VersionTupleSyntax should be parsed as (up to) 3 integer literals

[bnbarham]

Issue Kind

Other

Source Code

@available(macOS 10.4p2.1, *)

Description

The given source should produce an error, rather than a valid parse with 10.4.p2 as the majorMinor and 1 as the patch on VersionTupleSyntax. Ideally VersionTupleSyntax would contain major, minor, and patch.

[rintaro]

The structure of VersionTupleSyntax should be like:

    children: [
      Child(name: "Major",
            kind: .token(choices: [.token(tokenKind: "IntegerToken")])),
      Child(name: "PeriodBeforeMinor",
            kind: .token(choices: [.token(tokenKind: "PeriodToken")]),
            isOptional: true),
      Child(name: "Minor",
            kind: .token(choices: [.token(tokenKind: "IntegerToken")]),
            isOptional: true),
      Child(name: "PeriodBeforePatch",
            kind: .token(choices: [.token(tokenKind: "PeriodToken")]),
            isOptional: true),
      Child(name: "Patch",
            kind: .token(choices: [.token(tokenKind: "IntegerToken")]),
            isOptional: true),
    ]

And the parser should reject integer literals with non-decimal characters e.g. 0x01

[grynspan]

I imagine "1." or "1.2." should also be rejected, so it's more like (I, .I?, .I?) than (I, .?, I?, .?, I?), right?

[ahoppen]

rdar://107152155

[rintaro]

I imagine "1." or "1.2." should also be rejected, so it's more like (I, .I?, .I?) than (I, .?, I?, .?, I?), right?

Right, or more like (I, (.I, (.I)?)?). The parser will diagnose those malformed things, either by consuming it as an unexpected token, or leave it and let the "outer" parser do things.
Unfortunately, we don't have a way to declare grouped optional without introducing a nominal syntax type, the declaration will be flatten.

[grynspan]

Gotcha. Okay, works for me! Thanks for taking a look at this.