[LinuxPPS] Linux issues discussed earlier this year.

[William S. Brasher]

On Wed, 29 Jul 2009, Hal V. Engel wrote:

> On Wednesday 29 July 2009 12:28:17 pm William S. Brasher wrote:
> > See:
> >  http://sourceware.org/bugzilla/show_bug.cgi?id=9690
> >
> > The current glibc-2.10.1 has the NANO related changes present.  It does
> > not have the TAI changes, since an ABI change is required.
> 
> 
> At one point I thought that the TAI change would affect the ABI but the 
> maintainer of glibc says it does not.  The reason that this is the case is 
> that the calls that uses this pass it in a structure that has padding in case 
> the structure needs new fields.  So adding TAI to the structure and reducing 
> the size of the padding so that the total size of the structure is the same 
> takes care of the issue and apps built against the old structure definition 
> will continue to work as before since these will treat the tai field like it 
> is part of the padding.
> 
> Please see comment #9 of the above bug report.
>  


Note that Ulrich Drepper didn't install the following patch to 
ntp_gettime.c:

Index: glibc-2.7/sysdeps/unix/sysv/linux/ntp_gettime.c
===================================================================
--- glibc-2.7.orig/sysdeps/unix/sysv/linux/ntp_gettime.c        2008-03-04 21:28:45.000000000 +0100
+++ glibc-2.7/sysdeps/unix/sysv/linux/ntp_gettime.c     2008-03-04 21:29:13.000000000 +0100
@@ -38,5 +38,6 @@
   ntv->time = tntx.time;
   ntv->maxerror = tntx.maxerror;
   ntv->esterror = tntx.esterror;
+  ntv->tai = tntx.tai;
   return result;
 }



I think that is where he had the "ABI change" problem, though I too 
thought the extra unused bits in the structure ntptimeval that ntv 
points to were there so new fields could be added without changing the 
size of the structure.

The patch hasn't been applied to ntp_gettime.c in eglibc-2.10.1, either.