"Rename failed: RLS found nothing to rename" when attempting to rename a struct

[norru]

I am unable to rename a struct and I get this in my RLS log:

LSP4E to org.eclipse.corrosion.rls:{"jsonrpc":"2.0","id":"727","method":"textDocument/rename","params":{"textDocument":{"uri":"file:///home/nico/Projects/itadinanta/ofx-rs/ofx/src/handle.rs"},"position":{"line":40,"character":18},"newName":"ImageEffectHandle"}}
org.eclipse.corrosion.rls to LSP4E:{"jsonrpc":"2.0","method":"window/showMessage","params":{"message":"Rename failed: RLS found nothing to rename - possibly due to multiple defs","type":2}}
org.eclipse.corrosion.rls to LSP4E:{"jsonrpc":"2.0","id":"727","result":{}}

how to reproduce:

git clone -btags/rls_1322_rename_failed https://github.com/itadinanta/ofx-rs.git
cd ofx-rs/ofx
rls --cli
rename src/handle.rs 40 18 ImageEffectHandle

Related to https://github.com/rust-lang/rls/issues/1233 but I get a different error.

+@mickaelistria

[lijinpei]

Thank you reporting the problem. Do you have an example code which causes RLS to failed? That would be very helpful for pinning down the problem.

[norru]

I can't rename ANY struct. Is RLS refactoring being tested only with trivial cases? What does the test suite look like?

[lijinpei]

Here you go. https://github.com/rust-lang/rls/blob/7ec19054394ee1ab29b106d1f2c3723a10e7ea6a/tests/tests_old.rs#L619

[norru]

Yes, it does appear that test only covers a trivial case. Feel free to correct me if I'm wrong.

[norru]

Hi @lijinpei Here's a non-trivial repro case for you:

git clone -btags/rls_1322_rename_failed https://github.com/itadinanta/ofx-rs.git
cd ofx-rs/ofx
rls --cli
rename src/handle.rs 40 18 ImageEffectHandle

Output:

{"jsonrpc":"2.0","method":"window/showMessage","params":{"message":"Rename failed: RLS found nothing to rename - possibly due to multiple defs","type":2}}
2: WorkspaceEdit {
    changes: None,
    document_changes: None
}

$rls --version
rls 1.34.0 (0d6f53e 2019-02-14)

[norru]

Hi @lijinpei. I am trying to run the RLS tests from source but this is what I get:

nico@devuan-3xS:~/Projects/3rdParty/rls$ cargo test

error[E0658]: use of unstable library feature 'str_escape': return type may change to be an iterator (see issue #27791)
   --> /home/nico/.cargo/registry/src/github.com-1ecc6299db9ec823/rustc-ap-graphviz-373.0.0/lib.rs:525:52
    |
525 |             LabelStr(ref s) => format!("\"{}\"", s.escape_default()),
    |                                                    ^^^^^^^^^^^^^^
    |
    = help: add #![feature(str_escape)] to the crate attributes to enable

error[E0658]: use of unstable library feature 'str_escape': return type may change to be an iterator (see issue #27791)
   --> /home/nico/.cargo/registry/src/github.com-1ecc6299db9ec823/rustc-ap-graphviz-373.0.0/lib.rs:540:27
    |
540 |                     (&*s).escape_default().to_string().into()
    |                           ^^^^^^^^^^^^^^
    |
    = help: add #![feature(str_escape)] to the crate attributes to enable

error: aborting due to 2 previous errors

For more information about this error, try `rustc --explain E0658`.
error: Could not compile `rustc-ap-graphviz`.
warning: build failed, waiting for other jobs to finish...
error: build failed

EDIT: Never mind, it appears that building RLS requires nightly

[lijinpei]

@norru building RLS do require nighlty, because only nighlty allows #[feature(rustc_private)]

[lijinpei]

Hi, @norru . After a little bit of digging around, I think the symptom probably is you can't rename things you have used in object_property! macro, but renaming other variables probably works fine. The cause of the problem is RLS(or rls-analysis) has some mapping between source code location, reference, and definition. Surprisingly, one source code location can corresponds to multiple definitions, this is especially the case when a macro expands to some definitions, and all the expanded results are represented by the same location(where the macro use occurs). And when the same location corresponds to multiple definitions, rls-analysis/RLS doesn't know what to do, so renaming fails. In short, renaming maybe problematic with macros. BTW, I think macro in statement position can expand to nothing and perhaps your implementation of object_properties! can be simplified.

[norru]

So to summarize, renames don't work with macros? Kind of a bummer.

Any suggestion on how to simplify object_properties! without changing the macro usage? At the moment I am tidying the code of that library up, hence the need to rename.

I am also evaluating whether to replace macro_rules with procedural to make the implementation clearer.

On Mon, 25 Feb 2019, 03:30 lijinpei, [email protected] wrote:

Hi, @norru https://github.com/norru . After a little bit of digging around, I think the symptom probably is you can't rename things you have used in object_property! macro, but rename other variables probably works fine. The cause of the problem is RLS(or rls-analysis) has some mapping between source code location, reference, and definition. Surprisingly, one source code location can corresponds to multiple definitions, this is especially the case when a macro expands to some definitions, and all the expanded results are represented by the same location(where the macro use occurs). And when the same location corresponds to multiple definitions, rls-analysis/RLS doesn't know what to do, so renaming fails. BTW, I think macro in statement position can expand to nothing and perhaps your implementation of object_properties! can be simplified.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/rust-lang/rls/issues/1322#issuecomment-466860045, or mute the thread https://github.com/notifications/unsubscribe-auth/AA2Ry2l8-avDZ8UwVTFibW4VFUja0DmAks5vQ1jRgaJpZM4bOF_h .

[lijinpei]

It's just some reduction base cases are unnecessary.

macro_rules! object_properties {
	(@tail $trait:ty =>) => {
        };

	(@tail $trait:ty => $property:ident read+write, $($tail:tt)*) => {
		impl $property::CanGet for $trait {}
		impl $property::CanSet for $trait {}
		object_properties!(@tail $trait => $($tail)*);
	};

	(@tail $trait:ty => $property:ident write, $($tail:tt)*) => {
		impl $property::CanSet for $trait {}
		object_properties!(@tail $trait => $($tail)*);
	};

	(@tail $trait:ty => $property:ident read, $($tail:tt)*) => {
		impl $property::CanGet for $trait {}
		object_properties!(@tail $trait => $($tail)*);
	};

	(@tail $trait:ty => $capability:ident inherit, $($tail:tt)*) => {
		impl $capability for $trait {}
		object_properties!(@tail $trait => $($tail)*);
	};

	($trait:ty { $($tail:tt)* }) => {
		object_properties!(@tail $trait => $($tail)*);
	};
}

I admit renaming can't work with macros is a big bummer. I need to look into whether rustc exposes enough information to fix this, if it does, then we can fix this problem on RLS side, if it doesn't, then I don't expect this to be fixed in a short time.

[norru]

Thanks for the reduction tip, I'll try it. I think I can live with the macro accepting an empty properties list :)

[Xanewok]

Unfortunately we don't support the macro flow entirely when it comes to analyzing defs/refs for the identifiers :cry:

[Diggsey]

Can't RLS do a best-effort thing here? Rename the uses outside macros, and provide a list of the places which RLS could not rename?

[anticrisis]

Hi there, I'm a new Rust user. Here is a minimal case (without macros) where Rename fails using the RLS extension in Visual Studio Code, and a simple change which allows Rename to succeed. Hopefully this helps to track down this bug.

If you attempt to rename MyLoop to Loop using F2 or the right-click popup menu, RLS reports "RLS found nothing to rename - possibly due to multiple defs"

If you remove the body of main(), the Rename operation is successful. Obviously this is not a helpful workaround, but it might help diagnose the problem or create a better test case.

pub enum MyLoop {
    Continue,
    Quit,
}

pub struct MyResponse {
    pub loop_continue: MyLoop,
}


fn main() {
    let mut loop_continue = MyLoop::Continue;
    while let MyLoop::Continue = loop_continue {
        println!("Hello, world!");

        loop_continue = MyLoop::Quit;

    }
}

Version info. Running on Windows 10.

$ cargo --version
cargo 1.35.0 (6f3e9c367 2019-04-04)

$ rustc --version
rustc 1.35.0 (3c235d560 2019-05-20)

$ rls --version
rls 1.35.0 (49efc06 2019-04-06)

[0X1A]

@anticrisis's Test case is rather minimal and perfectly exemplifies the issues described, hoping we can get some one to pick this up as this has been a long standing issue with RLS that I've had but never reported :pray:

[DzmitryFil]

Rename never works for me with structs that use derive, so this feature is mostly useless at this point. I can reliably rename only local vars inside function.

[iago-lito]

@DzmitryFil As a workaround, I delete the derive attributes, rename, and then paste the derive attributes back in place. This works fine :)

[wyhinton]

Any updates on this issue or plans for a proper patch? I'm still experiencing it with the VSCode extension. Issue also seems to apply to Enums.

[Testare]

Found this issue, one year later...