12:24 p.m.
So Moji has burned through all of the armv5tel builds at least once[0];
we have 4092 successful builds and 915 error builds at this point. Jon
and Paul are pulling in some lovingly hand-crafted builds done on the
weekend (thanks tinker915 and Henrick) and working through the error'd
builds -- any help is appreciated if you have some free cycles.
I think we're about ready to...
- compare the armv5tel and armv7hl package sets again
- solve any discrepancies
- double-check that patches have been upstreamed
...and then do the koji external repo setup.
-Chris
[0] http://moji.proximity.on.ca/moji
1:46 p.m.
>> - compare the armv5tel and armv7hl package sets again
>
> Do we need another pretty chart? ;-)
yes please
Is there a suitable rsync target we can use to get a local copy of the
appropriate files?
Note: the answer to that partly depends on what kinds of comparisons
you're interested in for this script... Hint, hint.
6:06 p.m.
>> - compare the armv5tel and armv7hl package sets again
>
> Do we need another pretty chart? ;-)
yes please
http://djdelorie.fedorapeople.org/armv5tel-vs-armv7hl.html
7:48 p.m.
On 10/25/2011 04:06 PM, DJ Delorie wrote:
Cool!
What is the plan for reconciling the differences?
A few packages stand out as being particularly concerning:
gcc: There are patches in the armv7hl build that affect volatile
bitfields and memory consumption. The lack of the volatile bitfield
patch may result in a few broken packages on armv5tel while the absence
of the memory consumption patch may result in some packages failing to
build.
glibc: Both have arm patches, but are they the same patches?
rpm: Special armv7hl handling was required. Is this as upstream
acceptable as it's going to be for the foreseeable future?
--
Brendan Conoboy / Red Hat, Inc. / blc(a)redhat.com
10:41 a.m.
On Tue, 2011-10-25 at 17:48 -0700, Brendan Conoboy wrote:
On 10/25/2011 04:06 PM, DJ Delorie wrote:
> http://djdelorie.fedorapeople.org/armv5tel-vs-armv7hl.html
Cool!
What is the plan for reconciling the differences?
A few packages stand out as being particularly concerning:
...
glibc: Both have arm patches, but are they the same patches?
I started looking at 2.14.7.arm0 (v5) vs 2.13.90.9.0.arm1 (v7) this
morning. It is not clear to me if glibc-2.14.7.arm0 does have arm
specific patches. 2.13.90.9.0.arm1 does have:
glibc-arm-mathdefs.patch
glibc-arm-tzdata.patch
glibc-arm-clone-unwind.patch
It looks like glibc-arm-tzdata.patch isn't needed in 2.14.7.
Upstream commit 998832a46688b9fb3a101eccae77bc45e7c1d7ab:
Author: Andreas Schwab <schwab(a)redhat.com>
Date: Fri Nov 4 14:39:05 2011 +0100
Always provide declarations of long double math functions when !_LIBC
appears to fix the problem glibc-arm-mathdefs.patch tries to work
around. It is a simple patch which easily backports to 2.14.7.
The arm-clone-unwind is still needed and applies to 2.14.7.
So, I'm in the process of building the 2.14.7 glibc for armv7hl with
the long double math and unwind patches added.
--Mark
11 a.m.
Always provide declarations of long double math functions when
!_LIBC
appears to fix the problem glibc-arm-mathdefs.patch tries to work
around. It is a simple patch which easily backports to 2.14.7.
Yes, the "correct" fix for glibc-arm-mathdefs is not the patch I
originally put in, but an alternate (upstream?) one that just maps the
declarations to existing functions.
11:04 a.m.
On Fri, 2011-11-04 at 12:00 -0400, DJ Delorie wrote:
> Always provide declarations of long double math functions
when !_LIBC
>
> appears to fix the problem glibc-arm-mathdefs.patch tries to work
> around. It is a simple patch which easily backports to 2.14.7.
Yes, the "correct" fix for glibc-arm-mathdefs is not the patch I
originally put in, but an alternate (upstream?) one that just maps the
declarations to existing functions.
Here's the relevant bz:
http://sourceware.org/bugzilla/show_bug.cgi?id=10103
11:16 a.m.
On 11/04/2011 03:41 PM, Mark Salter wrote:
On Tue, 2011-10-25 at 17:48 -0700, Brendan Conoboy wrote:
> On 10/25/2011 04:06 PM, DJ Delorie wrote:
>> http://djdelorie.fedorapeople.org/armv5tel-vs-armv7hl.html
>
> Cool!
>
> What is the plan for reconciling the differences?
>
> A few packages stand out as being particularly concerning:
>
...
> glibc: Both have arm patches, but are they the same patches?
I started looking at 2.14.7.arm0 (v5) vs 2.13.90.9.0.arm1 (v7) this
morning. It is not clear to me if glibc-2.14.7.arm0 does have arm
specific patches. 2.13.90.9.0.arm1 does have:
glibc-arm-mathdefs.patch
glibc-arm-tzdata.patch
glibc-arm-clone-unwind.patch
It looks like glibc-arm-tzdata.patch isn't needed in 2.14.7.
Correct, I hope that the patch for tzdata from
https://bugzilla.redhat.com/show_bug.cgi?id=708452#c14 has been included.
Andreas closed the bug after the last comment, so I assume it was accepted.
Cheers,
Niels
10:44 a.m.
On Fri, 2011-11-04 at 11:41 -0400, Mark Salter wrote:
On Tue, 2011-10-25 at 17:48 -0700, Brendan Conoboy wrote:
> On 10/25/2011 04:06 PM, DJ Delorie wrote:
> > http://djdelorie.fedorapeople.org/armv5tel-vs-armv7hl.html
>
> Cool!
>
> What is the plan for reconciling the differences?
>
> A few packages stand out as being particularly concerning:
>
...
> glibc: Both have arm patches, but are they the same patches?
I started looking at 2.14.7.arm0 (v5) vs 2.13.90.9.0.arm1 (v7) this
morning. It is not clear to me if glibc-2.14.7.arm0 does have arm
specific patches. 2.13.90.9.0.arm1 does have:
glibc-arm-mathdefs.patch
glibc-arm-tzdata.patch
glibc-arm-clone-unwind.patch
I have been running an armv7hl glibc 2.14-7 for the last few days
and haven't found any problems with it. It is the same as the
2.14-7.arm0 package with the addition of the arm-clone-unwind.patch
and the patch for http://sourceware.org/bugzilla/show_bug.cgi?id=10103
I put the src rpm at http://msalter.fedorapeople.org/arm/
If someone could build this and try it out on armv5tel and it doesn't
break anything, then I think we can start using it for both builds.
--Mark
1:48 p.m.
On Tue, 2011-10-25 at 14:41 -0400, DJ Delorie wrote:
> Ideally, crontab'd!
Heh. The trick is finding a machine that has both access to the repos
*and* access to a web server, and is powerful enough to run the script
in a reasonable time.
But I'll see what I can do ;-)
We can host on Australia -- send me an ssh public key for access.
-Chris
1:55 p.m.
On 10/25/2011 11:52 AM, DJ Delorie wrote:
> We can host on Australia -- send me an ssh public key for access.
Will do. What sort of information do we want this new chart to convey?
Packages only in armv7hl build (latest version)
Packages only in armv5tel build (latest version)
Changes only in armv7hl build
Changes only in armv5tel build
--
Brendan Conoboy / Red Hat, Inc. / blc(a)redhat.com
2 p.m.
On Tue, 2011-10-25 at 14:52 -0400, DJ Delorie wrote:
> We can host on Australia -- send me an ssh public key for
access.
Will do. What sort of information do we want this new chart to convey?
I think we need to see if the package sets are congruent, so a display
of the differences between the package sets would be good: package
missing in A or in B, EVR mismatches.
I have scripts for http://hongkong.proximity.on.ca/f13-arm-beta2/ type
output if you want them for a starting point.
-Chris
4551
days inactive
4565
days old
16 comments
6 participants
participants (6)
-
Brendan Conoboy -
Chris Tyler -
DJ Delorie -
Mark Salter -
Niels de Vos -
Peter Robinson