Segmentation fault on VADSself

Segmentation fault on VADSself

[FAVIER PIERRE]

We have a program compiled using rational VADSself 6.2.3d on HP-UX/9.07
which has always been working very well.

Now, when recompiling on HP-UX/10.20 (with the same version of the compiler)
we get random segmentation faults. Nevertheless, when running the program
under the debugger it runs fine.

Any clue on how to solve this?

Re: Segmentation fault on VADSself

[Robert Dewar]

In article <7qer2f$j2a$1@mars.worldonline.fr>,
  "FAVIER PIERRE" <pierre.favier@worldonline.fr> wrote:
> Any clue on how to solve this?

Does the vendor say the compiler should work with the later
OS version? If not, then no reason to think it will. If so,
then I would say let the vendor sort out the problem!
Generally you cannot assume version X of a compiler that works
on version Y of an OS will work on version Y+1 of the OS,
even if upwards compatibility is promised by the OS vendor.

Robert Dewar
Ada Core Technologies


Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.

Re: Segmentation fault on VADSself

[David Tannen]

Hi,

>We have a program compiled using rational VADSself 6.2.3d on 
>HP-UX/9.07 which has always been working very well.

Hey this sounds familiar.
 
>Now, when recompiling on HP-UX/10.20 (with the same version 
>of the compiler) we get random segmentation faults. 

As Robert asked - you do have the VADSself 6.2.3d for 
HPUX 10.20?

>Nevertheless, when running the program under the debugger 
>it runs fine.

Yeah the VADS debugger is kinds of a pain on this.  Are
you using the command line debugger or a GUI one? We 
found the behaved in different manners.

>Any clue on how to solve this?

Yes.  First of all make sure you have the latest HP/UX 
ACE patches installed on your system.  If you dig deep
enough using the debugger you will eventually find that
the problem lies someplace in either the VADS runtime or
the HP/UX OS code.  Our segmentation faults went away
when we installed the ACE patches.

Another area that could be causing problems would be 
anyplace where you *might* have variant records that
are being allocated space by the following method:
1.  You have a standard record the same size as your 
    largest variant.
2.  You have access types for the standard & variant
    record.
3.  You use 'is new' to create space with the standard
    record.
4.  You then use unchecked conversion between the 
    two different access types.

You will find that the moment you try to access a
part of memory associated with the variant part of
the record, you get a core dump.

The fix is easy - do a pragma byte pack on both records.

There you go - that should save you a couple months worth
of digging.  You will be getting my bill in the morning <G>.

An ambassador for Christ
David Tannen (bluesguy@jcdisciples.org)
John 14:21 "Whoever has my commands and obeys them, he is the one who
            loves me.  He who loves me will be loved by my Father, 
            and I too will love him and show myself to him."
Check Out:   http://www.jcdisciples.org/davidtannen/index.html

Re: Segmentation fault on VADSself

[Simon Wright]

David Tannen <bluesguy@jcdisciples.org> writes:

> Another area that could be causing problems would be 
> anyplace where you *might* have variant records that
> are being allocated space by the following method:
> 1.  You have a standard record the same size as your 
>     largest variant.
> 2.  You have access types for the standard & variant
>     record.
> 3.  You use 'is new' to create space with the standard
>     record.
> 4.  You then use unchecked conversion between the 
>     two different access types.
> 
> You will find that the moment you try to access a
> part of memory associated with the variant part of
> the record, you get a core dump.

This sounds like bitter exerience speaking :-)

My advice, FWIW, would have been to find another way of solving this
problem that didn't involve variant records; it sounds horribly
non-portable!

Re: Segmentation fault on VADSself

[David Tannen]

Simon,

>This sounds like bitter exerience speaking :-)

It certainly is.

 
>My advice, FWIW, would have been to find another way of 
>solving this problem that didn't involve variant records; 
>it sounds horribly non-portable!

In my case I inherited a lot of code that is running on VAX,
Solaris & HP/UX.  The funny thing is that the very same code
w/ the very same version of the VADS compiler works perfectly
on Solaris.  Bizzare!

Just the joys of s/w development.

-- 
An ambassador for Christ
David Tannen (bluesguy@jcdisciples.org)
John 14:21 "Whoever has my commands and obeys them, he is the one who
            loves me.  He who loves me will be loved by my Father, 
            and I too will love him and show myself to him."
Check Out:   http://www.jcdisciples.org/davidtannen/index.html