VW parser bug? + feature request

> On 10/20/2011 08:22 AM, Steffen Märcker wrote:
>> Hi,
>>
>> I've just tried to define a method with the binary selector #|@ but I  
>> does
>> not compile. Are binary selectors that include $| not allowed in general
>> or is this a parser bug? At least in my projects domain this message  
>> would
>> be very convenient.
>
> The ANSI spec calls for $| to be a valid binary selector character, so  
> #|@ would be allowed under ANSI. I can confirm that it doesn't seem to  
> work in VW, but does work in at least two other Smalltalk  
> implementations. So I'd be inclined to agree that this is a bug, or at  
> least a less-than-desirable feature.
>
>>
>> Also I'd like to ask whether there is a chance to get #:: and other  
>> binary
>> selectors including $: could be supported in a future release?
>
> $: is more problematic (and not specified by ANSI). For instance, if $:  
> were considered a general-purpose binaryCharacter, := would be a valid  
> binary selector, colliding with assignment. Even if := was  
> special-cased, the use of $: in keyword selectors would probably break  
> things somewhere. It's probably possible, but non-trivial and  
> non-portable, to allow use of $: in binary selectors, with restrictions.  
> I'd probably rather keep the rules simple and leave this one the way it  
> is.
>
> Regards,
>
> -Martin

> Hi,
>
>> The ANSI spec calls for $| to be a valid binary selector character, so
>> #|@ would be allowed under ANSI. I can confirm that it doesn't seem to
>> work in VW, but does work in at least two other Smalltalk
>> implementations. So I'd be inclined to agree that this is a bug, or at
>> least a less-than-desirable feature.
>
> I didn't try it in other dialects but I am not surprised that it works in
> some of them. I'd like to hear whether Cincom will support those  
> selectors
> in the next release. =)
>
>> I'd probably rather keep the rules simple and leave this one the way it
>> is.
>
> I totally agree. Actually I asked for $: out of curiosity. =)
>
> Regards, Steffen
>
>
>
> Am 20.10.2011, 23:33 Uhr, schrieb Martin McClure
> <[hidden email]>:
>
>> On 10/20/2011 08:22 AM, Steffen Märcker wrote:
>>> Hi,
>>>
>>> I've just tried to define a method with the binary selector #|@ but I
>>> does
>>> not compile. Are binary selectors that include $| not allowed in  
>>> general
>>> or is this a parser bug? At least in my projects domain this message
>>> would
>>> be very convenient.
>>
>> The ANSI spec calls for $| to be a valid binary selector character, so
>> #|@ would be allowed under ANSI. I can confirm that it doesn't seem to
>> work in VW, but does work in at least two other Smalltalk
>> implementations. So I'd be inclined to agree that this is a bug, or at
>> least a less-than-desirable feature.
>>
>>>
>>> Also I'd like to ask whether there is a chance to get #:: and other
>>> binary
>>> selectors including $: could be supported in a future release?
>>
>> $: is more problematic (and not specified by ANSI). For instance, if $:
>> were considered a general-purpose binaryCharacter, := would be a valid
>> binary selector, colliding with assignment. Even if := was
>> special-cased, the use of $: in keyword selectors would probably break
>> things somewhere. It's probably possible, but non-trivial and
>> non-portable, to allow use of $: in binary selectors, with restrictions.
>> I'd probably rather keep the rules simple and leave this one the way it
>> is.
>>
>> Regards,
>>
>> -Martin
> _______________________________________________
> vwnc mailing list
> [hidden email]
> http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

>
> On 10/20/2011 08:22 AM, Steffen Märcker wrote:
> > Hi,
> >
> > I've just tried to define a method with the binary selector #|@ but I does
> > not compile. Are binary selectors that include $| not allowed in general
> > or is this a parser bug? At least in my projects domain this message would
> > be very convenient.
>
> The ANSI spec calls for $| to be a valid binary selector character, so
> #|@ would be allowed under ANSI. I can confirm that it doesn't seem to
> work in VW, but does work in at least two other Smalltalk
> implementations. So I'd be inclined to agree that this is a bug, or at
> least a less-than-desirable feature.
>

> Martin McClure <martin.mcclure <at> vmware.com> writes:
>
>>
>> On 10/20/2011 08:22 AM, Steffen Märcker wrote:
>> > Hi,
>> >
>> > I've just tried to define a method with the binary selector #|@ but I  
>> does
>> > not compile. Are binary selectors that include $| not allowed in  
>> general
>> > or is this a parser bug? At least in my projects domain this message  
>> would
>> > be very convenient.
>>
>> The ANSI spec calls for $| to be a valid binary selector character, so
>> #|@ would be allowed under ANSI. I can confirm that it doesn't seem to
>> work in VW, but does work in at least two other Smalltalk
>> implementations. So I'd be inclined to agree that this is a bug, or at
>> least a less-than-desirable feature.
>>
>
> Squeak suffers from the same limitations.
>
> Technically, such binary selectors cannot be formed at scan time,  
> because they
> ambiguously could be an empty temp vars specification, or a local temp  
> var
> specification in a block of arity>0.
>
> bugMe
>   ||
>   ^self
>
> bugMeToo
>   ^[:null || andVoid | andVoid := null]
>
> But of course, it would be possible to find a workaround...
>
> Nicolas
>
>
>
>
> _______________________________________________
> vwnc mailing list
> [hidden email]
> http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

>
> Actually the workaround already exists, since at least in VW #| is a valid  
> binary selector, e.g. see Boolean. But I am surprised that longer  
> selectors, e.g. #|@ or #@| are not accepted.
> When using a scanner, e.g. in LR parser, it might be indeed tricky to  
> build the parser. But if you're using a scanner-less parser, e.g. PEG  
> parsers, the ambiguity is resolved, since variable declarations appear  
> always at the beginning of a method/block.
>
> Regards, Steffen
>