Opened 4 years ago
Last modified 16 months ago
#2040 new task
buildbot switch to docker
Reported by: | Antoine Martin | Owned by: | Smo |
---|---|---|---|
Priority: | critical | Milestone: | 5.0 |
Component: | packaging | Version: | 2.4.x |
Keywords: | Cc: |
Description
Based on #2028, we can have a base image with the main build dependencies for each distro supported.
ie for centos7, here's a list of packages, not just xpra but also the build dependencies for some of the other packages we have to provide:
xorg-x11-server-devel sudo epel-release krb5-devel python-devel rpm-build pkgconfig which #gstreamer plugin: gstreamer1-devel gstreamer1-plugins-base-devel #(py)gtkgl: gtk2-devel libGLU-devel libXmu-devel libXt-devel pangox-compat-devel ncurses-devel libXinerama-devel pango-devel pygtk2-devel #pycuda: numpy python-distribute boost-devel openssl-devel python-setuptools #python-uinput: libudev-devel python-six xmlto #python-pillow: tkinter tk-devel libjpeg-devel lcms2-devel PyQt4
Not sure if all the dependencies should be in the base image, or only added when we build the package that requires them. (so many packages..)
Then we have to install the packages we build for the next step:
- pangox-compat
- gtkglext
- pygtkglext
or:
- lz4
- python-lz4
So some form of scripting is likely to be needed to manage those dependencies.
Other distros may not need to build some packages (ie: python-pillow is in the Fedora repository).
This can then be used to build the master image for building xpra.
Attachments (1)
Change History (20)
comment:1 Changed 4 years ago by
Status: | new → assigned |
---|
comment:2 Changed 3 years ago by
Example of something difficult to handle: http://winswitch.org/trac/ticket/307#comment:6: we need to build lz4 on centos, except on centos 7.4 and later..
comment:3 Changed 3 years ago by
Another problem: mock 1.4.2-1.el7 fails to build el6 packages and systemd-nspawn cannot change user with EL6 chroot.
I was getting getent initgroups failed with error code 1.
, worked around that by using nspawn as root and running su -l buildbot
from there..
comment:4 Changed 3 years ago by
And another: we need a recent version of nasm to build x264, r21315 adds a spec file so we can package the latest release. But on centos6, it does not build:
configure.ac:3: error: Autoconf version 2.69 or higher is required ...
comment:5 Changed 3 years ago by
more:
- pycuda 2014.1 is the last version that can be built against the version of numpy found in the centos6 repos, and the specfile for 2014.1 is found in our 0.15.x release!
- cuda 7.5 is the last release suported by pycuda 2014 / centos6
And installing older versions of CUDA on fedora is not easy because we're missing attachment/ticket/2040/InstallUtils.pm... so run it like so:
INSTALLER_DIR=/opt/cuda-5.5-installer INSTALL_DIR=/opt/cuda-5.5 sh cuda_5.5.22_linux_64.run -verbose -override -extract $INSTALLER_DIR cd $INSTALLER_DIR wget https://xpra.org/trac/raw-attachment/ticket/2040/InstallUtils.pm PERL5LIB=. ./cuda-linux64-rel-5.5.22-16488124.run --prefix=$INSTALL_DIR -noprompt
comment:6 Changed 3 years ago by
Another issue similar to comment:2 is when the distro ships a broken package (ie: either out of date version or buggy packaging): in this case we want to ship a working package, but stop doing so if and when the distro fixes things.
Example: python-uinput broken in Fedora 24 to Fedora 29! (~3 years without a fix!)
comment:7 Changed 3 years ago by
Out of the 40+ builds, one slipped through with the wrong ffmpeg dependency: Error: Package: ffmpeg-xpra-4.1-1.el7_6.x86_64 (Xpra)..
comment:8 Changed 3 years ago by
new constraint: some versions may require a package (ie: versions before 2.5 require websockify and python-websocket-client - see #2121), but not after. Yet we need to keep the RPM specfile updated and update the repository from that branch - until the distribution ships it / updates that version.
comment:11 Changed 3 years ago by
Just when you think that things can't possibly get more interesting: Ubuntu ships two incompatible versions of Xorg for the same distro version, see ticket:2190#comment:7.
comment:12 Changed 3 years ago by
Priority: | major → critical |
---|
comment:13 Changed 3 years ago by
Another one that should have been spotted: #2240, xpra 1.0.13 builds for centos 6.9 are MIA.
comment:15 Changed 3 years ago by
Owner: | changed from Antoine Martin to Smo |
---|---|
Status: | assigned → new |
comment:16 Changed 3 years ago by
Yes I've been working on a few instances of docker to build these. I have committed the stuff that I was working on for centos 7.6 which works mostly well.
I'm not too happy with the results as every time I build I have to go through all the packages that I need and build them in the correct order.
This has a few issues.
1) No caching of source tarballs which is quite wasteful
2) Rebuilding all rpms even when we don't need to
Should probably change this so we can call the shell script with a name of spec file to build that way we can do one offs.
That doesn't solve the problem completely but at least we can built each one 1 at a time.
The problem is that some things need to be installed in the docker container for building other things. Which is possible but do we want to install all these every time we build 1 rpm that would be costly for time.
comment:17 Changed 3 years ago by
Milestone: | 3.0 → 4.0 |
---|
comment:18 Changed 2 years ago by
Milestone: | 4.0 → 5.0 |
---|
comment:19 Changed 16 months ago by
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/2040
See also
el7
vsel7_6
in ticket:2028#comment:13