Temporary workaround for compiler bugs in impl Trait feature.

[tikue]

We will want to revert this as soon as possible. There's not really a good way to test whether it's fixed because we don't even have a rustc bug right now to track.

cc @jonhoo

[mcmathja]

Was this about a linker error when trying to implement FutureClient from a separate crate? Was running into issues with that the other night. Tracked this down, if it's of any assistance: rust-lang/rust#35870. Modifying the macro to mark service methods inline resolved the issue in my case.

[compressed]

That's interesting. That issue also talks about some visibility issues. I wonder if that is actually the underlying problem since the full type has some types that needed to be public with this change.

Maybe just making those public would make it go away.

[jonhoo]

@mcmathja yeah, that bug came up in our discussion on gitter yesterday too — thanks for confirming!

@compressed I just tried locally reverting the signature change in 9d552e4, but keeping the visibility modifiers, and that then still causes the linking error for me unfortunately :/

[tikue]

Thanks for the info @mcmathja! Do you happen to have a reproduction you can share? I'd love to open a rustc issue.

[mcmathja]

Played around with this for a bit and the issue is more subtle than I thought. The calls to unreachable!() here and here are getting expanded to the standard panic call, which runs something like:

::rt::begin_panic("internal error: entered unreachable code",
{
    static _FILE_LINE: (&'static str, u32) = ("src/lib.rs", 7u32);
     &_FILE_LINE
})

For me at least, it seems to be those _FILE_LINE references that are creating difficulties:

  = note: Undefined symbols for architecture x86_64:
            "service::FutureClient::test::_$u7b$$u7b$closure$u7d$$u7d$::_FILE_LINE::hd619a832b2c2aa1b", referenced from:
                service::FutureClient::test::_$u7b$$u7b$closure$u7d$$u7d$::h4a4885791567c19d in client-b336e68e002de9fe.0.o
            "service::FutureClient::test::_$u7b$$u7b$closure$u7d$$u7d$::_FILE_LINE::hcb35c9d631e99707", referenced from:
                service::FutureClient::test::_$u7b$$u7b$closure$u7d$$u7d$::h4a4885791567c19d in client-b336e68e002de9fe.0.o

Replacing the macros with concrete errors resolves the issue. It's beyond me why only those two calls need to be replaced, I haven't been able to set up a more generic reproduction. (Have the one I used to generate those errors here). Looks like a different bug from the one I linked earlier in any case.

[jonhoo]

Yup, that looks like my error too.

[jonhoo]

Interestingly, I get a bunch of these errors if I try to compile alacritty on beta or nightly (after a cargo update; haven't tried without).

[jonhoo]

So here's something interesting: I now have mit-pdos/noria@b7abc6e compiling on one machine, and failing with a linker error on another, just using impl Trait in my crate, and on the same nightly (2017-03-03). Digging into it now. Will report back what I find.

[jonhoo]

Travis also complains: https://travis-ci.org/mit-pdos/distributary/builds/208844288#L682

[jonhoo]

A cargo update seems to have inexplicably fixed the problem, even though no dependencies changes. cargo build --release still fails though. And on more machines..

[jonhoo]

Following this up in rust-lang/rust#35870 (comment)