`convert_named_struct_to_tuple_struct` produces intersecting edits when a macro expands a field access twice

[Albab-Hasan]

rust-analyzer version: 0.3.2929-standalone (panic message below is from a source build of master with force-always-assert; on the release build the assist silently does nothing and only logs the error)

rustc version: rustc 1.97.0-nightly (ca9a134e0 2026-04-26)

editor or extension: VSCode, rust-analyzer extension 0.3.2929

relevant settings: none

repository link (if public, optional): n/a

code snippet to reproduce:

macro_rules! m { ($e:expr) => { ($e, $e) }; }

struct S { a: i32 }

fn main() {
    let s2 = S { a: 1 };
    m!(s2.a);
}

reproduce:
cursor on S in struct S { a: i32 }
Convert to tuple struct

expected:
struct S(i32); with the usage rewritten to m!(s2.0);, or the field reference inside the macro left alone. either way no crash.

actual:
on the release build the assist is offered, applies nothing, and logs the error. on a build with debug assertions the request handler panics:

request handler panicked: some replace change ranges intersect!
left change:  ReplaceWithMany(Token(IDENT@299..300 "a"), [Node(NAME_REF@0..1)]) a0
equals
right change: ReplaceWithMany(Token(IDENT@299..300 "a"), [Node(NAME_REF@0..1)]) a0

panicked at crates/syntax/src/syntax_editor/edit_algo.rs:478:5:
some replace change ranges intersect!

cause:
the macro expands $e twice, so the field-reference search returns two FileReferences for the single a token in the source. edit_field_references maps both back to the same source token and registers two identical ReplaceWithMany changes, tripping stdx::always!(false, "some replace change ranges intersect!") in edit_algo.rs . the whole edit is discarded.

note:
same root cause as the destructure_tuple_binding / destructure_struct_binding issues, also affects convert_tuple_struct_to_named_struct.

env:

• rustc: 1.97.0-nightly
• rust-analyzer: 0.3.2929
• OS: Ubuntu
• Editor: VS Code

[Albab-Hasan]

@rustbot claim