Bug? Datetime creation produces both NaT and 1-Jan-1970 values.

When I run the following code in either R2014b or R2016a I get:

 clear d, d(:,2) = datetime(randi(10,5,3))
 d = 
    NaT           03-Feb-0007
    01-Jan-1970   10-Mar-0009
    01-Jan-1970   04-Sep-0010
    01-Jan-1970   02-Mar-0006
    01-Jan-1970   03-Sep-0002

and with 3 columns instead of 2:

 clear d, d(:,3) = datetime(randi(10,5,3))
 d = 
    NaT           01-Jan-1970   06-Feb-0008
    NaT           01-Jan-1970   03-Feb-0008
    01-Jan-1970   01-Jan-1970   08-May-0003
    01-Jan-1970   01-Jan-1970   03-Oct-0007
    01-Jan-1970   01-Jan-1970   06-Apr-0007

and with a column vector of 6 elements instead of 5:

 clear d, d(:,3) = datetime(randi(10,6,3))
 d = 
    NaT           01-Jan-1970   08-Jul-0008
    NaT           01-Jan-1970   08-Apr-0001
    01-Jan-1970   01-Jan-1970   02-Oct-0003
    01-Jan-1970   01-Jan-1970   05-Jan-0001
    01-Jan-1970   01-Jan-1970   05-May-0001
    01-Jan-1970   01-Jan-1970   07-Apr-0009

So, what this does is create a column vector of some random datetime values and assign it to a non-existent matrix in its last column. If you'd do this with a vector of numeric values, all other values in the matrix would become zeroes by default, e.g.:

 clear M, M(:,2) = ones(3,1)
 M =
      0     1
      0     1
      0     1

As we can see, the values in the datetime matrix that are not set are not zeroes. This cannot be, because a datetime must always obey a format such as dd-MM-yyyy like in the default behaviour. It appears MATLAB tries to fill the rest of the matrix with either

 datetime(0,'ConvertFrom','posixtime')
 ans = 
    01-Jan-1970 00:00:00

or NaT (Not-a-Time), the equivalent of the numeric NaN (Not-a-Number).

Why does MATLAB behave like this? Is this a bug, or am I causing it?

Answer 1

[Peter Perkins]

Erik, that does appear to be a bug. You are making an indexed assignment to only some elements of a datetime array that does not yet exist. I will make a note to have this looked at. The work-around would be to pre-allocate the array with NaT:

>> clear d, d = NaT(5,2); d(:,2) = datetime(randi(10,5,3))
d = 
   NaT           09-Oct-0008
   NaT           03-Jun-0003
   NaT           09-Feb-0006
   NaT           03-Feb-0007
   NaT           10-Mar-0009

[Erik]

What is MATLAB's original intention with its currently bugged auto-preallocation of datetime arrays: NaT or some ‘zero’ datetime? Numeric matrices get zeroes, not NaN. Should the default datetime be NaT then?

I know it can't be zero, but maybe using NaT creates problems when the datetime matrix is used by other functions?

Thanks for looking into this, I hope to see it fixed in an upcoming release!

[Philip Borghesani]

Defaulting to NaT is well established by simpler use cases that work correctly:

 clear d
 d(4)=datetime

 d = 
   NaT         NaT         NaT       25-Mar-2016 14:34:51

Note that empty also works around the problem:

 >> clear d, d = datetime.empty(5,0); d(:,2) = datetime(randi(10,5,3))
d = 

   NaT           03-Oct-0007
   NaT           06-Apr-0007
   NaT           07-Jun-0002
   NaT           09-Mar-0002
   NaT           10-Aug-0005