHashMap<String, String> = FxHashMap::default(),

need their types changed to FxHashMap<String, Option<String>>

Here we need to also insert an extra entry for that type (but keep the previous)

rust-analyzer/crates/rust-analyzer/src/config.rs

Lines 3222 to 3224 in a09a550

	"FxHashMap<String, String>" => set! {
	"type": "object",
	},

And the rest the compiler should guide one through via diagnostics.

[jyn514]

@Veykril should we also improve the config parsing to log an error when it can’t deserialize the environment?

[Veykril]

Well we should already either log them or show them in the status bar item, so if we don't that's a bug somewhere.

[jyn514]

i think the bug is that .find(Result::is_ok).and_then(/* handle_error */) doesn't work, the value in handle_error will always be an Ok value because it's filtered out everything else:

rust-analyzer/crates/rust-analyzer/src/config.rs

Lines 3094 to 3098 in a09a550

	.find(Result::is_ok)
	.and_then(|res| match res {
	Ok(it) => Some(it),
	Err((e, pointer)) => {
	tracing::warn!("Failed to deserialize config field at {}: {:?}", pointer, e);

[jyn514]

ok yeah that fixed it, i'll make a pr