about summary refs log tree commit diff
path: root/tests/codegen/inline-function-args-debug-info.rs
AgeCommit message (Collapse)AuthorLines
2025-04-10replace `//@ compile-flags: --edition` with `//@ edition`Pietro Albini-3/+4
2024-08-12Rework MIR inlining debuginfo so function parameters show up in debuggers.Kyle Huey-0/+1
Line numbers of multiply-inlined functions were fixed in #114643 by using a single DISubprogram. That, however, triggered assertions because parameters weren't deduplicated. The "solution" to that in #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 (8b3870784 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.
2024-02-22[AUTO_GENERATED] Migrate compiletest to use `ui_test`-style `//@` directives许杰友 Jieyou Xu (Joe)-1/+1
2023-10-17Automatically enable cross-crate inlining for small functionsBen Kimock-2/+3
2023-04-11Preserve argument indexes when inlining MIRDavid Lattimore-0/+20
We store argument indexes on VarDebugInfo. Unlike the previous method of relying on the variable index to know whether a variable is an argument, this survives MIR inlining. We also no longer check if var.source_info.scope is the outermost scope. When a function gets inlined, the arguments to the inner function will no longer be in the outermost scope. What we care about though is whether they were in the outermost scope prior to inlining, which we know by whether we assigned an argument index.