Loading variable names from DWARF and PDB

[SjRNMzU]

DWARF

Radare2 does not output variable/attribute names (DW_AT_name). As a result, local function variables in binaries with debuginfo aren't named and still use "local_XX".

For example:

1. Compile a binary with "-g3"
2. Compare the output of objdump -g a.out, readelf --debug-dump=info a.out and rabin2 -d a.out

It would be fantastic if we could automatically rename "local_XX" by thier DW_AT_name for binaries with debuginfo.

PDB

Same applies here

[XVilka]

@HoundThe apart from types - please be sure to add tests for variable and function names as well.

[HoundThe]

^ The DWARF PR added integration of the DWARF variable info into functions, but it's still far from perfect as DWARF is very complex format (So I wouldn't close it)

I'll just specify some things that are still in my head for future reference of what's missing to DWARF variables.

• Very often, especially unoptimized binaries are using location in Call frame information (CFA stored in .debug_frame section, for some binaries it's in the .eh_frame), but we don't even parse this section yet.
• There are sometimes multiple scopes (lexical_block in DWARF) that we ignore and take all variables from a function (not sure if there is even ability to show scope information in the disassembly), so we can have a name collision so the last one added in the function scope wins, so some cache to keep track of existing variables and creating some unique name (maybe just checking if it can be successfully added, if not -> append maybe an offset or smth should work too).
• There is even some problem of variable existing on multiple locations based on the place of the function, DWARF has a location lists for this, right now we take a longest range the variable lives in and show that, I couldn't think of better solution to this, when variables can exists on different places at different times. Maybe this isn't possible to show in disassembly but might be good in future for decompiler?
• I've added parsing for .debug_loc (where the location lists are) but GCC has a GNU extension to those with location views. maybe adding support for those? But this is probably just a detail