A very serious bug introduced in the last update of CVF 6.6

jeanvezina · ‎10-12-2004

To all:

The last update of CVF 6.6, the C3 update, has introduced a

very serious bug. This was discussed recently in the

comp.lang.fortran newsgroup. I bring it to your attention here.

The bug is that array strides ARE NO LONGER FUNCTIONAL when an allocatable array is used.

Here is a test program

Code:

program test
implicit none
real, allocatable :: x(:)
allocate (x(10))
x = 1.0
print *,x(::2)
print *, size(x(::2))
end program test

If you compile and run this test program with the latest

update of CVF ( 6.6-4088-47D3B), you will see that

10 values are printed (instead of 5) and that the size

of the array section returned by the SIZE function is 10!

This bug was absent in the two previous C1 and C2 updates.

In my opinion, HP should fix this.

In our case, switching to the Intel 8.1 compiler is not an option

because it is stillunable (hang) to compile the large files produced by the Module Wizard .

Best regards,

Jean Vezina

Steven_L_Intel1 · ‎10-12-2004

I find it curious that it has taken eight months for this problem to be noticed. I don't recall seeing it reported to HP before this. In any event, I don't think anyone from HP reads this forum - please report it, if you haven't already, to vf-support@hp.com.

The long compile problem is being worked on.

jeanvezina · ‎10-13-2004

Thanks,

Best regards,

Jean Vezina

jeanvezina · ‎10-13-2004

Well, the vf-support@hp.com e-mail address appears to bedisabled as

my report bounced three times with the message:

host mail.compaq.com[161.114.64.101] said: 554
: Recipient address rejected: Access denied (in reply to RCPT TO command)

Jean Vezina

anthonyrichards · ‎10-13-2004

The solution would appear to be to go back one update. How is that done, Steve?

Intel_C_Intel · ‎10-13-2004

I cut and pasted the code into a test workspace, but was unable to reproduce the error.

Win2K SP2, CVF 6.6C on a P4.

Both the default Debug and Release Builds produce five vales printed and a size of 5.

The Help>About for MSVS does not give the full designation you mention for the 6.6C update (6.6-4088-47D3B). How can I confirm that I am [not] testing the same subversion of the compiler?

Cliff

jeanvezina · ‎10-13-2004

Open a DOS box (command prompt), and type:

df /what

The exact compiler version will be displayed.

Regards,

Jean Vezina

Steven_L_Intel1 · ‎10-13-2004

Try vf-support@compaq.com if the hp.com address doesn't work. It should.

The C3 update will show, in df-what:

Compaq Visual Fortran 6.6-4088-47D3B

I don't know if the earlier 6.6C updates are on the FTP site - possibly not. They both had VERY serious bugs. I suspect that this particular bug is less serious than it seems at first and may depend on some other particular aspect of the program being used.

If you need to back up, you have to uninstall CVF, reinstall 6.6 and then apply the update you want.

lklawrie1 · ‎10-13-2004

I couldn't make it work as stated in CVF 6.6C but I moved the size above the first print and, yes, it prints out 10, etc as stated.

I can verify the program segment works as the user expects in IVF 8.1 as well as Lahey 7.1.1

Linda

gregscvf · ‎10-13-2004

FWIW,

Using C3, I repeat the errorof 10 with the "Print *, size..." statement, but the "Print *, x(::2)"statement gives: Array bound exceeded.

But both commands work fine using x(:10:2), or x(1:10:2).

Greg

jeanvezina · ‎10-13-2004

To all:

The HP support e-mail address now works and I was able to send a problem report. It was assigned the problemnumber CVF22412.

The fact that the test program behaves differently from user to user

seems to show that the generated coderefersto unitialized memory

locations for the array descriptor.

Regards,

Jean Vezina

Intel_C_Intel · ‎10-14-2004

Jean,

C:>df /what

Compaq Visual Fortran 6.6-3780-47D3B

This subversion doesn't seem to have the stride problem. You could revert to that subversion, but earlier comments in this thread make it sound like it has other bugs I haven't run into yet.

Good Luck, Cliff

Jugoslav_Dujic · ‎10-14-2004

As I recall, the problems with two earlier versions were mostly related with F9x standards checking (that I don't care about). I happily use the same 3780 version (the first 6.6C release?) for more than a year. But then, I don't care about this stride problem in 4088 either :-).

Jugoslav

anthonyrichards · ‎10-14-2004

DF /what on my installationgives

Compaq Visual Fortran 6.6-4018-47D3B
C:Program FilesMicrosoft Visual StudioDF98BINdecfor90.exe

Am I ok? Or is the original poster misprinting 4018 as 4088 ?

Jugoslav_Dujic · ‎10-14-2004

In short:

1st version:

CVF-66-66C.exe (Nov 2003)

Compaq Visual Fortran 6.6-3780-47D3B

2ndversion:

CVF-66-6C.exe (Jan 2004)

Compaq Visual Fortran 6.6-4018-47D3B

3rfversion:

CVF-66-66C3.exe (Feb 2004)

Compaq Visual Fortran 6.6-4088-47D3B

Jugoslav

anthonyrichards · ‎10-14-2004

Thanks, Jugoslav.