Taskfile.yml

version: "3"

# Everything in this repository is driven by `task`.
# `task` with no argument lists what there is; `task verify` is the whole gate.
#
# # A cache that lies is worse than no cache
#
# Task skips work by CHECKSUMMING `sources`. That says nothing about whether the outputs still exist.
# Two real failures came out of that gap: `rm dist/krm-stream.js && task build-client` reported "up to
# date" and rebuilt nothing, and `npm install` silently pruned a devDependency when run from a branch
# that predated it, so a later build died on `sh: 1: esbuild: not found`. Both are the CI bug this all
# started from: a step reporting success about a thing it never looked at. Hence:
#
#   - Every cache-worthy task delivers a file, named individually in `generates`, never a glob.
#   - Every such task also carries a `status:` that probes the filesystem. That is the check a
#     checksum cannot do, and it makes a deleted artifact or a pruned dependency self-heal.
#   - `npm ci`, never `npm install`: the tree is then a pure function of two committed files.
#   - One definition of a thing. CI runs the same `task pack-client` a developer does.
#
# Go's build/test cache is deliberately left alone. It is content-addressed and sound, so a cached
# `ok` means those exact inputs really did pass. `task clean` does not touch it.

run: once

vars:
  CLIENT_DIR: packages/krm-stream
  REPLAY_ADDR: 127.0.0.1:8099
  # A throwaway cluster for proving the gateway against a real API server.
  # (The unit suites need no cluster at all — see `task test`.)
  SPIKE_CLUSTER: krm-stream-spike
  # Kubernetes 1.35+ is a REQUIREMENT of this library, not a preference: from 1.35, resourceVersion
  # orderability is a Certified Kubernetes conformance requirement, and the gateway relies on it to
  # promise consumers per-object monotonicity. So the cluster we verify against must be one.
  K3S_IMAGE: rancher/k3s:v1.36.4-k3s1
  KUBECONFIG_PATH: '{{.HOME}}/.kube/config' 

tasks:
  default:
    desc: "List the available tasks."
    cmds:
      - task --list
    silent: true

  # ---------------------------------------------------------------- fixtures --
  fixtures:
    desc: "Regenerate conformance/gen/ — the JSON both suites read, and the SSE both suites agree on."
    sources:
      - conformance/bodies/*.yaml
      - conformance/fixtures/*.yaml
      - conformance/scopes.yaml
      - conformance/generate.sh
      - gateway/*.go
      # The harness that replays each fixture into the goldens: change how a script is played, and the
      # bytes it records can change with it.
      - gateway/internal/conformance/*.go
      # The generator's own command line lives in this file, so it is a source too: changing WHICH
      # golden tests run, and then being told "up to date", is a trap I walked into writing this line.
      - Taskfile.yml
    generates:
      - conformance/gen/*.json
      - conformance/gen/sse/*.sse
    cmds:
      - bash conformance/generate.sh
      # The SSE goldens: the bytes the gateway REALLY writes, through its real sink. They are what
      # closes the seam — the TypeScript suite parses these exact files back and replays them into a
      # store, so "the client can read what the gateway wrote" is a test and not a hope.
      #
      # They are generated by the gateway rather than by generate.sh precisely BECAUSE the gateway is
      # the thing under test: a transcript hand-written next to the implementation would agree with it
      # by construction and prove nothing.
      # …and gen/protocol.json, written by the Go const that DEFINES the protocol version, so the
      # TypeScript suite can assert it rather than keeping its own hand-copied number in step.
      - cd gateway && go test -count=1 -run 'TestSSEGoldens|TestProtocolVersionIsPublished' -update ./...

  fixtures-check:
    desc: "Fail if conformance/gen/ is stale — the YAML is the source of truth, the JSON is built."
    deps: [fixtures]
    cmds:
      - git diff --exit-code -- conformance/gen || (echo "conformance/gen is stale — run 'task fixtures' and commit" && exit 1)

  # ------------------------------------------------------------------ verify --
  verify:
    desc: "The whole gate, in the order CI runs it. If this passes, CI passes."
    cmds:
      # Sequential on purpose: the first failure is the one you want to read, not the fourth.
      - task: fixtures-check
      - task: lint
      - task: test
      # After the suites, not beside them: it rewrites go.work while it runs.
      - task: release-rehearsal
      - task: e2e-wire
      - task: e2e-browser
      # Last, because nothing else proves it: what we upload contains what the exports map advertises.
      - task: pack-client

  # -------------------------------------------------------------------- test --
  test:
    desc: "Run both suites against the shared conformance fixtures."
    deps: [test-gateway, test-kube, test-client, test-vue, test-compare]

  test-gateway:
    desc: "Go: the gateway + its half of the conformance suite."
    deps: [fixtures]
    dir: gateway
    cmds:
      # -race, and it stays: SharedBackend fans one upstream watch out to N subscribers, so the
      # gateway now has a pump goroutine, a per-scope cache and N bounded queues. A data race in
      # THAT is a browser being shown another tenant's object, not a flaky test.
      - go test -race ./...

  # A SEPARATE MODULE, so `go test ./...` in gateway/ does not reach it — which means that until it
  # was listed here, nothing tested it at all. The Kubernetes adapter is opt-in for adopters; it is
  # not optional for us.
  test-kube:
    desc: "Go: the Kubernetes adapter (no cluster — see test-cluster for the real thing)."
    dir: gateway/kube
    cmds:
      - go test ./...

  release-rehearsal:
    desc: "Build the Go adapter as a release PR would, with a core version that is not tagged yet."
    cmds:
      - node .github/scripts/release-rehearsal.mjs

  test-client:
    desc: "TypeScript: the client + its half of the conformance suite (node --test, no deps)."
    deps: [fixtures, _client-deps]
    dir: "{{.CLIENT_DIR}}"
    cmds:
      - node --test

  test-vue:
    desc: "Typecheck and test the optional Vue example."
    deps: [_vue-deps]
    dir: examples/vue
    cmds:
      - npm test

  test-compare:
    desc: "Typecheck the comparison page and driver, and test their field sync and measurement gates (no cluster)."
    deps: [_client-deps]
    cmds:
      # The page and the driver are the only TypeScript outside the client package that nothing else
      # typechecks: build.mjs and node both strip types without checking them.
      - packages/krm-stream/node_modules/.bin/tsc -p examples/comparison
      - node --test examples/comparison/fields.test.ts examples/comparison/gates.test.ts

  _vue-deps:
    internal: true
    dir: examples/vue
    sources: ["package.json", "package-lock.json"]
    generates: ["node_modules/.package-lock.json"]
    status:
      - test -x node_modules/.bin/tsc
      - test -f node_modules/vue/package.json
    cmds:
      - npm ci --no-audit --no-fund

  # --------------------------------------------------------------------- e2e --
  replay:
    desc: "Serve the conformance corpus over real SSE — a cluster you can point a browser at."
    dir: gateway
    cmds:
      - go run ./cmd/replay --addr {{.REPLAY_ADDR}} {{.CLI_ARGS}}

  e2e-wire:
    desc: "End to end, no cluster: the real Go gateway over a real socket, into the real TS store."
    deps: [fixtures]
    cmds:
      # Two processes, two languages, one contract. Everything else in the suite feeds the store bytes
      # that never left the process — which cannot prove that the server flushes, that the frames
      # survive arbitrary chunk boundaries, or that a terminal error really closes the connection.
      - |
        set -e
        log="$(mktemp -t krm-replay-XXXX.log)"

        # BUILD, then run the binary — do not `go run` it. `go run` execs the real program as a
        # CHILD, so killing the pid we know about leaves the server holding the port, and the next
        # run of this task fails with a confusing "address already in use".
        #
        # And redirect its output to a file rather than inheriting ours: a background process holding
        # our stdout open makes `task` (and any pipe we are in) wait for it forever, which looks
        # exactly like a hung test and is not one.
        bin="$(mktemp -t krm-replay-XXXX)"
        (cd gateway && go build -o "$bin" ./cmd/replay)
        "$bin" --addr {{.REPLAY_ADDR}} --corpus conformance >"$log" 2>&1 &
        pid=$!
        trap 'kill $pid 2>/dev/null || true; rm -f "$bin"' EXIT

        for _ in $(seq 50); do
          curl -sf http://{{.REPLAY_ADDR}}/healthz >/dev/null && break || sleep 0.2
        done
        curl -sf http://{{.REPLAY_ADDR}}/healthz >/dev/null || { echo "replay server never came up:"; cat "$log"; exit 1; }

        cd {{.CLIENT_DIR}} && REPLAY_URL=http://{{.REPLAY_ADDR}} node e2e/wire.ts

  e2e-browser:
    desc: "End to end in a REAL browser: native EventSource, unbundled ESM AND the bundle, no cluster."
    # build-client, because playwright's webServer rebuilds the library from packages/krm-stream —
    # and it cannot do that with no node_modules there. On a developer's machine that directory is
    # already populated by `task test`, which is exactly why this gap survived until the first CI run
    # on a clean checkout found it.
    deps: [fixtures, build-client, build-native-editor, _example-deps]
    dir: examples/vanilla-browser
    cmds:
      # The only place the library's central promise is actually tested: that the published ESM
      # imports in a browser with no bundler. Node importing it proves nothing — Node is not a browser.
      # Playwright lives in the EXAMPLE's package, so the library keeps its four devDependencies.
      # Runs the suite twice, once per entry point (per-module and ./bundle).
      - npx --no-install playwright install chromium --with-deps
      - npx --no-install playwright test {{.CLI_ARGS}}

  demo:
    desc: "Serve the browser demo — watch a status reconcile while you edit spec. No cluster."
    deps: [build-client]
    cmds:
      - |
        echo "→ http://127.0.0.1:8100/?fixture=status-follow-live&pace=800ms"
        echo "  (also: conflict-and-converge, secret-redaction, edit-vs-unrelated-change — /fixtures lists them)"
        cd gateway && go run ./cmd/replay --addr 127.0.0.1:8100 --corpus ../conformance \
          --static ../examples/vanilla-browser --dist ../packages/krm-stream/dist

  # -------------------------------------------------------------------- lint --
  lint:
    desc: "Vet + lint everything: both languages, the workflows, and the Dockerfile."
    deps: [lint-gateway, lint-kube, lint-client, lint-actions, lint-dockerfiles]

  # A workflow YAML and a Dockerfile were the only files here with no linter, and a workflow is where
  # the bundle-that-was-never-built bug lived. actionlint also shellchecks every `run:` block.
  lint-actions:
    desc: "Lint the GitHub Actions workflows (actionlint: expressions, needs/runs-on, shellcheck on every run: block)."
    sources:
      - .github/workflows/*.yml
      - .github/workflows/*.yaml
    cmds:
      # No path argument: actionlint discovers every workflow itself, so a new one is linted the day it
      # lands rather than the day someone remembers to list it.
      - actionlint

  lint-dockerfiles:
    desc: "Lint the devcontainer Dockerfile (hadolint)."
    sources:
      - .devcontainer/Dockerfile
      - .hadolint.yaml
    cmds:
      - hadolint .devcontainer/Dockerfile

  lint-gateway:
    dir: gateway
    cmds:
      - go vet ./...
      - golangci-lint run ./...

  lint-kube:
    dir: gateway/kube
    cmds:
      # --build-tags e2e, or the real-cluster suite is the one Go file in the repo nobody vets.
      - go vet -tags e2e ./...
      - golangci-lint run --build-tags e2e ./...

  lint-client:
    deps: [_client-deps]
    dir: "{{.CLIENT_DIR}}"
    cmds:
      # Two tsconfigs, on purpose. The first is the BUILD (src only — its rootDir is what ships);
      # the second is the only thing that ever typechecks test/, because `node --test` STRIPS types
      # rather than checking them. Without it the conformance suite — the file whose whole job is to
      # be the contract check — would be the one piece of unverified TypeScript in the repo.
      - npx --no-install tsc --noEmit
      - npx --no-install tsc --noEmit -p tsconfig.test.json
      # biome = the TypeScript half of gofmt + go vet + golangci-lint. `task fmt-client` fixes.
      - npx --no-install biome check

  fmt-client:
    desc: "Format and auto-fix the TypeScript (the gofmt of this side)."
    deps: [_client-deps]
    dir: "{{.CLIENT_DIR}}"
    cmds:
      - npx --no-install biome check --write

  # -------------------------------------------------------------------- deps --
  # node_modules is shared mutable state across git branches and nothing in git tracks it.
  # `npm install` reconciles it to the CURRENT branch's package.json, which means it prunes: run it
  # from a branch predating a devDependency and that dependency is gone, leaving a later build to die
  # on `sh: 1: esbuild: not found`. `npm ci` installs exactly the lockfile instead, so the tree cannot
  # drift from the branch you are on. `generates` is npm's own record of what it installed.
  _client-deps:
    internal: true
    dir: "{{.CLIENT_DIR}}"
    sources: ["package.json", "package-lock.json"]
    generates: ["node_modules/.package-lock.json"]
    status:
      # The checksum never looks at the filesystem, so without this an `rm -rf node_modules` is
      # invisible to Task and the next build fails on a missing binary.
      - test -x node_modules/.bin/tsc
      - test -x node_modules/.bin/esbuild
      - test -x node_modules/.bin/biome
    cmds:
      - npm ci --no-audit --no-fund

  _example-deps:
    internal: true
    dir: examples/vanilla-browser
    sources: ["package.json", "package-lock.json"]
    generates: ["node_modules/.package-lock.json"]
    status:
      - test -x node_modules/.bin/playwright
    cmds:
      - npm ci --no-audit --no-fund

  # ------------------------------------------------------------------- build --
  build-client:
    desc: "Emit the dependency-free ESM a browser can <script type=module> import — per-module, and flattened."
    deps: [_client-deps]
    dir: "{{.CLIENT_DIR}}"
    sources: ["src/**/*.ts", "tsconfig.json", "package.json"]
    # Named individually, not as a `dist/**/*.js` glob. A glob is only a checksum key, and Task never
    # asks whether the files exist: `rm dist/krm-stream.js && task build-client` used to say "up to
    # date" and rebuild nothing.
    generates:
      - dist/index.js
      - dist/krm-stream.js
    status:
      # The existence check a checksum cannot do.
      - test -f dist/index.js
      - test -f dist/krm-stream.js
    cmds:
      # Two artifacts, both published: dist/index.js keeps its relative imports (tree-shakeable, for
      # bundlers), and dist/krm-stream.js is the same API flattened into one file, for a host that
      # vendors the library with no bundler.
      #
      # `npm run build`, not `npx tsc`: one definition of the build, shared with package.json and CI.
      # A hand-copied `npx tsc` kept doing half of it once esbuild was added, and CI packed a tarball
      # whose exports map pointed at a file that was not in it.
      - npm run --silent build

  build-native-editor:
    desc: "Build the library and compile the native editor page's controller and recipes into examples/native-editor/dist."
    deps: [build-client]
    sources:
      - examples/native-editor/build.mjs
      - examples/native-editor/editor.ts
      - examples/editor-recipes/keepLocal.ts
      - examples/editor-recipes/recoveryCopy.ts
    generates:
      - examples/native-editor/dist/editor.js
      - examples/native-editor/dist/keepLocal.js
      - examples/native-editor/dist/recoveryCopy.js
    status:
      - test -f examples/native-editor/dist/editor.js
      - test -f examples/native-editor/dist/keepLocal.js
      - test -f examples/native-editor/dist/recoveryCopy.js
    cmds:
      # Types stripped, library import renamed to the published package; see build.mjs.
      - node examples/native-editor/build.mjs

  pack-client:
    desc: "Pack the npm tarball and PROVE it contains every entry point the exports map advertises."
    deps: [build-client]
    vars:
      # CI passes PACK_DIR=/tmp/npm and uploads what lands there; release.yml publishes those bytes
      # without rebuilding.
      PACK_DIR: '{{.PACK_DIR | default "/tmp/krm-stream-pack"}}'
    cmds:
      # The gap between "the build works" and "the thing we upload contains what we said it does".
      # Every other rung builds its own dist and never looks at the tarball, so a missing ./bundle
      # packs green, publishes green, and fails for the first consumer who imports it. CI runs this
      # same task, so the local gate and the release gate cannot drift.
      - |
        set -euo pipefail
        mkdir -p "{{.PACK_DIR}}"
        rm -f "{{.PACK_DIR}}"/*.tgz
        npm pack --pack-destination "{{.PACK_DIR}}" ./{{.CLIENT_DIR}} >/dev/null
        tarball="$(ls "{{.PACK_DIR}}"/*.tgz)"

        # List once into a variable. `tar | grep -q` under pipefail reports failure on a MATCH: grep
        # exits early, tar takes SIGPIPE, and the pipeline inherits tar's status. This guard failed a
        # perfectly good tarball that way on its first run.
        listing="$(tar -tzf "$tarball")"
        for entry in package/dist/index.js package/dist/krm-stream.js; do
          if ! grep -qxF "$entry" <<<"$listing"; then
            echo "✗ ${entry#package/} is in the exports map but NOT in the tarball. Publishing this ships a broken import."
            echo "$listing"
            exit 1
          fi
        done
        echo "✓ every entry point in the exports map is present in $(basename "$tarball")."

  # ------------------------------------------------------------------- clean --
  # The escape hatch for when a cache lies to you. It should be rare: the `status:` guards above exist
  # so the two failures that used to need it (a pruned node_modules, a half-deleted dist/) self-heal.
  clean:
    desc: "Drop every local cache: Task's fingerprints, node_modules, dist. Next run is cold."
    cmds:
      # .task/ is Task's checksum store: what makes a task say "up to date". Remove it when one is wrong.
      - rm -rf .task
      - rm -rf {{.CLIENT_DIR}}/node_modules {{.CLIENT_DIR}}/dist examples/native-editor/dist examples/comparison/dist
      - rm -rf examples/vanilla-browser/node_modules
      # Go's test cache is content-addressed and sound: a cached `ok` means those inputs really did
      # pass. Use `go test -count=1` to watch a suite run; `go clean -cache` buys nothing but minutes.
      - echo "clean. (Go's test cache is deliberately untouched.)"

  # ---------------------------------------------------------------- cluster --
  # The one rung a fake watch cannot reach. Everything the gateway BELIEVES about Kubernetes — that a
  # streaming list ends with an `initial-events-end` bookmark, that a 410 arrives as a watch error,
  # that resourceVersions are orderable decimals — is unverified until something asks a real API
  # server. See CONTRIBUTING.md#test-levels and docs/facts/observed-v1.36.2+k3s1.md.
  cluster-up:
    desc: "A real Kubernetes (k3d, {{.K3S_IMAGE}}) — with real etcd, because that is what we are verifying."
    status:
      - k3d cluster list {{.SPIKE_CLUSTER}} >/dev/null 2>&1
    cmds:
      # --cluster-init gives k3s its EMBEDDED ETCD. Without it k3s stores in kine (SQLite), and then
      # `resourceVersion` and compaction/410 semantics would be kine's rather than etcd's — which are
      # exactly the two things this rung exists to check. Verifying them against a substitute storage
      # layer would be worse than not verifying them at all.
      - k3d cluster create {{.SPIKE_CLUSTER}} --image {{.K3S_IMAGE}} --k3s-arg "--cluster-init@server:0" --wait
      - task: cluster-kubeconfig
      - kubectl --context k3d-{{.SPIKE_CLUSTER}} get --raw /healthz
      - kubectl --context k3d-{{.SPIKE_CLUSTER}} version -o json | jq -r '.serverVersion.gitVersion'

  cluster-kubeconfig:
    desc: "Point the kubeconfig at the cluster in a way that works from INSIDE the devcontainer."
    cmds:
      # Docker-outside-Docker. k3d publishes the API server on the HOST's docker daemon, and writes a
      # kubeconfig saying `https://0.0.0.0:<port>` — which is the host's address, and means nothing in
      # here. From this container the host is `host.docker.internal`.
      #
      # But the API server's certificate has no SAN for that name (it has `localhost`, `127.0.0.1`,
      # `0.0.0.0` and the k3d container IPs), so simply rewriting the host fails TLS verification.
      # `--tls-server-name` is the honest fix: connect to the host, verify the name the cert actually
      # carries. The alternative — turning off certificate verification — would be trading a real check
      # for a convenient one, in a repo whose entire point is not doing that.
      - |
        set -e
        port="$(docker port k3d-{{.SPIKE_CLUSTER}}-serverlb 6443/tcp | head -1 | sed 's/.*://')"
        kubectl config set-cluster k3d-{{.SPIKE_CLUSTER}} \
          --server "https://host.docker.internal:${port}" --tls-server-name localhost >/dev/null
        echo "kubeconfig → https://host.docker.internal:${port} (tls-server-name: localhost)"

  cluster-down:
    desc: "Delete the throwaway cluster."
    cmds:
      - k3d cluster delete {{.SPIKE_CLUSTER}}

  cluster-aggregated-api:
    desc: "Install Kubernetes' own sample-apiserver (wardle Flunders) as a REAL aggregated API."
    deps: [cluster-up]
    cmds:
      # The one upstream whose behaviour Kubernetes' conformance requirements do NOT cover — so the only
      # place OrderingLenient could be needed, and (as it turns out) the only place the gateway's
      # streaming-list assumption actually breaks. See docs/facts/observed-*.md, F6.
      - kubectl apply -f test/cluster/sample-apiserver/sample-apiserver.yaml
      - kubectl -n wardle rollout status deploy/wardle-server --timeout=180s
      - kubectl wait --for=condition=Available apiservice/v1alpha1.wardle.example.com --timeout=120s

  cluster-facts:
    desc: "Ask a REAL API server the questions the docs do not answer. Writes docs/facts/observed-*.md."
    deps: [cluster-up, cluster-aggregated-api]
    dir: gateway/kube
    cmds:
      - go run ./cmd/facts --kubeconfig {{.KUBECONFIG_PATH}} --out ../../docs/facts

  # The backend, against the real thing. The unit suite stubs the API server — and a stub is a thing we
  # wrote, so it agrees with us. This drives the REAL stream loop over the REAL backend against a REAL
  # API server, on BOTH paths: the streaming list against kube-apiserver, and list-then-watch against
  # the aggregated API that refuses it (F6). It does not run in per-PR CI: it needs Docker-in-Docker
  # and it takes minutes, while the fixtures gate every PR in milliseconds.
  # The save path, composed against a real API server: the real gateway, a real host endpoint and the
  # real browser store (run by node).
  test-real-api:
    desc: "Real-API composition: save path (status churn, overlapping recovery, UID race) and native viewing and editing through a host proxy."
    deps: [cluster-up]
    dir: gateway/kube
    env:
      KUBECONFIG: '{{.KUBECONFIG_PATH}}'
    cmds:
      - go test -tags e2e -v -count=1 -run '^TestRealAPI' .

  test-cluster:
    desc: "The kube backend against a real API server on both watch paths."
    deps: [cluster-up, cluster-aggregated-api]
    dir: gateway/kube
    env:
      KUBECONFIG: '{{.KUBECONFIG_PATH}}'
    cmds:
      # -p 1: the shared-host example measures a CLUSTER-WIDE API-server watch gauge, which the
      # backend suite's own real watches corrupt if their packages run concurrently.
      - go test -tags e2e -v -count=1 -p 1 -timeout 10m ./...

  # ------------------------------------------------------------- comparison --
  # One host serves the same Widgets and Secrets through every source — native /k8s, krm-full/v1 and
  # krm-spec/v1, unshared and shared — so a page and a driver can compare them under identical churn.
  # Port 8110, so it never collides with `task demo` or a browser suite on 8100. See
  # examples/comparison/README.md and the dated runs in docs/facts/comparison-*.md.
  build-compare-page:
    desc: "Build the library and compile the comparison page's controller, with the editors and recipe it reuses, into examples/comparison/dist."
    deps: [build-client]
    sources:
      - examples/comparison/build.mjs
      - examples/comparison/page.ts
      - examples/comparison/fields.ts
      - examples/native-editor/editor.ts
      - examples/conditional-save/editor.ts
      - examples/editor-recipes/keepLocal.ts
    generates:
      - examples/comparison/dist/page.js
    status:
      - test -f examples/comparison/dist/page.js
    cmds:
      # Types stripped, library import renamed to the published package; see build.mjs.
      - node examples/comparison/build.mjs

  compare:
    desc: "Serve the comparison page against the spike cluster: native vs krm-full/v1 vs krm-spec/v1, with editing and counters."
    deps: [cluster-up, build-compare-page]
    dir: gateway/kube
    env:
      KUBECONFIG: '{{.KUBECONFIG_PATH}}'
    cmds:
      # The browser address is the TLS one: HTTP/2, so the page's six streams do not use up the six
      # HTTP/1.1 connections a browser allows one origin. The certificate is self-signed at startup.
      - echo "→ https://127.0.0.1:8111/   (also ?entry=bundle, ?variant=shared; plain http://127.0.0.1:8110/ is for drivers)"
      # Foreground, so Ctrl-C reaches the host, which deletes its scratch namespace on the way out.
      - go run ./examples/comparison/cmd/compare --addr 127.0.0.1:8110 {{.CLI_ARGS}}

  compare-browser:
    desc: "The comparison page in Chromium against the real comparison host, on both entry points (needs the spike cluster)."
    deps: [cluster-up, build-compare-page, _example-deps]
    env:
      KUBECONFIG: '{{.KUBECONFIG_PATH}}'
    cmds:
      # BUILD the host and hand Playwright the binary: `go run` would leave its child holding the port
      # and keep the scratch namespace alive when Playwright stops the parent.
      - |
        set -e
        bin="$(mktemp -t krm-compare-XXXX)"
        trap 'rm -f "$bin"' EXIT
        (cd gateway/kube && go build -o "$bin" ./examples/comparison/cmd/compare)
        cd examples/vanilla-browser
        npx --no-install playwright install chromium --with-deps >/dev/null
        COMPARE_BIN="$bin" npx --no-install playwright test -c compare.config.ts {{.CLI_ARGS}}

  compare-measure:
    desc: "Measure every source under every workload against the spike cluster (CLI_ARGS go to measure.ts)."
    # Not cached, and it must not be: it measures a cluster, not files.
    deps: [cluster-up, build-client]
    env:
      KUBECONFIG: '{{.KUBECONFIG_PATH}}'
    cmds:
      - bash examples/comparison/measure.sh {{.CLI_ARGS}}

  compare-native-baseline:
    desc: "Native before/after: measure the native source with the library built at BASELINE_REF (default main), then with this tree's."
    deps: [cluster-up, build-client]
    vars:
      # Name the commit BEFORE the change under test. Once main contains it, main is not a baseline.
      BASELINE_REF: '{{.BASELINE_REF | default "main"}}'
    env:
      KUBECONFIG: '{{.KUBECONFIG_PATH}}'
    cmds:
      - bash examples/comparison/measure.sh --baseline-ref {{.BASELINE_REF}} {{.CLI_ARGS}}