Re: Javascript and konq crash

Luciano Montanaro wrote:

>On Friday 24 March 2006 14:26, Yan Seiner wrote:
>  
>
>>Eva Brucherseifer wrote:
>>    
>>
>>>Hi Yan,
>>>
>>>can you have a look at it? We are pretty busy with fixing other stuff...
>>>
>>>Greetings,
>>>eva
>>>      
>>>
>>OK, I've looked at it.... It's a scary mess.
>>
>>In your javascript that fails, are you calling any floating point?
>>sin/cos? log?
>>
>>The code that fails for us uses a good bit of floating point....  It
>>would be good to confirm that suspicion....
>>
>>    
>>
>
>The dtoa code is really quite intricated.
>It seem to be called from a single point in kjs, in ustring.
>Perhaps for testing purposes another conversion method (with sprintf or a 
>stringstream) could be used. If that works, the problem is indeed in dtoa.
>
>Luciano
>
>  
>
OK, some testing....

Floating point in Javascript on ARM seems to be badly broken -

document.getElementById("xa"+idx).value =
                depth * 100 /
                document.getElementById("ha"+idx).value;

which should be

0.5 * 100 / 3 = 16.66667 - and in fact shows up as 16.6667 in Firefox on 
x86 - shows up as -0.0003 just before konq crashes....

The big question is where is it broken?  Can the kjs folks look at it?

I can tarball the code I am using if that will help....  But for now I 
am going to walk around and clear my head....

--Yan