Demangle Rust symbols

[rui314]

Currently, we demangle only C++ symbols, but we should be able to demangle (i.e. pretty print) mangled Rust symbols.

Unfortunately, it looks like there's no C++ library that can demangle Rust symbols. Maybe we need to rewrite https://github.com/rust-lang/rustc-demangle in C++.

Note that this is not high priority.

[Alcaro]

LLVM has something named RustDemangle.cpp, in case that helps.

Don't know how complete it is, though. Nor how stable its API is. AFAIK LLVM offers no stability guarantees other than their C API (which only exports a subset of functionality), but char *llvm::rustDemangle(const char *MangledName, char *Buf, size_t *N, int *Status) doesn't look very likely to change.

Another option would be tossing a C-compatible API onto rustc-demangle and calling that. This has the drawback that mold needs rustc nearby to demangle Rust symbols, but the advantage that there's no need to wait for the alternate demangler to update if Rust starts doing something new (C++ lambdas took a while to be demangled correctly).

[rui314]

I don't want to depend on LLVM nor rustc, so I was thinking of implementing it ourselves. It looks like it's not too hard to write our own implementation.

[bjorn3]

GNU libiberty (as used by binutils and gdb) has support for demangling both the current legacy symbol mangling scheme and the new v0 symbol mangling scheme. In any case if you want to implement it yourself the v0 symbol mangling scheme is documented at https://rust-lang.github.io/rfcs/2603-rust-symbol-name-mangling-v0.html

[a-lafrance]

If there's still interest in implementing this ourselves, I'd be interested in giving it a try. I looked over the RFC linked above and it seems doable.

If that sounds alright, I'm just wondering what approach to take for writing the new demangler. I was thinking of either writing a new Rust demangling function in demangle.cc, or extending the existing one to cover Rust symbols (or doing something else entirely if another approach is better). In any case, just looking for tips on the overall approach here, since I have very limited knowledge of how the demangler interacts with the rest of the code.

Thanks in advance!

[rui314]

The current situation isn't as bad as I originally thought when I filed this bug. I misunderstood that the new mangling scheme is already in use, but in fact, Rust seems to be still using the old modified Itanium mangling scheme, and that can be demangled fairly well with the existing C++ demangler.

We need to support the new Rust mangling scheme eventually if Rust will eventually switch to that scheme, though. Does anyone have any idea when that will happen?

@a-lafrance
A demangler for Rust written in C++ will be useful not only for mold but for other C++ programs, so here is what I was planning:

1. Create a separate repository for a new demangler,
2. write a demangler in C++ that demangles argv[1] and print out the demangled name to stdout,
3. release it with a liberal license such as the MIT license, and
4. let other project to take the source code from that repo to theirs.

[a-lafrance]

Interesting, makes sense. In that case, if you're looking for help with that new demangler project, I'd be interested.

[bjorn3]

Rust seems to be still using the old modified Itanium mangling scheme, and that can be demangled fairly well with the existing C++ demangler.

Indeed.

We need to support the new Rust mangling scheme eventually if Rust will eventually switch to that scheme, though. Does anyone have any idea when that will happen?

You can already opt in on stable. All blockers towards switching to it by default seem to have been resolved so we may switch the default soonish: rust-lang/rust#60705 (comment) After that it takes between 6 and 12 weeks before the change lands on stable depending on where in the release cycle we switch.

[rui314]

@a-lafrance Feel free to start a project. If you start it, it'll be your project, and not only mold but other projects will find it useful.

[a-lafrance]

@rui314 Sounds good, I'll give it a try then. I'll revisit this thread when I have something to show for it I guess.

[eddyb]

This would be the fourth non-Rust demangler for Rust (libiberty, IllumOS, LLVM off the top of my head).

I would suggest reusing the libiberty one (that I wrote - let me know what you need from me to re-license it, surely I am not limited to GPL just because I contributed it to the GCC repo?), or the LLVM one if its license is agreeable. AFAIK both are ports of the Rust code, though in different styles.

The former has been used in several other projects, and it was supposed to be a self-contained single-file demangler implemented in C from the start, that everyone used, but perhaps we failed on the messaging front.

[rui314]

I wasn't aware that there are already multiple implementations for demangling Rust symbols. Let me take a look.

As to the one in the libiberty, I think you can relicense it unless you assign its copyright to GNU, but until the last year, GNU project had required all contributors to assign copyright to them, so the code you wrote may no longer be owned by you.

If the Rust demangler in libiberty were in its own repo and can be compiled to a simple command that just demangles a mangled symbol, it would have been much easier for me to find.

[bjorn3]

https://www.fsf.org/bulletin/2014/spring/copyright-assignment-at-the-fsf

Thus, we grant back to contributors a license to use their work as they see fit. This means they are free to modify, share, and sublicense their own work under terms of their choice. This enables contributors to redistribute their work under another free software license. While this technically also permits distributing their work under a proprietary license, we hope they won't.

If I understand correctly @eddyb could still relicense it under any license, the FSF would just be the copyright holder.

[rui314]

I'm not a lawyer, but I don't think you can (re-)license software as you want unless you have the copyright of that software. You can still ask the current owner of the software to relicense, though.

[bjorn3]

As I understand it they give you a license back that allows you to sublicense it under any terms you want. I can't easily find the exact legal text though.