Implement RFC 1925

[mattico]

cc #44101

[rust-highfive]

r? @steveklabnik

(rust_highfive has picked a reviewer for you, use r? to override)

[mattico]

Is there a better span to use here?

I suppose used_beginning_vert could become Option<Span>...

[petrochenkov]

It should be possible to call emit_feature_err immediately from parse_arm, the precise span is available there.

In addition, changes to AST (used_beginning_vert: bool) will likely break rustfmt and you'll have to send PR to rustfmt repo as well, do submodule updates, etc, so I recommend against it.

[mattico]

Oh, I didn't know that features were already available during parsing!

[petrochenkov]

I don't know if they are available as well, but it's worth trying, because the alternative is not especially pleasant.

[petrochenkov]

This needs a // gate-test-match_beginning_vert annotation, so the test is counted as a feature gate test.

[mattico]

That shouldn't be necessary as long as the test is named correctly, no? That's what tidy said, and most of the other tests (ex) don't have them.

[petrochenkov]

Okay, if this passes testing, then it's good.

[arqunis]

This should be just

let beginning_vert = ifself.eat(&token::BinOp(token::Or)){Some(self.prev_span)}else{None};

edit: Nevermind, that can work as well.

[mattico]

I can't find any way to query features in the parser, and I'm pretty sure it doesn't exist. Nothing else in the parser (outside of libsyntax_ext) uses features, and the features need to be parsed! I see a few paths forward:

1. Enable this unconditionally. It's not too dangerous of a change, IMO.
2. Enable this but only on nightly. This way we'd have a chance to catch any strange edge cases before it goes to stable, but it's not gated in the traditional sense. (currently implemented in this PR)
3. Make the changes to have a real feature gate. Altering the AST to store this information is the path of least resistance, but it has some issues as pointed out by Implement RFC 1925 #44108 (comment)

[petrochenkov]

I can't find any way to query features in the parser, and I'm pretty sure it doesn't exist

Ok, sorry for pointing the wrong way.

Altering the AST to store this information is the path of least resistance, but it has some issues

Not issues, just possible inconveniences with updating rustfmt and rls.

[carols10cents]

r? @arielb1

[arielb1]

r? @petrochenkov - he's our parser guy

[arielb1]

But currently, I think a feature gate is wide-open

[00:49:52] ---- [compile-fail] compile-fail/feature-gate-match_beginning_vert.rs stdout ----
[00:49:52] [00:49:52] error: compile-fail test compiled successfully!
[00:49:52] status: exit code: 0
[00:49:52] command: "/checkout/obj/build/x86_64-unknown-linux-gnu/stage2/bin/rustc" "/checkout/src/test/compile-fail/feature-gate-match_beginning_vert.rs" "-L" "/checkout/obj/build/x86_64-unknown-linux-gnu/test/compile-fail" "--target=x86_64-unknown-linux-gnu" "--error-format" "json" "-C" "prefer-dynamic" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/test/compile-fail/feature-gate-match_beginning_vert.stage2-x86_64-unknown-linux-gnu" "-Crpath" "-O" "-Lnative=/checkout/obj/build/x86_64-unknown-linux-gnu/native/rust-test-helpers" "-L" "/checkout/obj/build/x86_64-unknown-linux-gnu/test/compile-fail/feature-gate-match_beginning_vert.stage2-x86_64-unknown-linux-gnu.compile-fail.libaux" "-A" "unused"
[00:49:52] stdout:
[00:49:52] ------------------------------------------
[00:49:52] [00:49:52] ------------------------------------------
[00:49:52] stderr:
[00:49:52] ------------------------------------------
[00:49:52] [00:49:52] ------------------------------------------
[00:49:52] [00:49:52] thread '[compile-fail] compile-fail/feature-gate-match_beginning_vert.rs' panicked at 'explicit panic', /checkout/src/tools/compiletest/src/runtest.rs:2435:8
[00:49:52] note: Run with `RUST_BACKTRACE=1` for a backtrace.
[00:49:52] [00:49:52] [00:49:52] failures:
[00:49:52] [compile-fail] compile-fail/feature-gate-match_beginning_vert.rs
[00:49:52] [00:49:52] test result: FAILED. 2724 passed; 1 failed; 13 ignored; 0 measured; 0 filtered out
[00:49:52] 

[arielb1]

You should use sess.features.borrow().match_beginning_vert here, otherwise the feature gate won't work

[mattico]

sess is a ParseSess which does not have a features field.

[mattico]

@arielb1 yep! waiting for confirmation that a feature gate is necessary/desired before I reimplement it by mucking with the AST.

[petrochenkov]

@mattico
Citing@nikomatsakis:

Basically every time I've skipped a feature gate I've regretted it.

Let's add a feature gate.

[mattico]

@petrochenkov sounds good, will do!

[mattico]

Interesting failures. Taking a look.

[mattico]

I ran a build of rustfmt, it works fine with this change. They only access single fields of Arm, so it doesn't break anything. RLS also works. Clippy was already broken before these changes.

[petrochenkov]

@bors r+

[bors]

📌 Commit 22ca03b has been approved by petrochenkov

[bors]

⌛ Testing commit 22ca03b with merge 6f66730...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: petrochenkov
Pushing 6f66730 to master...