Factor out gsym crate

[d-e-s-o]

We should consider factoring out the gsym crate with all the necessary logic for reading and interpreting Gsym files. This would make this building block easily reusable for others.

[bobrik]

It would be super cool to have an alternative to llvm-gsymutil to convert dwarf to gsym too.

I'm seeing non-deterministic output form llvm-gsymutil:

ivan@131m38:~$ llvm-gsymutil-19 --convert /usr/lib/debug/.build-id/fc/e446c9d4ad48e2b0c90cce1a11722897805281.debug --out-file /tmp/libc.gsym >/dev/null
ivan@131m38:~$ llvm-gsymutil-19 --address 0x1107b7 /tmp/libc.gsym
Looking up addresses in "/tmp/libc.gsym":
0x00000000001107b7: __clone3 + 39 @ ./misc/../sysdeps/unix/sysv/linux/x86_64/clone3.S:78

ivan@131m38:~$ llvm-gsymutil-19 --convert /usr/lib/debug/.build-id/fc/e446c9d4ad48e2b0c90cce1a11722897805281.debug --out-file /tmp/libc.gsym >/dev/null
ivan@131m38:~$ llvm-gsymutil-19 --address 0x1107b7 /tmp/libc.gsym
Looking up addresses in "/tmp/libc.gsym":
0x00000000001107b7: __GI___clone3 + 39 @ ./misc/../sysdeps/unix/sysv/linux/x86_64/clone3.S:78

Which in turn makes it hard to aggregate stacks:

I'm not as excited about digging in C++ to fix this as I would've been with a rust implementation.

[d-e-s-o]

Agreed, that would be useful to have. However, this issue is really only about moving the existing decoding code into a separate crate. Supporting encoding is a different beast altogether, and while it could live in the same crate, we don't really have plans for reimplementing the logic.

[d-e-s-o]

Did you report the non-determinism upstream @bobrik ? If not, if you can share the binary I can reach out to the Gsym creator.

[bobrik]

The issue was fixed in LLVM 21, so it's all good now: llvm/llvm-project#122921.

If the Gsym creator can stamp my relevant PR (llvm/llvm-project#177524) that would be nice, but no worries if not.