Re: program crash

Larry Ducie wrote:
> 
> So, it would appear to be a balancing act - increased level of information
> vs reduced program performance. In some circumstances it may be better to
> code a PSDS in a program which uses only the first 80 bytes, and include
> subfields past byte 80 only if specifically required - such as the job name
> or user name. Of course, the performance hit is felt when the program is
> initialised, so it is much less of an impact if the program is called
> several times and returns with LR off, or if it is a long-running program.
> It's just a case of getting the right balance.
> 

Larry, there are only a few subfields that aren't available in the
formatted dump.  The dump routine fetches most of the information if it
hasn't already been fetched during initialization.  The job name and
user name would be in the dump even if the PSDS wasn't long enough.  The
only subfields that wouldn't be available in the dump is the ones that
are meaningless if not obtained during initialization (like the time the
program started).

Try running this program and reply 'D' to the divide-by-zero error, and
then check the dump to see what's not available.

C                   div       zero          zero              5 0   
C                   seton                                        lr 

But actually the performance hit for having a PSDS is pretty small.  I
don't think it wouldn't be noticeable unless you repeatedly call
programs that do very little, AND set LR on every time.  I would code a
PSDS of whatever size I wanted, whenever I wanted, and not worry about
*INIT performance.

I just ran a little test.  If I have program that has a couple of loops
with 500,000 iterations, first calling a program with a PSDS that just
sets on LR, and then calling a program without a PSDS that just sets on
LR, then it takes about 20 times longer to do the has-PSDS loop (on my
system, 79 seconds vs 4 seconds).  I added an F spec for a printer file
to the both programs, (adding an open and close of one file) and
adjusted the number of loop iterations to 2000 (guessing correctly that
it would take way longer to make the calls).  The no-psds loop took 15
seconds and the has-psds loop took 16 seconds.

Conclusion: don't worry about the time it takes to initialize the PSDS.