Special case DUMMY_SP to emit line 0/column 0 locations on DWARF platforms.

[khuey]

Line 0 has a special meaning in DWARF. From the version 5 spec:

The compiler may emit the value 0 in cases
where an instruction cannot be attributed to any
source line.

DUMMY_SP spans cannot be attributed to any line. However, because rustc internally stores line numbers starting at zero, lookup_debug_loc() adjusts every line number by one. Special casing DUMMY_SP to actually emit line 0 ensures rustc communicates to the debugger that there's no meaningful source code for this instruction, rather than telling the debugger to jump to line 1 randomly.

[rustbot]

r? @nnethercote

rustbot has assigned @nnethercote.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[jieyouxu]

Is it possible to add a test for this?

[klensy]

And this probably need comment in code too?

[nnethercote]

I agree a comment would be useful, particularly for explaining how/why MSVC is treated differently.

[khuey]

I added a comment and a test. The test is potentially fragile, if the code sample I used to emit a dummy span ever stops getting dummy spans the test will no longer test anything meaningful. But here's an example of the debug experience.

Before:

Breakpoint 1, tmp::main () at src/main.rs:5
5	    let args = env::args();
(gdb) n
6	    let number_str = args.skip(1).next().unwrap();
(gdb)
6	    let number_str = args.skip(1).next().unwrap();
(gdb)
7	    let number = number_str.parse::<i32>()?;
(gdb)
8	    if number % 7 == 0 {
(gdb)
9	        return Ok(());
(gdb)
1	use std::env;
(gdb)
13	}
(gdb)

After

Breakpoint 1, tmp::main () at src/main.rs:5
5	    let args = env::args();
(gdb) n
6	    let number_str = args.skip(1).next().unwrap();
(gdb) 
6	    let number_str = args.skip(1).next().unwrap();
(gdb) 
7	    let number = number_str.parse::<i32>()?;
(gdb) 
8	    if number % 7 == 0 {
(gdb) 
9	        return Ok(());
(gdb) 
13	}
(gdb) 

The inexplicable (to the debugger user) jump to line 1 goes away.

[nnethercote]

It would be nice to have an is_dwarf() method rather than relying on the current-but-not-guaranteed-to-forever-be-true equivalence with is_like_msvc. But that's beyond the scope of this PR.

[nnethercote]

@bors r+ rollup

[bors]

📌 Commit 4e9725c has been approved by nnethercote

It is now in the queue for this repository.