forums.silverfrost.com

[yair999]

The following program yields the value of 1.00000 under debug win32
mode of FTN95 compiler version 5.20.1.
The correct value should be 2.00000
I do get the correct value when the 'OPTIONS(CHECK)' is commented out.

Can someone check it out?


yair


PRINT *,F2()
END
FUNCTION F1(X)
F1=X
END
OPTIONS(CHECK)
FUNCTION F2()
REAL X(1)/1./
F2=ABS(F1(X(1)))*2.
END

[JohnCampbell]

I tested the function F2, using a temporary variable and found when it did and did not work. I've listed the program below. Surprisingly the F1 call using a temporary variable doesn't works in :-
y = ABS (F1(X(1))) * 2.
f2 = y

I used ftn95 Ver 4.9.1, so it is not a recent bug.

Regards John

! Test code
real f2
PRINT *,F2()
END

real FUNCTION F1 (X)
real x
F1=X
END

OPTIONS(CHECK)
real FUNCTION F2()
real f1 , y
external f1
REAL X(1)/1./
!
! this works
! y = F1(X(1))
! f2 = abs(y) * 2.
!
! this works
! y = ABS (F1(X(1)))
! f2 = y * 2.
!
! this doesn't works
y = ABS (F1(X(1))) * 2.
f2 = y
!
! this doesn't work
! F2 = ABS (F1(X(1))) * 2.
END

[PaulLaidler]

Thanks for the bug report.
It appears to be a new bug rather than a regression.
A temporary work-around is to put the 2 at the beginning.

[yair999]

I'd like to add that my example was the minimal program that demonstrates my original problem: A 12,000 lines legacy code which yielded 'slightly' different results under CHECKMATE mode than under DEBUG or RELEASE (the last two being identical).

Yair

[yair999]

[PaulLaidler]

We have only just fixed this bug so I don't think that the fix will have got into the current release.