Add as_c_str method on strings

[bstrie]

Formerly this was a free function rather than a method. I've left it in place for now, although redefined it so that it just calls the method.

[pcwalton]

I really dislike this API. Can we not make it something like this:

enum CStringBundle {
    allocation: Option<~[u8]>,
    c_str: *u8,
}

fn as_c_str(&self) -> CStringBundle {
    if I'm null terminated {
        CStringBundle { None, &self[0] }
    } else {
        let new_allocation = copy and null terminate me;
        CStringBundle { Some(new_allocation), &new_allocation[0] }
    } 
}

---

HELLO WORLD LOL, THIS IS BEN EDITING PATRICK'S COMMENT TO PROVE THAT GITHUB ALLOWS ME TO EDIT ANY ARBITRARY COMMENT ON ANY PULL REQUEST THAT I'VE OPENED! NOTHING IS SAFE!

R+ BSTRIE/REFLECTIONS_ON_TRUSTING_TRUST

[bstrie]

@Aatch Good catch, I completely glossed over the extra indirection there.

@pcwalton I'm not attached to the current API, but how would you expect to use a CStringBundle for C interop? Right now my Lua interop code looks like this:

let ret1 = str::as_c_str("function foo (x,y) return x+y end",
                         |s| luaL_loadstring(L, s));

With this pull I was expecting to change it to this:

let ret1 = do "function foo (x,y) return x+y end".as_c_str |s| {
    luaL_loadstring(L, s);
}

This still seems roundabout, I'd be perfectly happy with something like:

let ret1 = luaL_loadstring(L, "function foo (x,y) return x+y end".as_c_str());

With your proposed function, would "foo".as_c_str() become "foo".as_c_str().c_str ?

[brson]

@pcwalton What I don't like about this API most is that the performance characteristics are not obvious. What if we instead changed the signature to as_c_str(&~str) so you are guaranteed to be null terminated and not allocate?

[brson]

Redesigning this interface probably doesn't need to block this pull request though.

[brson]

Oh, what I really like the least about this api is that it is possible to read uninitialized memory when checking for null.

[Kimundi]

@brson Are you talking about the current as_c_str? how can it read uninitialized memory?

[bstrie]

I've opened #6869 for addressing the API change. In the meantime, this is still the direction that we'd like to go wrt changing functions to methods where it makes sense (#6045).

[brson]

@Kimundi perhaps it can't if we maintain the invariant that all slices are overallocated. There was a bug once because some code didn't and I've had a vendetta for it since.