Bug with zero size arrays

I should be able to pass zero size arrays to a function. But the compiler removes the call, with the comment:

comment 338 - This expression contains a zero-sized rank, so will do nothing

However, this is not the correct behaviour according to the standard. One might want to do something useful if the array is zero size. In particular, one might want to write defensive code which copes with such cases.

I have checked the standard, and you should definitely allow zero sized arrays to be passed.

I discovered this while implementing quickselect.

Example code (quick select with parts removed):

module statistics
contains
   ! Finds kth smallest (i.e. kth order statistic) amongst an array of values
   function kth_smallest(x, k)

      integer, intent(in) :: k
      real, intent(in) :: x(:)
      real kth_smallest

      ! Return maximum value if k is larger then the array size.
      ! This special case should also be triggered if the array has zero size
      ! in which case, maxval should return -HUGE(1.0) for the default real
      ! kind.

      if (k > size(x)) then
         kth_smallest = maxval(x)
         return
      end if

      ! Return minimum value if k == 1

      if (k == 1) then
         kth_smallest = minval(x)
         return
      end if
     
      ! Assert 1<k <= size(x).
      ! Find kth order statistic using the quickselect algorithm
    
      !** Code not shown

   end function kth_smallest

end module statistics

program test
   use statistics
   real :: x(7) = (/ 1.0, 3.0, 2.0, 4.0, 7.0, 6.0, 5.0/)
   real :: y = 0.0
   ! Test with zero size array section
   y = kth_smallest(x(2:1), 3)  !!<< Bug. When I step with the debugger this is not called.
   ! Should print -Huge(1.0) but doesn't
   print *, y
end program test

David.

Works OK for me. Which version of FTN95 are you running and under what configuration (CHECKMATE etc; Win32 or .NET)?

Quoted from PaulLaidler Works OK for me. Which version of FTN95 are you running and under what configuration (CHECKMATE etc; Win32 or .NET)?

First, I apologize for missing of the contains statement before the function definition -- I have edited the above now to correct this.

I am using version 6.10.0.

I get the same behaviour with Checkmate, Debug, Release configurations for Win32 and .NET.

I get the [u:7aa6078f25]comment[/u:7aa6078f25] message above and the code prints 0.00000 [u:7aa6078f25]instead of -3.402823E+38[/u:7aa6078f25].

I don't mind the comment, its the fact that it removes the call that is not correct. When I step in the debugger, the line is just passed over. When I look at the assembler the code is not there!

If I change things so the compiler cannot know its a zero size array at compile time then it works, e.g.

integer i
print *, 'Enter upper bound of 1'
read *, i !<< enter 1

! ...

y = kth_smallest(x(2:i), 3)
print *, y !< prints -3.402823E+38

David.

That's strange. I am using XP. I will try other operating systems.

The problem does not occur in my current development version of FTN95 but it does occur for me in version 6.10. I cannot think of a connection with a recent fix but there we are.

Thank you. Hopefully it was connected with another fix or development you have made and so will work correctly with the next release.

David.