Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp
Path: utzoo!watmath!clyde!bonnie!akgua!whuxlm!harpo!decvax!decwrl!sun!guy
From: guy@sun.uucp (Guy Harris)
Newsgroups: net.sources.games
Subject: Re: Robots for XENIX Rel 3.0
Message-ID: <2317@sun.uucp>
Date: Mon, 17-Jun-85 00:23:03 EDT
Article-I.D.: sun.2317
Posted: Mon Jun 17 00:23:03 1985
Date-Received: Wed, 19-Jun-85 02:01:40 EDT
References: <612@intelca.UUCP>
Distribution: net
Organization: Sun Microsystems, Inc.
Lines: 73

A lot of the changes "ifdef"fed with XENIX reflect the way the code should
have been in the first place:

1) Contrary to popular belief, "int"s are not guaranteed to be 32 bits long.
All of the variables that are declared with LONG in the XENIX version should
be declared as long in *all* versions (yes, even under 4.xBSD).  All
"printf" and "scanf" formats that print or read into these variables should
use "%ld" format, not "%d".  The second argument to "lseek" should always be
a "long", so you should say

	lseek(fd,0L,0);

not

	lseek(fd,0,0);

even on 4.xBSD systems.

2) Contrary to popular belief, not all UNIX systems have "_doprnt".  System
III doesn't have it; PDP-11 V7 and System Vs Releases 1 and 2 have it, but
don't document it; 4.2BSD has it and made the mistake of documenting it.
Zilog's ZEUS doesn't have it because the Zilog calling sequence makes
routines that take a variable number of arguments a pain to implement.

If you take Chuck McManus' corrected version and:

	Change all the "ifdef"fed sections involving "int" vs. "long int"
	stuff or the "msg" routine to *only* contain the code compiled if
	XENIX is defined,

	Change the section that reads

		# ifdef XENIX
		# include <sys/types.h>
		# endif
		# include <sys/file.h>

	to read

		# ifdef BSD4_2
		# include <sys/file.h>
		# endif

	because the inclusion of "sys/file.h" seems to be solely for
	the benefit of the "flock" system call which only exists on
	4.2BSD,

	Change all other occurrences of # ifdef XENIX to # ifndef BSD4_2,
	because all the other # ifdef XENIXes are there to conditionally
	compile in code using 4.2BSD features (if you feel *really*
	ambitious, you can throw in #ifdefs for the file-locking stuff to
	use the various flavors of "lockf" derived from the original John
	Bass version for those versions of UNIX that support it),

	And run the whole thing through "lint" to catch trash like

	       seed = time(0)+pass->pw_uid;

	and replace it with the *correct* version which is

	       seed = time((time_t *)0)+pass->pw_uid;

	(I'd classify this as a venial sin, except that I've fixed
	that kind of stuff enough times on CCI's 16-bit-"int"/32-bit-pointer
	machines that I think it deserves to be a mortal sin).

Could people take a little more care *NOT* to make assumptions about the
version of UNIX or machine they're writing for when writing code?  If so,
there'll be a lot fewer "has anybody gotten 'Hump The Hostess' running under
{V7,4.1BSD,4.2BSD,System III,System V,Xenix,Zeus,Penix}" questions, and a
lot fewer repeated postings of programs until a portable version emerges.

	Guy Harris