Statically ban `return continue eval("...")`

[littledan]

Closes #14

[bterlson]

Sounds good!

[claudepache]

That looks buggy to me:

• StringValue is not defined for |MemberExpression|s;
• It does not handle the case continue (((eval)))("..."), which is still a direct eval (unless a nontrivial definition of StringValue is provided),

I think that some more static analysis is needed. Something like the following:

• It is a SyntaxError if MayBeDirectEvalReference of |MemberExpression| is true.

And:

Static semantics: MayBeDirectEvalReference

Identifier : IdentifierName

    1. If StringValue of IdentifierName is "eval" return true.
    2. Return false.

CoverParenthesizedExpressionAndArrowParameterList[Yield] :  (  Expression[In, ?Yield]  )

    1. Return MayBeDirectEvalReference of Expression[In, ?Yield].

SomeNonTerminal : SomeOtherNonTerminal

    1. Return MayBeDirectEvalReference of SomeOtherNonTerminal.

Anything else

    1. Return false.        

[littledan]

I guess we have more detailed cases like continue (1, eval)("...") as well. We implementations statically detect direct eval possibilities, so this should be possible somehow, but it's not clear how the spec text should be worded. @claudepache wanna write a PR?

[claudepache]

I guess we have more detailed cases like continue (1, eval)("...") as well.

(1, eval) triggers an indirect eval.

implementations statically detect direct eval possibilities, so this should be possible somehow, but it's not clear how the spec text should be worded. @claudepache wanna write a PR?

Detailing MayBeDirectEvalReference for each production, following the model of IsValidSimpleAssignmentTarget, is easy but tedious. However, one could instead use a formulation similar to what is done for the delete operator (12.5.3.1). OK, I'll write a PR soon.