Comment Embedded Directives

I get some nasty error for weeks which is hard to find because the compiler somehow hides it. I am getting Access Violation crash in some subroutine when compiled with /NOCHECK and when I turn all checks and undefs ON the damn error disappears (of course the code in debug mode works much slower)

So i turned to Comment Embedded Directives which i never used and even did not know before (when they appeared in FTN95? at the time of NET development?) in the hope to turn checking (or not checking) one part by one only for small portions of the code inside faulty subroutine. And unfortunately it does not work like i hoped . Will appreciate the clarifications how these directives work.

Here is smallest example of some artificial code where intentionally array elements 4 and 5 are not set and we will be trying to catch that. We compile the code this way

C>FTN95 A.FOR /DEBUG /LINK >z

and inside the code push the directive not to check anything before DO loop and only check DO loop on possible undefined variables turning just before it the UNDEF key ON

	Real a(5), b(5), c(5)

!FTN95$OPTIONS(nocheck)
	a(1)=1
	a(2)=2
	a(3)=3

	b(1)=10
	b(2)=20
	b(3)=30

	d=c(5)

!FTN95$OPTIONS(undef)
	do i=1,5
	 c(i) = a(i)+b(i)
	enddo

	print*, c
	end 

But what happening is that UNDEF starts to work for entire code before and after this directive and for example finds undefined variable at d=c(5) line.

What i am doing wrong ?

If i switch these directives, things become even more weird - i do not get the source code shown in the debugger which confirms that last directive suppresses previous ones in the entire code

This is just a quick reply without any testing but my understanding is that comment embedded directives must be applies before the subprogram to which they relate. /LIST will clarify this because it should indicate the options in force.

Using them before the first line of subroutine was how older OPTIONS(....) command worked. I used it and it is helpful. But I thought this new syntax also means new functionality (and it would be cool by the way and extremely helpful one) when you can compile some parts of the subroutine/function using different compiler options. Syntax kind of resembling OpenMP parallelization instructions ...If this is not done it should be done!

Tried your suggestion with /LIST. It shows that program was compiled with UNDEF in both cases (when either nocheck or undef is used first in source code). The run though reveals that this is not true, the program above stops with error on undefined variable, while if change places of these comment embedded directives it runs without hiccup.

I have noted your suggestion.