[ncl-talk] An "2d" Attribute Array Conundrum

[David Brown]

Hi Alan,
Agreed. Seg faults are never good. We will look into this issue.
Thanks for pointing it out.
 -dave

On Fri, May 15, 2015 at 11:24 AM, Alan Brammer <abrammer at albany.edu> wrote:
> To piggyback on Phillipe's email. I just coincidentally came across this
> problem myself, worse though I get seg faults.
>
> Easy work around but seg faults are never good.
>
> e.g.
>
> res = True
> res at test := new( (/10,10/) , float)
> ; print(res at test(0,3)) ;;; we can't do this, so is it actually a 1d array?
> print(res at test(3))  ;; Seg fault.
>
>
> ~alan
>
>
>
> On 13 May 2015 at 17:01, Philippe Papin <ppapin at albany.edu> wrote:
>>
>> Hi David,
>>
>> Thanks for the detailed reply and suggestions. I can easily assign the
>> attribute values to a local variable! Also good to know that 2d arrays are
>> allowed in attributes even if the are not fully supported!
>>
>> Have a great day!
>> -Philippe
>>
>> On Wed, May 13, 2015 at 2:17 PM, David Brown <dbrown at ucar.edu> wrote:
>>>
>>> Hi Phillipe,
>>> That's a good question. Actually the error message, as you discovered,
>>> is a bit misleading. NCL was originally designed to only allow 1D
>>> attributes in keeping with the NetCDF data model. Later, however, when
>>> we needed the ability to transmit 2D coordinate information to the
>>> graphics routines (mainly), we loosened the restriction to allow NCL
>>> variables to have multi-d attributes as a convenience. However, as you
>>> have found, we did not implement full support for this feature. Partly
>>> this is because you cannot write such an attribute to any of the
>>> output file formats we support -- so it can never be truly integrated
>>> into the data model.
>>> Practically speaking, if you wanted to have a routine that handles
>>> subscripting of 2D attributes, I'm guessing you could do it by
>>> assigning the attribute value to a local variable, subscripting the
>>> variable, and using the 'reassign' syntax to give the attribute its
>>> new subscripted value.
>>> We will try to make the error message a little more informative.
>>> Eventually maybe we can make subscripting work in this situation.
>>>  -dave
>>>
>>> On Wed, May 13, 2015 at 10:56 AM, Philippe Papin <ppapin at albany.edu>
>>> wrote:
>>> > Hi NCL-talk,
>>> >
>>> > I came across a strange conundrum today when testing some code. I use
>>> > attributes in my code to send 1d variables through attributes into
>>> > subroutines without explicitly defining it in the new subroutine. This
>>> > practice is limited to 1d arrays since attributes in theory can only be
>>> > 1d.
>>> > However, if this is the case, why can you "create" a 2d attribute array
>>> > and
>>> > then print this attribute array without error?
>>> >
>>> > Here is a code snippet of an example of what I am talking about. I'm
>>> > using
>>> > NCL version 6.2.1
>>> >
>>> > opt = True
>>> > array2d = (/ (/1,2,3/) , (/4,5,6/) /)
>>> > opt at array2d = array2d
>>> > print(opt at array2d) ;;; output below seem to print out array correctly
>>> > in 2d
>>> >
>>> > (0,0) 1
>>> >
>>> > (0,1) 2
>>> >
>>> > (0,2) 3
>>> >
>>> > (1,0) 4
>>> >
>>> > (1,1) 5
>>> >
>>> > (1,2) 6
>>> >
>>> > However if you try to actually subscript the data, you get an error
>>> > message
>>> > that attributes only have one dimension.
>>> >
>>> > subarray2d = opt at array2d(0,1)
>>> >
>>> > fatal:Attributes only have one dimension, 2 subscripts used
>>> >
>>> > fatal:["Execute.c":8576]:Execute: Error occurred at or near line 6
>>> >
>>> > This is a bit counterintuitive to me. Perhaps the 2d attribute array I
>>> > created is really 1d, but just gives the appearance of a 2d array when
>>> > printed. Sorry if this is a trivial question, but hopefully that means
>>> > it
>>> > won't take much to figure out what is going on here.
>>> >
>>> > Thanks for any additional information!
>>> >
>>> > -Philippe
>>> >
>>> >
>>> > --
>>> > -------------------------------------
>>> > Philippe P. Papin
>>> > Dept. of Atmospheric and Environmental Sciences
>>> > University at Albany, SUNY
>>> > 1400 Washington Ave, Albany, NY 12222
>>> > http://www.atmos.albany.edu/student/ppapin/
>>> >
>>> > _______________________________________________
>>> > ncl-talk mailing list
>>> > List instructions, subscriber options, unsubscribe:
>>> > http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>>> >
>>> _______________________________________________
>>> ncl-talk mailing list
>>> List instructions, subscriber options, unsubscribe:
>>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>>
>>
>>
>>
>> --
>> -------------------------------------
>> Philippe P. Papin
>> Dept. of Atmospheric and Environmental Sciences
>> University at Albany, SUNY
>> 1400 Washington Ave, Albany, NY 12222
>> http://www.atmos.albany.edu/student/ppapin/
>>
>> _______________________________________________
>> ncl-talk mailing list
>> List instructions, subscriber options, unsubscribe:
>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>>
>