Variable declarations inside with statements

On Jul 7, 2008, at 10:40 AM, Allen Wirfs-Brock wrote:

> Herman,
>
> I think you are correct, according to the ES3 standard.  The rules  
> of variable instantiation would host the variable to the scope  
> level of the function and initially set it to undefined.  However,  
> the algorithm you quoted, if executed inside a with statement  
> scoped to an object that includes a property with that identifier  
> name should assign the value to the property, not the hosted variable.
>

s/hosted/hoisted/g

> We need to check out all the common browser implementations to see  
> what the other ones do and we need to add the results to the our  
> discrepancies doc.  If a majority of browsers follow the JScript  
> lead in not implementing what the specification says we can  change  
> the spec. in ES3.1 (although, is isn’t clear that we have a good  
> specification mechanism available for describing the actual  
> behavior you observe).
>

There's no "follow[ing] JScript lead" here. The first JavaScript  
implementation in Netscape 2, and others followed it. The following  
page:

<script>
     var x = 42;
     var o = {x:33};
     with (o)
         var x = 21;
     alert('var x is ' + x + ', o.x is ' + o.x);
</script>

alerts "var x is 42, o.x is 21" in Firefox, Opera, and Safari.

> Of course, this is just further reinforcement of  why the ES 3.1  
> working group wants eliminate with statements from the ES3.1  
> “cautious” subset.
>

That shows poor reasoning. If 'with' in the wild never contains var  
declarations, but remains excessively popular, then the argument  
fails. But there is no argument from evidence here, AFAICT, only from  
aesthetic and best-practices ideas that may not be shared enough to  
be worth trying to legislate from the standard.

Also, please stop splitting strict mode preemptively. Let's argue  
about what strict mode should forbid, and why, sooner rather than  
later. Adding more modes won't get agreement by default. Cc'ing es4- 
discuss at mozilla.org.

/be

On 2008-07-07, at 17:01 EDT, Brendan Eich wrote:

> <script>
>     var x = 42;
>     var o = {x:33};
>     with (o)
>         var x = 21;
>     alert('var x is ' + x + ', o.x is ' + o.x);
> </script>
>
> alerts "var x is 42, o.x is 21" in Firefox, Opera, and Safari.

Nice to see so many implementations followed the spec., but... ouch!   
I bumped my head on that silly `var`.

On Jul 7, 2008, at 2:29 PM, P T Withington wrote:

> On 2008-07-07, at 17:01 EDT, Brendan Eich wrote:
>
>> <script>
>>     var x = 42;
>>     var o = {x:33};
>>     with (o)
>>         var x = 21;
>>     alert('var x is ' + x + ', o.x is ' + o.x);
>> </script>
>>
>> alerts "var x is 42, o.x is 21" in Firefox, Opera, and Safari.
>
> Nice to see so many implementations followed the spec., but... ouch!
> I bumped my head on that silly `var`.

Yes, it's silly, or worse.

The original Netscape implementation, IIRC (and I should, but I'm  
getting old and recycling RAM all the time), did not hoist var in  
global code, only in function code. This optimization was cloned in  
IE JScript and extended to work in global code too. That change fed  
into ECMA-262 Edition 1, since we all agreed after some gnashing of  
teeth that it was better to hoist always than only sometimes. (The  
Borland guy wanted var to introduce a fresh binding at runtime, only  
if evaluated -- that's clearly the wrong answer, and it had no  
support in the committee. :-/)

That's hoisting. The interaction with 'with' was a problem in  
Netscape 2 for function code, and can be laid at the feet of 'with'  
as much as blamed on hoisting.

The whole "ban with from strict mode" idea is plausible if with  
causes bugs and confusions in the wild, in any measurable way -- and  
if 'with' is otherwise used so little or so trivially that the  
migration tax exacted by strict mode can be paid. This is one of the  
bones of contention between different people working on ES3.1 and ES4.

Another bone of contention, from what I can see, is about the proper  
role of a standard specification in either steering people clear of  
trouble, or to take the other side, in trying to teach pigs to sing.

/be