Compatibility

Plugin compatibility contract

Goal: any app folder from splitflap-os drops into this project's apps/ and works unmodified. That inbound direction is the contract this document holds the companion to: its engine and UI are original, but the plugin ABI is a faithful, behavior-identical port.

The reverse direction is not promised. An app authored here loads on a stock splitflap-os — every companion helper is opt-in by signature, so nothing breaks — but it runs without them (no get_weather, no i18n, no caps), unoptimized, and we make no guarantees of it.

Pinned upstream: splitflap-os commit 12df2773cbbe9890a7d6f92fdc60d2be920129bd (VERSION 0.3.0).

What must stay identical

• Layout & loading: apps/<id>/{manifest.json, app.py} (+ bundled data files). app.py is loaded as a single standalone module — the app's directory is not put on sys.path, so relative or sibling-file imports do not resolve; keep the code in one file and read bundled data relative to __file__. Both functional (has app.py) and channel (data-only) app types load.

• fetch(settings, format_lines, get_rows, get_cols) -> list[str] — same argument order and semantics. format_lines(*lines, cols=None), get_rows(), get_cols() are behaviour copies (padding/centering/width; each page is rows × cols chars), with one deliberate divergence:

  Vertical centering. When an app returns fewer lines than the wall is tall, splitflap-os pads only at the bottom; we pad above and below. On a 3-row wall — what splitflap-os targets — the two are identical, because apps fill it. On a taller wall (a 5×15 MatrixPortal, say) bottom-padding strands a 3-line app at the top with two dead rows beneath it. An app is unaffected either way: it still returns the same lines, and get_rows() still tells it how much room it has.

  Case is the display's business, not the app's. The vendored apps no longer call .upper() on the text they show. normalize() folds a non-raw page for a wall that cannot show lowercase — which every split-flap is — so the flaps are unchanged; a Matrix Portal (whose index-addressed API can show lowercase) gets the text as written. A splitflap-os app that still uppercases its own strings is unaffected: it is simply folding text that was already folded.

  The divergence is opt-out. An app that wants the original behaviour — or that builds its own layout and needs its rows left where it put them — declares it in its manifest:

  { "vertical_align": "top" }      // "center" (default) | "top" | "bottom"

  "top" is byte-for-byte splitflap-os padding. The key is additive: absent means "center", so every existing app — and every unmodified splitflap-os app — behaves exactly as it does today without being touched. format_lines's signature does not change: the runtime hands each app a format_lines already bound to that app's alignment, so apps keep calling format_lines(*lines).

  Declaring "top" is what an app needs if it places its own blank rows. Without it, an app that centres its own block gets centred a second time and drifts below the middle — which is exactly what happened to three vendored apps in 1.9.0-beta.5.

• settings dict — manifest defaults merged with saved user values, same key resolution, including any global keys apps read (e.g. currency_symbol).

• Optional caps argument — like i18n and get_location, injected by parameter name and only when the app asks for it. It reports what the wall can show (lowercase, pictographs, named_colours, instant), so an app can offer ↑ where the wall has a real arrow and HIGH where it does not. Additive and defaulted to None, so an app that uses it still runs on stock splitflap-os — where None correctly means "a plain reel".

• Triggers — trigger(settings, conditions) -> bool on a trigger_interval, with trigger_display_seconds / trigger_cooldown / trigger_conditions, and the setattr(fetch, '_state', …) state pattern.

• Manifest schema / settings fields — text, number, password, datetime-local, textarea, select, toggle, search_chips, computed, inline_toggle, plus visible_when, sync_values / sync_parent, stepper, string/object options, and the LUCIDE_APP_ICONS opt-in.

• search_chips searchUrl endpoints — served at the same paths with the same response shapes: /location_search, /location_timezone, /timezones, /stocks_search, /crypto_search, /sports_search (leagues and teams in one call — the endpoint the vendored sports app declares).

• Rendering — the emoji→colour-code map and all transition orderings (apps emit colour codes and expect the same animation styles).

• Caching / paging — results cached per refresh_interval; each page shown loop_delay seconds.

Where the companion intentionally differs

Final character normalization is companion-specific and deliberately not a port of splitflap-os. Because the gateway and modules support the full Windows-1252 set, the companion sends accented letters, € and punctuation through verbatim (upper-cased in a cp1252-aware way, so ß and accents survive) instead of policing them against a fixed FLAP_CHARS set or substituting the currency $/"→q. This changes only the final glyphs on the wire, never what an app sees — the inbound ABI above is untouched, so splitflap-os apps stay drop-in compatible here.

How it's enforced

• backend/tests/test_renderer.py guards normalization (cp1252-aware upper-casing + verbatim Windows-1252 passthrough) and the animation orderings (a drift here is a compatibility regression).
• backend/tests/test_plugins.py::test_every_app_loads imports every apps/* and asserts its manifest + fetch/data satisfy this contract. The manual drop-in check still holds: copy a not-pre-vendored splitflap-os app into apps/ and confirm it appears, configures, and runs with no changes.

When upstream changes

If splitflap-os adds a field type, helper, or endpoint, bump the pinned commit above, port the addition, and extend the conformance test. Never diverge the ABI silently.