Re: [Angstrom-devel] State of package feed for 2015.06
> Op 26 okt. 2015, om 21:13 heeft Stefan Agner het volgende > geschreven: > > On 2015-10-26 10:29, Koen Kooi wrote: >>> Op 26 okt. 2015, om 18:21 heeft Stefan Agner het volgende >>> geschreven: >>> >>> On 2015-10-22 04:55, Koen Kooi wrote: > Op 20 okt. 2015, om 22:34 heeft Stefan Agner het > volgende geschreven: > > Hi, > > It seems that the feeds of version 2015.06 currently contains a lot less > packages than the 2014.12 feeds did. > > At least this is true for armv7at2hf-vfp/ and armv7at2hf-vfp-neon/. E.g. > the first package at the very top which is missing is a52dec, but there > is a lot of stuff like binutils or mono which are missing. > > Is this intentional? build-feeds.sh in meta-angstrom/contrib/ seems not > to be altered recently, so it should produce the same packages…? Great timing! The feeds are now being populated using Jenkins CI and there is a big rebuild for 2015.06 underway due to the big DISTRO changes that went in last week (i.e. blacklisting bluez5). Eyeballing the jenkins page there are 5 architectures left in the queue, good for 16h builds each spread over 2 machines, so I’d say “will finish this weekend”. >>> >>> It seems the packages are online now, thanks Koen! >>> >>> Was there any specific reason why those packages were not part of the >>> previous build for 2015.06? >> >> Mostly lack of time to sit down and check a few architectures on real >> devices. I was also concerned about ABI changes and heisenbugs like >> the bluez5 blacklisting last week. The prserv support needs to get >> updated as well to keep feed updates consistent and working. The good >> news is that all the work going into v2015.06 wrt feed updates works >> for v2015.12 and next as well :) >> >>> Just to understand your setup a bit better: Is your Jenkins using the >>> scripts under meta-angstrom/contrib/ or is that no longer the way the >>> packages get built? >> >> For v2015.06 and newer it wraps around >> https://github.com/Angstrom-distribution/angstrom-manifest/ and does >> some magic for reusing sstate. I’ll paste the build step of the >> console-feed run at the bottom of this email. It’s not pretty, but it >> gets the job done. > > Ok, and you call these script for each machine? So you basically build > all packages for all machines, and then upload just the packages which > are not up there yet.. Yes and no :) For image builds, yes, all machines get built. For feed builds every arch has one or two machines that gets built. > I guess you do this to make sure that all the packages built by the > recipe list in the build step are available for _all_ the machines you > build... But with that, you probably rebuild armv7at2hf-vfp-neon several > times right? And the images are another build step? See above, every arch specific package gets built once or twice. The second build is used to have buildhistory catch problems. > Would you mind add colibri-vf to the list of machines? Since fido the > machine is in meta-fsl-arm-extra which is already part of your > bblayers.conf… I’m doing a build for colibri-vf now to see what kind of surprises it has, since it is in a freescale BSP layer... regards, Koen ___ Angstrom-distro-devel mailing list Angstrom-distro-devel@linuxtogo.org http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/angstrom-distro-devel
Re: [Angstrom-devel] State of package feed for 2015.06
On 2015-10-26 10:29, Koen Kooi wrote: >> Op 26 okt. 2015, om 18:21 heeft Stefan Agner het volgende >> geschreven: >> >> On 2015-10-22 04:55, Koen Kooi wrote: Op 20 okt. 2015, om 22:34 heeft Stefan Agner het volgende geschreven: Hi, It seems that the feeds of version 2015.06 currently contains a lot less packages than the 2014.12 feeds did. At least this is true for armv7at2hf-vfp/ and armv7at2hf-vfp-neon/. E.g. the first package at the very top which is missing is a52dec, but there is a lot of stuff like binutils or mono which are missing. Is this intentional? build-feeds.sh in meta-angstrom/contrib/ seems not to be altered recently, so it should produce the same packages…? >>> >>> Great timing! The feeds are now being populated using Jenkins CI and >>> there is a big rebuild for 2015.06 underway due to the big DISTRO >>> changes that went in last week (i.e. blacklisting bluez5). >>> Eyeballing the jenkins page there are 5 architectures left in the >>> queue, good for 16h builds each spread over 2 machines, so I’d say >>> “will finish this weekend”. >> >> It seems the packages are online now, thanks Koen! >> >> Was there any specific reason why those packages were not part of the >> previous build for 2015.06? > > Mostly lack of time to sit down and check a few architectures on real > devices. I was also concerned about ABI changes and heisenbugs like > the bluez5 blacklisting last week. The prserv support needs to get > updated as well to keep feed updates consistent and working. The good > news is that all the work going into v2015.06 wrt feed updates works > for v2015.12 and next as well :) > >> Just to understand your setup a bit better: Is your Jenkins using the >> scripts under meta-angstrom/contrib/ or is that no longer the way the >> packages get built? > > For v2015.06 and newer it wraps around > https://github.com/Angstrom-distribution/angstrom-manifest/ and does > some magic for reusing sstate. I’ll paste the build step of the > console-feed run at the bottom of this email. It’s not pretty, but it > gets the job done. Ok, and you call these script for each machine? So you basically build all packages for all machines, and then upload just the packages which are not up there yet.. I guess you do this to make sure that all the packages built by the recipe list in the build step are available for _all_ the machines you build... But with that, you probably rebuild armv7at2hf-vfp-neon several times right? And the images are another build step? Would you mind add colibri-vf to the list of machines? Since fido the machine is in meta-fsl-arm-extra which is already part of your bblayers.conf... -- Stefan ___ Angstrom-distro-devel mailing list Angstrom-distro-devel@linuxtogo.org http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/angstrom-distro-devel
Re: [Angstrom-devel] State of package feed for 2015.06
> Op 26 okt. 2015, om 18:21 heeft Stefan Agner het volgende
> geschreven:
>
> On 2015-10-22 04:55, Koen Kooi wrote:
>>> Op 20 okt. 2015, om 22:34 heeft Stefan Agner het volgende
>>> geschreven:
>>>
>>> Hi,
>>>
>>> It seems that the feeds of version 2015.06 currently contains a lot less
>>> packages than the 2014.12 feeds did.
>>>
>>> At least this is true for armv7at2hf-vfp/ and armv7at2hf-vfp-neon/. E.g.
>>> the first package at the very top which is missing is a52dec, but there
>>> is a lot of stuff like binutils or mono which are missing.
>>>
>>> Is this intentional? build-feeds.sh in meta-angstrom/contrib/ seems not
>>> to be altered recently, so it should produce the same packages…?
>>
>> Great timing! The feeds are now being populated using Jenkins CI and
>> there is a big rebuild for 2015.06 underway due to the big DISTRO
>> changes that went in last week (i.e. blacklisting bluez5).
>> Eyeballing the jenkins page there are 5 architectures left in the
>> queue, good for 16h builds each spread over 2 machines, so I’d say
>> “will finish this weekend”.
>
> It seems the packages are online now, thanks Koen!
>
> Was there any specific reason why those packages were not part of the
> previous build for 2015.06?
Mostly lack of time to sit down and check a few architectures on real devices.
I was also concerned about ABI changes and heisenbugs like the bluez5
blacklisting last week. The prserv support needs to get updated as well to keep
feed updates consistent and working. The good news is that all the work going
into v2015.06 wrt feed updates works for v2015.12 and next as well :)
> Just to understand your setup a bit better: Is your Jenkins using the
> scripts under meta-angstrom/contrib/ or is that no longer the way the
> packages get built?
For v2015.06 and newer it wraps around
https://github.com/Angstrom-distribution/angstrom-manifest/ and does some magic
for reusing sstate. I’ll paste the build step of the console-feed run at the
bottom of this email. It’s not pretty, but it gets the job done.
regards,
Koen
env
repo init -u git://github.com/Angstrom-distribution/angstrom-manifest -b
angstrom-v2015.06-yocto1.8
repo sync -j16
repo manifest -r
# Jenkins slaves use different UIDs, so set umask to 002 for sstate sharing
umask 002
chmod 0777 .
# reset to known state
git reset --hard
rm -rf cache/bb_codeparser.dat* bitbake.lock build/tmp* deploy || true
ANGSTROMVERSION="${MD_ANGSTROMVERSION}"
if [ -z "${ANGSTROMVERSION}" ] ; then
echo "Angstrom version not set, exiting"
exit 1
fi
SHARED_DL_DIR="$(echo $WORKSPACE | awk -F'/v201' '{print
$1}')/angstrom-${ANGSTROMVERSION}-shared/downloads"
SHARED_SSTATE_DIR="$(echo $WORKSPACE | awk -F'/v201' '{print
$1}')/angstrom-${ANGSTROMVERSION}-shared/sstate-cache"
WORKSPACEROOT="$(echo $WORKSPACE | awk -F'/v201' '{print $1}')"
mkdir -p ${WORKSPACEROOT}/angstrom-${ANGSTROMVERSION}-shared || true
# Check if sstate mirror and source cache are mounted
if ! [ -e ${WORKSPACEROOT}/angstrom-${ANGSTROMVERSION}-shared/mounted ] ; then
sshfs beast.local:/build/jenkins/angstrom-${ANGSTROMVERSION}-shared
${WORKSPACEROOT}/angstrom-${ANGSTROMVERSION}-shared -o
nonempty,reconnect,ServerAliveInterval=15,ServerAliveCountMax=3
fi
# Point to shared download dir
sed -i -e s:'${OEROOT}/sources/downloads':${SHARED_DL_DIR}:
setup-environment-internal
sed -i -e s:'${OEROOT}/sources/downloads':${SHARED_DL_DIR}:
.repo/manifests/setup-environment-internal
# Point to shared sstate-dir
sed -i -e s:'${OEROOT}/build/sstate-cache':${SHARED_SSTATE_DIR}:
setup-environment-internal
sed -i -e s:'${OEROOT}/build/sstate-cache':${SHARED_SSTATE_DIR}:
.repo/manifests/setup-environment-internal
rm -rf buildhistory || true
git clone --single-branch -b ${ANGSTROMVERSION}
g...@github.com:Angstrom-distribution/buildhistory.git
( cd buildhistory && git checkout origin/${ANGSTROMVERSION} -b
${ANGSTROMVERSION} || true )
MACHINE=$machine . ./setup-environment
for recipe in \
accel-ppp \
alsa-utils \
angstrom-skel \
apache2 \
apcupsd \
asterisk \
at \
avahi \
bc \
bonnie++ \
btrfs-tools \
canutils \
cherokee \
cloud9 \
cmake \
collectd \
connman \
connman-tests \
cpio \
cpufreq-tweaks \
crda \
cronie \
curl \
dcraw \
ddrescue \
devmem2 \
dfu-util \
dosfstools \
dpkg \
dracut \
dropbear \
dtc \
dvb-apps \
e2fsprogs \
elfutils \
enblend-enfuse \
evtest \
exiftool \
exiv2 \
fbida \
fuser \
gateone \
gdb \
gdm \
gettext \
git \
glamor \
gperf \
gphoto2 \
gptfdisk \
grep \
gst-ffmpeg \
gst-plugins-bad \
gst-plugins-base \
gst-plugins-gl \
gst-plugins-good \
gst-plugins-ugly \
gstreamer1.0-plugins-bad \
gstreamer1.0-plugins-base \
gstreamer1.0-plugins-good \
gstreamer1.0-plugins-ugly \
gstreamer1.0-rtsp-server \
Re: [Angstrom-devel] State of package feed for 2015.06
On 2015-10-22 04:55, Koen Kooi wrote: >> Op 20 okt. 2015, om 22:34 heeft Stefan Agner het volgende >> geschreven: >> >> Hi, >> >> It seems that the feeds of version 2015.06 currently contains a lot less >> packages than the 2014.12 feeds did. >> >> At least this is true for armv7at2hf-vfp/ and armv7at2hf-vfp-neon/. E.g. >> the first package at the very top which is missing is a52dec, but there >> is a lot of stuff like binutils or mono which are missing. >> >> Is this intentional? build-feeds.sh in meta-angstrom/contrib/ seems not >> to be altered recently, so it should produce the same packages…? > > Great timing! The feeds are now being populated using Jenkins CI and > there is a big rebuild for 2015.06 underway due to the big DISTRO > changes that went in last week (i.e. blacklisting bluez5). > Eyeballing the jenkins page there are 5 architectures left in the > queue, good for 16h builds each spread over 2 machines, so I’d say > “will finish this weekend”. It seems the packages are online now, thanks Koen! Was there any specific reason why those packages were not part of the previous build for 2015.06? Just to understand your setup a bit better: Is your Jenkins using the scripts under meta-angstrom/contrib/ or is that no longer the way the packages get built? -- Stefan ___ Angstrom-distro-devel mailing list Angstrom-distro-devel@linuxtogo.org http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/angstrom-distro-devel
Re: [Angstrom-devel] State of package feed for 2015.06
> Op 20 okt. 2015, om 22:34 heeft Stefan Agner het volgende > geschreven: > > Hi, > > It seems that the feeds of version 2015.06 currently contains a lot less > packages than the 2014.12 feeds did. > > At least this is true for armv7at2hf-vfp/ and armv7at2hf-vfp-neon/. E.g. > the first package at the very top which is missing is a52dec, but there > is a lot of stuff like binutils or mono which are missing. > > Is this intentional? build-feeds.sh in meta-angstrom/contrib/ seems not > to be altered recently, so it should produce the same packages…? Great timing! The feeds are now being populated using Jenkins CI and there is a big rebuild for 2015.06 underway due to the big DISTRO changes that went in last week (i.e. blacklisting bluez5). Eyeballing the jenkins page there are 5 architectures left in the queue, good for 16h builds each spread over 2 machines, so I’d say “will finish this weekend”. regards, Koen PS: here’s the list of what’s currently in from your requests above. The sama5d3 console feed build is still in the queue, which should explain the missing mono and binutils. a52dec was probably pulled in by the octoprint-image builds earlier this week. koen@beast:/data/www/angstrom/feeds/v2015.06/ipk/glibc/unsorted$ find . -name "a52dec_*.ipk" ./a52dec_0.7.4-r4.0_armv7at2hf-vfp.ipk ./a52dec_0.7.4-r4.0_arm1176jzfshf-vfp.ipk ./a52dec_0.7.4-r4.0_xscaleteb.ipk ./a52dec_0.7.4-r4.0_corei7-64.ipk ./a52dec_0.7.4-r4.0_core2-32.ipk ./a52dec_0.7.4-r4.0_armv5e.ipk ./a52dec_0.7.4-r4.0_mips32r2.ipk ./a52dec_0.7.4-r4.0_core2-64.ipk ./a52dec_0.7.4-r4.0_i586.ipk ./a52dec_0.7.4-r4.0_aarch64.ipk ./a52dec_0.7.4-r4.0_armv7at2hf-vfp-neon.ipk koen@beast:/data/www/angstrom/feeds/v2015.06/ipk/glibc/unsorted$ find . -name "mono_*.ipk" ./mono_4.0.3.13-r0.0_corei7-64.ipk ./mono_4.0.3.13-r0.0_core2-32.ipk ./mono_4.0.3.13-r0.0_armv5e.ipk ./mono_4.0.3.13-r0.0_mips32r2.ipk ./mono_4.0.3.13-r0.0_core2-64.ipk ./mono_4.0.3.13-r0.0_i586.ipk ./mono_4.0.3.13-r0.0_armv7at2hf-vfp-neon.ipk koen@beast:/data/www/angstrom/feeds/v2015.06/ipk/glibc/unsorted$ find . -name "binutils_*.ipk" ./binutils_2.24-r0.0_xscaleteb.ipk ./binutils_2.24-r0.0_corei7-64.ipk ./binutils_2.24-r0.0_core2-32.ipk ./binutils_linaro-2.25-r2015.01.0_armv5e.ipk ./binutils_2.24-r0.0_mips32r2.ipk ./binutils_2.24-r0.0_core2-64.ipk ./binutils_2.24-r0.0_i586.ipk ./binutils_linaro-2.25-r2015.01.0_armv7at2hf-vfp-neon.ipk koen@beast:/data/www/angstrom/feeds/v2015.06/ipk/glibc/unsorted$ ___ Angstrom-distro-devel mailing list Angstrom-distro-devel@linuxtogo.org http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/angstrom-distro-devel
[Angstrom-devel] State of package feed for 2015.06
Hi, It seems that the feeds of version 2015.06 currently contains a lot less packages than the 2014.12 feeds did. At least this is true for armv7at2hf-vfp/ and armv7at2hf-vfp-neon/. E.g. the first package at the very top which is missing is a52dec, but there is a lot of stuff like binutils or mono which are missing. Is this intentional? build-feeds.sh in meta-angstrom/contrib/ seems not to be altered recently, so it should produce the same packages...? -- Stefan ___ Angstrom-distro-devel mailing list Angstrom-distro-devel@linuxtogo.org http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/angstrom-distro-devel
