[Laszlo-dev] [Platform-team] NaN vs. Infinity

[J Crowley]

Approved!

P T Withington wrote:
> I updated my change:
>
> http://svn.openlaszlo.org/openlaszlo/patches/20081111-ptw-E.tar
>
> If you can review it, I will check it in!
>
> On 2008-11-11, at 15:34EST, J Crowley wrote:
>
>> Ahh, I see.  I won't file the bug, then, since it's an Adobe issue.
>>
>> Since you've already changed it, I'd say check it in as well, since 
>> there's not much point in me just copy-pasting your work.  :)
>>
>> Thanks!
>>
>> P T Withington wrote:
>>> Ok, it's actually not our bug.  swf8 is broken.  In swf8 
>>> `Number(null)` is `NaN`, and by ECMA (9.3) it is supposed to be `0`.
>>>
>>> Here is a revised version of the program that will work the same in 
>>> all runtimes (by constraining the visibility of the green view to be 
>>> only when `order` is not null):
>>>
>>> <canvas height="150">
>>>  <debug x="50%" width="45%" y="5%" height="95%" />
>>>  <dataset name="onion">
>>>    <layer order="1"><layer><layer>core</layer></layer></layer>
>>>  </dataset>
>>>
>>>  <view datapath="onion:/layer">
>>>    <simplelayout spacing="5"/>
>>>
>>>    <button text="Peel next layer">
>>>      <handler name="onclick"><![CDATA[
>>>      with (parent.datapath) {
>>>      // Go down one layer
>>>      selectChild();
>>>      // If there are no more layers to go, disable ourself
>>>      if (getNodeCount() == 0) this.setAttribute('enabled', false);
>>>      // Finally, set the order attribute of the new layer to one 
>>> more than it's parent
>>>      setNodeAttribute('order', 
>>> String(Number(p.parentNode.attributes['order']) + 1));
>>>      }
>>>      ]]></handler>
>>>    </button>
>>>    <view width="${100 / parent.order}" height="${100 / parent.order}"
>>>          bgcolor="0x09d055"
>>>          opacity="${Math.min(1, parent.order / 3)}"
>>>          visible="${parent.order != null}">
>>>    </view>
>>>    <attribute name="order" value="$path{'@order'}" type="number" />
>>>    <handler name="onorder" args="value">
>>>      Debug.debug("order: %w", value);
>>>    </handler>
>>>    <text text="${this.escapeText(this.formatToString('order: %w, 
>>> data: %#w', parent.order, parent.datapath.data))}" />
>>>
>>>  </view>
>>> </canvas>
>>>
>>> I think you should just use that.  I have a patch out for review 
>>> that makes this work without the debugger too.  Or, if you like, I 
>>> can just check this version of the test in with my change?
>>>
>>> On 2008-11-11, at 12:49EST, J Crowley wrote:
>>>
>>>> Ah, excellent, thanks!  And I'll file the bug.
>>>>
>>>> P T Withington wrote:
>>>>> I think there is a bug here in the LFC, but to try to make sense 
>>>>> out of it, I re-worked the example to be both more modern, and 
>>>>> more understandable.  I think I have made it do what it was 
>>>>> intended to do, maybe you'd like to use this to update the 
>>>>> example.  I did a number of things:
>>>>>
>>>>> 1) I broke out the databinding as an attribute (named 'order') and 
>>>>> gave it a type, so that it will be automatically coerced (rather 
>>>>> than poking directly at the datapath's data field).
>>>>> 2) I fixed the logic in the click handler to disable itself when 
>>>>> there are no more children to select, instead of _after_ it fails 
>>>>> because there are no more children.
>>>>> 3) I fixed the computation of the data set 'order' field to be a 
>>>>> string when setting it.
>>>>> 4) I added a text field to display the current values of 'order' 
>>>>> and the datapath, rather than writing it out to the debugger
>>>>>
>>>>> <canvas height="250" debug="true">
>>>>> <debug x="50%" width="45%" y="5%" height="95%" />
>>>>> <dataset name="onion">
>>>>>   <layer order="1"><layer><layer>core</layer></layer></layer>
>>>>> </dataset>
>>>>>
>>>>> <view datapath="onion:/layer">
>>>>>   <simplelayout spacing="5"/>
>>>>>
>>>>>   <button text="Peel next layer">
>>>>>     <handler name="onclick"><![CDATA[
>>>>>       with (parent.datapath) {
>>>>>         // Go down one layer
>>>>>         selectChild();
>>>>>         // If there are no more layers to go, disable ourself
>>>>>         if (getNodeCount() == 0) this.setAttribute('enabled', false);
>>>>>         // Finally, set the order attribute of the new layer to 
>>>>> one more than it's parent
>>>>>         setNodeAttribute('order', 
>>>>> String(Number(p.parentNode.attributes['order']) + 1));
>>>>>       }
>>>>>     ]]></handler>
>>>>>   </button>
>>>>>
>>>>>   <attribute name="order" value="$path{'@order'}" type="number" />
>>>>>   <view width="${100 / parent.order}" height="${100 / parent.order}"
>>>>>         bgcolor="0x09d055"
>>>>>         opacity="${Math.min(1, parent.order / 3)}">
>>>>>   </view>
>>>>>   <text text="${this.escapeText(this.formatToString('order: %d, 
>>>>> data: %#w', parent.order, parent.datapath.data))}" />
>>>>>
>>>>> </view>
>>>>> </canvas>
>>>>>
>>>>> This should have made it so that I could run the example without 
>>>>> debugging on, but it blows up somewhere.  I'm filing a bug to 
>>>>> myself on that.  But please file a bug about the NaN vs. 0 
>>>>> behavior for missing data.
>>>>>
>>>>> On 2008-11-11, at 02:50EST, J Crowley wrote:
>>>>>
>>>>>> So, I'm running the 
>>>>>> docs/src/developers/programs/databinding-$15.lzx example, fixing 
>>>>>> a few warnings, and I noticed something:  The width of the square 
>>>>>> is supposed to be 100/this.data, where this.data can be 3, 2, 1 
>>>>>> or core.  In DHTML and SWF9, 100/this.data is Infinity when it 
>>>>>> tries to divide 100 by core.  In SWF8, it's NaN.  This is 
>>>>>> actually resulting in different behavior between the runtimes:  
>>>>>> In DHTML and SWF9, when you reach the core (where the data is 
>>>>>> 100/core), the cube disappears.  In SWF8, the cube retains its 
>>>>>> previous dimensions.
>>>>>>
>>>>>> Is this expected behavior?  Should I file a bug?
>>>>>>
>>>>>> Thanks!
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>
>>
>
>