penpot/.serena/memories/frontend/composable-component-tests.md
Andrey Antukh 5c22f5bfb7
Build the frontend bundle once for all E2E suites (#11792)
*  Build the frontend bundle once for all E2E suites

Merge tests-integration, tests-composable-suite and tests-plugin-api-suite
into one "CI: E2E" workflow. Each of the three ran its own full
frontend/scripts/build on every PR, so one PR paid the build three times.

The new build-bundle job restores actions/cache key frontend-bundle-<sha>,
runs frontend/scripts/build only on a miss and saves the key before the
job ends. The integration shards, the composable suite and the mocked
Plugin API suite now all need build-bundle and restore the same key with
fail-on-cache-miss, so none of them builds. A workflow re-run of the same
SHA reuses the cached bundle instead of rebuilding it.

Triggers become the union of the previous paths (frontend, common,
render-wasm, plugins): the bundle embeds the built plugins, so a plugins
change runs the whole set. workflow_dispatch keeps running the
integration job only, as before.

Job names are kept identical on purpose: they are the GitHub check
contexts and branch protection may match them by name.

Docs: new mem:frontend/e2e-ci-workflow records the build-once contract,
referenced from mem:frontend/core and mem:frontend/testing; the composable
memory and both suite READMEs are updated.

AI-assisted-by: deepseek-v4.1-flash

* 🐛 Fix mocked plugin suites crashing without frontend deps

The mocked CI drivers shelled out to frontend/scripts/e2e-server.js,
which imports express from frontend/node_modules. CI jobs install
only plugins/ deps, so the import failed with ERR_MODULE_NOT_FOUND
and the run timed out waiting for localhost:3000.

Serve the prebuilt bundle with a zero-dependency static server
built into each driver (ci/static-server.ts, kept in sync in both
suites) plus node:test coverage for it.

AI-assisted-by: muse-spark-1.3-contributor
2026-09-22 10:23:15 +02:00

10 KiB
Raw Blame History

Composable component tests

A framework concept for systematically testing Penpot's component subsystem (synchronisation/propagation, swaps, variant switches, nesting, overrides), implemented in TWO test suites that share the principles below:

  1. ClojureScript suite — in the frontend test tree (frontend/test/frontend_tests/ composable_tests/), driving a minimally-assembled real app headlessly. The original.
  2. TypeScript suite — a Penpot plugin (plugins/apps/composable-test-suite/), driving the FULL production app end-to-end through the Plugin API, with a slightly more elaborate set of abstractions. Runs interactively (panel), remotely (Playwright), and headlessly in CI. Its README is the authoritative operational reference.

Shared core idea

A test is a composition of operations over a starting configuration, plus assertions. You describe a test as data (a setup + a sequence of operations) rather than writing bespoke imperative code, and coverage grows by COMPOSITION: a new variation is one combinator wrapped around existing pieces, not a copied test. Choice points (one-of alternatives, optional steps) EXPAND the composition into a full sweep of variants — one written case stands for a whole matrix of concrete tests.

Shared principles

  • Every producing object is the accessor interface to what it produces downstream. An operation — and related objects such as content-creation strategies — is not merely an action: the SAME object instance the case holds is the typed interface through which everything it created or changed is later retrieved, checked, and asserted, parameterized by the situation. A foundation operation exposes accessors for the participants it built; an edit operation exposes its dual check (assertHasChangedProperty / has-property-of); a choice is recovered by asking the one-of object (getChoice/get-choice); "did this step run" is asked of the step (wasApplied/applied?). NEVER reach into a situation (or the document) for something an upstream object produced — ask the producer. This is what keeps sweeps sound (object identity ties the question to the exact node that ran) and what keeps retrieval logic in exactly one place. Particularly explicit in the TS OOP implementation, where these accessors are methods on the operation/strategy classes; repeatedly violating it (reading the document directly, duplicating retrieval) was the most common review correction while building the suites.
  • Operations are data with identity. Each operation node has a unique id at construction and records what it did under that id; interrogation is by identity. Bind an operation to a value ONCE and reuse it in the composition and in every query about it.
  • Drive the real production pipeline. Operations route through genuine Penpot logic — real change functions / real workspace events / the real Plugin API, never raw field writes — so the production watcher's AUTOMATIC propagation is what's under test.
  • Roles, not internals. A starting configuration names its participants (roles). Role→id capture happens when the configuration is built; operation TARGETS resolve at apply-time and may be re-bound, so an operation targeting a role follows it as state-building ops re-point it — which lets a single operation be swept across depth.
  • Enumeration is authored, not exhaustive. Compose only VALID cases, so outcomes are just pass / fail / error — no not-applicable cells.
  • Naming discipline. Penpot domain nouns ("component", "variant") must not name framework abstractions; an operation may name the domain ACTION it performs.
  • Operator algebra (same in both suites): sequence (cartesian product of the steps' variants), one-of (union, choice recorded), optional(X) = one-of([X, skip]), inline assertion ops, trailing asserters.
  • Case authoring: a case carries a CamelCase identifier and a plain-terms description in three parts — situation setup, actions/variations, asserted requirement.

ClojureScript suite (frontend test tree)

Test-only .cljs code in the frontend test tree (nothing "common" about it). A situation = the in-memory file value + named roles + :vars + an ordered applied-log. Operations are records implementing IOperation/apply-to (apply collides with core). Assertions = inline Test ops and/or a trailing asserter; the runner makes no judgment. Failures carry describe-applied (the transcript), which is what makes a failing variant in a sweep identifiable.

Layout: core.cljs (the domain-agnostic engine: situation, identity/transcript, roles/targets, operators, runners), comp/setups.cljs (setups + role accessors), comp/nodes.cljs (the component operations and their check duals), interpreter.cljs (runs cases against the real frontend), comp/sync_test.cljs (the cases; registered in frontend_tests/runner.cljs). Case letters B..N; the sweeps (K: depth × edit-precedence; L: swaps; M: variant switches; N: rotated-instance geometry, on the #10109 fix branch until merged) are the flagship pattern — read them before writing a new sweep.

Scenario lineage model (behind the sweeps): scenario ops track named component lineages as objects under :vars, each holding the FIXED deepest origin (:remote-*), the ADVANCING outer main (:main-*), and per-nesting-level data whose :nested-head (the deepest instance at that level, found by descending the :shape-ref chain — matching chain MEMBERSHIP, not terminus) is the swap/switch target, anchored by its swap-stable parent. Nesting seeks the FIXED origin, not the advancing main — that is what makes each level's :nested-head land on the deepest instance. A variant nesting re-points the lineage's remote to the chosen member. Construction lesson: cross-level propagation requires progressively NESTED levels (one variant + plain wraps); sibling nestings do not propagate between each other.

Interpreter: installs the situation's files into the global st/state (aux files tagged :library-of), starts the real watch-component-changes (+ harness watch-undo-stack), maps event-ops to REAL workspace events (dwsh/update-shapes, dwl/component-swap, dwv/variants-switch, dwt/increase-rotation — which runs the check-delta placement classification — dwt/update-dimensions, dwu/undo, dwl/sync-file, …) and runs sync-ops' apply-to against the live store file; awaits settlement (idle-gap heuristic + per-op grace) and re-reads :file each step so the shared accessors keep working. STORE-SWAP IMMUNITY: other test namespaces set! st/state/st/stream and never restore, while the app.main.refs lenses stay bound to the ORIGINAL atoms — propagation then dies silently. The interpreter captures the atoms at namespace-load time and re-set!s them per variant.

Running: cd frontend && pnpm run build:test, then node target/tests/test.js --focus frontend-tests.composable-tests.comp.sync-test (var-level focus for one case).

Fidelity warning: the harness drives a MINIMALLY-ASSEMBLED app — only some initialize-workspace subscriptions are wired. Risk = SILENT UNDER-WIRING (e.g. undo needs the harness watch-undo-stack). When a case needs app behaviour beyond a raw edit, check for an unwired subscription and verify by PROBING store state, not by trusting a green assertion.

Caveats: inline Test exceptions are UNCAUGHT on the frontend (crash the runner — assert in the trailing asserter). (optional (in-sequence …)) is not flattened for the interpreter — use independent optionals. The Serena/clj-kondo cache for nodes.cljs goes stale (phantom symbols) — trust the build. Cross-namespace global-state leaks land in this suite first; suspect them before the framework on inexplicable full-run-only failures. Case H's sync-file schedules a delayed RPC that fails headless (benign; absorbed by per-op grace).


TypeScript suite (the plugin) — full e2e

plugins/apps/composable-test-suite/ — same principles against the FULL production app through the Plugin API (real frontend, real propagation). Continuation of the CLJS suite per issue #10584. Operational details (build/run, connect URL, remote control, reading logs, auto-reload, CI): the plugin README.

Distinguishing abstractions (the OOP articulation of the shared principles):

  • TestCase {identifier, description, operation} with the three-part description mandated in the constructor docstring.
  • The accessor-interface principle is class-level: foundation operations (e.g. OpCreateSimpleComponentWithCopy) expose the roles they build; content-creation strategies (pluggable: what content a foundation builds around) expose accessors for the content they created; edit operations expose their checks (OpChangeProperty.assertHasChangedProperty); OpOneOf/OpOptional are queried for what ran. Tests never grope the document for something a producer can be asked for.
  • ShapeProp model: property duals with numeric tolerance; rotation is a writable attr, height goes via resize (readonly in the Plugin API).
  • TestSuite enumerates cases into a TestTree with stable per-test ids; run(ids, TestRunObserver) is the ONLY output channel — the framework is UI-free by construction. plugin.ts (panel adapter), main.ts (panel UI) and src/ci/headless.ts (CI adapter) are three thin consumers.
  • Cases live in src/composable-tests/cases/ as case<Identifier>.ts (e.g. MainEditSyncs — the sweep that found #10109).
  • Panel checkboxes carry stable DOM ids (case identifier / Identifier-N composites) for remote control via Playwright; recipe in the README.

CI

Headless per-PR gate: the composable-test-suite job in .github/workflows/tests-e2e.yml runs pnpm --filter composable-test-suite run test:ci — mocked backend (frontend e2e static server + Playwright RPC fixtures, no backend/login), the in-sandbox bundle injected via ɵloadPlugin, results streamed via console markers, TEST_FILTER by identifier substring. The mocked backend is NOT a limitation for this suite (everything asserted is frontend store logic; empirically confirmed against the interactive runs). Architecture mirrors plugin-api-test-suite's CI driver; the mock harness exists in THREE places that must stay in sync (provenance note in ci/run-ci.ts). The frontend bundle is built once per SHA and restored by this job (mem:frontend/e2e-ci-workflow). Details: README, "Running in CI".

Substrate

mem:common/test-setup, mem:common/component-data-model, mem:common/component-swap-pipeline, mem:frontend/testing.