Re: BSD/OS 2.0.1 and Zmailer 2.99.36

[Matti Aarnio <mea@nic.funet.fi>]

Andrew bugged me about forgetting his logs before makeing 2.99.37..

> BSDI BSD/OS 2.0.1, shlicc2 (gcc 2.6.3 using shlibs), Zmailer 2.99.36
> 
> Nothing really major here.
> 
> 1) Working mmap() test gives 'yes' if I replace (long) with (off_t)
> 	in two places in test program inside configure.

	This needs a fix to GNU autoconf.  The ``configure'' script is
	generated by the autoconf macroes, and I am unwilling to go to
	fiddle with those macro source files (not that I would now know
	M4, I don't, but it would not stop me :) )

> 2) libsh/builtins.c includes sys/param.h and this makes RE_DUP_MAX
> 	redefined. Removed line #include <sys/param.h> because
> 	it's already included in hostenv.h . Now regex.h really
> 	undefs RE_DUP_MAX from previous param.h anf defines its own.

	Ok.

> 3) Line 387 in config.h.in contains typo in macro HAVE_VARARGS_H.

	Oops..

> 4) I got million warnings that const is redefined. I don't know
> 	why gcc 2.6.3 can't pass const test, I just commented out
> 	#define const from config.h.

	??

> 5) libc/allocate.c has ALIGN macro redefined (BSD/OS has that).
> 	Inserted ifdef-undef-endif combination.

	Ok.

> 6) libc/strsignal.c has declaration of sys_siglist conflicting
> 	with unistd.h . This is wrapped with #ifndef __FreeBSD__.
> 	Better solution is to have configure check existence
> 	of sys_siglist. And it tried, but got wrong result.
> 	Again, (like in 1) test program has hardwired declaration
> 	of sys_siglist which is wrong on BSD/OS, so it never
> 	compiled. If test program includes right headers, there
> 	no need to declare sys_siglist. If that will be done,
> 	#ifndef __FreeBSD__ may be removed.

	I must take you on your proposal of BSD/OS account
	to figure that out.  Somehow I do it myself far
	faster, than by listening on people's comments...

	OTOH.. The same problem is at FreeBSD, isn't it ?
	I have one here near me.

> Now it compiles. Still there are problems. lseek() often is
> used with long second argument (in about every program),
> which surprisingly compiles but causes SIGSEGV of mailbox and smtp.
> long is 32 bit, and off_t is 64 bit in BSD/OS versions >= 2.0.

	Weird..
	I have now scattered  (off_t) casts at strategic locations.

> -- 
> /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
> Joint stock company "GVC Energeticy", North-Caucasian Branch, Pyatigorsk
> Andrew Petrovich Kokarev, postmaster         andrew@skfgvc.pyatigorsk.su

	/Matti Aarnio