Re: [Pyobjc-dev] Is it possible to create a CFData without copying bytes?

[Uliano G. <gue...@gm...>]

Il giorno 11/nov/07, alle ore 17:24, Ronald Oussoren ha scritto:

>
> On 11 Nov, 2007, at 16:31, Uliano Guerrini wrote:
>
>>
>> def newf():
>>   a=numpy.zeros((256,256),dtype='float32')
>>   for x in  range(0,256):
>>       a[:,x]=x/255.0
>>   return a.tostring()
>>
>> def cff():
>>   return Foundation.CFDataCreateWithBytesNoCopy(None,newf(), 
>> 65536*4,None)
>>
>> changing between CFDataCreate and CFDataCreateWithBytesNoCopy makes  
>> the difference from expected image and crappy noise
>
> That's the expected behaviour, albeith one that is a bit annoying.

well, I'm newbie to pyobjc and to some extent also to python, however  
I didn't expect (and can't explain) it:

it seems to work in a way when I try simple example on the python  
console and in another inside the image synthesis routine

however I found that also NSData.dataWithBytes.NoCopy_Length_ has the  
same (unexpected for me) behaviour


> There is however a better way to convert a python object to an  
> NSData/CFData without copying: if the object supports the buffer  
> protocol, which strings and numpy arrays do, you can use  
> buffer(myObject) to get a buffer object. That object is bridged as  
> an NSData object, without copying data.

this works in that contex, however I don't understand to what extent  
it is bridged: I passed it to a CF routine and worked as a CFData but  
does not respond to NSData methods like .bytes()

I guess that as this bridged buffer there are other bags of tricks  
about PyObjc, where can I look for them?

uliano