Debian 9 (stretch) comes with Python 3.5 by default and End of Life is going to be
on 30th June 2022, let's update it to Debian 10 (buster).
Signed-off-by: Josef Schlehofer <pepe.schlehofer@gmail.com>
E.g. some packages are target, or arch specific, skip compiling them if
they cannot be enabled for current sdk. This should reduce false
positives for packages like docker-ce etc.
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Commit 200f2666fb1c ("sdk: fix GCC and Python dangling symlinks") in
master tree has forced prereq build step, which is now failing on the CI
as the Docker CI images are missing Python3 support, so now every build
fails with the following error:
make: *** [staging_dir/host/.prereq-build] Error 1
which is not helpful due to the silenced output of `make defconfig >
/dev/null`, so let's add `make prereq` build step in order to get
something usable next time.
Reported-by: Karl Palsson <karlp@etactica.com>
Signed-off-by: Petr Štetiar <ynezz@true.cz>
Due to concurrent build, it's currently not that easy to find the proper
reason for the package build failure, so let's just build the failing
package with -j1 in order to provide more human readable build log
failures.
Ref: https://github.com/openwrt/packages/pull/9584#issuecomment-518038631
Signed-off-by: Petr Štetiar <ynezz@true.cz>
It appears snapshot target builds have switched from GPG signatures
(sha256sums.asc) to usign signatures (sha256sums.sig).
This adds support for verifying these usign signatures. (GPG signatures
will also be verified if found.)
This also restores the alphabetical ordering of packages to be installed
by apt-get.
Signed-off-by: Jeffery To <jeffery.to@gmail.com>
Installing all packages via ./scripts/feeds install -a produces a lot of
error messages, it is also possible to only install the packages which
were touched by the PR.
Signed-off-by: Paul Spooren <mail@aparcar.org>
Change docker hub repository to 'openwrtorg'.
Create and use non-root user 'build'. Add xz-utils and unzip.
Use version numbers for docker images (SHA has no sense of time).
Disable pipefail in build step
Signed-off-by: Ted Hess <thess@kitschensync.net>
Checking:
- Pull request does not contain unwanted merges
- signed-off-by tag exists and matches author
- Subject line has package name
- Author name has 'firstname lastname' (no nicknames)
Signed-off-by: Ted Hess <thess@kitschensync.net>
[Use git instead of CircleCI variables]
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
I used podman/buildah to build this image, and the local sha256 is not the same than
the docker hub sha256. The layers are the same, so maybe just docker hub changing the manifest
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
Even when configured to only build PR, CircleCI always build the default branch
`only: /pull.*/` seems like a good idea, but it fails when:
- you are testing on the same repo, because github doesn't create a pull branch
- when your repo is configured, because it see the push on your repo,
prevent building this commit because it doesn't match the regexp, then when you
do the PR CircleCI doesn't work
In the end what we really want is `ignore: master`
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
The big advantages are:
- everyone can download the build logs and the ipks
- we use our own docker image
- people with commit access can ssh into the build env
The disadvantages:
- need to push new commits to restart the build
I haven't reimplemented the commit message checks as this should be replaced
with a separate script doing only that so we can require it
(https://help.github.com/articles/enabling-required-status-checks/)
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
Josef Schlehofer
Daniel Engberg
Jeffery To
Yousong Zhou
Petr Štetiar
Paul Spooren
Rosen Penev
Ted Hess
Etienne Champetier