Snow Leopard users: Turn off automatic date and time in System Preferences immediately

WZZZ wrote:

...maybe it is a coding error. Would that be in the compiler itself?

I tried to explain that in my previous post.

Btw, I never bothered to get the Xcode 3.2.6 update. Think it might make a difference, or does it contain the same compiler as the plain 3.2?

The compiler is always being updated with every release of Xcode. But as I said I really am not blaming the compiler here.

Yeah, I have never seen that error, either with flatsixrsacer's original command line version or from the update built from the MI site. And, inexplicably, I have the same ntpd-wrapper on my Snow as you. I posted the one from my Snow here way back I don't know how many pages ago. I remember we went through that and that was, and still is, left unexplained. I think you asked flatsix to stick the one from 10.8 in his installer to correct that error.

Lets' just leave it as a mystery. We know that -v is invalid for the newly built sntp so its not really worth pursuing.

shiekh wrote:

Any chance of compiling Universal binaries? and making them available?

Its possible. But it takes some investigation on how to specify the CXXFLAGS and CPPFLAGS compiler flags to the build configure step in such a way so as to not remove any flags those would normally be defined with (for example, I already know its using -g and -O2 on the build lines -- would need to check what else in order to add -arch ppc,i386,x86_64 -- and is it 32 bit or 64 bit for the intel archs?).

If you don't know what I am talking about, well, it's ok, never mind. But for anyone reading this who does, I found an interesting (well, I thought it was interesting to me) configure option I didn't know about but I thought worth mentioning. It's --disable-silent-rules.

When doing this ntp build I thought it was rather strange that the terminal was showing compilation lines like just "CC foo.o". What happened to the actual command lines? Because that certainly wasn't correct to compile a file! Well, it's the way this configure tool (and thus I assume most newer configure tools for other sources) was built, i.e., to just show an abbreviated log line to indicate what is happening. If you do a configure --help you will see the documentation that says using --disable-silent-rules allows you to see the full command line for each file being compiled! Cool.

(I apologize for the little off-topic diversion)

Where might I get some guidance using XCode to compile the NTP code? I see something about GNU Make in XCode

http://www.scorec.rpi.edu/~littld/xcode_makefile.html

but this did not take me far (I have XCode 3.1.4)

I thought XCode might be one way to set architecture

Yes, I'm a total neophyte

I was fine getting binaries for PPC, but thought it might be nice to have Universal binaries, as I am patching 10.5

Maybe I wasn't clear above. The (your) problem is not with how write makefiles. Your problem is how to get ntp's configure to generate the appropriate gcc command line options that you want. The (Apple) gcc command line option to generate multiple architectural binaries is -arch (do gcc -v --help to get the gcc command line options).

A simple way that may get what you want is to do this configure step (with bash syntax) is something as follows:

export CXXFLAGS='-O2 -g -arch ppc -arch=386 -arch=x86_64'

export CFLAGS="$CXXFLAGS"

./configure --prefix=/usr

I haven't tried this and I don''t know that overriding the flags like this will affect configure's use of these flags (the exports will override configure's definitions).

As I said the "problem" is how to get configure to do what you want, not on how to do makefiles. Configure builds the makefile for you as a function of the options you send to configure. It helps of course to know what options you can use for gcc. That's where the gcc -v --help might, well, help :-) But be warned. There are a lot of options for gcc so the help info is long.

Hi.

I applied your patch and I have a problem : ntpd just won't start anymore. At boot "automatic update" is of in the system preference and whenever I tick it, it seems to work but switching away and back to the panel it shows it is of.

Commande like "ntpd -v" or "ntpd --help" all respond "Bad CPU type in executable".

Could it be you included only 64bits version ? I have an old macmini1,1, late 2006 wich is 32 bits.

Many thanks for all the help; for now I'll just lipo together the binaries for the i386 and ppc; I can see I am over my head at the moment.

WZZZ wrote:

Forgot to mention that, after I applied the MI one over the flatsix rev4, I also didn't get the mlockall(): Function not implemented

You have me confused now. Theoretically the only difference between flatsixracer's installer and MI''s explicit build is that flatsixracer's installs the modified ntp-restrict.conf and ntpd-wrapper. If you previously installed with flatsixracer's installer you installed ntp-restrict.conf and ntpd-wrapper. If you then explicitly build ntp and installed (e.g., like shown in MI) after and install from flatsixracer then those updated ntp-restrict.conf and ntpd-wrapper are still in your system.

On the one hand you show me a log that has the mlockall error but no sntp -v error and then you say you are not getting the mlockall error. The log indicates to me that at that time you had the updated ntpd-wrapper (which fixes the sntp -v error) but did not have the updated ntp-restrict.conf which gets rid of the KOD and mlockall log messages. On the other hand you say you didn't get the mlockall message.

I cannot see you having a log which reports the KOD messages and not having the mlockall error. However, as I type this I seem to now recall that briefly flatsixracer posted an installer where he addressed to mlockall message by fixing it from /private/etc/ntp.conf instead of from ntp-restrict.conf. I posted a recommendation he move the fix (rlimit memlock 0) into ntp-restrict.conf to keep all the changes in the one file plus I think that any changes to ntp.conf would get clobbered if you change the ntp server in Date&Time preferences. So maybe you still have his rlimit change in /private/etc/ntp.conf.

If you want to go back to the beginning you would need to make sure you have the original OSX ntp.conf, ntp-restrict.conf, and ntpd-wrapper. Then you can use the normal explicit build as described by MI. You should then see the -v message, KOD messages, and mlockall message in the log. But why would you want to do that? If you use flatsixracer's latest installer (or just add the fixed/updated ntp-restrict.conf and ntpd-wrapper) you should not see those log messages.

I recommend you use flatsixracer's installer. Because there is (or should be) no difference between the ntp files it installs and the ones you would get from an explicit build. But with flatsixracer's installer you do get the updated ntp-restrict.conf and ntpd-wrapper that MI doesn't address. You really want those updates (well, or at a minimum the ntpd-wrapper update since that sntp -v really does need to be fixed while ntp-restrict.conf just addresses suppressing a bunch of log warnings).

As for that very rare crash, here's a portion of that log again with one line highlighted:

Application Specific Information:

*** error for object 0x100200268: incorrect checksum for freed object - object was probably modified after being freed.

Thread 0 Crashed: Dispatch queue: com.apple.main-thread

0 libSystem.B.dylib 0x00007fff8382f5d6 __kill + 10

1 libSystem.B.dylib 0x00007fff838cfcd6 abort + 83

2 libSystem.B.dylib 0x00007fff838be90d szone_error + 519

3 libSystem.B.dylib 0x00007fff837eabe6 tiny_free_list_remove_ptr + 158

4 libSystem.B.dylib 0x00007fff837e81ae szone_free_definite_size + 2005

5 libcrypto.0.9.8.dylib 0x00007fff867142a5 CRYPTO_free + 37

6 libcrypto.0.9.8.dylib 0x00007fff86710979 lh_free + 57

7 libcrypto.0.9.8.dylib 0x00007fff86770c10 int_err_del + 48

8 libSystem.B.dylib 0x00007fff837f389f __cxa_finalize + 214

9 libSystem.B.dylib 0x00007fff837f37ac exit + 18

10 ntpd 0x0000000100011918 finish + 88

11 libSystem.B.dylib 0x00007fff8384166a _sigtramp + 26

12 libSystem.B.dylib 0x00007fff837e8af2 tiny_free_list_add_ptr + 314

13 libSystem.B.dylib 0x00007fff837e81fc szone_free_definite_size + 2083

14 ntpd 0x00000001000336f3 uninit_util + 99

15 libSystem.B.dylib 0x00007fff837f389f __cxa_finalize + 214

16 libSystem.B.dylib 0x00007fff837f37ac exit + 18

17 ntpd 0x0000000100051a74 exit_worker + 68

18 ntpd 0x0000000100051ec4 send_blocking_req_internal + 1092

19 ntpd 0x000000010004f54b queue_blocking_request + 187

20 ntpd 0x000000010004e584 getaddrinfo_sometime + 372

21 ntpd 0x0000000100002c5c config_peers + 1036

22 ntpd 0x0000000100004eb9 config_ntpd + 6521

23 ntpd 0x00000001000078fc getconfig + 380

24 ntpd 0x000000010001221b ntpdmain + 1707

25 ntpd 0x0000000100001344 start + 52

My current theory is that it was a function named finish() that failed. There is a function name finish() in the ntpd.c source file.

/*

* finish - exit gracefully

*/

static RETSIGTYPE

finish(

int sig

)

{

const char *sig_desc;

sig_desc = NULL;

#ifdef HAVE_STRSIGNAL

sig_desc = strsignal(sig);

#endif

if (sig_desc == NULL)

sig_desc = "";

msyslog(LOG_NOTICE, "%s exiting on signal %d (%s)", progname,

sig, sig_desc);

/* See Bug 2513 and Bug 2522 re the unlink of PIDFILE */

# ifdef HAVE_DNSREGISTRATION

if (mdns != NULL)

DNSServiceRefDeallocate(mdns);

# endif

exit(0);

}

I'm guessing that "finish+88" in the crash log is that DNSServiceRefDeallocate() call but that's only a guess since I didn't look any deeper into this. It's interesting that that piece of code is trying do address a couple of bug reports :-) This is all I have to say on this. This post is getting way too long and now going off topic.

shiekh wrote:

Many thanks for all the help; for now I'll just lipo together the binaries for the i386 and ppc; I can see I am over my head at the moment.

Out of curiosity I looked into this a little more. A configure line like this does work:

CFLAGS='-g -O2 -arch ppc' configure --prefix=/usr

Add --disable-silent-rules and you will see it adds those flags to each build line and a ppc version is generated (didn't test it of course).

Unfortunately you cannot add multiple -arch's to build fat binaries in one build. The gcc command lines have other gcc options which don't permit specifying multiple -arch's that aren't controlled by CFLAGS. So to build fat binaries you would need to build a ppc version and say a x86_64 version and lipo each respective code file together.

strangely, this is making a lot of sense; the fog is leaving. Many thanks for all the help.

I looked at flatsixracers installer, which initially installed

/usr/bin/ntp-keygen

/usr/bin/ntpq

/usr/bin/sntp

/usr/sbin/ntpd

/usr/sbin/ntpdate

/usr/sbin/ntpdc

/usr/sbin/ntptrace

/usr/libexec/ntpd-wrapper

/System/Library/LaunchDaemons/org.ntp.ntpd.plist

/etc/ntp.conf

/etc/ntp-restrict.conf

but then (rev 4) dropped installing /etc/ntp.conf

I also see that NTP are now (2nd Jan 2015) on beta 3 of patch 1, so I might wait till things have settled some and someone more adapt than myself might have an installer that covers 10.5 (PPC and Intel)

Yup, that's what I recall happening with the conf fix as I mentioned above.

Thanks for the heads up on the beta3.

This stuff is addictive

Using the Intel binaries of flatsixracer I put together a universal installer for OS 10.5

http://www.cubeowner.com/forums/index.php?s=&showtopic=14866&view=findpost&p=104 503

As usual this is 'use at your own risk' and I have not tested it on an Intel Mac.