inactive-code not returned for inactive cfg blocks inside function that has been proc_macro_attribute'd

[d3d9]

rust-analyzer version: current master (not version specific)

rustc version: 1.97.0-nightly

editor or extension: Zed 1.0.1, VSCode

This has been bugging me for a while, it seems like something that might already have an open issue but I couldn't manage to find it.

This works:

With the proc_macro from embassy applied, it doesn't:

It's confusing because the code doesn't get dimmed, while it only gets partial highlighting, no inlay hints etc. (which would be fine and not confusing if it were also dimmed..) and a bunch of unresolvedReference etc. as well in some cases.

[ChayimFriedman2]

This was a deliberate decision:

rust-analyzer/crates/ide-diagnostics/src/handlers/inactive_code.rs

Lines 13 to 16 in 8f678f2

	// If there's inactive code somewhere in a macro, don't propagate to the call-site.
	if d.node.file_id.is_macro() {
	return None;
	}

And it makes sense: macros can make some code disabled that they don't want to be shown as such, and there is no way to disable this diagnostic.

[ChayimFriedman2]

For context, #6715 introduced this behavior.

[d3d9]

And it makes sense: macros can make some code disabled that they don't want to be shown as such, and there is no way to disable this diagnostic.

So the reasoning is mainly about code that wouldn't normally be disabled, and if a macro makes it end up be disabled, it should not be shown as disabled?
How about the case where code would normally (if not inside the macro) be disabled, could this be processed normally or does the reasoning also exist in reverse (that a macro could make disabled code non disabled, that's why the diagnostic should still be suppressed?)

[d3d9]

I just looked at the original issue and I guess I understand now, basically if anything inside the macro definition is cfg'd out it ended up marking the whole macro call as inactive, is that what lead to that change back then?

(Just noticed, I guess I have something similar in my current project where there is an unused variable and an unreachable call in an unused fn defined in an impl inside my macro_rules and these diagnostics also "leak out" to the place where I call the macro, they even make it dimmed for some reason)