Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Rework MIR inlining debuginfo so function parameters show up in debuggers. #128861

Merged
merged 1 commit into from
Aug 15, 2024

Commits on Aug 13, 2024

  1. Rework MIR inlining debuginfo so function parameters show up in debug…

    …gers.
    
    Line numbers of multiply-inlined functions were fixed in rust-lang#114643 by using a
    single DISubprogram. That, however, triggered assertions because parameters
    weren't deduplicated. The "solution" to that in rust-lang#115417 was to insert a
    DILexicalScope below the DISubprogram and parent all of the parameters to that
    scope. That fixed the assertion, but debuggers (including gdb and lldb) don't
    recognize variables that are not parented to the subprogram itself as parameters,
    even if they are emitted with DW_TAG_formal_parameter.
    
    Consider the program:
    
    use std::env;
    
    fn square(n: i32) -> i32 {
        n * n
    }
    
    fn square_no_inline(n: i32) -> i32 {
        n * n
    }
    
    fn main() {
        let x = square(env::vars().count() as i32);
        let y = square_no_inline(env::vars().count() as i32);
        println!("{x} == {y}");
    }
    
    When making a release build with debug=2 and rustc 1.82.0-nightly (8b38707 2024-08-07)
    
    (gdb) r
    Starting program: /ephemeral/tmp/target/release/tmp
    [Thread debugging using libthread_db enabled]
    Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
    
    Breakpoint 1, tmp::square () at src/main.rs:5
    5	    n * n
    (gdb) info args
    No arguments.
    (gdb) info locals
    n = 31
    (gdb) c
    Continuing.
    
    Breakpoint 2, tmp::square_no_inline (n=31) at src/main.rs:10
    10	    n * n
    (gdb) info args
    n = 31
    (gdb) info locals
    No locals.
    
    This issue is particularly annoying because it removes arguments from stack traces.
    
    The DWARF for the inlined function looks like this:
    
    < 2><0x00002132 GOFF=0x00002132>      DW_TAG_subprogram
                                            DW_AT_linkage_name          _ZN3tmp6square17hc507052ff3d2a488E
                                            DW_AT_name                  square
                                            DW_AT_decl_file             0x0000000f /ephemeral/tmp/src/main.rs
                                            DW_AT_decl_line             0x00000004
                                            DW_AT_type                  0x00001a56<.debug_info+0x00001a56>
                                            DW_AT_inline                DW_INL_inlined
    < 3><0x00002142 GOFF=0x00002142>        DW_TAG_lexical_block
    < 4><0x00002143 GOFF=0x00002143>          DW_TAG_formal_parameter
                                                DW_AT_name                  n
                                                DW_AT_decl_file             0x0000000f /ephemeral/tmp/src/main.rs
                                                DW_AT_decl_line             0x00000004
                                                DW_AT_type                  0x00001a56<.debug_info+0x00001a56>
    < 4><0x0000214e GOFF=0x0000214e>          DW_TAG_null
    < 3><0x0000214f GOFF=0x0000214f>        DW_TAG_null
    
    That DW_TAG_lexical_block inhibits every debugger I've tested from recognizing
    'n' as a parameter.
    
    This patch removes the additional lexical scope. Parameters can be easily
    deduplicated by a tuple of their scope and the argument index, at the trivial
    cost of taking a Hash + Eq bound on DIScope.
    khuey committed Aug 13, 2024
    Configuration menu
    Copy the full SHA
    1c5e3c9 View commit details
    Browse the repository at this point in the history