| Issue |
76193
|
| Summary |
DWARF with -gmlt and -flto is missing name-mangled C++ symbol names for inlined functions
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
bhamiltoncx
|
This is part of https://bugs.chromium.org/p/google-breakpad/issues/detail?id=883 (DWARF symbols are missing the full C++ name-mangled info when built with `-gmlt`).
I initially fixed https://bugs.chromium.org/p/google-breakpad/issues/detail?id=883 by falling back to the STABS table (which has the full C++ name-mangled version of the symbol) when the symbol name stored in DWARF doesn't match the STABS name:
https://chromium-review.googlesource.com/c/breakpad/breakpad/+/4453650
However, that technique doesn't work for inlined functions — they have no entry in the STABS table, so the name-mangled C++ information is simply missing, causing all sorts of challenges e.g. with clustering symbolicated crash traces (you have to handle both the fully name-mangled C++ version as well as the minimal C-style symbol name which is missing namespace and type information).
Here's a repro:
https://gist.github.com/bhamiltoncx/83895a77882c96c081d200cf37505036
You can see the symbol information for `foo::f()` is present in the STABS (since it's marked `noinline`), but `foo::g()` and `foo::h()` are not present in STABS.
In addition, the minimal DWARF information is missing the namespace and type info for `foo::g()` and `foo::h()`:
```
0x00000043: DW_TAG_subprogram
DW_AT_name ("g")
DW_AT_inline (DW_INL_inlined)
0x00000049: DW_TAG_subprogram
DW_AT_name ("h")
DW_AT_inline (DW_INL_inlined)
0x00000060: DW_TAG_inlined_subroutine
DW_AT_abstract_origin (0x0000000000000049 "h")
DW_AT_low_pc (0x0000000100003f80)
DW_AT_high_pc (0x0000000100003f84)
DW_AT_call_file ("/Users/benhamilton/test.cc")
DW_AT_call_line (21)
DW_AT_call_column (0x03)
0x00000074: DW_TAG_inlined_subroutine
DW_AT_abstract_origin (0x0000000000000043 "g")
DW_AT_low_pc (0x0000000100003f80)
DW_AT_high_pc (0x0000000100003f84)
DW_AT_call_file ("/Users/benhamilton/test.cc")
DW_AT_call_line (15)
DW_AT_call_column (0x03)
```
I think we'll need to use the name-mangled symbol for inlined functions, since there's no representation of the symbol in the STABS to fall back on in this case.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
