Objective-C interop is too slow

[kvark]

I can see us spending quite a bit of time in querying the obj-c selectors on each message send.
Upstream issue - SSheldon/rust-objc#49
One possible solution (that needs to be revived, cc @mystor): https://github.com/mystor/objc_methname

[jrmuizel]

@gankro was looking at stabilizing const string length or something like that. Hopefully, he can remind us what the current status is.

[Gankra]

#![feature(const_slice_len)]

static string: &[u8; b"hello".len()] = b"hello";

fn main() {}

compiles on nightly; we just need to push to stabilize this feature (which should be trivial?)

[vincentisambart]

For the time being, have you tried using for example lazy_static for selectors? Last time I tried it didn't seem to be much slower from having the system resolving them a launch, and way faster than resolving the selector at each call (as msg_send! does).
Something like that:

use objc::Message;

#[allow(non_upper_case_globals)]
mod selectors {
    use objc::runtime::Sel;

    lazy_static! {
        pub static ref isEqual_: Sel = Sel::register("isEqual:");
    }
}

fn is_equal(object: *mut objc::runtime::Object, compared_to: *mut objc::runtime::Object) -> bool {
{
    let ret_val: BOOL = unsafe {
        (*object)
            .send_message(*selectors::isEqual_, (compared_to,))
            .unwrap()
    };
    ret_val != 0
}

Also changing it later on to use link_section instead should be pretty straightforward.

[kvark]

@vincentisambart honestly, this is not as urgent to get in stable as it is important to just see what impact it's going to have. So while your suggestion is completely reasonable, I'm leaning towards rolling out support for it under "nightly" feature and checking out how it helps in the meantime. Those few of us who do the benchmarking could easily use the nightly, while for others it's not a blocker in any way.

Also, cc @pcwalton as requested.

[jrmuizel]

@gankro, do you mind filing an issue about stabilizing const_slice_len?

[Gankra]

there's some issues with const fn stuff that means they're not quite ready for it; but nika wrote up a version that doesn't need it (kvark has it)