Debugger step-over runs backward

[vlad20012]

Setup

main.rs:

fn main() {
    let a = 2; // 1
    let b = 3; // 3
    if b > 0 {
        let b = 10; // 2
        println!("a + b = {}", a + b);
    }
    println!("a = {}", b);
}

Reproducing in terminal:

cargo build
lldb-19 target/debug/debugger-bug

(lldb) breakpoint set --file main.rs --line 2
Breakpoint 1: where = debugger-bug`debugger_bug::main::h53ce27eca21116c8 + 7 at main.rs:2:9, address = 0x0000000000015aa7
(lldb) run
Process 551489 launched: '/home/vlad20012/RustroverProjects/debugger-bug/target/debug/debugger-bug' (x86_64)
Process 551489 stopped
* thread #1, name = 'debugger-bug', stop reason = breakpoint 1.1
    frame #0: 0x0000555555569aa7 debugger-bug`debugger_bug::main::h53ce27eca21116c8 at main.rs:2:9
   1    fn main() {
-> 2        let a = 2;
   3        let b = 3;
   4        if b > 0 {
   5            let b = 10;
   6            println!("a + b = {}", a + b);
   7        }
(lldb) step
Process 551489 stopped
* thread #1, name = 'debugger-bug', stop reason = step in
    frame #0: 0x0000555555569ab2 debugger-bug`debugger_bug::main::h53ce27eca21116c8 at main.rs:5:13
   2        let a = 2;
   3        let b = 3;
   4        if b > 0 {
-> 5            let b = 10;
   6            println!("a + b = {}", a + b);
   7        }
   8        println!("a = {}", b);
(lldb) step
Process 551489 stopped
* thread #1, name = 'debugger-bug', stop reason = step in
    frame #0: 0x0000555555569abd debugger-bug`debugger_bug::main::h53ce27eca21116c8 at main.rs:3:13
   1    fn main() {
   2        let a = 2;
-> 3        let b = 3;
   4        if b > 0 {
   5            let b = 10;
   6            println!("a + b = {}", a + b);
   7        }

Reproducing in RustRover:

Screencast.from.2025-05-21.15-04-15.mp4

Env

LLDB:

lldb version 19.1.7

Rust:

rustc 1.89.0-nightly (bc82152 2025-05-20)
or
rustc 1.87.0 (17067e9 2025-05-09)

[vlad20012]

The bug is not reproduced with -Zmir-enable-passes=-SingleUseConsts, so the optimization is most likely the cause.

In disassembly it's clear that the assignments were reordered:

rustc -Zdump-mir=main src/main.rs:

main.main.3-2-015.SingleUseConsts.before.mir:

    bb0: {
        _1 = const 2_i32;
        _2 = const 3_i32;
        _4 = copy _2;
        _3 = Gt(move _4, const 0_i32);
        switchInt(move _3) -> [0: bb5, otherwise: bb1];
    }

    bb1: {
        _5 = const 10_i32;
        _31 = const main::promoted[1];
        _9 = copy _31;
        _8 = copy _9;
        _17 = copy _1;
        _18 = copy _5;
        _19 = AddWithOverflow(copy _17, copy _18);
        assert(!move (_19.1: bool), "attempt to compute `{} + {}`, which would overflow", move _17, move _18) -> [success: bb2, unwind continue];
    }

main.main.3-2-015.SingleUseConsts.after.mir:

    bb0: {
        nop;
        _2 = const 3_i32;
        _4 = copy _2;
        _3 = Gt(move _4, const 0_i32);
        switchInt(move _3) -> [0: bb5, otherwise: bb1];
    }

    bb1: {
        nop;
        nop;
        _9 = const main::promoted[1];
        _8 = copy _9;
        _17 = const 2_i32;
        _18 = const 10_i32;
        _19 = AddWithOverflow(copy _17, copy _18);
        assert(!move (_19.1: bool), "attempt to compute `{} + {}`, which would overflow", move _17, move _18) -> [success: bb2, unwind continue];
    }

c.c. @scottmcm #125910

[scottmcm]

FWIW, I think what SingleUseConsts is doing here is entirely fine.

There isn't an assignment in the MIR, so that assembly is an artifact of how the debug info is being emitted.

[vlad20012]

сс @wesleywiser - can be related to #73210

[khuey]

This is a duplicate of #113819 I believe. (Or at least the same root cause)

[Enselic]

Triage: No objections so let's close as duplicate.