[ On Friday, November 28, 2003 at 08:19:32 (+1100), Luke Mewburn wrote: ]
> Subject: Re: Fwd: libc without YP
>
> Some points:
> 
>     *	The compat routine needs to exist _especially_ in the default
> 	case, since the default for passwd: is "compat".

Well, no, not for me it's not!  ;-)

If I'm not mistaken the comments in my updated template nsswitch.conf
correctly reflect the implemenation as I build it:

	$ cat /usr/src/etc/nsswitch.conf
	#	$NetBSD: nsswitch.conf,v 1.5 1999/10/24 12:36:52 lukem Exp $
	#
	# nsswitch.conf(5) -
	#	name service switch configuration file
	#
	
	# NOTE: rename or remove this file if you're not changing anything in it
	
	# These are the defaults in libc
	#
	#group:		files
	#hosts:		files dns
	#networks:	files dns
	#passwd: 	files
	#shells: 	files
	
	# List of supported sources for each database
	#
	# group:		files
	# hosts:		dns, files
	# networks:		dns, files
	# passwd:		files
	# shells:		files

On i386:

	$ size getpwent.o
	text    data    bss     dec     hex     filename
	1300    0       96      1396    574     getpwent.o

	$ nm getpwent.o
	000002dc t __hashpw
	00000254 t __initdb
	         U _dbopen
	00000224 T _endpwent
	00000134 T _getpwent
	0000016c T _getpwnam
	000001b8 T _getpwuid
	00000000 t _local_getpw
	         U _nsdispatch
	00000048 b _pw_db
	00000054 b _pw_flags
	0000004c b _pw_keynum
	00000020 b _pw_passwd
	00000050 b _pw_stayopen
	000001f4 T _setpassent
	00000210 T _setpwent
	         U _syslog
	         U abort
	00000008 b buf.38
	00000000 r compatsrc
	00000018 r dtab.12
	00000040 r dtab.16
	00000064 r dtab.20
	00000224 W endpwent
	         U free
	         U geteuid
	00000134 W getpwent
	0000016c W getpwnam
	000001b8 W getpwuid
	00000004 b max.37
	         U memmove
	         U realloc
	000001f4 W setpassent
	00000210 W setpwent
	00000000 b warned.33


>     *	There is no point doing
> 		#ifdef YP
> 			NS_NIS_CB(_nis_getpw, NULL)
> 		#endif
> 	since the definition for NS_NIS_CB() already does the right
> 	thing if YP isn't defined.  (See <nsswitch.h>)

I'm not sure if I noticed this, or whether I just thought it was too
much magic to depend upon.  I'm not sure what to think now.

>     *	The compat code is for more than just YP (it's also for HESIOD),
>     	so doing
> 		#ifdef YP
> 			NS_NIS_CB(_nis_getpw, NULL)
> 			NS_COMPAT_CB(_compat_getpwent, NULL)
> 		#endif
> 	is not correct _anyway_.

Yes, well, given I turn off both HESIOD and YP, I don't want any of it
wasting time and space for me.  I should have noted that....

This, and probably more places, might be fixed to allow HESOID without
YP and vice versa with a judiciously placed #ifdef _PASSWD_COMPAT

All the #ifdef's elsewhere still get really rather ugly though, even
with the magic disappearing act in those macros.

>     *	We generally try to avoid "whitespace cleanups" in the same
> 	commit as "functionality changes".

Yeah, well I sure as heck wasn't going to edit my local changes just to
undo cleanups that were in the same diff hunks.  I was even going to
leave in the isolated hunks containing whitespace cleanups too, but in
the end I whacked them out.

>     *	"cvs diff -p" is our friend :-)

Hmmmm.....

> In short, I don't think this diff is necessary or correct.

I guess it's correct for all combinations of -D defines, but it sure
works well for me!  ;-)

Without my changes the size of the resulting code is not sufficiently
reduced to meet my goals.  I really only need/want "hosts" and
"networks" sources to be configurable though even those I don't really
expect to have to change on any production server.

-- 
						Greg A. Woods

+1 416 218-0098                  VE3TCP            RoboHack <woods@robohack.ca>
Planix, Inc. <woods@planix.com>          Secrets of the Weird <woods@weird.com>