#!/usr/bin/env bash # Exercise the pipeline scripts end to end against a stub DealSage, without # Gitea. Builds a throwaway app repo, then runs these scenarios: # 1. auto policy: checks pass (one advisory fails), build, upload, promote → live # 2. a required check fails: nothing is allocated # 3. manual policy: built and kept on the branch channel, not promoted # 4. installing dependencies fails: the run closes as build_failed # 5. a step dies without closing the run (a missing variable) # 6. Set up Node.js (a stock action) fails # Every scenario also checks the step list the pipeline reports for the build page. # Usage: templates/ci-actions/test/run-local.sh set -euo pipefail HERE="$(cd "$(dirname "$0")" && pwd)" SCRIPTS="$HERE/../pipeline/scripts" PORT="${PORT:-8765}" WORK="$(mktemp -d)" FIXTURE_APP_ID="11111111-2222-4333-8444-555555555555" trap 'kill ${STUB_PID:-} 2>/dev/null || true; [ -n "${KEEP_WORK:-}" ] || rm -rf "$WORK"' EXIT pass=0; fail=0 check() { if eval "$2"; then echo " ok $1"; pass=$((pass+1)); else echo " FAIL $1"; fail=$((fail+1)); fi; } make_app() { # dir required_check_cmd local d="$1" reqcmd="$2" mkdir -p "$d/.dealsage" "$d/.gitea/workflows" "$d/src/lib" cat > "$d/package.json" <ok' > dist/index.html && echo 'console.log(1)' > dist/assets/app.js", "typecheck": "$reqcmd" } } JSON cat > "$d/.dealsage/app.json" < "$d/.gitea/workflows/publish.yml" : > "$d/src/lib/dealsage.ts"; : > "$d/src/Root.tsx"; : > "$d/src/main.tsx" printf '# Agent guide\n\n\n' > "$d/AGENTS.md" git -C "$d" init -q -b feature/x && git -C "$d" -c user.name=t -c user.email=t@t add -A \ && git -C "$d" -c user.name=t -c user.email=t@t commit -qm "Add fixture app" } load_env() { # GITHUB_ENV file -> exports; composite steps share state through it (heredoc form) eval "$(python3 - "$1" <<'PY' import re,sys,shlex txt=open(sys.argv[1]).read() for m in re.finditer(r'^(\w+)<<__DS_EOF__\n(.*?)\n__DS_EOF__$', txt, re.S|re.M): print(f"export {m.group(1)}={shlex.quote(m.group(2))}") PY )" } # run_pipeline DIR -> runs the action's steps in order, stopping at the first # that fails, then finish-steps.sh (if: always()); echoes the step that failed # or "ok". Knobs, as environment: INSTALL_COMMAND, UNSET_ENV (names to unset), # FAIL_NODE=1 (Set up Node.js fails; nothing of ours runs in that step). run_pipeline() { local d="$1" ghenv ghout ghsum ghenv="$(mktemp)"; ghout="$(mktemp)"; ghsum="$(mktemp)" export GITHUB_ENV="$ghenv" GITHUB_OUTPUT="$ghout" GITHUB_STEP_SUMMARY="$ghsum" export RUNNER_TEMP; RUNNER_TEMP="$(mktemp -d)" export GITHUB_SHA="$(git -C "$d" rev-parse HEAD)" GITHUB_REF_NAME="feature/x" GITHUB_ACTOR="tester" export DEALSAGE_FUNCTIONS_URL="http://127.0.0.1:$PORT" PUBLISH_TOKEN="test-publish-token" export SUPABASE_ANON_KEY="anon" SUPABASE_URL="https://example.supabase.co" CUSTOM_APPS_WORKER_URL="http://127.0.0.1:$PORT" export MANIFEST_PATH=".dealsage/app.json" INSTALL_COMMAND="${INSTALL_COMMAND:-echo 'up to date, audited 0 packages'}" unset APP_ID MANIFEST_JSON CHECKS_JSON CHECKS_RESULT_JSON BUILD_ID local v; for v in ${UNSET_ENV:-}; do unset "$v"; done local stage failed="" for stage in ensure-tools manifest start-build node install run-checks build publish; do if [ "$stage" = node ]; then if [ "${FAIL_NODE:-0}" = 1 ]; then failed=node; break; fi continue fi ( cd "$d" && bash "$SCRIPTS/$stage.sh" ) > "$WORK/$stage.log" 2>&1 || failed="$stage" load_env "$ghenv" [ -z "$failed" ] || break done if ! ( cd "$d" && bash "$SCRIPTS/finish-steps.sh" ) > "$WORK/finish-steps.log" 2>&1; then failed="${failed:-finish-steps}" fi cp "$ghout" "$WORK/last.out"; cp "$ghsum" "$WORK/last.summary" echo "${failed:-ok}" } start_stub() { # policy : > "$WORK/requests.jsonl" STUB_LOG="$WORK/requests.jsonl" STUB_POLICY="$1" python3 "$HERE/stub_server.py" "$PORT" & STUB_PID=$! for _ in $(seq 1 30); do curl -s -o /dev/null "http://127.0.0.1:$PORT/" && break; sleep 0.1; done } stop_stub() { kill "$STUB_PID" 2>/dev/null || true; wait "$STUB_PID" 2>/dev/null || true; } req() { jq -c "$1" "$WORK/requests.jsonl"; } # The step lists the pipeline posted (record-app-build, action "steps"). first_steps_post() { jq -cs 'map(select(.body.action=="steps"))[0].body' "$WORK/requests.jsonl"; } last_steps() { jq -c 'select(.body.action=="steps") | .body.steps' "$WORK/requests.jsonl" | tail -n 1; } statuses_of() { jq -r 'map(.id + "=" + .status) | join(" ")'; } statuses() { last_steps | statuses_of; } step() { last_steps | jq -r --arg id "$1" "map(select(.id == \$id))[0] | $2"; } finishes() { req 'select(.body.action=="finish")' | wc -l | tr -d ' '; } finish_is() { req 'select(.body.action=="finish") | .body' | jq -e --arg o "$1" --arg r "$2" '.outcome == $o and .failureReason == $r' >/dev/null; } ALL_PASSED="tooling=passed manifest=passed record=passed node=passed install=passed checks=passed build=passed publish=passed" echo "Scenario 1: auto policy" make_app "$WORK/app1" "true"; start_stub auto r="$(run_pipeline "$WORK/app1")" check "pipeline completed" "[ '$r' = ok ]" check "four checks recorded (typecheck, lint, conventions, build)" "[ \"\$(req 'select(.path|endswith(\"/record-app-check\"))' | wc -l)\" = 4 ]" check "advisory lint failure recorded as failed" "req 'select(.body.checkId==\"lint\") | .body.status' | grep -q '\"failed\"'" check "conventions passed" "req 'select(.body.checkId==\"conventions\") | .body.status' | grep -q '\"passed\"'" check "check payload carries commit, branch and log tail" "req 'select(.body.checkId==\"typecheck\") | .body | (.commitSha|length)==40 and .branch==\"feature/x\" and (.logTail|length)>0' | grep -q true" check "allocate sent branch, message, author, manifest and check results" "req 'select(.body.action==\"allocate\") | .body | .branch==\"feature/x\" and .commitMessage==\"Add fixture app\" and .author==\"t\" and (.manifest.dependencies.objectTypes|length)==1 and (.checks|length)==3' | grep -q true" check "two files uploaded through the Worker with the upload token" "[ \"\$(req 'select(.method==\"PUT\" and .auth==\"Bearer upl-token\")' | wc -l)\" = 2 ]" check "promote requested after upload" "req 'select(.body.action==\"promote\") | .body.version' | grep -q 7" check "outputs: version=7 promoted=true" "grep -q '^version=7$' '$WORK/last.out' && grep -q '^promoted=true$' '$WORK/last.out'" check "build start recorded with commit, branch and author" "req 'select(.body.action==\"start\") | .body | (.commitSha|length)==40 and .branch==\"feature/x\" and .author==\"t\"' | grep -q true" check "allocate carries the build id" "req 'select(.body.action==\"allocate\") | .body.buildId' | grep -q '\"bld-1\"'" check "build closed as built, promoted, v7" "req 'select(.body.action==\"finish\") | .body | .buildId==\"bld-1\" and .outcome==\"built\" and .promoted==true and .version==7' | grep -q true" check "steps first sent with the build record, for this build and app" \ 'first_steps_post | jq -e --arg app "$FIXTURE_APP_ID" ".buildId == \"bld-1\" and .appId == \$app" >/dev/null' check "the first post has the steps so far and Set up Node.js running" \ '[ "$(first_steps_post | jq -c .steps | statuses_of)" = "tooling=passed manifest=passed record=passed node=running install=pending checks=pending build=pending publish=pending" ]' check "every step passed, in action.yml order" '[ "$(statuses)" = "$ALL_PASSED" ]' check "each step has a start, a finish and a duration" \ 'last_steps | jq -e "all(.[]; (.startedAt|type) == \"string\" and (.finishedAt|type) == \"string\" and (.durationMs|type) == \"number\" and .durationMs >= 0)" >/dev/null' check "internal fields stay out of the payload" '! req "select(.body.action==\"steps\") | .body.steps[] | select(has(\"startedMs\"))" | grep -q .' check "summaries: manifest, checks (the advisory failure named), build, publish" \ '[ "$(step manifest .summary)" = "3 checks declared" ] && [ "$(step checks .summary)" = "2 of 3 passed (Lint failed, not required)" ] && [ "$(step build .summary)" = "2 files in dist/" ] && [ "$(step publish .summary)" = "v7 is live" ]' check "no log tails on a green run" '[ "$(last_steps | jq "map(select(has(\"logTail\"))) | length")" = 0 ]' check "the finalizer did not close the run again" '[ "$(finishes)" = 1 ]' stop_stub echo "Scenario 2: required check fails" make_app "$WORK/app2" "echo 'type error' && exit 1"; start_stub auto r="$(run_pipeline "$WORK/app2")" check "stopped at run-checks" "[ '$r' = run-checks ]" check "failure still recorded" "req 'select(.body.checkId==\"typecheck\") | .body.status' | grep -q '\"failed\"'" check "nothing allocated" "! req 'select(.body.action==\"allocate\")' | grep -q ." check "advisory and platform checks still ran" "[ \"\$(req 'select(.path|endswith(\"/record-app-check\"))' | wc -l)\" = 3 ]" check "build closed as checks_failed with the reason" "req 'select(.body.action==\"finish\") | .body | .outcome==\"checks_failed\" and (.failureReason|test(\"TypeScript\"))' | grep -q true" check "steps: Checks failed, Build and Publish skipped" \ '[ "$(statuses)" = "tooling=passed manifest=passed record=passed node=passed install=passed checks=failed build=skipped publish=skipped" ]' check "Checks names the required check that failed" 'step checks .summary | grep -q "Required check(s) failed: TypeScript"' check "skipped steps carry no timings" '[ "$(step build ".startedAt // \"none\"")" = none ]' check "the finalizer did not close the run again" '[ "$(finishes)" = 1 ]' stop_stub echo "Scenario 3: manual policy" make_app "$WORK/app3" "true"; start_stub manual r="$(run_pipeline "$WORK/app3")" check "pipeline completed" "[ '$r' = ok ]" check "outputs: promoted=false, channel=branch/feature/x" "grep -q '^promoted=false$' '$WORK/last.out' && grep -q '^channel=branch/feature/x$' '$WORK/last.out'" check "summary mentions the channel" "grep -q 'channel' '$WORK/last.summary'" check "build closed as built but not promoted" "req 'select(.body.action==\"finish\") | .body | .outcome==\"built\" and .promoted==false and .version==7' | grep -q true" check "every step passed" '[ "$(statuses)" = "$ALL_PASSED" ]' check "Publish says where the build was kept" '[ "$(step publish .summary)" = "v7 kept on branch/feature/x; live stays at v6" ]' stop_stub echo "Scenario 4: installing dependencies fails" make_app "$WORK/app4" "true"; start_stub auto r="$(INSTALL_COMMAND="echo 'npm error 404 Not Found - GET https://registry.npmjs.org/left-pad-nope' >&2; exit 1" run_pipeline "$WORK/app4")" check "stopped at install" '[ "$r" = install ]' check "build closed as build_failed, saying why" \ 'finish_is build_failed "Installing dependencies failed (exit code 1); nothing was published."' check "steps: Install dependencies failed, the rest skipped" \ '[ "$(statuses)" = "tooling=passed manifest=passed record=passed node=passed install=failed checks=skipped build=skipped publish=skipped" ]' check "Install dependencies carries the tail of its output" 'step install .logTail | grep -q "npm error 404"' check "no checks ran and nothing was allocated" \ '! req "select((.path|endswith(\"/record-app-check\")) or .body.action == \"allocate\")" | grep -q .' check "the run was closed once" '[ "$(finishes)" = 1 ]' stop_stub echo "Scenario 5: a step dies without closing the run" make_app "$WORK/app5" "true"; start_stub auto r="$(UNSET_ENV="SUPABASE_URL" run_pipeline "$WORK/app5")" check "stopped at build" '[ "$r" = build ]' check "the finalizer closed the run as failed, naming the step and the cause" \ 'finish_is failed "Build: SUPABASE_URL is not set"' check "steps: Build failed, Publish skipped" \ '[ "$(statuses)" = "tooling=passed manifest=passed record=passed node=passed install=passed checks=passed build=failed publish=skipped" ]' check "the run was closed once" '[ "$(finishes)" = 1 ]' stop_stub echo "Scenario 6: Set up Node.js fails" make_app "$WORK/app6" "true"; start_stub auto r="$(FAIL_NODE=1 run_pipeline "$WORK/app6")" check "stopped at node" '[ "$r" = node ]' check "the finalizer failed the step that never finished and closed the run" \ 'finish_is failed "Set up Node.js did not finish."' check "steps: Set up Node.js failed with a duration, the rest skipped" \ '[ "$(statuses)" = "tooling=passed manifest=passed record=passed node=failed install=skipped checks=skipped build=skipped publish=skipped" ] && [ "$(step node ".durationMs | type")" = number ]' stop_stub echo; echo "$pass passed, $fail failed" [ "$fail" = 0 ]