adcm-bundle-validate

Validating an ADCM bundle change

The upload response is the test

Real QA suites for these bundles live in a separate, usually private, repository. What you can do from the bundle repository alone is stronger than it sounds, because ADCM validates the entire bundle definition when you upload it: every prototype, every config field, every action's state machine, every requires:, every component constraint.

So bundle testing here is three steps, and the third is the one that matters:

ADCM_SCRIPTS="${CLAUDE_PLUGIN_ROOT}/scripts"
PY=.adcm/venv/bin/python

# 1. rebuild — always, even for a one-line change; it takes a second and guarantees
#    you are testing what you edited
$PY "$ADCM_SCRIPTS/bundle_build.py" --dev-version --skip-pymod

# 2. offline static checks
$PY "$ADCM_SCRIPTS/bundle_check.py"

# 3. let ADCM judge it (no argument: uses last_build from .adcm/state.json)
python3 "$ADCM_SCRIPTS/adcmctl.py" upload

Success reads:

UPLOADED: bundle id=42 name=<prototype> version=<version>_dev_<epoch> edition=enterprise

The bundle id is recorded in .adcm/state.json so adcm-cluster-deploy picks it up with no arguments.

The static layers, and what each one cannot see

bundle_check.py runs four independent layers over the files the selected edition actually ships:

Layer	Checks	Blind to
yaml	every shipped *.yaml / *.yml parses	whether ADCM accepts the schema
python	every shipped *.py parses (library/, module_utils/, callback_plugins/, filter_plugins/, roles)	imports, runtime behaviour
jinja	every runtime *.j2 parses as a template	undefined variables — the context comes from ADCM at job time
gen	a generated file edited without its auto_bump source	whether the mirrored content is correct

Two exclusions are deliberate and worth knowing so a "missing" check does not look like a gap:

• Splitter sources named in spec.yaml are skipped by the first three layers. They are build-time templates; their rendered output is what must be valid, and bundle_build.py validates that.
• templates/ inside an Ansible role is skipped by the python layer. Those files are rendered on the target host and may be python 2, or Jinja that is not valid python.