Page carousel navigation with long-press jumps (#60) #85

Merged

robert merged 1 commit from area/page-carousel into main

2026-09-04 09:48:04 +02:00

robert commented

2026-09-04 09:47:10 +02:00

Owner

Closes #60.

Replaces page_view.c's issue #8 placeholder with the real button model from docs/DESIGN.md section 5, wired on top of #58's page descriptors (PR #81) and #11's ride state machine (PR #84).

What changed

Up/Down step to the previous/next visible page, wrapping at both ends. "Visible" skips a page with no live field (page_has_live_field(), #58's primitive built for exactly this) — unless none of the three default pages has data yet (cold boot, before the phone has sent anything), in which case every page counts as visible rather than leaving a blank ring. Only the landing page is ever rendered — a run of skipped pages never flickers past on screen.
Long Up jumps straight to page 1, deliberately bypassing the skip rule (an explicit rider request, not a ring step — matches DESIGN.md's "manual navigation always wins" precedent for Nav auto-switching).
Long Down is supposed to jump to the map. There is no map page or view anywhere in this codebase yet, so this is a documented no-op: an APP_LOG line and a comment at the call site, not a fake fallback to page 1 (which would make long Up and long Down indistinguishable) and not a silent, unexplained nothing.
Select starts/pauses (ride_link_start_pause()); long Select begins a stop confirmation (ride_link_stop_request()); Back leaves the app, confirmed first only if a ride is running or paused. Both confirmations share one on-screen prompt and the same Select/Back roles reversing between "confirm" and "cancel". The stop confirmation reads ride_state_stop_pending() directly as its single source of truth (per state.h's own documented expectation that a stray Select during confirmation must route to confirm/cancel, not resume); the exit confirmation is a local carousel.c flag, since state.h has no "leave the app" primitive and doesn't need one.
Checked against the current Core Devices SDK headers on disk (SDK 4.33.1, not assumed from older docs): window_single_click_subscribe()'s own doc says that when a long-click handler is also registered on a button, the single-click handler is held back until the system knows the press wasn't a long one — that's what gives "long-press doesn't also fire the short-press action" for free, no hand-rolled debounce needed. window_long_click_subscribe()'s doc says the back button cannot take a long-click handler at all ("a long press will always terminate the app"), which is why there's no long-Back handling here — and DESIGN.md's own button table has no long-Back row either, so this isn't a gap.
Current page persisted via a new PERSIST_KEY_ACTIVE_PAGE_INDEX (added to page.h's key registry, persist_read_int()/persist_write_int()), alongside #58's existing PERSIST_KEY_PAGE_DESCRIPTORS.

What's honestly still absent, and why

Navigation page insertion ("only when a route is loaded"): there is no route-loaded state anywhere in this codebase (that's Meridian/companion-side work, much later in the tracker). Treated as always-false — the Nav page never appears. Not faked with a placeholder flag.
Disabled pages: "disabled" has no concrete meaning yet (no settings surface, #61). Only the auto-hidden half of "disabled and auto-hidden pages are skipped" is implemented (page_has_live_field()); the carousel has nothing to source "disabled" from until #61 lands.
"Buttons stay responsive while map frames are arriving": there's no map-frame-arrival code in this codebase yet, so this criterion is currently vacuous — nothing to test it against. Not simulated with synthetic traffic.

page_view.c vs. carousel.c

Renamed page_view.c/.h to carousel.c/.h in place (git history preserved via git mv) rather than adding a new carousel.c alongside them. page_view.c's own header comment already said "issue #60 replaces this handler wholesale" — keeping both would just be two files owning one Window and one job. ride_link.h and state.h's comments describing page_view.c as the still-current placeholder are updated to point at carousel.c instead, so nothing in the tree keeps citing stale facts.

Verification

pebble build clean on emery, gabbro and basalt (all three targetPlatforms).
All 68 existing host tests (test_fields/test_page/test_state) still pass unchanged — page.h's enum addition doesn't touch anything they assert on.
Exercised live in the emery and gabbro emulators via pebble emu-button (click and long-click, both buttons) with pebble logs and pebble screenshot as ground truth:
- Up/Down cycles Ride/Effort/Progress, wrapping both directions.
- Long Up jumps to Ride from anywhere; long Down logs the documented no-op.
- Select start -> pause -> resume, each queuing and sending the right CMD (ride_link.c's own logging).
- Long Select shows "Stop ride? / Select=Y / Back=N"; Select confirms (ride stops, prompt clears); Back cancels (prompt clears, ride still running) — screenshotted both platforms, no clipping on gabbro's round display.
- Back while running shows "Leave app? / Back=Y / Select=N"; Select cancels back to the page; Back again confirms and exits — the firmware's own exit log reported Still allocated <0B>, i.e. no leak on exit.
- Back while stopped/idle exits immediately with no prompt.
- heap_bytes_free() logged identical before/after on every single transition throughout the whole run (107512 -> 107512 the entire time on emery).
- Reinstalling after exit reopened the app directly on the last-viewed page (Effort), confirming persistence survives a restart.

https://claude.ai/code/session_01DAoXbRmJUf2uxNYBfdAXPt

Closes #60. Replaces page_view.c's issue #8 placeholder with the real button model from docs/DESIGN.md section 5, wired on top of #58's page descriptors (PR #81) and #11's ride state machine (PR #84). ### What changed - **Up/Down** step to the previous/next *visible* page, wrapping at both ends. "Visible" skips a page with no live field (`page_has_live_field()`, #58's primitive built for exactly this) — unless *none* of the three default pages has data yet (cold boot, before the phone has sent anything), in which case every page counts as visible rather than leaving a blank ring. Only the landing page is ever rendered — a run of skipped pages never flickers past on screen. - **Long Up** jumps straight to page 1, deliberately bypassing the skip rule (an explicit rider request, not a ring step — matches DESIGN.md's "manual navigation always wins" precedent for Nav auto-switching). - **Long Down** is supposed to jump to the map. There is no map page or view anywhere in this codebase yet, so this is a **documented no-op**: an `APP_LOG` line and a comment at the call site, not a fake fallback to page 1 (which would make long Up and long Down indistinguishable) and not a silent, unexplained nothing. - **Select** starts/pauses (`ride_link_start_pause()`); **long Select** begins a stop confirmation (`ride_link_stop_request()`); **Back** leaves the app, confirmed first only if a ride is running or paused. Both confirmations share one on-screen prompt and the same Select/Back roles reversing between "confirm" and "cancel". The stop confirmation reads `ride_state_stop_pending()` directly as its single source of truth (per state.h's own documented expectation that a stray Select during confirmation must route to confirm/cancel, not resume); the exit confirmation is a local `carousel.c` flag, since state.h has no "leave the app" primitive and doesn't need one. - Checked against the current Core Devices SDK headers on disk (SDK 4.33.1, not assumed from older docs): `window_single_click_subscribe()`'s own doc says that when a long-click handler is also registered on a button, the single-click handler is held back until the system knows the press wasn't a long one — that's what gives "long-press doesn't also fire the short-press action" for free, no hand-rolled debounce needed. `window_long_click_subscribe()`'s doc says the back button cannot take a long-click handler at all ("a long press will always terminate the app"), which is why there's no long-Back handling here — and DESIGN.md's own button table has no long-Back row either, so this isn't a gap. - Current page persisted via a new `PERSIST_KEY_ACTIVE_PAGE_INDEX` (added to page.h's key registry, `persist_read_int()`/`persist_write_int()`), alongside #58's existing `PERSIST_KEY_PAGE_DESCRIPTORS`. ### What's honestly still absent, and why - **Navigation page insertion** ("only when a route is loaded"): there is no route-loaded state anywhere in this codebase (that's Meridian/companion-side work, much later in the tracker). Treated as always-false — the Nav page never appears. Not faked with a placeholder flag. - **Disabled pages**: "disabled" has no concrete meaning yet (no settings surface, #61). Only the auto-hidden half of "disabled and auto-hidden pages are skipped" is implemented (`page_has_live_field()`); the carousel has nothing to source "disabled" from until #61 lands. - **"Buttons stay responsive while map frames are arriving"**: there's no map-frame-arrival code in this codebase yet, so this criterion is currently vacuous — nothing to test it against. Not simulated with synthetic traffic. ### page_view.c vs. carousel.c Renamed page_view.c/.h to carousel.c/.h in place (git history preserved via `git mv`) rather than adding a new carousel.c alongside them. page_view.c's own header comment already said "issue #60 replaces this handler wholesale" — keeping both would just be two files owning one Window and one job. ride_link.h and state.h's comments describing page_view.c as the still-current placeholder are updated to point at carousel.c instead, so nothing in the tree keeps citing stale facts. ### Verification - `pebble build` clean on **emery**, **gabbro** and **basalt** (all three `targetPlatforms`). - All 68 existing host tests (test_fields/test_page/test_state) still pass unchanged — page.h's enum addition doesn't touch anything they assert on. - Exercised live in the **emery** and **gabbro** emulators via `pebble emu-button` (click and long-click, both buttons) with `pebble logs` and `pebble screenshot` as ground truth: - Up/Down cycles Ride/Effort/Progress, wrapping both directions. - Long Up jumps to Ride from anywhere; long Down logs the documented no-op. - Select start -> pause -> resume, each queuing and sending the right CMD (`ride_link.c`'s own logging). - Long Select shows "Stop ride? / Select=Y / Back=N"; Select confirms (ride stops, prompt clears); Back cancels (prompt clears, ride still running) — screenshotted both platforms, no clipping on gabbro's round display. - Back while running shows "Leave app? / Back=Y / Select=N"; Select cancels back to the page; Back again confirms and exits — the firmware's own exit log reported `Still allocated <0B>`, i.e. no leak on exit. - Back while stopped/idle exits immediately with no prompt. - `heap_bytes_free()` logged identical before/after on every single transition throughout the whole run (107512 -> 107512 the entire time on emery). - Reinstalling after exit reopened the app directly on the last-viewed page (Effort), confirming persistence survives a restart. https://claude.ai/code/session_01DAoXbRmJUf2uxNYBfdAXPt

robert added 1 commit

2026-09-04 09:47:10 +02:00

Page carousel navigation with long-press jumps (#60 )

dev-artifact / build-pbw (push) Has been cancelled

Details

dev-artifact / build-apk (push) Has been cancelled

Details

fast-lane / host-c-tests (push) Has been cancelled

Details

fast-lane / jvm-tests (push) Has been cancelled

Details

fast-lane / pebble-build (push) Has been cancelled

Details

fast-lane / lint-and-secrets (push) Has been cancelled

Details

fast-lane / meta-declares-required-jobs (push) Has been cancelled

Details

fast-lane / host-c-tests (pull_request) Has been cancelled

Details

fast-lane / jvm-tests (pull_request) Has been cancelled

Details

fast-lane / pebble-build (pull_request) Has been cancelled

Details

fast-lane / lint-and-secrets (pull_request) Has been cancelled

Details

fast-lane / meta-declares-required-jobs (pull_request) Has been cancelled

Details

dev-artifact / publish (push) Has been cancelled

Details

7b78029231

Replaces page_view.c's issue #8 placeholder with the real button model
from docs/DESIGN.md section 5, wired on top of #58's page descriptors
and #11's ride state machine:

- Up/Down step to the previous/next *visible* page, wrapping at both
  ends. "Visible" skips a page with no live field (page.h's
  page_has_live_field(), issue #58's primitive built for exactly this)
  unless literally none of the three default pages has data yet (cold
  boot), in which case every page counts as visible rather than
  leaving a blank ring. Only the landing page is ever rendered, so a
  run of skipped pages never flickers past on screen.
- Long Up jumps straight to page 1, bypassing the skip rule (an
  explicit request, not a ring step). Long Down is meant to jump to
  the map; there is no map page anywhere in this codebase yet, so this
  is a documented no-op (APP_LOG + a comment at the call site) rather
  than a fake destination or a silent do-nothing.
- Select starts/pauses (ride_link_start_pause()); long Select begins a
  stop confirmation (ride_link_stop_request()); Back leaves the app,
  confirmed first only if a ride is running or paused. Both
  confirmations share one on-screen prompt and the same Select=confirm
  / Back=cancel (or vice versa) shape. The stop confirmation's pending
  state is read directly from ride_state_stop_pending() (single source
  of truth, per state.h's own documented expectation); the exit
  confirmation is a local flag, since state.h has no primitive for
  "leave the app" and doesn't need one.
- Confirmed against the current Core Devices SDK (pebble.h, SDK
  4.33.1, read locally, not assumed from old docs):
  window_single_click_subscribe()'s own doc says a long-click handler
  on the same button defers the single-click handler until the press
  is known not to be long, which is what gives "long-press doesn't
  also fire the short-press action" for free. Back cannot take a
  long-click handler at all ("a long press will always terminate the
  app") - matches DESIGN.md's table, which has no long-Back row.
- Current page persisted via a new PERSIST_KEY_ACTIVE_PAGE_INDEX
  (page.h's registry), alongside #58's existing
  PERSIST_KEY_PAGE_DESCRIPTORS.
- Navigation page insertion ("only when a route is loaded") and
  disabled-page skipping are both honestly absent: there is no
  route-loaded state and no settings surface (#61) in this codebase
  yet, so implementing either now would mean inventing a fake flag.
  Documented at the point where each would plug in.
- page_view.c/.h are renamed to carousel.c/.h in place (git history
  preserved) rather than adding a new carousel.c alongside them —
  page_view.c's own header comment already said "#60 replaces this
  handler wholesale"; having both would just be two files owning one
  Window and one job. ride_link.h and state.h's comments describing
  page_view.c as the current placeholder are updated to point at
  carousel.c.

Verified: `pebble build` clean on emery, gabbro and basalt. All 68
existing host tests (test_fields/test_page/test_state) still pass
unchanged. Exercised live in the emery and gabbro emulators via
`pebble emu-button`: Up/Down cycling, long Up jump, long Down no-op,
Select start/pause/resume, long Select stop-confirm, Back
cancel-stop/exit-confirm/cancel-exit/confirm-exit, and Back exiting
immediately with no prompt once stopped — heap_bytes_free() flat
across every transition, and the firmware's own exit log confirmed
"Still allocated <0B>" on quit. Active-page persistence confirmed by
reinstalling and observing the app reopen on the last-viewed page.

Claude-Session: https://claude.ai/code/session_01DAoXbRmJUf2uxNYBfdAXPt