Does AroCC support outputting C code ?

[heysokam]

I'm writing a transpiler for another lang in Zig, and I was wondering if AroCC has the ability to convert its Tokens or AST into valid C code at some point in its process.
The idea would be to use AroCC as a library, generate the data expected by AroCC manually in my transpiler, and then send it to AroCC in a form it understands.

Is C source code output possible with the existing toolset of AroCC in some way?

[Vexu]

It does not since it doesn't implement a C formatter. I'm also not sure something like that is even needed for outputting C. The Zig compiler's C backend just appends text to some buffers. You could probably find useful parts to copy from there to get started.

[heysokam]

Somebody else sent me to the C backend. And I landed there myself before they sent me to the file.
But its like my first week in Zig, and my first real compiler project ever... so getting deep into a 7k sloc file, right at the deepest depths of the Zig compiler, just to know if the fully mangled C output of the Zig.C backend is even usable at all in some partial way is not very approachable at all.

The main issue I had reading that file is that I have no clue how to tell if the mangling is happening as an implicit/mandatory process of that file (which I tried to understand a couple of times before you sent me there, I was aware of it) or if the Zig compiler is doing that mangling somewhere else in its really difficult to follow AST and compilation process.

Do you know anything about that mangling?
How could I tell if the C backend is creating its Zig-like code there in that file, or if that's just a code generator/formatter and the Zig-to-C is happening somewhere else?

[Vexu]

If you're talking about identifier mangling then that is required because C has reserved identifiers (keywords, type names, library functions) and because Zig allows arbitrary byte sequences to be used as identifiers with the @"" syntax.

And if the size is daunting you could always start by looking at an older, simpler version of the C backend like the version shipped with 0.9.0 (note though that it is missing some major optimizations like ziglang/zig#13768.

[heysokam]

Good point, didn't think of looking at earlier versions. Might even be good to take a look at early commits in whatever PR introduced the feature, just to see how the code evolved in the early days.
Good thinking ✍️

[heysokam]

The Zig's C backend doesn't seem to be usable for non-mangled output.
From my limited understanding so far, I believe their formatter functions are what mangle the resulting symbols.