Testing parameters with N_ELEMENTS and !null

The N_ELEMENTS function is typically used to test whether an input parameter has data. For example, here's a program that doubles a number:

function double_it, x
   compile_opt idl2
   on_error, 2

   if n_elements(x) eq 0 then $
      message, 'Need input number to double.'

   return, x*2
end

N_ELEMENTS is used to test whether the user actually passed anything when calling DOUBLE_IT. If not, an error message is thrown:

IDL> a = double_it(4)
IDL> print, a
           8
IDL> b = double_it()
% DOUBLE_IT: Need input number to double.
% Execution halted at: $MAIN$

In IDL 8, we have the option of instead comparing parameters with the null variable, !null. In DOUBLE_IT, this looks like:

function double_it, x
   compile_opt idl2
   on_error, 2

   if x eq !null then $
      message, 'Need input number to double.'

   return, x*2
end

Now, which is faster: using N_ELEMENTS or !null? Here's a simple test program:

pro test_nullparameter, param
   compile_opt idl2

   n_iter = 1e7

   t0 = systime(/seconds)
   for i=1, n_iter do a = n_elements(param) eq 0
   t1 = systime(/seconds)
   print, 'N_ELEMENTS:', t1-t0, format='(a15,f12.8,1x,"s")'

   t0 = systime(/seconds)
   for i=1, n_iter do a = param eq !null
   t1 = systime(/seconds)
   print, '!null:', t1-t0, format='(a15,f12.8,1x,"s")'
end

and here's a sample result from my laptop:

IDL> test_nullparameter
    N_ELEMENTS:  1.07800007 s
         !null:  0.84400010 s

It turns out that it's more efficient to compare against !null. The syntax is more compact, too. (Thanks to Jim Pendleton, who initially pointed out this to me.)