Re: status of PL/I as a viable language

[Preben Randhol <randhol+news@pvv.org>]

Hyman Rosen wrote:
> 
> My point was that checks should be disabled when it is
> important for programs to continue operation. Security
> situations are just the opposite. In that case, you want
> to abort at any sign of error, perhaps even with timeouts
> to kill things which run too long, because any such issue
> can be a sign of attack.

Yes, but in the case you want to have a program running continously it
won't help if there are errors in the program which makes the program
behave in strange ways. Say a nuclear powerplant. You don't want an
untreated integer overflow to cause the rods to be pulled up causing a new
nuclear meltdown. You want that the program to catch this error and
go into a backup/safe-mode program IMHO.

In my simple program I have an outer exception handling. If an exception
is caught here the program will try to save the data in an EMERGENCY
file before it quits. It will also give the user a notice of the error
of course. When the program starts it always checks if there is an
EMERGENCY file which if it exists will be loaded and the user will be told
that the data is restored and ought to be checked. This works.

Another example, when I'm running my experiments (chemistry) in the lab,
I'm logging the data from an electrode onto a computer. The logging
software worked fine except that I sometimes found that the log had
stopped at some random place in the experiment. These experiments can
be 10 or more hours and I'm not there during this time. After some checking
I found out that what made the program crash was that the some strange
feed came from the logger causing the logging program to crash. I
changed the program so that in case it got this feed it should simply
write -1000 (as this is not a value I would get from my electrode and
thus I can remove it later from the log) and then I got the program to
continue. I can see in my logs that in some experiments I get this
strange error, but after 3 years I still have no idea what causes it,
and it is not important as it doesn't influence my experiment now that
the logging is working.  Note this was not written in Ada, but
some language of the logger.  So my point is that I don't think that one
are more safe turning off all checks in the distributed code. You newer
know the side-effects that this will have. It is better to make an
outer safty net which can catch the any error and deal with it and
notify the user.

-- 
Preben Randhol ---------------- http://www.pvv.org/~randhol/ --
"Violence is the last refuge of the incompetent", Isaac Asimov