svn commit: r282415 - head/usr.sbin/ntp

[Ian Lepore]

On Fri, 2015-05-22 at 12:57 -0700, John-Mark Gurney wrote:
> Cy Schubert wrote this message on Sun, May 17, 2015 at 17:53 -0700:
> > In message <20150517190810.4006b7bb at bender.Home>, Andrew Turner writes:
> > > On Mon, 4 May 2015 12:42:53 +0000 (UTC)
> > > Cy Schubert <cy at FreeBSD.org> wrote:
> > > 
> > > > Author: cy
> > > > Date: Mon May  4 12:42:52 2015
> > > > New Revision: 282415
> > > > URL: https://svnweb.freebsd.org/changeset/base/282415
> > > > 
> > > > Log:
> > > >   Restore CPU dependent compile time conditionals.
> > > >   
> > > >   MFC after:	1 month (with r281143 and r282408)
> > > 
> > > Do you have any plans on fixing this before you MFC it? Specifically:
> > >  * sizeof(char *) == 8 on arm64, powerpc64, and mips64
> > >  * sizeof(long) == 8 on arm64, powerpc64, and mips64
> > >  * sizeof(time_t) == 8 on arm, arm64, powerpc64, mips, and mips64
> > >  * STR_SYSTEM is incorrect on arm, arm64, powerpc64, mips, and mips64,
> > >    however this is now less broken as it was previously also broken on
> > >    i386, and sparc64.
> > > 
> > > This change is only correct for amd64, i386, and sparc64.
> > 
> > That is correct.
> > 
> > > 
> > > I suspect all of the above will also be relevant on ia64. We still have
> > > support in 10-STABLE for this so should be fixed before an MFC.
> > 
> > We should support ia64 at least until it's no longer supported in -stable.
> > 
> > > 
> > > I have heard reports that NTP is broken on some of the above platforms
> > > so they don't keep very accurate time. Can you fix this, or revert this
> > > change until a fix can be written.
> > 
> > Unfortunately -stable is currently as broken as it is in -current. The 
> > quick fix is to add all supported platforms. The other unfortunate fact is 
> > that, scanning the tree for other examples there's more code as broken as 
> > this is. I propose:
> > 
> > 1. Fix NTP -- I will do this. I have some free time coming up this week to
> >    focus on this.
> > 2. Identify other parts of the tree needing a little love and either putting
> >    it up on the wiki or in phabricator.
> > 3. Fix #2 (either me or others as we see fit).
> > 
> > Alternatively, a common solution for all parts of the tree, if possible. 
> > This is probably a long term solution though. (The purist in me prefers 
> > this but I am practical too.)
> 
> Why can't we use sizeof for these?
> 

They are used in a preprocessor context.

I'm not sure I understand the comments about various kinds of brokeness.
ntpd works just fine on arm and other platforms and has done so for
about 10 years on all branches, until a new version was imported a
couple weeks ago.

Leaving the non-x86 platforms without functioning ntp for weeks is
mind-boggling (but I haven't had time to find and fix the problem
myself).

-- Ian