zig fmt leaves @"_" alone in enums, but not in tagged unions

[mk12]

Zig Version

0.12.0-dev.5+fc6e57568

Steps to Reproduce and Observed Behavior

$ echo 'const A = enum { @"_", _ }; const B = union(enum) { @"_", _ };' > test.zig
$ zig fmt test.zig
test.zig
$ cat test.zig
const A = enum { @"_", _ };
const B = union(enum) { _, _ };

Expected Behavior

I expected zig fmt to treat @"_" in enum and union(enum) the same:

const A = enum { @"_", _ };
const B = union(enum) { @"_", _ };

The behavior was changed for enums in #15617 (cc @jacobly0).

(Although this seems like an obscure corner case, it actually matters for my project. I'm writing a Markdown parser, and enjoy naming my token variants like .@"* * *" instead of e.g. .thematic_break, since that's a needless extra mapping you have to keep in your head. So I want to use .@"_", not .emphasis.)

[mk12]

@nektro Could you explain the 👎? This seemed like it was clearly an oversight, to me. I can't think of any rationale for the behavior being different in enums than in tagged unions.

[andrewrk]

I do wish that people would refrain from using the thumbs down emoji. This isn't a democracy and there is no voting. It just adds negative energy without contributing to the discussion.

[jacobly0]

const E = enum {
    @"_",
    @"-",
};

const U = union(E) {
    _: u8,
    @"-": u16,
};

const U2 = union(enum) {
    _: u8,
    @"-": u16,
};

comptime {
    @compileLog(U{ ._ = 5 });
    @compileLog(U{ .@"-" = 13 });

    @compileLog(U2{ ._ = 5 });
    @compileLog(U2{ .@"-" = 13 });
}

Without a use case for having both _ and @"_" fields in a union, I don't see a reason to change zig fmt. The idea is that you can only have the non-canonical form if it has a semantic meaning, and currently putting both _ and @"_" in a union results in error: duplicate union field: '_' (aka, there is currently no such thing as a non-exhaustive tagged union).

[mk12]

Oh interesting, I didn't realize that. I think non-exhaustive tagged unions are useful, so I hope #8162 gets accepted and then my request here should follow. If it gets rejected, then I would instead propose that using _ in a tagged union should be a compile error, just like this already is for variables:

$ echo 'pub fn main() void { var _: bool = true; }' > test.zig
$ zig build-exe test.zig
test.zig:1:26: error: '_' used as an identifier without @"_" syntax
pub fn main() void { var _: bool = true; }
                         ^

That way there's no confusion over whether the _ is special (as it very often is) or just a regular variant name.

[ianprime0509]

As a side note, this might be a point in favor of #4859: using ... to indicate non-exhaustiveness would avoid colliding with anything that could possibly be interpreted as an identifier. Consistently banning _ from use as an identifier in any context (including enum and field names) would also achieve the same goal, I suppose.