Location generation in debuginfo treats mangled names as short names

[tempdragon]

Problems

Currently, at least for GDB, demangling for variable names and function names are not supported by the debuginfo generated in cg_gcc.

Causes

For the short test program, https://github.com/rust-lang/rustc_codegen_gcc/blob/b5d61f1c9e2177f66f2c80c63ced76d8cc66d3db/example/std_example.rs, cg_gcc-generated debuginfo, as is dumped by abi-dumper:
ABI_gcc.txt
While in LLVM:
ABI_llvm.txt
Upon inspection, it's evident that the current debuginfo generation treats mangled names as short names in the DWARF file.

Potential Solutions

Add a new function to generate a Location from mangled names and change the internal representation of Location in gccjit to allow storage of names as demangled ones.

[antoyo]

Is this something you want to investigate yourself or would you want me to look into how to do this in libgccjit?

[tempdragon]

Either is ok. If you are in a hurry(of solving this in 14 days), you can do it yourself.
I opened this to see if there are any other suggestions.
Oh, if I shouldn't even post this, please also tell me. Thanks！

[tempdragon]

As far as I know at this time, there are two ways of generating a (de)mangle-aware DWARF:

1. By adding a duplicated DW_AT_linkage_name identical to DW_AT_name to DW_TAG_subprogram (rustc_codegen_cranelift)
2. By adding a different linkage name, as is done in rustc_codegen_llvm

As is mentioned in rustc_codegen_cranelift/src/debuginfo/emit.rs, it is a MUST to have a linkage name in GDB, or the whole entry will be ignored.

In theory, the second one may result in a smaller size but seems harder to implement. However, in GCC, DW_AT_linkage is generated from DECL_ASSEMBLER_NAME with the prerequisite of NOT being identical to DECL_NAME which in turn becomes the DW_AT_name(as is called the "ShortName" sometimes), making it hard to generate without modifying the dwarf generation functions. Meanwhile, a short name will also require adding new structs and modifying the dwarf since the function names must NOT repeat.

It seems that I will have to modify the dwarf generation anyway.