I'm trying to compile Sendmail 8.13.4 on Mac OS X 10.4.1, using GCC 3.3.

I'm getting the message that GCC no longer supports varargs.h and code 
should be updated to use stdarg.h. I know this is a long-standing 
problem, and I used to be able to solve it (with Sendmail 8.12.x on Mac 
OS X 10.2.x and 10.3.x) by reverting to GCC 3.1.

Unfortunately, GCC 3.1 is not installed on Mac OS X 10.4. The weird 
thing is that the last machine I build this way was Sendmail 8.13.3 on 
Mac OS X 10.3.7 and it compiled just fine with GCC 3.3. Possibly this 
had something to do with the fact that GCC 3.1 was also installed, but I 
confirmed today that it was *definitely* GCC 3.3 that I was using by 
compiling a second time, but specifically running "gcc_select 3.3" first.

Can anybody give me a better idea of what's going on here? I'm going to 
check if GCC 3.1 is even an option in the Developer package (the machine 
I'm building on has no DVD-ROM to mount the 10.4 disc), but it seems odd 
that GCC 3.3 would work fine with 8.13.3 and not with 8.13.4, even if 
there was an OS upgrade in the meantime.

Should I attempt to compile on 10.3.9 and install on 10.4.1? Should I 
try to copy the GCC 3.1 files from 10.3.9 to 10.4.1? And anyway, why is 
Sendmail still using varargs.h? GCC 3.3 came out a long time ago, and 
4.0 is out already!

-Michael Amper <amper@mac.com>

Michael Craig Amper  wrote:
> I'm trying to compile Sendmail 8.13.4 on Mac OS X 10.4.1, using GCC 3.3.

> Should I attempt to compile on 10.3.9 and install on 10.4.1? Should I 
> try to copy the GCC 3.1 files from 10.3.9 to 10.4.1? And anyway, why is 
> Sendmail still using varargs.h? GCC 3.3 came out a long time ago, and 
> 4.0 is out already!

sendmail 8 is portable to hundreds of different OS versions,
some of which have varargs.h, some of which have stdargs.h.

If you take a look at the source code, you'll see that sendmail
uses stdargs.h (include/sm/varargs.h).  Obviously your compiler
doesn't define __STDC__ for some obscure reason. Maybe you can
find out why?

-- 
Note: please read the netiquette before posting. I will almost never
reply to top-postings which include a full copy of the previous
article(s) at the end because it's annoying, shows that the poster
is too lazy to trim his article, and it's wasting my time.

Claus A�mann wrote:

> If you take a look at the source code, you'll see that sendmail
> uses stdargs.h (include/sm/varargs.h).  Obviously your compiler
> doesn't define __STDC__ for some obscure reason. Maybe you can
> find out why?
> 

Hmm...I see. Well, just for shits and giggles, I added a little hack to 
include/sm/varargs.h:

#ifndef __STDC__
# define __STDC__

which from the following lines would seem to be all that is needed to 
trick the compiler into using <stdarg.h> instead.

....running ./Build -n (under GCC 4.0.0) produces no errors, so we'll see 
how this goes!

-Michael Amper

Michael Craig Amper  wrote:
> Claus A�mann wrote:

> > If you take a look at the source code, you'll see that sendmail
> > uses stdargs.h (include/sm/varargs.h).  Obviously your compiler
> > doesn't define __STDC__ for some obscure reason. Maybe you can
> > find out why?

> Hmm...I see. Well, just for shits and giggles, I added a little hack to 
> include/sm/varargs.h:

> #ifndef __STDC__
> # define __STDC__

Well, that's a hack that can't be used in general of course.

I really would like to know why gcc on your OS does not define
__STDC__

Is it a bug in MacOS X? Does the compiler need some extra option
(if so: which)?

-- 
Note: please read the netiquette before posting. I will almost never
reply to top-postings which include a full copy of the previous
article(s) at the end because it's annoying, shows that the poster
is too lazy to trim his article, and it's wasting my time.

  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--0-1508012799-1119716974=:9946
Content-Type: TEXT/PLAIN; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 8BIT

Sat, 25 Jun 2005 (14:50 -0000 UTC) Claus A�mann wrote:

> Michael Craig Amper  wrote:
>> Claus A�mann wrote:
>
>>> If you take a look at the source code, you'll see that sendmail
>>> uses stdargs.h (include/sm/varargs.h).  Obviously your compiler
>>> doesn't define __STDC__ for some obscure reason. Maybe you can
>>> find out why?
>
>> Hmm...I see. Well, just for shits and giggles, I added a little hack to
>> include/sm/varargs.h:
>
>> #ifndef __STDC__
>> # define __STDC__
>
> Well, that's a hack that can't be used in general of course.
>
> I really would like to know why gcc on your OS does not define
> __STDC__
>
> Is it a bug in MacOS X? Does the compiler need some extra option
> (if so: which)?

I saw this error also, but I think it's not sendmail exactly--maybe in the 
siteconfig settings.  I say that because I was able to fix the error by 
changes to siteconfig (below).  However, there were other errors I could not 
fix without changing the source.

Here is my own Site Config for Darwin 8.1.  (and see the following note):
#	$Id: Darwin,v 8.0.0 2005/04/30 14:47:30 rroyar Exp $

define(`confCC', `gcc')
define(`confMAPDEF', `-DSASL=2 -DMILTER -DMAP_REGEX -DNETINFO -DNETINET6 -DNIS -DAUTO_NETINFO_ALIASES -DAUTO_NETINFO_HOSTS -DMATCHGECOS -DSCANF -DXDEBUG -DSTARTTLS -DNEWDB -DUSERDB')
APPENDDEF(`conf_sendmail_LIBS', ` -L/opt/local/lib -lsasl2.2.0.20 -lssl -lcrypto -L/usr/lib -lpthread -lresolv')
APPENDDEF(`confINCDIRS', `-I/opt/local/include')
APPENDDEF(`conf_sendmail_ENVDEF', `-DSASL=20120 -DSTARTTLS -DBIND_8_COMPAT')
APPENDDEF(`confCCOPTS', `-w -pipe')
define(`confDEPEND_TYPE', `CC-M')
define(`confOPTIMIZE', `-Os')
define(`confUBINGRP', `smmsp')
define(`confSBINOWN', `root')
define(`confSBINGRP', `smmsp')

I also had to change one line in sendmail/conf.c by forcing #define SPT_TYPE 
SPT_REUSEARGV at line 2350.  Without that the build appears to work but does 
not create an executable because gcc4 has moved sys/exec.h

gcc -Os -I. -I../../include  -I/opt/local/include -DSASL=2 -DMILTER 
-DMAP_REGEX -DNETINFO -DNETINET6 -DNIS -DAUTO_NETINFO_ALIASES 
-DAUTO_NETINFO_HOSTS -DMATCHGECOS -DSCANF -DXDEBUG -DSTARTTLS -DNEWDB 
-USERDB -DDARWIN -DSASL=20120 -DSTARTTLS -DBIND_8_COMPAT -w -pipe     -c -o 
conf.o conf.c
conf.c:2363:24: error: sys/exec.h: No such file or directory
make[1]: *** [conf.o] Error 1

You can get around the error by including the 3.3SDK as part of the 
includes, but that causes errors when you try to get SASL to work.
I do not know if this is a change for all gcc4.x or just Mac's version.


-- 
  Robert Delius Royar       The email address is valid as it is written.
--0-1508012799-1119716974=:9946--

Claus A�mann wrote:

> Well, that's a hack that can't be used in general of course.
> 
> I really would like to know why gcc on your OS does not define
> __STDC__
> 
> Is it a bug in MacOS X? Does the compiler need some extra option
> (if so: which)?


OK, the hack didn't work, anyway.

What *does* work is to copy the devtools/OS/Darwin.7.x file to 
devtools/OS/Darwin.8.x. Apparently, it must be the old NetInfo 
compatibility stuff that needs varargs.h. Darwin 7.x and up no longer 
uses those NetInfo "-D" switches (Mac OS X accesses it through PAM, 
anyway), which is why my last compile on 10.3.7 worked flawlessly.

Then the only thing that needs to be corrected is that XTools 2.x 
doesn't supply the /usr/include/sys/exec.h, /usr/include/machine/exec.h, 
or /usr/include/ppc/exec.h. Copying these from an Xtools 1.5 
installation on Mac OS X 10.3.x will take care of this.

Also, in Mac OS X 10.4.x, the smmsp user is no longer supplied with the 
system, so a trip through the NetInfo Manager (or the niload, etc. 
utilities) solves that.

So, I'm compiled and installed, there's just some permissions to clean 
up and a new sendmail.cf and associated files to create, and I should be 
off and running!


Thanks for the help,

Michael Amper <amper@mac.com>

r royar wrote:

> I saw this error also, but I think it's not sendmail exactly--maybe in 
> the siteconfig settings.  I say that because I was able to fix the error 
> by changes to siteconfig (below).  However, there were other errors I 
> could not fix without changing the source.
> 
> Here is my own Site Config for Darwin 8.1.  (and see the following note):
> #    $Id: Darwin,v 8.0.0 2005/04/30 14:47:30 rroyar Exp $
> 
> define(`confCC', `gcc')
> define(`confMAPDEF', `-DSASL=2 -DMILTER -DMAP_REGEX -DNETINFO -DNETINET6 
> -DNIS -DAUTO_NETINFO_ALIASES -DAUTO_NETINFO_HOSTS -DMATCHGECOS -DSCANF 
> -DXDEBUG -DSTARTTLS -DNEWDB -DUSERDB')
> APPENDDEF(`conf_sendmail_LIBS', ` -L/opt/local/lib -lsasl2.2.0.20 -lssl 
> -lcrypto -L/usr/lib -lpthread -lresolv')
> APPENDDEF(`confINCDIRS', `-I/opt/local/include')
> APPENDDEF(`conf_sendmail_ENVDEF', `-DSASL=20120 -DSTARTTLS 
> -DBIND_8_COMPAT')
> APPENDDEF(`confCCOPTS', `-w -pipe')
> define(`confDEPEND_TYPE', `CC-M')
> define(`confOPTIMIZE', `-Os')
> define(`confUBINGRP', `smmsp')
> define(`confSBINOWN', `root')
> define(`confSBINGRP', `smmsp')
> 
> I also had to change one line in sendmail/conf.c by forcing #define 
> SPT_TYPE SPT_REUSEARGV at line 2350.  Without that the build appears to 
> work but does not create an executable because gcc4 has moved sys/exec.h
> 
> gcc -Os -I. -I../../include  -I/opt/local/include -DSASL=2 -DMILTER 
> -DMAP_REGEX -DNETINFO -DNETINET6 -DNIS -DAUTO_NETINFO_ALIASES 
> -DAUTO_NETINFO_HOSTS -DMATCHGECOS -DSCANF -DXDEBUG -DSTARTTLS -DNEWDB 
> -USERDB -DDARWIN -DSASL=20120 -DSTARTTLS -DBIND_8_COMPAT -w -pipe     -c 
> -o conf.o conf.c
> conf.c:2363:24: error: sys/exec.h: No such file or directory
> make[1]: *** [conf.o] Error 1
> 
> You can get around the error by including the 3.3SDK as part of the 
> includes, but that causes errors when you try to get SASL to work.
> I do not know if this is a change for all gcc4.x or just Mac's version.
> 
> 

Mr. Royer:

I saw your post to a German site (and the reply) that tipped me to the 
missing exec.h files. So, are you saying that even copying the exec.h's 
doesn't entirely fix the problem? I haven't fired the new build up yet, 
because I had to move the machine (it is in my wife's way!). I'm going 
to be using Cyrus SASL 2.1.21 and Cyrus IMAPD 2.2.12 for this build, 
both of which are already compiled and installed, with seemingly no 
problems.

-Michael Amper <amper@mac.com>

  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--0-1783834445-1119724397=:28862
Content-Type: TEXT/PLAIN; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 8BIT

Sat, 25 Jun 2005 (13:14 -0400 UTC) Michael Craig Amper wrote:

> Claus A�mann wrote:
>
>>  Well, that's a hack that can't be used in general of course.
>>
>>  I really would like to know why gcc on your OS does not define
>>  __STDC__
>>
>>  Is it a bug in MacOS X? Does the compiler need some extra option
>>  (if so: which)?
>
>
> OK, the hack didn't work, anyway.
>
> What *does* work is to copy the devtools/OS/Darwin.7.x file to 
> devtools/OS/Darwin.8.x. Apparently, it must be the old NetInfo compatibility 
> stuff that needs varargs.h. Darwin 7.x and up no longer uses those NetInfo 
> "-D" switches (Mac OS X accesses it through PAM, anyway), which is why my 
> last compile on 10.3.7 worked flawlessly.
>
> Then the only thing that needs to be corrected is that XTools 2.x doesn't 
> supply the /usr/include/sys/exec.h, /usr/include/machine/exec.h, or 
> /usr/include/ppc/exec.h. Copying these from an Xtools 1.5 installation on Mac 
> OS X 10.3.x will take care of this.
>
> Also, in Mac OS X 10.4.x, the smmsp user is no longer supplied with the 
> system, so a trip through the NetInfo Manager (or the niload, etc. utilities) 
> solves that.
>
> So, I'm compiled and installed, there's just some permissions to clean up and 
> a new sendmail.cf and associated files to create, and I should be off and 
> running!
>
>
> Thanks for the help,
>
> Michael Amper <amper@mac.com>

I'll respond to both your last posts, here.

I found that sendmail gave me a bus error when compiled with -Os and the old 
exec if I used a smart host that needed sendmail to authenticate through 
SASL.  Local mail that stayed on the machine went through.  There were so 
many variables, that I am not sure which of them fixed my problem.  But I 
have been using sendmail fine (with SASL self-compiled to be a later version 
than Apple supplies) for about two months.  I recently moved it over to 
launchd and will see how well that works.  I am including a copy of my 
current devtools/cf/Site/site.config.m4 and a patch to sendmail/conf.c that 
fixes my compile problems and avoids the varargs error you saw before. 
Moving the exec.h file into the /usr/include could be a problem later if 
there are conflicts.  If you have the Xcode SDK for 10.3.9 installed, and 
the method I describe below does not work, then you might try adding
APPENDDEF(`confINCDIRS', `-I/Developer/SDKs/MacOSX10.3.9.sdk/usr/include')dnl
to your site.config.m4 file

I used that technique before I found the bus error problem.  And that 
problem may only hit when there is a smart host.  I cannot run without the 
SM, or rather cannot send mail past my firewall, so I do not know if the 
error was just in the SASL negotiation--I suspect it was more SASL than 
sendmail because sendmail has always either worked for me or failed at the 
outset, more the former than the latter.

[Note to make the patch work, you need the -DDARWIN8 in 
conf_sendmail_ENVDEF.]

sendmail-8.13.4/devtools/Site/site.config.m4
#	$Id: Darwin,v 8.1.0 2005/06/25 14:05:20 rroyar Exp $

define(`confCC', `gcc')
define(`confMAPDEF', `-DSASL=2 -DMILTER -DMAP_REGEX -DNETINFO -DNETINET6 -DNIS -DAUTO_NETINFO_ALIASES -DAUTO_NETINFO_HOSTS -DMATCHGECOS -DSCANF -DXDEBUG -DSTARTTLS -DNEWDB -DUSERDB')
APPENDDEF(`conf_sendmail_LIBS', ` -L/opt/local/lib -lsasl2.2.0.20 -lssl -lcrypto -L/usr/lib -lpthread -lresolv')
APPENDDEF(`confINCDIRS', `-I/opt/local/include')
APPENDDEF(`conf_sendmail_ENVDEF', `-DSASL=20120 -DSTARTTLS -DBIND_8_COMPAT -DDARWIN8')
APPENDDEF(`confCCOPTS', `-w -pipe')
define(`confDEPEND_TYPE', `CC-M')
define(`confOPTIMIZE', `-Os')
define(`confUBINGRP', `smmsp')
define(`confSBINOWN', `root')
define(`confSBINGRP', `smmsp')

sendmail-8.13.4/sendmail/darwin8.patch
*** sendmail/conf.c.~1~	Mon Mar  7 12:18:44 2005
--- sendmail/conf.c	Sat Jun 25 14:09:37 2005
***************
*** 2359,2365 ****
   # endif /* SPT_TYPE == SPT_PSTAT */
   # if SPT_TYPE == SPT_PSSTRINGS
   #  include <machine/vmparam.h>
! #  include <sys/exec.h>
   #  ifndef PS_STRINGS	/* hmmmm....  apparently not available after all */
   #   undef SPT_TYPE
   #   define SPT_TYPE	SPT_REUSEARGV
--- 2359,2367 ----
   # endif /* SPT_TYPE == SPT_PSTAT */
   # if SPT_TYPE == SPT_PSSTRINGS
   #  include <machine/vmparam.h>
! #  ifndef DARWIN8
! #   include <sys/exec.h>
! #  endif
   #  ifndef PS_STRINGS	/* hmmmm....  apparently not available after all */
   #   undef SPT_TYPE
   #   define SPT_TYPE	SPT_REUSEARGV

-- 
  Robert Delius Royar       The email address is valid as it is written.
--0-1783834445-1119724397=:28862--

r royar wrote:

> I'll respond to both your last posts, here.
> 
> I found that sendmail gave me a bus error when compiled with -Os and the 
> old exec if I used a smart host that needed sendmail to authenticate 
> through SASL.

> I used that technique before I found the bus error problem.  And that 
> problem may only hit when there is a smart host.  I cannot run without 
> the SM, or rather cannot send mail past my firewall, so I do not know if 
> the error was just in the SASL negotiation--I suspect it was more SASL 
> than sendmail because sendmail has always either worked for me or failed 
> at the outset, more the former than the latter.
> 

Well, I've just finished setting up my build, with only the copied 
exec.h files, no other changes. So far, everything seems to be working 
properly...but I have no need to use a smart host with SASL, so I don't 
know if that will be a problem.

The final build is BerkeleyDB 4.3.28, Cyrus SASL 2.1.21, Cyrus IMAPD 
2.2.12, Sendmail 8.13.4 on Mac OS X 10.4.1 with all patches, compiled 
with GCC 3.3 from XCode Tools 2.0 (except for BDB, which is compiled 
with GCC 4.0.0 from XCode Tools 2.0).

Local and non-local mail delivery is working with SMTP AUTH through 
saslauthd. If everything runs well for a while, I'll finally get around 
to updating my blog with new build instructions.

-Michael Amper <amper@mac.com>

r royar  wrote:

> define(`confUBINGRP', `smmsp')

> define(`confSBINGRP', `smmsp')

smmsp should not be used for that. Why are you changing the default?

> I also had to change one line in sendmail/conf.c by forcing #define SPT_TYPE 
> SPT_REUSEARGV at line 2350.  Without that the build appears to work but does 
> not create an executable because gcc4 has moved sys/exec.h

What do you mean by "moved"? To /dev/null or renamed?

-- 
Note: please read the netiquette before posting. I will almost never
reply to top-postings which include a full copy of the previous
article(s) at the end because it's annoying, shows that the poster
is too lazy to trim his article, and it's wasting my time.

  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--0-1491054104-1119998026=:11378
Content-Type: TEXT/PLAIN; charset=ISO-8859-1
Content-Transfer-Encoding: 8BIT

Mon, 27 Jun 2005 (22:37 -0000 UTC) Claus A�mann wrote:

> r royar  wrote:
> 
> > define(`confUBINGRP', `smmsp')
> 
> > define(`confSBINGRP', `smmsp')
> 
> smmsp should not be used for that. Why are you changing the default?

OK, I removed those lines, did a 
../Build clean && ./Build -c && sudo ./Build install

The install did the following:
install -c -o root -g smmsp -m 2555 sendmail /usr/sbin

I haven't changed any files in devtools except for the site.config.m4
Where is it getting the value?  Do I need to delete some other files?  And 
what should the UBINGRP and SBINGRP be (see note below about problems using 
bin).

I had changed that for complex reasons.  The Darwin 8 system did not seem to 
be pulling in either OS/Darwin or OS/Darwin.7.x

Also, Apple's install did not handle -g wheel.  I later pulled in ginstall 
to handle the -g settings (-g wheel is what Darwin and Darwin.7 were 
setting). bin is what the docs say, but bin causes the submission agent on 
my system to error when it tries to open clientmqueue files.

I am not sure why I thought smmsp instead of wheel, but I suspect I found 
the site.config file on a site that had a working 
SASL/sendmail/SMTP-AUTH/SmartHost setup over two years ago, because I have 
versions of that site.config back to when I first installed OS 10.3 and 
sendmail with the MTA/MSP setup--when Apple stopped supporting sendmail and 
switched to postfix.

> > I also had to change one line in sendmail/conf.c by forcing #define SPT_TYPE 
> > SPT_REUSEARGV at line 2350.  Without that the build appears to work but does 
> > not create an executable because gcc4 has moved sys/exec.h
> 
> What do you mean by "moved"? To /dev/null or renamed?

The file no longer exists.  There is an exec.h in the php headers, but no 
where else.  Also, I haven't noticed problems from simply deleting reference 
to it in the few third-party programs I have that include it, so I suspect 
the defs are now in some of the other includes.  This is the case even in 
their latest SDKs so it seems to be on purpose.  I do not have any other 
gcc4 installations, but I assume that your question must mean that this is 
just an Apple-specific omission.

I will ask on the unix-porting list to see if one of the Apple developers 
can explain why it is gone.

-- 
 Robert Delius Royar       The email address is valid as it is written.
--0-1491054104-1119998026=:11378--

I haven't been able to keep up for the last couple of days, because my
provider's NNTP server is on the fritz again...grrr.

Anyway, my Sendmail build was picking up the devtools/OS/Darwin file
just fine before I forced it to use the copy of Darwin.7.x (renamed to
Darwin.8.x), as evidenced by the -DNETINFO switches that turn up at
compile time.

The exec.h files, as I mentioned before (but somewhat inaccurately),
are installed with XCode Tools v1.5 (the Mac OS X 10.3 version
available from the ADC site), but not with XCode Tools 2.0 (the version
on the Mac OS X 10.4 install disc). I simply copied them (as outlined
in the German site I mentioned before in a reply to one of Mr. Royar's
posts) to /usr/local/include.

Mr. Royar, it sounds to me as if you are using a home-rolled GCC 4.0 or
4.4? I don't think it's GCC that's responsible for these exec.h header
files, but Apple's developer toolkit. I've downloaded XCode Tools 2.1
form ADC, but I haven't installed it yet, so I don't know what changes
have been made there (other than the x86 support that has been widely
publicized).

My devtools/OS/Darwin.8.x is simply a copy of the
devtools/OS/Darwin.7.x file that ships with the Sendmail 8.13.4 source.
Here's a copy of my site.config.m4:

APPENDDEF(`confENVDEF', `-DSASL=20121 -DSTARTTLS')
APPENDDEF(`conf_sendmail_LIBS', `-lsasl2.2.0.21 -lssl -lcrypto')
APPENDDEF(`confINCDIRS', `-I/usr/local/include/sasl')
APPENDDEF(`confLIBDIRS', `-L/usr/local/lib')
APPENDDEF(`confLIBS',`-lresolv.9')

I'm not sure if the "-lresolv.9" is still necessary with Mac OS X
10.4.x, but it fixed something with early attempts on Mac OS X 10.3.x
(with Sendmail 8.12.x), so I've left it in there thus far, with
seemingly no detrimental side effects (that I can see, anyway). As far
as the SASL version info, that may not be strictly necessary, either,
but it makes me feel better to include it rather than just specifying a
generic SASLv2, seeing as Apple pre-installs an older version.

I've found that examining the packages from the Darwin source code site
to be helpful in determining the types of changes Apple has made to get
things working the way they like on Mac OS X (particularly as regards
support for Apple's Open Directory). Of course, the problems I've run
into over the past couple of years getting newer versions of Cyrus
SASL, Cyrus IMAP, etc. running make it pretty clear why Apple is
reluctant to move to newer releases of these packages.

My biggest problems have been in the areas of SSL support, Kerberos
support, MD5 support, and the like, all of which are essential for full
integration with Open Directory. I'm going to have to examine these
issues a bit more closely, as I'm not particularly fond of plaintext
passwords. Apparently Apple can make it work, but I haven't, yet. I
can, however, make my build authenticate to an existing Open Directory
(albeit un-Kerberized and with plain passwd's), as Mac OS X will
happily use PAM & Directory Acces.app to authenticate "saslauthd -a
pam" requests.

Hope this help clear things up a bit,

Michael Amper <amper@mac.com>

Wed, 28 Jun 2005 (22:22 -0700 UTC) amper@mac.com wrote:

> I haven't been able to keep up for the last couple of days, because my
> provider's NNTP server is on the fritz again...grrr.
> 
> Anyway, my Sendmail build was picking up the devtools/OS/Darwin file
> just fine before I forced it to use the copy of Darwin.7.x (renamed to
> Darwin.8.x), as evidenced by the -DNETINFO switches that turn up at
> compile time.
> 
> The exec.h files, as I mentioned before (but somewhat inaccurately),
> are installed with XCode Tools v1.5 (the Mac OS X 10.3 version
> available from the ADC site), but not with XCode Tools 2.0 (the version
> on the Mac OS X 10.4 install disc). I simply copied them (as outlined
> in the German site I mentioned before in a reply to one of Mr. Royar's
> posts) to /usr/local/include.
> 
> Mr. Royar, it sounds to me as if you are using a home-rolled GCC 4.0 or
> 4.4? I don't think it's GCC that's responsible for these exec.h header
> files, but Apple's developer toolkit. I've downloaded XCode Tools 2.1
> form ADC, but I haven't installed it yet, so I don't know what changes
> have been made there (other than the x86 support that has been widely
> publicized).

No, I received confirmation from an Apple source that sys/exec.h was removed 
in their own builds in October 2004.  It is not in Xcode's gcc-4, but is in 
the 10.2 and 10.3 SDKs.  It is not replaced by the Xcode 2.1 upgrade.  In 
fact, unless you are compiling code that uses obsolete Apple APIs (and 
defines a header variable to that effect), sys/exec.h is an empty file--or 
rather defines only its own I-HAVE-BEEN-SEEN variable.  So it shouldn't be a 
problem to use it.

There is a define that shows up with gcc.4.0 that may be usable to determine 
if one is compiling on a system without sys/exec
% touch /tmp/foo.h &&  cpp-4.0  -dM /tmp/foo.h
Shows that the seemingly meaningful define
#define __APPLE_CC__ 5026
is present but not when you replace cpp-4.0 with cpp-3.3.  Since I have not 
changed my default compiler my cpp just points to cpp-4.0, so the same 
command above with cpp substituted for cpp-4.0 gives the same define.

> 
> My devtools/OS/Darwin.8.x is simply a copy of the
> devtools/OS/Darwin.7.x file that ships with the Sendmail 8.13.4 source.
> Here's a copy of my site.config.m4:
> 
> APPENDDEF(`confENVDEF', `-DSASL=20121 -DSTARTTLS')
> APPENDDEF(`conf_sendmail_LIBS', `-lsasl2.2.0.21 -lssl -lcrypto')
> APPENDDEF(`confINCDIRS', `-I/usr/local/include/sasl')
> APPENDDEF(`confLIBDIRS', `-L/usr/local/lib')
> APPENDDEF(`confLIBS',`-lresolv.9')
> 
> I'm not sure if the "-lresolv.9" is still necessary with Mac OS X
> 10.4.x, but it fixed something with early attempts on Mac OS X 10.3.x
> (with Sendmail 8.12.x), so I've left it in there thus far, with
> seemingly no detrimental side effects (that I can see, anyway). As far
> as the SASL version info, that may not be strictly necessary, either,
> but it makes me feel better to include it rather than just specifying a
> generic SASLv2, seeing as Apple pre-installs an older version.
> 
> I've found that examining the packages from the Darwin source code site
> to be helpful in determining the types of changes Apple has made to get
> things working the way they like on Mac OS X (particularly as regards
> support for Apple's Open Directory). Of course, the problems I've run
> into over the past couple of years getting newer versions of Cyrus
> SASL, Cyrus IMAP, etc. running make it pretty clear why Apple is
> reluctant to move to newer releases of these packages.

Apple's decision to move daemons to launchd, a decision that appears to have 
begun about the same time they switched to postfix rather than sendmail, 
could be one reason for the switch from sendmail.  It has been difficult to 
get launchd and sendmail to play nice, and I am not certain that my system 
is yet configured exactly correct.  sendmail ran fine using the old rc-style 
startup on Tiger, but it consistently lost the ability to log its 
transactions.  Others have reported similar problems with non-launchd 
daemons.  Creating a way for sendmail to run as a launchd daemon *appears* 
to have fixed this. But the way I had to do it, makes me believe that it 
would not work so well on a real server with continual incoming/outgoing 
connections.

You appear to be running such a server.  Are you still using the 
systemstartup script method?  Do you see any problems keeping your sendmail 
transactions logged?  BTW I do not set my loglevel in my sendmail.mc, the 
level is left at the default, but upping it does not solve the syslog 
problem.  I have been running sendmail under launchd for five days, and logs 
are working correctly through three test resets.  Before moving to launchd 
logs would stop from sendmail at random times and not restart under any 
regular conditions--i.e. sometimes restarting the sendmail daemon would get 
logging back; sometimes restarting the syslog daemon would do it; other 
times only a system reboot would do it.

> My biggest problems have been in the areas of SSL support, Kerberos
> support, MD5 support, and the like, all of which are essential for full
> integration with Open Directory. I'm going to have to examine these
> issues a bit more closely, as I'm not particularly fond of plaintext
> passwords. Apparently Apple can make it work, but I haven't, yet. I
> can, however, make my build authenticate to an existing Open Directory
> (albeit un-Kerberized and with plain passwd's), as Mac OS X will
> happily use PAM & Directory Acces.app to authenticate "saslauthd -a
> pam" requests.
> 
> Hope this help clear things up a bit,
> 
> Michael Amper <amper@mac.com>


-- 
 Robert Delius Royar       The email address is valid as it is written.

r royar wrote:


> You appear to be running such a server.  Are you still using the 
> systemstartup script method?  Do you see any problems keeping your sendmail 
> transactions logged?  BTW I do not set my loglevel in my sendmail.mc, the 
> level is left at the default, but upping it does not solve the syslog 
> problem.  I have been running sendmail under launchd for five days, and logs 
> are working correctly through three test resets.  Before moving to launchd 
> logs would stop from sendmail at random times and not restart under any 
> regular conditions--i.e. sometimes restarting the sendmail daemon would get 
> logging back; sometimes restarting the syslog daemon would do it; other 
> times only a system reboot would do it.
> 


Yes, I have been using a /System/Library/StartupItems/Sendmail/Sendmail 
script copied from Mac OS X 10.2.x (modified to look for 
"SENDMAIL=-YES-" in /etc/hostconfig so as not to conflict with the 
Postfix installation).

I have installed several of these servers for clients under 10.3.x that 
have been running under relatively high loads without any 
problems--certainly not with logging transactions. I generally log at 
Log Level 32, unless I need to debug something. These servers have been 
running for as long as two years under 10.3.x and 10.2.x.

-Michael Amper <amper@mac.com>

Thu, 30 Jun 2005 (02:30 -0400 UTC) Michael Craig Amper wrote:

> r royar wrote:
> 
> 
> > You appear to be running such a server.  Are you still using the
> > systemstartup script method?  Do you see any problems keeping your sendmail
> > transactions logged?  BTW I do not set my loglevel in my sendmail.mc, the
> > level is left at the default, but upping it does not solve the syslog
> > problem.  I have been running sendmail under launchd for five days, and logs
> > are working correctly through three test resets.  Before moving to launchd
> > logs would stop from sendmail at random times and not restart under any
> > regular conditions--i.e. sometimes restarting the sendmail daemon would get
> > logging back; sometimes restarting the syslog daemon would do it; other
> > times only a system reboot would do it.
> > 
> 
> 
> Yes, I have been using a /System/Library/StartupItems/Sendmail/Sendmail script
> copied from Mac OS X 10.2.x (modified to look for "SENDMAIL=-YES-" in
> /etc/hostconfig so as not to conflict with the Postfix installation).
> 
> I have installed several of these servers for clients under 10.3.x that have
> been running under relatively high loads without any problems--certainly not
> with logging transactions. I generally log at Log Level 32, unless I need to
> debug something. These servers have been running for as long as two years
> under 10.3.x and 10.2.x.

I had mistakenly assumed your server was running 10.4.x and that you would 
eventually need to find a way to switch to launchd for sendmail.  I didn't 
see a performance problem possibility under the old StartupItems system, 
just loss of reliable logging with 10.4.  But launchd imposes restrictions 
on spawning children that would appear to lead to a performance hit with 
sendmail.

-- 
 Robert Delius Royar       The email address is valid as it is written.

r royar wrote:

> 
> I had mistakenly assumed your server was running 10.4.x and that you would 
> eventually need to find a way to switch to launchd for sendmail.  I didn't 
> see a performance problem possibility under the old StartupItems system, 
> just loss of reliable logging with 10.4.  But launchd imposes restrictions 
> on spawning children that would appear to lead to a performance hit with 
> sendmail.
> 

Well, the new box *is* running 10.4.1, but I don't see any problems with 
using the StartupItems scipt, so far. I'll have to look into the launchd 
issue further. My logging so far seems to be OK, as well.

  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--0-1534780763-1120489484=:18283
Content-Type: TEXT/PLAIN; charset=ISO-8859-1
Content-Transfer-Encoding: 8BIT

I am replying to this thread so future searches might find a solution to the 
original problem.  I tried sendmail 8.13.5.Alpha1
ftp://ftp.sendmail.org/pub/sendmail/.alpha/darwin/sendmail.8.13.5.Alpha1.tar.gz
and found that without any extra site config, other than what is needed to 
add features, and tell sendmail where to find libraries, I got a working 
version of sendmail (no compiler errors or crashes while running) on Mac OS 
X 10.4.1 (Darwin 8).  The changes to the code described in the release notes 
(with respect to the current release version) fixed the errors in the 
compile I and others were seeing.

Version 8.13.5.Alpha1
 Compiled with: DNSMAP LOG MAP_REGEX MATCHGECOS MILTER MIME7TO8 MIME8TO7
                NAMED_BIND NETINET NETINET6 NETINFO NETUNIX NEWDB NIS
                PIPELINING SASLv2 SCANF STARTTLS USERDB XDEBUG

Thank you Herr Claus A�mann for recommending this.

-- 
 Robert Delius Royar       The email address is valid as it is written.
--0-1534780763-1120489484=:18283--