Discuss
  • Rule 317

Rule 317

# David Teller (18 years ago)

Hello list,

I'm still fighting the syntax of JS2. Attempting to feed it into a parser generator (menhir, for the moment), has already allowed me to find a few typoes in the spec and a few useless variables in the reference implementation, which is all I have to show for the moment. That and the fact that all XML* productions seem absent from the RI.

At the moment, though, I'm being puzzled by rule 317:

ExpressionStatement | ListExpression (allowColon, allowIn) [if lookahead not in { "function", "let", "{" }]

Besides the fact that this rule is annoying to implement, does it actually mean what it intends ? In addition to blocks, it also prevents an assignment to an object pattern from starting the ListExpression. Is that desired ?

Thanks, David

P.S.: I forgot where I found that version of the grammar, it's dated 14/10/2007, so I assume it's the latest. Where are they to be found, in general ?

# liorean (18 years ago)

On 23/10/2007, David Teller <David.Teller at univ-orleans.fr> wrote:

Hello list,

I'm still fighting the syntax of JS2. Attempting to feed it into a parser generator (menhir, for the moment), has already allowed me to find a few typoes in the spec and a few useless variables in the reference implementation, which is all I have to show for the moment. That and the fact that all XML* productions seem absent from the RI.

At the moment, though, I'm being puzzled by rule 317:

ExpressionStatement | ListExpression (allowColon, allowIn) [if lookahead not in { "function", "let", "{" }]

Besides the fact that this rule is annoying to implement, does it actually mean what it intends ? In addition to blocks, it also prevents an assignment to an object pattern from starting the ListExpression. Is that desired ?

It's there to prevent the grammar from being ambiguous. "function", "let" and "{" have different meaning in statement context from expression context. If the ExpressionStatement construct allowed them, then they would be ambigous in statement context, they could have either the statement or the expression semantics. So, in order to prevent this ambiguity, they are not allowed in ExpressionStatement.

# Jeff Dyer (18 years ago)

-----Original Message----- From: es4-discuss-bounces at mozilla.org [mailto:es4-discuss- bounces at mozilla.org] On Behalf Of David Teller Sent: Tuesday, October 23, 2007 4:11 AM To: JS2 Subject: Rule 317

Hello list,

I'm still fighting the syntax of JS2. Attempting to feed it into a parser generator (menhir, for the moment), has already allowed me to find a few typoes in the spec and a few useless variables in the reference implementation, which is all I have to show for the moment. That and the fact that all XML* productions seem absent from the RI.

Correct. It is reserved syntax that is TBD.

P.S.: I forgot where I found that version of the grammar, it's dated 14/10/2007, so I assume it's the latest. Where are they to be found,

in

general ?

We keep an .xls file in the ./doc directory of the RI. You can find the latest there.

# David Teller (18 years ago)

On Tue, 2007-10-23 at 14:58 +0200, liorean wrote:

It's there to prevent the grammar from being ambiguous. "function", "let" and "{" have different meaning in statement context from expression context. If the ExpressionStatement construct allowed them, then they would be ambigous in statement context, they could have either the statement or the expression semantics. So, in order to prevent this ambiguity, they are not allowed in ExpressionStatement.

So I guess this could just be handled by making this rule have a lowest priority ?

# liorean (18 years ago)

On Tue, 2007-10-23 at 14:58 +0200, liorean wrote:

It's there to prevent the grammar from being ambiguous. "function", "let" and "{" have different meaning in statement context from expression context. If the ExpressionStatement construct allowed them, then they would be ambigous in statement context, they could have either the statement or the expression semantics. So, in order to prevent this ambiguity, they are not allowed in ExpressionStatement.

On 23/10/2007, David Teller <David.Teller at univ-orleans.fr> wrote:

So I guess this could just be handled by making this rule have a lowest priority ?

Not really. In statement context, a function without a name should be a syntax error, it shouldn't be handled as function expression because that'd be pointless. If the function object is only accessible through the return value of the function expression but it's in statement context - i.e. return values are thrown away - then compiling the function at all is totally pointless, you're just adding a non-reachable object for the garbage collector to deal with. In a statement context, "{" is a block opener and not the beginning of an object literal. In a block statement "Identifier :" is a label statement and not a property assignment. "string :" is a syntax error because ":" is not a valid operator there. (Not sure how type annotations play into this, I might be wrong here...)

So you see, they can't just be lowest priority. They must fail with a syntax error. For ES3 compatibility's sake, if nothing else. Also, it's better to not let nonsensical constructs be allowed.

# David Teller (18 years ago)

Ok, I think I'm now understanding something that I had completely missed about that syntax: it's LR(1). I was assuming it was something much more complicated with unlimited backtracking.

Dumping my code and restarting the parser with lighter tools...

Thanks and