Re: Using ATC to abort code written in Fortran

[Tucker Taft <stt@averstar.com>]

Duncan Sands wrote:
> 
> I have a heavy calculation to do which may have to be aborted in the
> middle,
> using ATC.  Thus
> 
> select
>    A.Trigger;
> then abort
>    Huge_Calculation;
> end select;
> 
> Most of the Huge_Calculation is performed by a numerical library written
> 
> in Fortran.  I am worried that ATC may not work correctly if the abort
> comes
> while in the Fortran code.  Does anyone have any idea?

I suspect it will work, but you need to check with the compiler vendor.

> I can see two kinds of potential problems.  The first comes from how the
> fortran
> code is written.  For example, if it stores information in global
> variables then these
> may be left corrupted after ATC.  I'm not so worried by this, because I
> can check
> the code and do some rewriting if necessary.  I am more worried about
> the second
> kind of potential problem: will ATC trash memory etc because the code
> produced
> by the fortran compiler is not compatible with ATC?  I've no idea how
> ATC is
> implemented, so I've no idea if there could really be code compatibility
> problems
> or not.  Perhaps someone out there has some experience with this kind of
> thing...
> 
> I am using g77 (egcs version) for compiling the fortran and GNAT 3.12p
> for the
> Ada.

I'm not sure how GNAT supports Abort and ATC.  There are generally two
approaches, one which involves propagating a special exception, the
other involves doing the equivalent of a "longjmp."  Longjmp should
work independently of what kind of code g77 generates.  Using a special
exception may or may not work, depending on how exceptions are propagated
through non-Ada code.  This sounds like a question for the ACT folks...

Note that ATC is not some nasty Ada "invention."  It was inspired by
the ANSI C setjmp/longjmp and sigsetjmp/siglongjmp.  There are some
dangers in using ATC, but essentially every interactive program that
catches "control-C" on Unix and then re-prompts uses setjmp/longjmp 
or equivalent, so there is plenty of code out there that uses ATC-like
constructs.

> Thanks for your help,
> 
> Duncan Sands.
> 
> PS: Of course, I could wrap the Fortran code up in a protected object,
> and exploit the abort deferred property of protected objects to avoid
> having the abort come during execution of the Fortran code.  But since
> it is the Fortran part that is likely to need to be aborted (because it
> can take a long time to complete), this is not very satisfactory.

-- 
-Tucker Taft   stt@averstar.com   http://www.averstar.com/~stt/
Technical Director, Distributed IT Solutions  (www.averstar.com/tools)
AverStar (formerly Intermetrics, Inc.)   Burlington, MA  USA