Detect if code is running via rust-analyzer

[rzvxa]

Is there any standard way to detect if the code is running with the rust analyzer or not?

I'd like to be able to compile parts of my proc macro only in actual builds and not when the rust analyzer is running it.

[alibektas]

How about adding a feature and not including it in rust-analyzer.runnables.extraArgs ?

[rzvxa]

@alibektas First off thanks for your quick response.

I've also found rust-analyzer.cargo.features; This would be an amazing solution if there is a way to enforce it to downstream users of our crate. However, I suppose it should be configured through the LSP client which is a local configuration.

It would've been awesome if we could get a rust_analyzer feature flag to configure our code based on this information, But we can't rely on local configurations and we need this to propagate correctly when our library is used as a dependency in another crate.

I'd appreciate any suggestions to help us achieve such a thing.

I'm up for contributing toward a way to expose a rust_analyzer feature or at the very least environment variable(if a feature is not possible due to how cargo works).

[alibektas]

@rzvxa Well there is one way actually. But I don't know if you would like it #17058. But beware : It is still in development. So basically you create a rust-analyzer.toml file near your top most Cargo.toml (omitting rust-analyzer prefix for the related configuration key ofc!). This is the only text-editor independent solution.

Let me know if I can close this issue 👋

And for the next time I would suggest using repo's Discussion for bringing up things like this.

[rzvxa]

Nice it seems like exactly what I want.

And for the next time I would suggest using repo's Discussion for bringing up things like this.

Yes sir, Thanks for the help.

[flodiebold]

I don't think rust-analyzer.toml will have any effect when the library is being used as a dependency? At least that would be very surprising to me. (At least for compilation-related settings, which will be determined by the top-level cargo metadata invocation.)

[rzvxa]

@flodiebold So do you have any suggestion to achieve this across crates? I also had some difficulties configuring the ratoml file; I couldn't find the reason in the source, but again I didn't spend too much time reading through the code.

I've asked about it here: #13529 (comment)
I'd appreciate any help.

---

On another note. If rust-analyzer.toml isn't going to cut it; Shall I reopen this issue?

[Veykril]

There is a cfg we set for rust-analyzer's analysis but its only set within the context of r-a (not when r-a invokes cargo) nor do we set it for dependencies, frankly because we don't want people to specialize on rust-analyzer right now.

More importantly though, toggling proc-macro compilation depending on whether r-a invoked cargo or not is a huge redflag as anyone using your crate would effectively get constant recompilations whenever they alternative between invoking cargo with and without rust-analyzer.

[Veykril]

(gonna re-open as a discussion)

[rzvxa]

Thanks for reopening this.

frankly because we don't want people to specialize on rust-analyzer right now.

I get the concerns behind this, But having such a flag would allow a few things that would be impossible otherwise.

• Skip some stuff to speed up the rust analyzer in big workspaces
• Control how and when the rust-analyzer expands macros
• Attributes/Derives can output explicit documentation to only be used with RA preview/tooltips.

I know it is a double-edged sword to have different behaviors with and without RA but it can unlock some crazy optimizations for people who need it.

For example, the use case we have for it is this:

We want our types to have #[repr(C)] but it causes them to have inefficient abi layouts. That's why we have created a lightweight proc-macro that would reorder fields to fix this issue. It allows us to have fully deterministic types with a well-optimized layout. It was going great until we encountered this issue with RA.

Sadly for us, RA uses the version after attribute expansion and it breaks our documentation/preview through it. That's why we want to find a way around it by outputting the unoptimized version for the rust analyzer(which has the field ordering as written in the source code) while doing this optimization in actual builds.

I hope you can see the value of it.

More importantly though, toggling proc-macro compilation depending on whether r-a invoked cargo or not is a huge redflag as anyone using your crate would effectively get constant recompilations whenever they alternative between invoking cargo with and without rust-analyzer.

You are right; Maybe an environment variable would be better to avoid shooting ourselves in the leg. Any way to infer this be it at compile time or runtime would do for us.

[Veykril]

I hope you can see the value of it.

I feel conflicted about that, we'd effectively show the user incorrect results then. Notably the layout information is just plain wrong in the hover if we did allow that.

I am certainly interested in giving proc-macro powers to interface with the IDE (to aid in completions and similar things), which would most likely allow this to be done in some capacity as well at some point, though I have yet to sit down and think about that more. Anyways, it won't allow for compilation to differ between plain cargo and r-a invoked cargo, if that ever ends up possible I will consider that a bug due to the recompilation problem.

[rzvxa]

I feel conflicted about that, we'd effectively show the user incorrect results then. Notably the layout information is just plain wrong in the hover if we did allow that.

Yeah, that's one issue that we have to compromise; I was going to combat this by explicitly adding a doc comment in RA runs to each type/field stating their actual layout data.

I was thinking about this flag as an unsafe tool that the implementor has to deal with based on their usage. Even if you omit some useless code for optimization in the RA pass you may cause confusion, But that can be sorted out with a lint rule to warn about it.

---

Is it possible to build the version for RA in a separate directory? What happens if we set the target dir to targets/rust_analyzer? With that can't we have a feature flag that wouldn't cause recompilations?

[rzvxa]

Another way to fix our specific issue is to have some sort of hint attribute that would allow RA to display a pre-expanded version while using the post-expansion version for inference.

---

Edit:

Having the ratoml work across crates would allow us to implement this without any change to RA internals. So may I ask for technical limitations behind it? Is there any way to achieve this?