ESBuild mistakes tuple label of "readonly" for a keyword

[matthew-dean]

Reproduction: https://esbuild.github.io/try/#dAAwLjI1LjQALS1sb2FkZXI9dHMAbGV0IGRlY2xDYW5kaWRhdGU6IFtkZWNsYXJhdGlvbjogRGVjbGFyYXRpb24sIHJlYWRvbmx5PzogYm9vbGVhbl0gfCB1bmRlZmluZWQ

Discovered this quite randomly by having a tuple with a label of "readonly". Since tuple labels cannot have a modifier keyword of readonly, TypeScript thinks this is perfectly fine. (Basically, any valid object JS identifiers should be valid tuple labels AFAIK.) ESBuild, however, fails on this valid TS syntax.

Note, the ESBuild code should probably have tests for other keyword identifiers which are valid in a tuple.

[evanw]

Thanks for this issue. The set of tokens that TypeScript allows here are really arbitrary. Why function but not class? Why import but not export? I'll have to do some research to model this behavior more appropriately, and I also need to handle ? after the token tuple label. Currently I think I'm only handling : after a tuple label.

Edit: It looks like the seemingly arbitrary set of allowed tuple labels is because isStartOfType must return true for a tuple label to be supported, and it's currently implemented like this:

function isStartOfType(inStartOfParameter?: boolean): boolean {
    switch (token()) {
        case SyntaxKind.AnyKeyword:
        case SyntaxKind.UnknownKeyword:
        case SyntaxKind.StringKeyword:
        case SyntaxKind.NumberKeyword:
        case SyntaxKind.BigIntKeyword:
        case SyntaxKind.BooleanKeyword:
        case SyntaxKind.ReadonlyKeyword:
        case SyntaxKind.SymbolKeyword:
        case SyntaxKind.UniqueKeyword:
        case SyntaxKind.VoidKeyword:
        case SyntaxKind.UndefinedKeyword:
        case SyntaxKind.NullKeyword:
        case SyntaxKind.ThisKeyword:
        case SyntaxKind.TypeOfKeyword:
        case SyntaxKind.NeverKeyword:
        case SyntaxKind.OpenBraceToken:
        case SyntaxKind.OpenBracketToken:
        case SyntaxKind.LessThanToken:
        case SyntaxKind.BarToken:
        case SyntaxKind.AmpersandToken:
        case SyntaxKind.NewKeyword:
        case SyntaxKind.StringLiteral:
        case SyntaxKind.NumericLiteral:
        case SyntaxKind.BigIntLiteral:
        case SyntaxKind.TrueKeyword:
        case SyntaxKind.FalseKeyword:
        case SyntaxKind.ObjectKeyword:
        case SyntaxKind.AsteriskToken:
        case SyntaxKind.QuestionToken:
        case SyntaxKind.ExclamationToken:
        case SyntaxKind.DotDotDotToken:
        case SyntaxKind.InferKeyword:
        case SyntaxKind.ImportKeyword:
        case SyntaxKind.AssertsKeyword:
        case SyntaxKind.NoSubstitutionTemplateLiteral:
        case SyntaxKind.TemplateHead:
            return true;
        case SyntaxKind.FunctionKeyword:
            return !inStartOfParameter;
        case SyntaxKind.MinusToken:
            return !inStartOfParameter && lookAhead(nextTokenIsNumericOrBigIntLiteral);
        case SyntaxKind.OpenParenToken:
            // Only consider '(' the start of a type if followed by ')', '...', an identifier, a modifier,
            // or something that starts a type. We don't want to consider things like '(1)' a type.
            return !inStartOfParameter && lookAhead(isStartOfParenthesizedOrFunctionType);
        default:
            return isIdentifier();
    }
}

[matthew-dean]

Thanks for fixing so quickly!