On Sun, 3 Mar 2013 08:51:34 -0500, John Gilmore <[email protected]> wrote:
>I come now to Tony Harminc's example: > ><begin extract> >But SETRP generates a NOPR with an expression (related to the SDWA, I think) >obviously intended (and I think commented) to fail if the length is >not 0. However HLASM doesn't think the expression is a likely register >value - a legal one, certainly - but still worth a warning if you have >registers EQUated with the GR or GR32 or GR64 option. ></end extract> > >It is very different. Register equates are ubiquitous. What we thus >have in this example is no or inadequate testing, and that is not >defensible. None of us expects IBM code to be error-free. None of us >writes such code. We do expect that IBM code will have been tested, >in effect that such errors as we find in it will be subtle and not >crudely obvious ones; and in this expectation we are now often >disappointed. I'll have to disagree with you, John. What we have there is (I believe) an old macro, using techniques that work perfectly well, unless someone uses an HLASM option that did not exist when the macro was written. If IBM has not needed to change the macro since HLASM created that option, then there has been no need to test the macro. Even if IBM has had to change the macro, there is nothing that would require IBM's testers to try it with all possible HLASM options and combinations of options. Note that I'm not saying the macro is as good as it could be. And I'm not saying that IBM shouldn't improve it. But claiming inadequate testing, or claiming that the macro definitively has an error, seems inappropriate to me. -- Walt (who is, of course, no longer an IBMer but once was) ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
