artemis-docker

Docker images in which Artemis grades the CommonRoad programming exercises. Each exercise's build runs in its image, which already has the fixed grading dependencies installed, so a build does not install them again.

Image	Exercise	What it contains
ghcr.io/commonroad/artemis-docker/bonus	Bonus Challenge	pytest, commonroad-io, commonroad-drivability-checker, commonroad-vehicle-models, triangle, commonroad-clcs. The build script installs nothing.
ghcr.io/commonroad/artemis-docker/competition	Motion Planning Competition	The grading dependencies in competition/requirements.txt. The build script installs the student's own dependencies into a separate venv at build time.
ghcr.io/commonroad/artemis-docker/search	CommonRoad Search (the Bonus Challenge without scenarios in its repositories)	The bonus image plus the 500 ground-truth scenarios in /opt/commonroad-search-data/scenarios/exercise. The build script installs nothing.

CI pushes each image to this project's registry and to the public GitHub Container Registry of the CommonRoad organization. Artemis pulls from ghcr.io without logging in; its build agents cannot pull from this registry, which listens on port 5005.

Updating an image

1. Change <name>/Dockerfile (or competition/requirements.txt) and push to main.
2. CI builds the image and pushes it with the tags <short commit SHA> and latest. The last line of the job log is the full ghcr.io image reference.
3. In Artemis, open the exercise's build configuration and set Docker Image to that reference.

Artemis always uses a SHA tag, so a new build never changes a running exercise until its Docker Image field is updated. A build can also be started by hand with Build → Pipelines → Run pipeline.

Exercise data of CommonRoad Search

The scenarios and motion primitives of the exercise never change during an edition and are about 400 MB of XML. Kept in the exercise's Artemis repositories, they are copied into every student repository and checked out for every build, so the search image and its exercise keep them here instead, as one archive: search/data/commonroad-search-data-<version>.tar.xz (about 20 MB).

• search/Dockerfile unpacks the archive and copies the exercise scenarios into the image, where the tests read them. Its base is a pinned bonus image, so the grading dependencies are defined in one place only, bonus/Dockerfile. After a change there, set the FROM tag in search/Dockerfile to the new bonus image.
• SMP/exercise_data.py in the exercise's Template repository downloads the archive from this repository into a student's checkout the first time SMP is imported. It pins the commit of the archive and its SHA-256.

The bonus image and exercises that use it are not affected: their Test repository still holds the scenarios.

To change the data (a new edition, or a corrected scenario):

1. Run python search/data/make_bundle.py <complete checkout of the exercise> <new version> and commit the new archive. Keep the old archive: checkouts of earlier Template versions still download it.
2. Set DATA_VERSION and DATA_SHA256 in search/Dockerfile to the printed values and push to main. Use the new image in Artemis as described above.
3. In the exercise's Template and Solution repositories, set DATA_VERSION, ARCHIVE_SIZE, ARCHIVE_SHA256 and _ARCHIVE_COMMIT (the commit of step 1) in SMP/exercise_data.py. Checkouts with the old data fetch the new archive on their next import of SMP.

ghcr.io access

The push to ghcr.io uses the CI/CD variables GHCR_USER and GHCR_TOKEN (masked): a GitHub account in the CommonRoad organization and a token of it with the write:packages scope. Without them, CI pushes to this project's registry only. Every package must be public on GitHub (package settings → Change visibility); a new image name, such as search, starts out private.