Plan WP-0017: whynot-design adoption (tokens + CSS only)

Add cross-framework analysis to history/ and WP-0017 workplan for the tokens+CSS phase of adopting ~/whynot-design. Component port deferred until upstream ships Lit web components and missing atoms (Card, Modal, Input, Table, Toast). Decisions captured: vendor (not npm), big-bang swap (not pilot), keep off-spec red for btn-danger until upstream defines one.
2026-05-23 19:09:18 +02:00
parent 739ffafedd
commit fbb8def9ce
2 changed files with 759 additions and 0 deletions
--- a/history/2026-05-23-whynot-design-cross-framework-analysis.md
+++ b/history/2026-05-23-whynot-design-cross-framework-analysis.md
@@ -0,0 +1,361 @@
+---
+date: 2026-05-23
+topic: whynot-design adoption — cross-framework analysis
+status: research / decision pending
+author: claude (opus-4.7)
+related:
+  - ~/whynot-design (the React DS in question)
+  - ~/whynot-control (the org control surface — explicitly out of scope)
+  - vergabe-teilnahme (the consuming Django app)
+follow-ups:
+  - workplan: VERGABE_TEILNAHME-WP-0017 (whynot-design adoption) — not yet drafted, depends on strategy decision below
+  - possible new repo: whynot-design-django — see §5
+---
+
+# whynot-design — cross-framework adoption analysis
+
+> Persisted from a research conversation on 2026-05-23. The conversation was triggered
+> by the question "should vergabe-teilnahme adopt the whynot design system, and how?"
+> The strategic question opened up because whynot-design ships React components
+> while vergabe-teilnahme is Django + HTMX + Tailwind.
+
+## 1. The headline insight
+
+There is **no community** around "React → Django component porting." That pattern
+doesn't exist as a tool, library, or codified practice. The reason: cross-framework
+UI sharing has, since ~2020, converged on **Web Components** as the standard answer.
+
+Every major design system that supports multiple frameworks (Shoelace / Web Awesome,
+IBM Carbon, Adobe Spectrum, Salesforce Lightning, Material Web) either ships Web
+Components or maintains parallel hand-rolled implementations per framework. There is
+no third option that the industry has validated.
+
+The W3C Design Tokens Community Group spec reached its **first stable version in
+October 2025**, with Style Dictionary, Figma, Penpot, Tokens Studio, and others as
+reference implementations. Tokens as a cross-framework contract are now genuinely
+portable; *components* still aren't, except via web components.
+
+The strategic question therefore reshapes itself: **do we accept the
+parallel-implementations cost, or do we change what `whynot-design` actually ships?**
+
+## 2. Research findings — existing approaches and their communities
+
+| Approach | What it is | Community / activity | Fit for whynot |
+|---|---|---|---|
+| **Web Components** (Lit, Stencil) | Browser-standard custom elements; works in any framework | Very active; React 19 finally scores perfect on custom-elements-everywhere.com | **High** — best general answer if we're willing to refactor whynot-design |
+| **Shoelace / Web Awesome** | Pre-built Web Components UI kit (Shoelace's successor) | Large, active OSS community | Not direct (competing DS), but proves the model. Has explicit HTMX integration guides |
+| **Stencil JS** | Compiler that emits framework-specific bindings from one source | Mature; IBM/Apple use it; declining mindshare vs. Lit | **Medium** — heavier than Lit but generates per-framework wrappers automatically |
+| **Style Dictionary + DTCG tokens** | One token file → CSS / Tailwind / iOS / Android / Flutter | W3C-backed, stable spec since 2025-10 | **High** — should be our token pipeline regardless of component choice |
+| **Carbon / Spectrum / Lightning / Material pattern** | Single design source, hand-maintained parallel packages per framework | Active but they're 100+ engineer teams | **Low** at our scale — the "expensive but principled" reference path |
+| **Single-spa / micro-frontends** | Multiple frameworks coexisting in one app | Niche, mostly enterprise integration | **Off-topic** — not about sharing components, about hosting heterogeneous apps |
+| **django-cotton / django-components / django-bird** | Django-template-native component libraries; HTMX-friendly | Active, growing in 2024–2026; mature enough to depend on | **High** as the *Django-side runtime* — but doesn't solve cross-framework, just gives Django a real component model |
+| **"Port React to Django" libraries** | (doesn't exist) | None | None |
+
+**Read of the field:** the world has settled on "tokens are cross-framework;
+components are framework-native." Most teams either pick web components (one
+implementation, runs anywhere with some friction) or accept parallel implementations
+from a shared design language.
+
+## 3. Three viable strategies for whynot's situation
+
+**Strategy A — Pivot whynot-design to Web Components.**
+Rewrite `Atoms.jsx` + `Chrome.jsx` as Lit components. Ship one runtime artefact
+(`@whynot/design`) that Django + HTMX templates can drop into a page via plain
+`<whynot-button>` tags. No React dependency anywhere. Tailwind / CSS variables
+continue to drive theming.
+
+**Strategy B — Keep whynot-design React-canonical; add a parallel `whynot-design-django` repo.**
+Treat the React JSX as the *visual + API specification*. Hand-port to Django
+partials in a separate, reusable repo so vergabe-teilnahme and future Django
+consumers share one implementation. Tokens stay in whynot-design (single source).
+
+**Strategy C — Tokens only; components are framework-local forever.**
+Don't promise component parity. Vergabe re-implements whynot's *look* using the
+tokens but its components live and die in vergabe-teilnahme. Other Django apps
+would do the same fork. Bernd implied this is unsatisfying — he wants a real
+component library for cross-project consistency.
+
+The real choice is **A vs. B**.
+
+## 4. Pros / cons — A (Web Components) vs. B (parallel Django repo)
+
+### Strategy A — Pivot to Web Components (Lit)
+
+**Pros**
+- *One* implementation. Zero divergence risk. Bug fixes ship once.
+- Standards-backed; outlives any framework choice. Works in vanilla HTML, Django
+  templates, future Vue/Svelte/Solid apps, marketing sites, claude.ai artifacts.
+- HTMX swaps work cleanly — web components are just DOM; HTMX doesn't care.
+- The community is *here*. Tutorials, debugging tools, Storybook integration,
+  accessibility testing infrastructure all exist.
+- The Claude atelier in claude.ai can still output JSX prototypes; Lit ports are
+  mechanical once the design is decided.
+
+**Cons**
+- One-time rewrite cost of the existing 11 React components. Real, but small
+  (~267 lines of JSX, not a library of 200 components).
+- Web components have Shadow DOM trade-offs: form-association, deep style scoping,
+  and slotting have learning curves. Workable but real friction.
+- React-shaped APIs (`<Button variant="primary">`) translate to attributes /
+  properties (`<wn-button variant="primary">`). Slightly less ergonomic in React,
+  equally ergonomic everywhere else.
+- Lose the claude.ai → npm symmetry: claude.ai design tool outputs JSX; you'd
+  convert each new component.
+- Whoever maintains whynot-design has to learn Lit (small surface area, but new
+  vocabulary).
+
+### Strategy B — Parallel `whynot-design-django` repo
+
+**Pros**
+- Each implementation is idiomatic for its stack. Django templates feel like
+  Django; React feels like React.
+- No new technology to learn. Both repos use mature, well-understood patterns.
+- Django consumers get to use `django-cotton` or `django-bird` — real ergonomics,
+  not a half-port.
+- claude.ai → JSX → Lit conversion step is avoided. The atelier output is
+  *directly* the React reference.
+
+**Cons**
+- *Two* implementations forever. Every component change needs both sides updated;
+  PRs need cross-repo review discipline.
+- API drift is inevitable. By v0.5 the React `<Button>` and Django `{% button %}`
+  will have diverged in props/slots/behavior unless enforced with conformance tests.
+- A *third* consumer stack (Vue, vanilla HTML, native mobile) means a *third* repo.
+  Cost scales linearly with the number of stacks; A's cost is roughly constant.
+- The "community" of "people maintaining a React→Django twin DS" is exactly Bernd
+  and whoever he recruits. No prior art to lean on.
+- Visual regression testing has to run twice (Playwright against examples in both
+  repos) to ensure parity.
+
+### Recommendation
+
+**Strategy A is the right long-term answer**; B is the right *interim* answer if
+we don't want to halt vergabe-teilnahme to do a Lit rewrite first.
+
+A real third path: **A scheduled, B today.** Build B as a reusable Django port now
+to unblock vergabe-teilnahme; treat it as the *bridge* until whynot-design pivots
+to Lit at, say, v0.5. The Django repo becomes obsolete when A lands — that's a
+feature, not a bug; it forces a clear retirement decision instead of accumulating
+tech debt.
+
+If we don't want to commit to a future Lit pivot, then B is the right choice on its
+own — just accept the linear-cost-per-stack reality.
+
+## 5. Naming, establishing, workflow — if we go with B
+
+### Naming proposals
+
+| Candidate | Pros | Cons |
+|---|---|---|
+| **`whynot-design-django`** (recommended) | Mirrors `@whynot/design`. Unambiguous about stack. Discoverable. | Slightly long. |
+| `whynot-django` | Short. | Ambiguous — sounds like a Django app, not a DS port. |
+| `whynot-design-py` | Hedges for Jinja2/Flask later. | Vague — Python doesn't pick a template engine. |
+| `whynot-dj` | Compact. | Cryptic. |
+| `whynot-templates` | Honest about what it is. | Doesn't carry the "design" meaning. |
+
+**Recommend `whynot-design-django`.** If a Jinja port appears, it gets its own repo
+(`whynot-design-jinja`), not a confusing rename.
+
+### Establishing the repo (concrete steps)
+
+1. **Gitea org placement:** `gitea-remote:whynot/whynot-design-django.git` — same
+   org as `whynot-design`, signaling sibling status.
+2. **Package shape:** a pip-installable Django app (`pyproject.toml`, importable as
+   `whynot_design`). Two install paths:
+   - `pip install git+ssh://...@v0.1.0` (matches whynot-design's tag-pinned discipline).
+   - For dev: editable install (`pip install -e ../whynot-design-django`).
+3. **Layout** (matches Django conventions; mirrors whynot-design's structure):
+   ```
+   whynot-design-django/
+   ├── README.md                 ← restates the "match the React spec" contract
+   ├── CONTRIBUTING.md           ← parity rules: any change here must follow a whynot-design change
+   ├── CHANGELOG.md
+   ├── pyproject.toml
+   ├── whynot_design/
+   │   ├── __init__.py
+   │   ├── apps.py
+   │   ├── templates/whynot_design/
+   │   │   ├── atoms/eyebrow.html, tag.html, button.html, stage_dot.html, stamp.html, icon.html
+   │   │   ├── chrome/top_nav.html, sidebar.html, page_header.html, pipeline_strip.html
+   │   │   └── _base.html        ← imports the CSS once
+   │   ├── templatetags/
+   │   │   └── whynot.py         ← {% wn_button variant="primary" %}…{% endwn_button %}, etc.
+   │   └── static/whynot_design/
+   │       └── colors_and_type.css  ← vendored from @whynot/design (synced)
+   ├── tests/
+   │   ├── test_components.py    ← Django test client, asserts rendered markup matches expected shape
+   │   └── parity/               ← Playwright comparison of Django render vs. React render
+   └── examples/                 ← a Django demo project rendering every component
+   ```
+4. **Token sync:** a `scripts/sync-from-whynot-design.py` that copies
+   `colors_and_type.css` and `tokens/*.json` from a pinned whynot-design ref into
+   the static dir + generates a Django settings constant for tokens. Run on every
+   whynot-design version bump.
+5. **Underlying component mechanism:** plain `{% include %}` partials (simplest, no
+   new dep), `django-cotton` (HTML-like syntax, very ergonomic, active community),
+   or `django-components` (most powerful, heaviest). My lean: **`django-cotton`** —
+   its `<c-button variant="primary">…</c-button>` syntax parallels React JSX best,
+   so the parity contract is more obvious.
+
+### What changes in the workflow
+
+Before:
+```
+claude.ai atelier ──► whynot-design ──► consumer
+```
+After:
+```
+claude.ai atelier ──► whynot-design ──┬─► whynot-design-django ──► vergabe-teilnahme (+ future Django apps)
+                                      │
+                                      ├─► future stacks: whynot-design-jinja / -svelte / -vue / …
+                                      │
+                                      └─► tokens flow into all child repos via a sync script
+```
+
+Concretely, this adds three new ongoing obligations:
+
+1. **Token sync gate.** When whynot-design ships a token change, the Django repo's
+   sync script must run before its next release. CI check: the static CSS in
+   `whynot-design-django` must hash-match the CSS at the pinned upstream tag.
+2. **Component parity gate.** When whynot-design adds or modifies a component, an
+   issue is opened automatically in `whynot-design-django` referencing the upstream
+   PR. A release is blocked until the issue closes. Encode in `CONTRIBUTING.md`
+   first; later, a CI cross-repo check.
+3. **Conformance tests.** Each component's Django partial has a Playwright snapshot
+   that's compared against the same component rendered in whynot-design's
+   `examples/whynot-control/`. Drift = failing build.
+
+Vergabe-teilnahme's workplan therefore depends on **whynot-design-django** existing
+first. The vergabe workplan becomes: install the new package, replace partials with
+cotton/include tags, retheme.
+
+## 6. Component-level inventory — whynot-design vs. vergabe-teilnahme
+
+| whynot-design (React) | vergabe-teilnahme (Django) | Match | Gap action |
+|---|---|---|---|
+| `Eyebrow` (mono uppercase label, fg-3) | — | Missing | New partial `atoms/eyebrow.html` |
+| `Tag` (mono uppercase pill: default / active / draft) | `status_badge.html` (different semantics — domain statuses) | Partial | Replace `status_badge` content with `Tag` variants; keep status→variant mapping |
+| `Button` (3 variants: primary / secondary / ghost; lucide icon support) | CSS classes `.btn-primary`, `.btn-secondary`, `.btn-danger`, `.btn-ghost` | Strong overlap; `btn-danger` not upstream | Adopt 3 base variants; propose `danger` upstream OR retire it (whynot's voice avoids red/destructive emphasis) |
+| `StageDot` (S0–S4 signal levels with grey-ramp + yellow S4) | `phase_nav.html` (numbered phase circles: todo/active/done/warn) | Semantically different — *whynot stages* ≠ *vergabe phases* | Keep both as distinct atoms. Propose `PhaseDot` upstream OR keep vergabe-local |
+| `Stamp` (yellow rotated "DRAFT") | — | Missing | New partial; useful for unfreigegeben items |
+| `Icon` (lucide via `data-lucide`) | inline SVGs / class-based | Missing as a component | New partial that wraps lucide; pull lucide JS into base.html |
+| `TopNav` | `topbar.html` | Both exist; styling and density differ significantly | Reskin vergabe's topbar to whynot's structure (search box + primary action right-aligned) |
+| `Sidebar` | `sidebar.html` | Both exist; nav model differs (vergabe has phase grouping; whynot has Work / Control docs grouping) | Adopt whynot's chrome (item style, eyebrow section headers) but keep vergabe's nav items |
+| `PageHeader` (eyebrow + h1 + lede + actions) | — (inline per template) | Missing | New partial; refactor all page templates to use it |
+| `PipelineStrip` (5-stage horizontal indicator) | `phase_nav.html` (vertical / different) | Semantically related, visually different | Decide: align on whynot's strip OR propose `PipelineStrip` variant upstream |
+| — | `field_row.html` (vergabe-specific) | n/a | Propose upstream — it's a generic key/value display row |
+| — | `breadcrumb.html` | n/a | Propose upstream as `Breadcrumb` |
+| — | `feedback_button.html`, `feedback_modal.html`, `feedback_success.html` | n/a (vergabe-specific UX flow) | Keep vergabe-local; product UI, not DS atoms |
+| — | `freigabe_modal.html`, `freigabe_success.html` | n/a (vergabe-specific) | Keep vergabe-local |
+| — | `search_results.html` (HTMX results target) | n/a | Probably vergabe-local, but a `Listbox`/`Combobox` atom could live upstream — defer |
+
+### Gap summary
+
+- **5 net-new atoms** in vergabe (Eyebrow, Stamp, PageHeader, Icon wrapper, PipelineStrip).
+- **4 atoms to align** (Tag ↔ status_badge, Button variants, TopNav, Sidebar).
+- **2 atoms to propose upstream** (Breadcrumb, FieldRow).
+- **1 semantic mismatch to resolve** (Stage vs Phase — they're different concepts).
+- **All vergabe-specific flow UI** (feedback, freigabe, search_results) stays
+  vergabe-local — that's correct, those are product UI, not design system.
+
+### Components missing from whynot-design — full gap list
+
+These are components that are absent from `whynot-design` today and would need to
+land there (or in `whynot-design-django`) before vergabe-teilnahme can fully adopt
+the system. Grouped by origin.
+
+#### Already exists in vergabe-teilnahme — candidates to promote upstream
+
+1. **`Breadcrumb`** — vergabe's `breadcrumb.html`. Trivially generic; no
+   vergabe-specific semantics.
+2. **`FieldRow`** — vergabe's `field_row.html` (label + value, 3-column grid). The
+   pattern recurs across detail pages; clearly a DS atom.
+3. **`PhaseDot` / `PhaseNav`** *(name TBD)* — vergabe's `phase_nav.html`.
+   Semantically distinct from whynot's `StageDot` (numbered phases with
+   todo / active / done / warn states, not S0–S4 signal levels). Either upstream
+   as a sibling component, or kept vergabe-local. **Open decision.**
+
+#### Doesn't exist anywhere yet, but the DS clearly needs them
+
+These are components that vergabe-teilnahme has as raw Tailwind classes or ad-hoc
+markup, and whynot-design has no equivalent — but any non-trivial app needs them,
+so they're real DS gaps:
+
+4. **`Card`** — whynot's house rules reference cards ("no shadows on cards,"
+   "0–4px radii for cards/sheets") but there is no `Card` JSX component. Vergabe
+   has `.card` as a Tailwind class. **Surprising omission given the explicit
+   design rules.**
+5. **`Input` / `Textarea` / `Select`** — whynot ships no form primitives at all.
+   Vergabe has `.form-input` and `.form-label` CSS classes. Required before any
+   form-based whynot artefact ships.
+6. **`Modal` / `Dialog`** — whynot ships none. Vergabe has two (`freigabe_modal`,
+   `feedback_modal`). Whynot's house rules even prescribe modal radius (8px),
+   confirming the intent — just the component is missing.
+7. **`Table`** — whynot ships none. Vergabe has `.table-base`, `.table-header`,
+   `.table-row` classes. Any data-heavy view (likely most of vergabe) needs it.
+8. **`Toast` / inline success banner** *(name TBD)* — whynot ships none. Vergabe
+   has `freigabe_success.html` and `feedback_success.html` as ad-hoc partials.
+   Even with whynot's quiet voice, success / error states need a defined visual.
+
+#### Worth flagging but defer until first real need
+
+9. **`SearchInput`** — currently inlined inside whynot's `TopNav`. Pulling it out
+   as a standalone atom would be useful (vergabe's `search_results.html` is a
+   target for an HTMX-driven search box).
+10. **`EmptyState`** — neither codebase has one; vergabe will need it for empty
+    list views. Worth designing once before either side fills the gap locally.
+11. **`Tabs`, `Dropdown`, `Tooltip`, `Pagination`, `Avatar`** — common DS atoms;
+    neither side has them. Don't add speculatively; add when the first concrete
+    use case appears.
+
+**Count:** 3 vergabe-existing to promote, 5 genuine DS-level gaps, 5 deferred.
+The 5 "genuine DS-level gaps" — especially `Card`, `Input`, `Modal`, `Table` — are
+**blockers for adoption**: vergabe can't replace its current chrome without them
+existing somewhere shared.
+
+### Stylistic gaps to address before "done"
+
+- Vergabe currently uses Tailwind utility classes pervasively
+  (`.btn-primary { @apply … }`); whynot uses inline-style CSS-variable assignment.
+  The Django port should commit to one — recommend CSS variable + utility classes
+  (re-derive whynot's component styles as Tailwind `@apply` blocks that read from
+  CSS vars), so HTMX-injected fragments don't need their own style scaffolding.
+- Lucide icons aren't wired into vergabe. Need a base.html script tag + an icon partial.
+- Font stack mismatch: vergabe is `ui-sans-serif`; whynot is IBM Plex Sans / Mono /
+  Serif. Importing whynot's CSS handles this — but it pulls Google Fonts at runtime.
+  Decide self-host vs. CDN.
+- Vergabe's brand blue (`#3b5bdb`) has to go entirely. No mid-state — the whynot
+  aesthetic forbids it.
+
+## 7. Recommendation and open decision
+
+**Recommendation:** Strategy B (parallel `whynot-design-django` repo) as the
+immediate move, with the README explicitly framing it as a **bridge** that becomes
+obsolete if whynot-design eventually pivots to Lit web components. Concretely this
+means **three workplans**, not one:
+
+1. **`whynot-design`**: small additions — tag v0.1.0, add upstream candidates
+   (`Breadcrumb`, `FieldRow`), document the parity contract for downstream Django port.
+2. **`whynot-design-django`** (new repo): bootstrap, ship v0.1.0 with the 11
+   components + 2 vergabe-contributed atoms, set up token-sync + parity tests.
+3. **`vergabe-teilnahme`** (`WP-0017`): consume `whynot-design-django`, retheme,
+   pilot on one page first, then full sweep.
+
+**Open decision (blocks workplan drafting):**
+
+Strategy A (pivot to Lit immediately) vs. Strategy B (parallel Django repo) vs.
+"B now, A later" — Bernd to choose. If "B now, A later," add a workplan stub for
+the Lit pivot in `whynot-design` itself so it doesn't get forgotten. If "A from the
+start," the Django repo doesn't exist and the workplans collapse to two.
+
+## Sources
+
+- [Shoelace / Web Awesome](https://shoelace.style/) — framework-agnostic web components DS
+- [Lit + Web Components for cross-framework UIs](https://thenewstack.io/how-to-build-framework-agnostic-uis-with-web-components/)
+- [HTMX + Shoelace example](https://binaryigor.com/htmx-with-shoelace-framework-agnostic-components-in-an-example-app.html)
+- [W3C Design Tokens spec first stable version (Oct 2025)](https://www.w3.org/community/design-tokens/2025/10/28/design-tokens-specification-reaches-first-stable-version/)
+- [Style Dictionary cross-platform tokens](https://styledictionary.com/info/tokens/)
+- [Django Cotton — HTML-like component syntax](https://django-cotton.com/)
+- [django-bird](https://github.com/joshuadavidthomas/django-bird)
+- [Salesforce — Beyond Components: multi-framework DS](https://medium.com/salesforce-ux/beyond-components-a-design-system-to-support-multiple-frameworks-cb1e4d511f66)
+- [AgnosticUI post-mortem (rewrite to Lit)](https://frontendmasters.com/blog/post-mortem-rewriting-agnosticui-with-lit-web-components/)