[ntp:hackers] GPS weirdness with ntp-dev-4.2.0a-200503xx

Harlan Stenn stenn at www.ntp.org
Tue Apr 19 22:02:34 PDT 2005


Dave,

> I don't think you understand the situation. The staff here will upgrade 
> all of our Sunses from time to time. You and I don't get to participate 
> in that process other than to suggest what might be upgraded in the 
> common department tool pool. If there is some issue with upgrading 
> common tools like gcc and that is important for the campus population, 
> then it is important that the staff know about that and work 
> appropriately. If it is an issue only with the NTP flock, then we are 
> entirely on our own. If gcc needs upgrade on pogo and the issue is 
> transparent to the general department population, you will need to make 
> the case. I am happy to carry the case to the department staff, but I 
> need a reason to do that. Note that I already have compiled the flock 
> after the upgrade and with the resident gcc.

I understand this perfectly well, and I accept it.  The situation with gcc
after a SunOS upgrade has happened a few times before.

If we will be using the vendor-provided gcc (and perl) as part of the upgrade
then there should be no issue.  If we are going to be using a *local* version
of gcc or perl then they must be upgraded after an OS upgrade.  There may be
more programs like this, but these are the two that I am aware of.

> As for the make dist issue, you will not be surprised that I take great 
> offense with that. The current build process is fragile, idiosyncratic 
> and plain wrong.

We disagree, and it would help if you would provide accurate examples.

Your issue with "make dist" is due to *your* cockpit error.  I do not expect
you to either believe or accept this, but I live in my world and you live in
yours.  The trick here is learning to work together and, at least from my
perspective, having that relationship be mutually "successful" and with a
minimum of headaches.

> I see little unexpected evils when building and 
> sometimes an incomplete build that I have to redo.

Please tell me when these happen and I will fix whatever underlying problem
there is (if possible) or get you back on-track.

> I see no reason 
> whatsoever why the build process should be anything other than like with 
> openssl and freebsd. That process is straightforward and with no surprises.

You seem to be missing the fact that you are *using* those packages and not
*making* those distributions.  To my knowledge you and I are the only ones
who try and "make dist".  Most of the time you have no problems.  To my
knowledge, you generally only have this problem if you try to "make dist"
outside of either A.whimsy or A.deacon, or after you have decided to "rm"
those directories and rebuild them without using the br-flock script.

There are special requirements for building a *releasable distribution* of
ntp, just as there are with FreeBSD (and most every other significant
package out there as well).

In the case of arlib, I expect that "soon" it will either become an active
part of the code or it will be replaced with something new.  In either of
those cases, this particular issue you face will be solved.

H



More information about the hackers mailing list