During refactoring, conflict_version was changed to version here, which
was wrong. The packages need to depend on the conflict_version of the
meta package, version is the version of the package itself.
Fix for:
The following packages have unmet dependencies:
libosmo-netif-dev : Depends: osmocom-nightly (= 202303160009) but it is not going to be installed
Depends: libosmocore but it is not going to be installed
Depends: libosmonetif11 (= 1.3.0.8.9e65.202303160009) but it is not going to be installed
libosmocore-dev : Depends: osmocom-nightly (= 1.8.0.79-b394d.202303170006) but it is not going to be installed
Depends: libosmocore but it is not going to be installed
Fixes: 0ed0e464 ("obs: don't pass conflict_version through functions")
Change-Id: I2326e3817c6f6887ef1196e603c3877768119a66
With recent changes in open5gs, building the source packages we
generated fails:
[ 45s] ../lib/metrics/meson.build:52:4: ERROR: Automatic wrap-based subproject downloading is disabled
[ 45s] dh_auto_configure: error: cd obj-x86_64-linux-gnu && LC_ALL=C.UTF-8 meson .. --wrap-mode=nodownload --buildtype=plain --prefix=/usr --sysconfdir=/etc --localstatedir=/var --libdir=lib/x86_64-linux-gnu --libexecdir=lib/x86_64-linux-gnu returned exit code 1
Fix this by downloading all subprojects. This is now possible as I added
a meson.build file to the prometheus-client-c repository that open5gs
uses: https://github.com/open5gs/prometheus-client-c/pull/2
Change-Id: If3910b520382b177a77b216c93771ea88414723d
Installing osmo-gsm-manuals-dev plus depends takes a long time. Don't do
this for every build, instead do it once when building a second docker
container and then use that.
Related: OS#2385
Change-Id: I8475bd954352b572197795ad4cd9461e39896d48
Add a new master feed, where packages are updated as soon as patches get
merged to master. Upload a commit_$COMMIT.txt file for each package in
this feed and use it to determine if the package needs to be updated or
not.
In most packages the commit is already part of the version in the dsc
file, e.g. "libosmocore_1.7.0.38.c3b90.dsc", but that's harder to parse
and is more likely to have a hash collision (just 5 characters).
Related: OS#2385
Change-Id: I3b0b4f4876b8c1eeb61f20d903a6f2cac6e99638
Having an option to explicitly checkout sources for package from gerrit
using numeric ID comes in handy for testing changes.
Related: OS#4107
Change-Id: I403c2083ef8c3f6f966c5b376fb13ea46d8a7a2b
If there's an empty debian/changelog, packaging needs to be fixed in the
git repository of the project. Print a meaningful error instead of:
File "/obs/lib/srcpkg.py", line 68, in get_version_for_feed
if ":" in ret:
TypeError: argument of type 'NoneType' is not iterable
Change-Id: I294d844ccccfa12599b6ba1def087e4a32d289e7
As Neels pointed out, with the current code the 2022q1 and similar feeds
do not get checked out properly if:
* the repository was not freshly cloned (like jenkins would do)
* the repository was fetched (--git-fetch)
Instead of the last revision of that branch from origin, we get the
revision that was previously checked out on the branch. Fix this by
prepending origin/.
Change-Id: I2e265fc84abd4596c386bb56cbabb2af121dc7ca
Harald requested that the OBS scripts should not stop if building one
specific source package fails, instead it should keep going and report
at the end a non-success exit code.
Given that the shell script code has historically grown and became hard
to maintain, I decided to rewrite the scripts for implementing this
feature. This rewrite solves additional problems:
* No full checkout of an OBS project like network:osmocom:latest
anymore, with lots of packages that won't get updated (e.g. the uhd
package has a uhd-images_3.14.1.1.tar.xz file that is 108 MB). With
the old code, developers had to wait minutes during the checkout
before the script reaches code that is currently being developed. Now
only single packages get checked out right before they get updated.
* No need to clone git repositories over and over. With the new code,
git repos only get cloned if needed (for latest it is not needed if
the remote git tag is the same as the version in OBS). During
development, the cloned git repositories are cached.
* Output from commands like "git tag -l" is not written to the log
unless they failed. This makes the log more readable, which is
especially important when a package fails to build, we keep going and
need to spot the build error in the middle of the log later on.
* No more duplicated code for nightly and latest scripts that worked
similar but had slight differences. Also the list of packages is not
duplicated for nightly and latest anymore; nightly uses all packages
and latest uses packages that have at least one git tag.
* Building source packages is decoupled from uploading them. A separate
script build_srcpkg.py can be used to just build the deb + rpm spec
source packages, without interacting with the OBS server.
* The scripts can optionally run in docker with a command-line switch,
and this is used by jenkins. This way we don't need to install
more dependencies on the host such as rebar3 which is now needed for
erlang/osmo_dia2gsup.
* Add erlang/osmo_dia2gsup and run its generate_build_dep.sh (SYS#6006)
I have done the new implementation in python to make use of argparse
and to be able to use try/except and print a trace when building one
package fails.
Example output:
* https://jenkins.osmocom.org/jenkins/job/Osmocom_OBS_nightly_obs.osmocom.org/48/console
* https://jenkins.osmocom.org/jenkins/job/Osmocom_OBS_latest_obs.osmocom.org/46/console
Change-Id: I45a555d05a9da808c0fe0145aae665f583cb80d9
Oliver Smith
Max
Neels Hofmeyr