Grammar question about ArrowFormalParameters

Thu Nov 13 16:44:55 PST 2014

That seems reasonable* Allen.

And I appreciate confirmation that our reading of the spec is in line with yours.

           -- Cyrus

* I personally prefer the approach that moving to new language constructs may come with additional restrictions that are more onerous.  But that's just me.  For example, I don't mind that Classes force 'strict mode' within them, and that you would not be able to copy "function (yield) { }" from outside the class to inside a class method.  But that's just me.  It wouldn't be javascript if there wasn't lots of fun complexity for us compiler writers :)

From: Allen Wirfs-Brock [mailto:allen at wirfs-brock.com]
Sent: Thursday, November 13, 2014 3:57 PM
To: Cyrus Najmabadi
Cc: Jason Freeman; es-discuss list
Subject: Re: Grammar question about ArrowFormalParameters

On Nov 13, 2014, at 3:21 PM, Cyrus Najmabadi wrote:

Hey Allen,

We have a followup question.  This time it's more about understanding the rationale behind things, rather than asking what the spec is explicitly stating.

Specifically, I'm curious why the spec says this in "14.2.1<http://people.mozilla.org/~jorendorff/es6-draft.html#sec-arrow-function-definitions-static-semantics-early-errors> Static Semantics: Early Errors":

If the [Yield] grammar parameter is present for CoverParenthesizedExpressionAndArrowParameterList[Yield] return the result of parsing the lexical token stream matched by CoverParenthesizedExpressionAndArrowParameterList[Yield] using ArrowFormalParameters[Yield, GeneratorParameter] as the goal symbol

Instead of:

If the [Yield] grammar parameter is present for CoverParenthesizedExpressionAndArrowParameterList[Yield] return the result of parsing the lexical token stream matched by CoverParenthesizedExpressionAndArrowParameterList[Yield] using ArrowFormalParameters as the goal symbol

This would permit:

function *g() {
   yield (yield) => 42;
}

Because 'yield' is an identifier if ArrowFormalParameters isn't parameterized with [Yield].  Because arrow functions are syntactically so tightly integrated into expressions I didn't think it would be a good idea to allow that usage.

Your example seems to exemplify the issue, but we're not sure we understand why arrow function are treated this way.  For example, you say (and we agree) that this should be illegal as per the spec:

var yield = 42;
function *g() {
   var f = (arg=yield) => arg;  //it  is a syntax error to use 'yield' within an arraw parameter list inside a generator function
   yield f();
}

However, it appears as if the following would be legal:

var yield = 42;
function *g() {
   var f = function (arg=yield) { return arg;}
   yield f();
}

yes, this is legal.  It is JS legacy that 'yield' is a regular identifier (not a reserved word) in non strict mode code.  We've tried to minimize the number of "micro-modes" so we didn't add additional special cases to make things like the above illegal. The function keyword strongly delimits the start of a new function definition.  Identical rules are applied to all non-strict FunctionDeclaration and FunctionExpressions .  We special case arrow functions because they lack that strong syntactic delimiter and seem more like part of the enclosing expression. Basically, definitions starting with 'function' can be though of as self-contained entities.

This is because of the following rule:

FunctionExpression : See 14.1
function BindingIdentifieropt ( FormalParameters ) { FunctionBody }

Which explicitly does not use [Yield] or [Generatorparameter] on the FormalParameters.

1)      do you think that code should be legal or not.  If not, what part of the spec does make it illegal?
2)      If it is legal, do you think that Arrow Functions and normal function expressions should behave differently here?  If so, why?

see above

Allen

Thanks!

            -- Cyrus

From: es-discuss [mailto:es-discuss-bounces at mozilla.org] On Behalf Of Cyrus Najmabadi
Sent: Thursday, November 13, 2014 2:45 PM
To: Allen Wirfs-Brock
Cc: Jason Freeman; es-discuss list
Subject: RE: Grammar question about ArrowFormalParameters

Hi Allen,

I see.  This was the part that was missed:

If the [Yield] grammar parameter is present for CoverParenthesizedExpressionAndArrowParameterList[Yield] return the result of parsing the lexical token stream matched by CoverParenthesizedExpressionAndArrowParameterList[Yield] using ArrowFormalParameters[Yield, GeneratorParameter] as the goal symbol

That does clear up the grammar question.

Thanks much!

             -- Cyrus

From: Allen Wirfs-Brock [mailto:allen at wirfs-brock.com]
Sent: Thursday, November 13, 2014 2:25 PM
To: Cyrus Najmabadi
Cc: es-discuss list; Jason Freeman
Subject: Re: Grammar question about ArrowFormalParameters

On Nov 12, 2014, at 4:46 PM, Cyrus Najmabadi wrote:

Hey ES6ers,

I'm currently implementing some of the ES6 support for the next version of TypeScript.  The part I'm looking at right now is generators and yield expressions.  So far we feel fairly comfortable with the grammar and understand the implications of the [Yield] and [GeneratorParameter].  One spec issue that is getting us to scratch our heads though is this section:

When the production
ArrowParameters[Yield] : CoverParenthesizedExpressionAndArrowParameterList[?Yield]

is recognized the following grammar is used to refine the interpretation of
CoverParenthesizedExpressionAndArrowParameterList:

ArrowFormalParameters[Yield, GeneratorParameter] :
    ( StrictFormalParameters[?Yield, ?GeneratorParameter] )

The issue relates to the [GeneratorParamater] parameter on ArrowFormalParameters.  We can't see any path through the grammar that could ever end up enabling this parameter.  While CoverParenthesizedExpressionAndArrowParameterList picks up the 'yield' parameter from ArrowParameters, there seems to be nothing related to 'GeneratorParameter'.

See static semantic rules of http://people.mozilla.org/~jorendorff/es6-draft.html#sec-arrow-function-definitions-static-semantics-early-errors and the second algorithm inhttp://people.mozilla.org/~jorendorff/es6-draft.html#sec-static-semantics-coveredformalslist

We also find the presence of this grammar parameter here to be somewhat odd as arrow function can't be generators.

It is dealing with code such as this:

var yield = 42;
function *g() {
   var f = (arg=yield) => arg;  //it  is a syntax error to use 'yield' within an arraw parameter list inside a generator function
   yield f();
}

Is this an issue with the spec?  Or is there some subtlety here that we've missed that enables this parameter?

It's subtle.

Allen

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/es-discuss/attachments/20141114/a8f51fd9/attachment-0001.html>