Parser mistakes macro argument for struct initialization?

[hannobraun]

When I updated to Rust master yesterday, some of my macro invocations broke in an interesting way. Here's a reduced example:

#![feature(macro_rules)]

macro_rules! my_macro(
    ($condition:expr $action:block) => ({})
)

fn main() {
    my_macro!(0 == y {
        y = 1;
    });
}

$ rustc main.rs
main.rs:10:5: 10:6 error: expected `:` but found `=`
main.rs:10      y = 1;
                  ^

With this workaround, it compiles without errors:

#![feature(macro_rules)]

macro_rules! my_macro(
    ($condition:expr $action:block) => ({})
)

fn main() {
    my_macro!(0 == y && true {
        y = 1;
    });
}

Looks to me like the parser sees "y { ..." and mistakes it for a struct initialization.

This problem occurs with commit 2f74325.
Before the upgrade, it worked fine with commit db29814.

[kmcallister]

This is an example of the problem that RFC 550 was meant to prevent, which indeed it does:

<anon>:2:6: 2:21 error: `$condition:expr` is followed by `$action:block`, which is not allowed for `expr` fragments
<anon>:2     ($condition:expr $action:block) => ({})
              ^~~~~~~~~~~~~~~

The ambiguity between struct initializers and blocks is a long-standing wart of the expression grammar. It's probably responsible for other bugs, but it should again not cause macro breakage like this, thanks to RFC 550.