On 24 Sep 2014 17:05:51 -0700, in bit.listserv.ibm-main you wrote: >As we're going to install IBM Enterprise COBOL 5.1 soon I'm looking to decide >what default compile options we should use, and if we should perhaps change >some. Plus we have to decide what the defaults for the new compile options >should be. We'll just discuss NUMPROC here so-as to not make the post TOO >long. > > >NUMPROC: To PFD or to not PFD >Currently we use NUMPROC=MIG, which is no longer even an available option. > >Looks like PFD generates more efficient code because it does not do 'numeric >fixup'. > >Do we only need to be concerned with the numeric class test, but not with >moves and arithmetic? My testing shows the following: > >with NUMPROC=PFD >- unsigned data in signed field: fails NUMERIC test >- signed data in unsigned field: fails NUMERIC test > >with NUMPROC=NOPFD and NUMCLS=PRIM >- unsigned data in signed field: fails NUMERIC test >- signed data in unsigned field: passes NUMERIC test > This looks like a serious bug which will affect all customers who still have CSP (Cross System Product) since CSP and probably its successors VisualGen, VisualAge Gen and Rational Business Developer forced an 'F' zone for positive numbers in signed fields. I ran into this in 1995 or 1996 while a contract programmer. Also signed data in an unsigned field always should be considered NOT NUMERIC. Getting rid of NUMPROC(MIG) was dumb if Rational Business Developer still forces 'F' zones for positive numbers.
Clark Morris > >So in either case numeric fails if there is unsigned data (last nibble = f) in >a field position we've defined as being signed. Since this is already >"wrong", if we're doing it (and I don't know that we are!), no harm going to >PFD in this case. > >But it appears that we could, with NOPFD, pass a numeric test if the data is >signed (either positive or negative) but the field is defined as unsigned. >And this would no longer pass with PFD. So the question is, do we do this? >Well, I don't know. And I can't imagine looking through maybe tens of >thousands of numeric tests (or may not that many...) to make sure the data >matches the definition. > >Should we risk it? Has anyone out there ever "migrated" to PFD with "legacy" >data files? I'm quite certain that we do have at least some cases where one >program wrote out data defined as signed and other programs read it with the >same area defined as unsigned. Curse it, but I know I've seen it. I think >the less likely, but not impossible, situation would be testing it for >numeric. Mostly we only test for numeric if: >- We sometimes intentionally store non-numeric (usually low-values; maybe >spaces sometimes) data in a numeric field in some situations >- We're getting data from an external source and want to validate it (though I >imagine we do this far less than we perhaps should!). > >Thanks, >Frank > >---------------------------------------------------------------------- >For IBM-MAIN subscribe / signoff / archive access instructions, >send email to [email protected] with the message: INFO IBM-MAIN ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
