PL/I compiler

[Bill Kennedy]

>dww at math.fu-berlin.de (Debora Weber-Wulff) writes:
>>PL/I? On an MS-DOS or CP/M (!!!) platform with close to full
>>implementation? Well, the beast compiler we used to use on our
 
cpcahil at virtech.uucp (Conor P. Cahill) replies:
>I had a PL/1 compiler for MS-DOS about 5 years ago.  I'll look around
>my basement to see if I can come up with the manufacturer (I know it
>was distributed by IBM).

The Digital Research PL/I compiler was a rather rich subset G implementation.
The G subset is far from "big" PL/I but it is quite adequate (I missed the
select) for most practical things.  Moreover, since PL/I is so unpopular, I
found the code to be very portable, nobody wants to enhance anything.  The
most robust implementation was for CP/M-80, worked great on 8080, 8085, and
Z80 machines, I never checked the later models.  The CP/M-86 version was a
hastily xlated (the xlate program was written in PL/I-80 and the turkeys
lost the source!) and although a little wobbly, also worked rather well.  When
it appeared that CP/M-86 was going to lose, big time, to DOS they did a quick
and dirty conversion to DOS 1.1 [sic] and it was the worst of the bunch.  It
was cantakerous and generally unreliable.  I don't know which one IBM shipped,
probably CP/M-86, I doubt that the DOS version could have passed even the
least stringent acceptance tests.

That bit of museum work out of the way, LPI has a rather fully implemented PL/I
for '386 UNIX and while I have no experience with it, the genes are good.  It,
like LPI's other languages, was shrunk down from "big" PL/I and was written
in PL/I.  Maybe their C compiler is written in C, but I'm pretty sure their
others are written in PL/I.  I'm a little sorry that the language fell into
such ill repute, if you avoid the chrome and tail fins it's a pretty good
language.  I lust for its BCD arithmetic for money calculations and I still
understand its structures better than C.  It's certainly possible to write a
bad program in PL/I but it's lots easier to write something unreadable in C.
If LPI did V/386 PL/I correctly, I have no reason to believe they didn't, it
should be trim and compact.  They've certainly got their code generator act
together by now and the linker shouldn't carry along any baggage you didn't
ask for.  I'll also speculate that it's subset G with some extensions (select
maybe?  huh? Huh? :-) which should be sufficient for as much work as you'd
want to do on a '386.
-- 
Bill Kennedy  internet  bill at ssbn.WLK.COM or ssbn!bill at attmail.COM
              uucp      {att,cs.utexas.edu,pyramid!daver}!ssbn.wlk.com!bill