For a student assignment, students wrote something like this
(I have stripped it down and changed some symbol names):

program SincTest
implicit none
real :: x, SINX_X
read *, x
print '(" sinc(", f6.2, ") =", f8.4)', x, SINX_X(x)
end program SincTest

real function SINX_X (x)
real :: x, z, small
small = 1.0E-6
if (ABS(x) < small) then
z = 1.0
else
z = (SIN(x))/x
endif
! SINX_X = z
end function SINX_X

Output for input x = 1:

Dieter Britz wrote:

(snip)

(snip)

It is certainly not obvious. Looking at the generated assembly code
in both cases might help understand what it is doing.

My guess is that the return value goes into a register, which
would normally be done by an assignment to SINX_X. Without that
assignment the value left in the register previously is returned.

Dieter Britz wrote:
[example elided]

I don't see that as being usefully answerable in this case. One can
speculate about how the compiler happens to assign addresses and
registers and how it might happen to get the "right" answer for that
particular compiler in this particular case, but I don't see a useful
generalization from this. It isn't as though there is a principle that
you can constructively use in coding.

Yes, it is possible for random coding errors to occasionally get the
"right" answer in particular cases on particular compilers. Plenty of
examples exist. In fact, it can be a bit unfortunate when the programmer
refuses to accept that there could be a programming error, since the
answer was "right"; I've run into that problem more than once.

I'll note, however, that I'd expect a good quality compiler to bitch
about the function never being assigned a value in this case. I have
seen compilers bitch about that.

Richard Maine wrote:

(snip)

Another possibility, which I don't really believe, is that the
compiler figures out that z was supposed to be assigned to SINX_X.
The old "do what I mean, not what I say" logic.

-- glen

As I once said (and disbelieved) on this Forum, the old F77 compiler I
use will allow a function to be called as a subroutine; in which case
nothing is done with the computed function result

In this particular case, the function returns an integer in AX (or
EAX); else the address of the value calculated by the function. If
called as a subroutine, the same happens.
So if there is no assignment of the function value to be returned
(before the END statment), what is in ax/EAX will be treated as the
computed result (or the address of same if not an integer).

Hence I think the corrected reason has already been suggested by
Dieter Britz.

glen herrmannsfeldt wrote:

Richard Maine wrote:

"Dieter Britz" wrote in message
news:fdtd5i$o0j$1@news.net.uni-c.dk...

The standard does not require that a compiler detect that a function result
is not defined, but it's a very helpful warning. Intel 10 says:

C:\test\test.f90(9) : Warning: The return value of this FUNCTION has not
been
defined. [SINX_X]

and gives an incorrect result if nevertheless forced into execution. That's
much better that what PGI manages.

Regards,

Mike Metcalf

glen herrmannsfeldt wrote:

Dieter Britz wrote: