[Laszlo-dev] For Review: Change 20081021-dda-p Summary: Add return type declarations to SWF9 [2nd review]

[Donald Anderson]

Okay - it's bigger than I have time to do this time.  It will be in a  
JIRA, with a TODO to mark the spot.
Thanks.

On Oct 24, 2008, at 3:46 PM, P T Withington wrote:

> I see.  Two different issues.
>
> 1) As André reminded me, we already have a bug, LPP-6884, to fix the  
> (void 0)/undefined issue.  That is, we just want `undefined` to be  
> an immutable constant in LZX, like it will be in future  
> Javascripts.  It is a bug in JS1 that it is not.
>
> 2) AS3 does not allow you to to return `undefined` from a function  
> that has a subtype of Object as a return type.  You either have to  
> return an object of the appropriate type or null.  But, at least in  
> the databinding case, it appears from the smoketest that there is a  
> contract to return `undefined` if a datapath matches nothing (or is  
> erroneous).  So that says to me (for whatever datapath call is  
> involved in that particular test case) the return type should not be  
> `String`, but must be `*` (or `(String | void)`, if that were  
> permitted), so that you can return `undefined`.
>
> Until LPP-6884 is fixed, it seems you will have to conditionalize  
> the return on the runtime, although fixing LPP-6884 should be fairly  
> easy.  If you look at the class VariableReference, you will see it  
> already tries to head you off from assigning to 'undefined'.  We  
> could just go one step further and for the JS1 backends, if you use  
> 'undefined' as an r-value, translate it to (void 0), ensuring that  
> it is a constant.  And, while you're there, you might as well handle  
> NaN and Infinity.
>
> On 2008-10-24, at 15:17EDT, Donald Anderson wrote:
>
>> Well, I'll file a JIRA on this (and the other items), but I'm also  
>> confused.
>>
>> I could change the String functions in LzDatapointer.lzs to return  
>> undefined
>> on any new return paths, and revert the smoke testcase.
>> This is what I did in the first review.  That compiles on all  
>> runtimes, and allows
>> the tests to pass unchanged.  I was under the impression that
>> this was not desirable - you said:
>>  `undefined` may be undefined in as3, but it is not in js1, so we  
>> can't rely on that.
>>
>> On the other hand, 'void 0' is not allowed, it gives a SWF9  
>> compilation error.
>>
>> - Don
>>
>> On Oct 24, 2008, at 1:33 PM, P T Withington wrote:
>>
>>> On 2008-10-24, at 13:20EDT, Donald Anderson wrote:
>>>
>>>>   The 'String returns null' case exposed a bug in test/smoke/ 
>>>> regression.lzl
>>>>   where an 'undefined' was expected, changed to a null expected.
>>>
>>> This is the main change that I think is potentially wrong.  I am  
>>> ok with it as a temporary work-around, but would like a p0 bug to  
>>> revert the test and find the cuplrit.  I think there is a place  
>>> where you need to relax a return type so you can return (String |  
>>> void) and return undefined on error.
>>>
>>> IWBNI you also filed Jira's for the other TODO's you mention.
>>
>>
>> --
>>
>> Don Anderson
>> Java/C/C++, Berkeley DB, systems consultant
>>
>> voice: 617-547-7881
>> email: dda at ddanderson.com
>> www: http://www.ddanderson.com
>>
>>
>>
>


--

Don Anderson
Java/C/C++, Berkeley DB, systems consultant

voice: 617-547-7881
email: dda at ddanderson.com
www: http://www.ddanderson.com



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.openlaszlo.org/pipermail/laszlo-dev/attachments/20081024/cd9b47fc/attachment.html