Allow empty identifier @"" where it can be represented

[wooster0]

Preface

I'm calling for a partial reversal of a463dc7 which made it so that @"" is disallowed, everywhere.
My question is, how do I parse this valid JSON data using std.json.parse?

{
    "": 123
}

This is no longer possible:

const std = @import("std");
const json = std.json;
const payload =
    \\{
    \\    "": 123
    \\}
;
const Config = struct {
    @"": u8,
};
const config = x: {
    var stream = json.TokenStream.init(payload);
    const res = json.parse(Config, &stream, .{});
    // Assert no error can occur since we are
    // parsing this JSON at comptime!
    break :x res catch unreachable;
};
pub fn main() !void {
    std.log.info("\"\": {d}", .{config.@""});
}

x.zig:9:5: error: identifier cannot be empty
    @"": u8,
    ^~~
x.zig:19:40: error: identifier cannot be empty
    std.log.info("\"\": {d}", .{config.@""});
                                       ^~~

I'm facing a problem like this right now. Especially with Zig not having something like tags/attributes/annotations (#1099, #14656) for fields, the only thing I can think of is to come up with a special representation for empty fields without clashes. It can be kind of annoying especially because Zig should be able to handle @"" in a lot of places like these just fine.

Proposal

I think @"" should be allowed whenever it can be handled properly:

• const @"" = 1 and var @"" = 1 should be perfectly fine
• test "" should be fine. It may not look clean and may look weird if the name is interpolated into a string, but so does a test name with a single space or unprintable characters so I think an empty test name should be allowed too.
• @"" as the field name in packed and normal structs
• @"" for the name of a non-export function

But it should not be allowed in places such as:

• extern const or export var
• extern struct
• export fn
  Basically it shouldn't be allowed in anything that interfaces with the outside world as it may not be possible to handle there. For example I don't think C has some way of defining a field with an empty name (correct me if I'm wrong).
  Things like @import("") should also stay disallowed, though.

Other Commentary

Another reason I have is that usually Zig avoids special cases like these where an empty string is not allowed so I find it odd that it is not allowed at all. Instead we should move the special cases over to mostly extern/export stuff where it simply cannot be represented properly.

Basically, I think it should be allowed everywhere where it is possible to properly handle an empty identifier.

Also, from #14534:

Even JSON has the ability to represent null bytes in map keys.

JSON also has the ability for empty keys. So allowing @"" would benefit ZON too and bring it on the same level as JSON in this regard. I do think ZON is currently lacking here.

[andrewrk]

Duplicate of #14534.

[wooster0]

that allows null bytes in symbols and length zero symbol names

Sorry, I totally missed that. I was scanning for "empty". I thought it was only about allowing \x00 in @"".

[andrewrk]

No problem. I edited the title to make it more clear. Thanks for typing up this proposal - it seems like we are thinking along similar lines.