Consider adding CString/*const i8 literals

[rust-highfive]

Issue by canndrew
Wednesday Oct 15, 2014 at 15:59 GMT

For earlier discussion, see rust-lang/rust#18065

This issue was labelled with: B-RFC, I-enhancement in the Rust repository

---

There's already b"foo" syntax for [u8] literals and I've seen talk of adding s"foo" for String literals (perhaps through a syntax extension). If we're going down that path we could also add c"foo" syntax which could be either a CString or a *const i8. This would be useful for interacting with foreign C code. Currently, to call a C function with a string constant I'm using

foo(b"my string\0".as_ptr() as *const i8)

But that's a little cumbersome (and error-prone if I accidentally leave the NUL off).

[ben0x539]

You can get close with a macro using concat!

macro_rules! c_str {
    ($s:expr) => { {
        concat!($s, "\0").as_ptr() as *const i8
    } }
}

foo(c_str!("my string"));

Doesn't look too bad imo. A limitation is that you can't initialize statics like that right now. Maybe there's some cast trickery to make that work but I haven't been able to figure that out.

Why not null-terminate the literals by default? It wouldn't affect Rust code at all.

[sfackler]

We used to null-terminate strings. It didn't really work, since any slice of a string that doesn't include the end will not be null terminated.

[comex]

That macro would be nice to have in the standard library.

[joshtriplett]

Rather than adding more magic syntax, why not add a trait similar to Haskell's IsString (used with overloaded string literals)? Then, "string" would mean fromString("string").

[ftxqxd]

@joshtriplett #143 proposed adding overloaded string literals, but was postponed.

[Parakleta]

I've created a compiler plugin which achieves this, piggy-backing on the binary strings (which must be used in place of regular strings). It's my first attempt at a plugin and I'm pretty much a beginner at Rust too so I may be doing something hideous, but it works for me. It requires libc for the type of c_char. This is currently targeted to the v1.3.0 nightly.

Using the plugin I have confirmed through LLVM-IR that the code compiles to the most obvious traditional C implementation.

The usage syntax is just

#![feature(libc, plugin)]
#![plugin(c_str)]
extern crate libc;
fn main()
{
    unsafe { libc::puts(c!(b"Hello, World!")); }
}

It would be ideal if this could just be available as suggested as a c prefixed string (i.e. c"Hello, World!") instead, but this is a good enough work-around for me for now.

[nagisa]

You do not even need a compiler plugin by doing something along the lines of:

macro_rules! c {
    ($str: expr) => { 
        concat!($str, "\0") as *const u8
    }
}

EDIT: I as always didn’t read backlog and there was a similar macro up there already at #400 (comment).

[Parakleta]

There's a missing .as_ptr() after the concat in your macro, but then the macro fails with "error: cannot concatenate a binary literal". Binary literals matter in C because it allows you to use characters in the 0x80..0xFF range, which you cannot with string literals.

[nagisa]

That’s a fair point which would be resolved by #1187.

There's a missing .as_ptr() after the concat in your macro

something along the lines of

[Parakleta]

Any solution would be fine, but since this issue has languished for nearly 12 months I thought I'd share what I'd been using as a workaround for other people tripped up by it.

[Parakleta]

It turns out my attempted implementation was naive, and I cannot work out if the problem is hard because I don't know what I'm doing, or if there just isn't any way to do it. I cannot find any way to cast a &[u8] to a &[i8] without using transmute but then transmute is not const so it cannot be used to define static bindings. If I try to bind statically a *const i8 instead of the &[i8] it complains that the trait Sync is not implemented for *const (and additionally that I cannot call as_ptr()).

Forgetting the issue of the null termination which it is apparent is already quite solvable, how do I get a statically bound reference (either &'static or *const) to an array or slice of type libc::c_char in the case that libc::c_char is i8? Or is the only option to always store/handle str or &[u8] types and do the conversion at the last moment with an .as_ptr() as *const libc::c_char as it is handed to a C function?

[Kimundi]

This seems to work:

const P: *const i8 = &b"test\0"[0] as *const u8 as *const i8;

[Parakleta]

@Kimundi what you have suggested compiles but results in undefined behaviour. It stores just the first character of the string and then takes the address of it and the rest is discarded.

@ref1537 = internal unnamed_addr constant i8 116