[ntp:questions] ntpd wedged again

A C agcarver+ntp at acarver.net
Thu Feb 9 03:47:37 UTC 2012


On 2/7/2012 19:12, Dave Hart wrote:
> On Tue, Feb 7, 2012 at 18:46, Dave Hart<hart at ntp.org>  wrote:
>> On Tue, Feb 7, 2012 at 18:38, A C<agcarver+ntp at acarver.net>  wrote:
>>> On 2/7/2012 10:21, Dave Hart wrote:
>>>> Thanks for the heads-up.  Assuming by "the C99 flag" you mean it was
>>>> configured using --enable-c99-snprintf, that flag didn't "take".  If
>>>> it had, you wouldn't be using libc's snprintf, you'd be using libntp's
>>>> rpl_snprintf() which would have called rpl_vsnprintf(), and based on
>>>> previous experience, it wouldn't have resulted in an infinite loop.
>>>
>>>
>>> I was wondering about that.  I recompiled it twice with that configure
>>> option.  Any idea why it would have been ignored this time?  I'll rebuild it
>>> and see what happens.
>>
>> No idea.  However, you don't have to wait for it to hang to see if the
>> option is working as intended.  After building, look in the build tree
>> for config.h and grep it for rpl_snprintf.  You should see #define
>> snprintf rpl_snprintf, otherwise it didn't take.  If it didn't, email
>> me the config.log.
>
> It appears to be a bug in code I added to sntp/m4/snprintf.m4 (from
> the C99-snprintf package, but not yet integrated upstream).  See
> http://bugs.ntp.org/2134 for details.  Thanks the agcarver for quickly
> helping identify it, and here's hoping it's the reason his NetBSD
> 5.1/sparc ntpd has been going haywire with its known-bad dtoa() used
> by its libc *snprintf routines.

The latest version is now compiled and running.  I'll let it go and see 
what happens over the next week.  Just as a point of reference, the 
system has been without ntpd (or any other clock discipline) for about 
three days and the clock was only off by about 1 millisecond when I 
started ntpd up again.  So it would appear that the system oscillator is 
reasonably stable and I don't have to be as concerned that it's doing 
something strange to cause the multi-second jumps.


More information about the questions mailing list