It happens that the integration branches are left over from provious
runs of the pipeline, when MRs are closed without merging, for example.

Also a bug in the gitlab-ci makes the pipeline fail, when an integration
branch should be reused. The new scripts just deletes all integration
branch in the given project tree without further checks.

Needs to be run manually.

Use:
```
cd script
source setup_venv.sh
```
to change to the venv.

Without the deep copy, the merge operation manipulates the dict it is
storing the new values in.

Dmitry Petrov authored 10 months ago and Jonas Höppner committed 10 months ago

The current solution makes too much assumptions about the pipeline and needs
to reimplement some of the gitlab yaml interpretion.

The new version of the script extracts MACHINE and DISTRO from the jobs themselves:
- MACHINE is part of the build job name ("build-$MACHINE");
- DISTRO is printed by bitbake and can be found in build log.

Typos

The credentials helper wasn't storing the credentials in the cache.
Seemingly, the echo added whitespaces at the beginning of the lines
which can't be interpreted by the git credential approve command.
Also the USERNAME was renamed to TOKEN_USER because the variable is
already filled with the system users name.

To use the generated json file in the json-fwr tooling, it helps to fill
the name.

Parse machines, pipeline names and distros from gitlab yaml file instead of
harcoding these values in the script.

The problem was found in pipeline:
https://git.seco.com/seco-ne/yocto/manifest/-/jobs/954246

Fix missing update of upstream pipeline status inside while loop.

The timeout in the script was not in sync with the default timeout for build jobs.
Set timeout in __get_build_pipelines() function to 8 hours.

Dmitry Petrov authored 1 year ago and Jonas Höppner committed 11 months ago

Several blocks of code are too long and don't fit on the page.
Words are not wrapped to the next line.
To fix this, adjust the span slacc in css.

Dmitry Petrov authored 1 year ago and Jonas Höppner committed 11 months ago

- fix h1 color
- fix missing gorizontal line above h1
- increased h2 font size
- fix missing borders for table
- restore link style

It seems that these jobs are not yet completly errorsafe and we could
ignore the failure without changing the build artifacts.

Dmitry Petrov authored 11 months ago and Jonas Höppner committed 11 months ago

In the latest commit in the manifest it often happens, that the pipeline
does not contain the artifacts itself, but just a link to the pipeline of the MR.

Added a python script with corresponding GitLab CI job to launch this script.
The script searches for the latest pipelines for default branch in manifest repo,
and then decides either to keep or delete artifacts for particular build.

The build artifacts are kept for:
- the latest commit
- the last successful build
- the tagged commits (e.g. "fngsystem/47.0", "kirkstone/20.0")

Dmitry Petrov authored 11 months ago and Jonas Höppner committed 11 months ago

Add a python script and a GitLab CI job to run it. The job is triggered
when the build is complete.

The script retrieves build artifacts for the default/main branch
(currently "kirkstone") in the manifest repo and build artifacts
for the integration branch of merge request.

After that, it compares those builds and creates a summary of the overall
changes. In particular:
- size of compressed artifacts (artifacts.zip)
- size of built image
- difference between manifest files

The summary is reported to stdout and as a comment to the merge request.
Each run of the pipeline means removing the old summary comment and
creating a new one when the build is finished.

The SSH setup is required for some private projects that pull in
external repos via SSH.

This reverts commit 4d5bca8f.

Even though the variable isn't used in the package_release script, it
seems to be required by the os.walk function. Underscore satisfies the
linter preventing it from causeing unused variable errors.

Item should end with "/" to be treated as a directory

See for error:
https://git.seco.com/seco-ne/yocto/infrastructure/ci-test/minimal-manifest/-/jobs/828261

The function that merges the variables of the different jobs, i.e:

`MACHINE="mx6"`

`MACHINE="mx6ull"`

used python `in` to check if a value already exists.
But `"mx6" in "mx6ull"` is true, so mx6 was dropped.
This rewrites it, storing the variables in a set (which is unique
automatically) instead of space seperated strings.
Space seperation is added before printing the variables to be used on
on the shell.

Dmitry Petrov authored 1 year ago and Jonas Höppner committed 1 year ago

The job checks bitbake recipes added/updated in MR using the
oelint-adv tool (https://github.com/priv-kweihmann/oelint-adv).

.oelint.json is a rulefile, created with the command
"oelint-adv --print-rulefile". It can be used during development.
It contains default rules for now, but will probably be customized later.

To get the list of changed files in MR, we adjusted the GIT_DEPTH variable.
The new value is 50, which should be fine for getting MR changes,
and still small enough to keep checkout fast for large projects.

Fix the problem introduced by commit 4f06516a "convert_md2html: fix id
fields for headings".

The text converted from markdown was written twice to the HTML document.

Mikhail Vanyulin authored 1 year ago and Jonas Höppner committed 1 year ago

Converter adds "user-content-" to "id" fields for headings.
As a result links from TOC do not work.
Fix this for now by removing "user-content-" from the resulting html.

Signed-off-by: Mikhail Vanyulin <mikhail.vanyulin@rtsoft.de>

Mikhail Vanyulin authored 1 year ago and Jonas Höppner committed 1 year ago

Update convert_md2html.py to use gitlab api with gitlab flavor.

Signed-off-by: Mikhail Vanyulin <mikhail.vanyulin@rtsoft.de>

On i.MX6 devices, the U-Boot file types are not set at all. Extent check for
the existance of the uboot_file_types var.

The sourcery plugin fails again due to a missing login. Disable it for
now.