forums.silverfrost.com

[KL]

Dear All,

I have noticed a difference between the two compiler options
/Checkmate and /Optimise. Please see the following little
program with neads the input file TestData.
When compiled with /Checkmate the result is "Correct",
when compiled with /Optimise the result is "False".
The program uses the do-loop index outside the do-loop
(which should be allowed), the better solution is commented out.

Program:
Winapp

Program Test43
Implicit None

! Integer :: i, k, nUnit
Integer :: i, nUnit

nUnit = 16
Open ( Unit = nUnit , File = 'TestData' )

Do i = 1, 999
! k = i
Read ( nUnit, *, end = 100 )
End Do

100 Continue

! If ( (k-1) .eq. 3 ) Then
If ( (i-1) .eq. 3 ) Then
Write (*,*) ' "Correct" Result'
Close ( Unit = nUnit )
stop
End If

Write (*,*) ' "False" Result'
Close ( Unit = nUnit )

End Program Test43

TestData (simply 3 lines in the data file):
1
2
3

Bat file to run the program:

del comp.lis
del *.obj
del *.exe

ftn95 Test43.f95 /Optimise /link >> comp.lis
Test43.exe

Does the program violate any rules? At least, the option /Checkmate
does not indicate an error.

Best regards,
Klaus

[JohnHorspool]

Klaus,

I always thought that a do-loop index should never be used outside the do-loop, although I may be wrong.

regards
John

[JohnCampbell]

Using the value of I outside the do loop is valid and often very useful, as you have done, to see where the loop stoped.

Why not simply print out the value of i when you exit the loop.

There may be something wrong with reading nothing, especially free format reads. I don't know, but /opt may differ from /check in this respect.

I'd read a variable or at least a character string.
I also prefer the more general "iostat=ierr" style of error trapping

character string*1

Read ( nUnit, fmt='(a)', iostat=ierr ) string
if (ierr /= 0) exit

John

[PaulLaidler]

There is a bug in FTN95 here. You should be able to use the do loop index in this context.

The problem only occurs when using /opt on the command line and almost any change to the code (e.g. printing the value of i immediately after 100 continue) fixes the problem.

We will aim to fix the bug for the next release of FTN95.

[KL]

Paul,

thank you very much. I had noticed that this error does not occur, when I modify the code sligtely. So this is certainly a minor problem (but it took some time to find it out when I had more or less finished a program development and changed the compiler option to /Optimise).

Best regards,

Klaus