Container image build
Why a container image instead of a zip
The Jolt sibling deploys a zip on provided.al2023, whose execution environment is Amazon Linux 2023 (AL2023) with glibc 2.34. jank's own CI builds on ubuntu-24.04, whose glibc is 2.39. The Jolt sibling hit the same mismatch with its own prebuilt binary (needing glibc >= 2.35) and solved it by building jolt from source inside an AL2023 container.
Doing that for jank would mean building its full LLVM/Clang toolchain from source on AL2023. That is much heavier than Jolt's ~4-minute Chez Scheme build, and nobody in the jank ecosystem had shown it working when this port was written.
So this port deploys as a Lambda container image. AWS's own documentation says a non-AWS base image just needs to "include a runtime interface client... if you're using a language that doesn't have an AWS-provided runtime interface client, you must create your own", which is what runtime.jank already is. ubuntu:24.04 plus jank's own APT package (ppa.jank-lang.org) avoids the glibc/LLVM-from-source problem entirely.
The build
FROM ubuntu:24.04
# ... install jank via its APT package ...
COPY src ./src
RUN jank compile --module-path src --target-dir target --name bootstrap net.b12n.lambda-mvp.main
ENTRYPOINT ["/var/task/target/bootstrap"]
bb image builds this for LAMBDA_ARCH (default amd64). Phase 0's spike confirmed jank's PPA for Ubuntu 24.04 publishes amd64 only: dists/noble/Release states Architectures: amd64 and there is no binary-arm64/ directory. That differs from the Jolt sibling's arm64-by-default choice; see the design spec's Open Risks for the full detail.
Deploying a container image needs Amazon ECR
Unlike the zip-based Jolt sibling (a direct --zip-file upload), Lambda container-image functions must be pushed to Amazon ECR. bb deploy creates the ECR repository if it doesn't exist, authenticates Docker to it, tags and pushes the local image, then creates or updates the Lambda function with --package-type Image --code ImageUri=.... bb teardown deletes the ECR repository along with the function and IAM role.
A plain docker push can produce an image Lambda rejects
If bb deploy fails with InvalidParameterValueException: The image manifest, config or layer media type for the source image ... is not supported, the cause is Docker Desktop's default behavior on a containerd-backed image store (the default since Docker Desktop enabled it), not an environment quirk like the README's "Restricted networks" issues. A plain docker build attaches a build-provenance attestation to every image, so a plain docker push writes the registry tag as an OCI image index (the single-platform image manifest plus a small attestation manifest) instead of a single image manifest, and Lambda's CreateFunction/UpdateFunctionCode reject that index outright. bb deploy's push-image! avoids this with docker push --platform <arch>, Docker's documented flag for pushing a single-platform manifest without the index/attestation wrapper. It was verified against a real deploy on 2026-09-11. If you adapt this project's deploy script elsewhere, keep that flag: any recent Docker Desktop install needs it.