Zig: solve the grammar ambiguity with enum literals inside array literals and struct literals

[schroffl]

Accepted Proposal

---

The new anonymous enum literals are great, but they don't to seem work with arrays. I assume that this is a bug.

Let's take this enum as an example:

const MyEnum = enum {Ok, Error};

Declaring a single variable works just fine:

var enum_instance: MyEnum = .Ok;

But an array of MyEnum does not:

var enum_instances: []MyEnum = {.Error};

Which leads to the following compiler output:

error: expected token '=', found '}'

[hryx]

I'd expect that to work too. Enum literals seem to be missing from the grammar - are they supposed to be included? That might make this easier to address.

[andrewrk]

Enum literals aren't finished because they are missing from the grammar and documentation: #683

This is invalid syntax:

var enum_instances: []MyEnum = {.Error};

Correct syntax would be:

test "aoeu" {
    var enum_instances = [1]@typeOf(.x){.Error};
}

Which would give a compile error of error: variable of type '(enum literal)' must be const or comptime, which you could fix with using const instead of var.

However there is a grammar issue here, which is that {.Error is getting parsed as the beginning of a struct literal rather than an array literal with an enum literal in it.

Related: #208

[andrewrk]

cc @Hejsil I think you might have an opinion on how to resolve this.

[thejoshwolfe]

my proposal:

• implement anonymous array literals.
• remove current syntax for array literals.
• then T{ is always the start of a struct initializer.

[schroffl]

@thejoshwolfe I'm assuming by "anonymous array literals" you mean array initializers without the type prefix, right? Like this:

const arr: []u8 = {4, 2};

That would be really nice to have when the type can be inferred by the compiler, since you wouldn't have to specify it manually then. This actually sounds like a great solution to me.

Off topic: Is there an existing issue about anonymous array literals? I couldn't find any.

[Hejsil]

@thejoshwolfe might be a good idea, but i haven't thought much about this. I'm out travelling and will have time to look more into this in a weeks time.

[andrewrk]

I'm accepting https://github.com/ziglang/zig/issues/2235#issuecomment-481941256

[Hejsil]

@andrewrk well, if we are gonna have inferred struct initializes as well, then this is not gonna work. Related #208.

[daurnimator]

@andrewrk well, if we are gonna have inferred struct initializes as well, then this is not gonna work. Related #208.

A better link might be https://github.com/ziglang/zig/issues/685#issuecomment-481806770

[Hejsil]

If we add enum literals with this patch:

--- a/grammar/grammar.y
+++ b/grammar/grammar.y
@@ -121,6 +121,7 @@ PrimaryTypeExpr
     <- BUILTINIDENTIFIER FnCallArguments
      / CHAR_LITERAL
      / ContainerDecl
+     / DOT IDENTIFIER
      / ErrorSetDecl
      / FLOAT
      / FnProto

then the peg parser parses []Enum {.Tag} succesfully as an array literal. So this is not a grammar problem, and just an implementation problem with the stage 1 parser. This is therefor a bug.

[Hejsil]

No ambiguity. This was just a bug in the implementation of the parser. You probably want to implement this in your stage 2 parser PR @hryx :)

[hryx]

'Tis now done