Synthesize a TokenStream for StmtKind::Empty

[Aaron1011]

Fixes #80760

[rust-highfive]

r? @lcnr

(rust-highfive has picked a reviewer for you, use r? to override)

[lcnr]

r? @petrochenkov probably

[petrochenkov]

StmtKind::Empty is only for diagnostics they said, we only want to report a lint they said.
1.5 years pass and ; is already a full-fledged statement accepted by stmt matcher, just through the power of natural bug evolution.

At least

#[attr]
;

is not supported yet.

[petrochenkov]

At least #[attr] ; is not supported yet.

Oh well.

macro_rules! m {
    ($stmt:stmt) => { #[allow(bad_style)] $stmt }
}

fn main() {
    m!(;);
}

[petrochenkov]

I can't complain about the end result, but the process here was funny.
Of course, to align with the token-based expansion model better ; would eventually become a statement, with some lang team process and approval, but until then it was still supposed to be only a "filler" between other statements, a part of the block syntax.

[Aaron1011]

@petrochenkov: The current behavior w.r.t to attributes is very odd - any attributes added via #[attr] $empty_stmt are not actually invoked. I've added a test to ensure that we detect if this changes.

[petrochenkov]

r=me after moving the common part prepend_attrs(sess, stmt.attrs(), nt, ...) to outside of the condition.

[Aaron1011]

@bors r=petrochenkov

[bors]

📌 Commit 3ed6caa has been approved by petrochenkov

[bors]

⌛ Testing commit 3ed6caa with merge 080ee6f...

[bors]

☀️ Test successful - checks-actions
Approved by: petrochenkov
Pushing 080ee6f to master...