Rust: Allow macros in type positions

[pnkfelix]

Tracking issue for rust-lang/rfcs#873

[alex]

The feature gate for this points at #27336, should it point here instead?

[durka]

How is this going? Is anything on fire yet?

[regexident]

Is there any review/stability status on this feature by any chance?

[tikue]

Any update on this?

[huonw]

cc @rust-lang/lang. This seems like a constrained extension to macros, that we could consider stabilising.

[durka]

There's an issue re interaction with deriving, see #32950, but I think it's minor.

[aturon]

Going to FCP for this feature seems good to me. If people are using it successfully (or have found bugs!) please say so :-)

[nrc]

+1 to stabilising. There are still a few issues with type macros, but just bugs, and nothing major that I'm aware of.

[durka]

Did the lang team decide to FCP or wait?

[aturon]

We went to FCP. Apologies @durka for the dropped administrative work recently -- all of the Mozilla folks just traveled to London for a week and the preparations for that resulted in some balls being dropped.

[aturon]

This feature is in its _final comment period_ for stabilization (in 1.11).

[slash3g]

rustc 1.11 still reports type macros as unstable. Is there any chance of having this feature stabilized in 1.12?

[aturon]

Yes, it's ready for immediate stabilization. Anyone should feel free to open a stabilization PR and ping me on it.

[nikomatsakis]

Indeed I am slow in writing it up, but the @rust-lang/lang team decided formally to stabilize this feature (as @aturon commented). Horray!

[nikomatsakis]

I am tagging the issue with E-mentor because I would be happy to mentor someone through the process of implementing a stabilization PR.

[slash3g]

@nikomatsakis I don't have experience with rustc internals, but I'd be happy to try to implement the stabilization PR. :-)

[nikomatsakis]

@slash3g ok, basically all you have to do is to modify expand_type in libsyntax/ext/expand.rs and basically remove the if from this code:

pub fn expand_type(t: P<ast::Ty>, fld: &mut MacroExpander) -> P<ast::Ty> {
    let t = match t.node.clone() {
        ast::TyKind::Mac(mac) => {
            if fld.cx.ecfg.features.unwrap().type_macros {
                expand_mac_invoc(mac, None, Vec::new(), t.span, fld)
            } else {
                feature_gate::emit_feature_err(
                    &fld.cx.parse_sess.span_diagnostic,
                    "type_macros",
                    t.span,
                    feature_gate::GateIssue::Language,
                    "type macros are experimental");

                DummyResult::raw_ty(t.span)
            }
        }
        _ => t
    };

    fold::noop_fold_ty(t, fld)
}

you also need to change this declaration from libsyntax/feature_gate.rs and move it to be "accepted" (there is a section of "accepted" features below in the file):

    // Allows macros to appear in the type position.
    (active, type_macros, "1.3.0", Some(27245)),

[durka]

Also remove #![feature(type_macros)] from any tests that mention it.

[slash3g]

Ok, I've removed the check on the feature gate in expand_type, updated the tests and moved the declaration in libsyntax/feature_gate.rs.
Just two questions:
1) should I bump the version number in the declaration to "1.12.0"? (or maybe "1.13.0", it's not clear to me if this commit will be backported to the current beta)
2) should I reference this issue in the commit message? I've put "Closes #27245" for now, I don't know if this convention is ok for this project.
Thanks!