[Numpy-discussion] Array data and struct alignment

[Travis Oliphant]

Travis Oliphant wrote:
> Albert Strasheim wrote:
>> Hello all
>>
>> I'm busy wrapping a C library with NumPy. Some of the functions 
>> operate on a
>> buffer containing structs that look like this:
>>
>> struct node {
>>   int index;
>>   double value;
>> };
>>
>>   
>
> In my previous discussion I was wrong.  You cannot use the 
> array_descriptor format for a data-type and the align keyword at the 
> same time.  You need to use a different method to specify fields.
>
> This, for example:
>
> descr = dtype({'names':['index', 'value'], 
> 'formats':[intc,'f8']},align=1)
>
> On my (32-bit) system it doesn't produce any difference from align=0. 
>
> -Travis
>
>

However notice the difference with

 >>> dtype({'names':['index', 'value'], 'formats':[short,'f8']},align=1)
dtype([('index', '<i2'), ('', '|V2'), ('value', '<f8')])

 >>> dtype({'names':['index', 'value'], 'formats':[short,'f8']},align=0)
dtype([('index', '<i2'), ('value', '<f8')])


There is padding inserted in the first-case.  This corresponds to how 
the compiler packs a short; double struct on my system.   The default is 
align=0.  You need to use the dtype() constructor to change the 
default.   The auto-constructor used in dtype= keyword calls will not 
change the alignment from align=0.


-Travis