SPB Git

Repository Public

IMPLEMENTATION_STATUS.md 9.5 KB

# Implementation status — 2.0.0 (2), branch next, 2026-11-19 (« SP Note Next Generation »)

Not released: nothing was archived, exported or uploaded. Built against ../SPNotesCore branch next @ 0be0497 (2.0.0 final, with SPNotesFeatures). The 1.0.0 (1) status is in git history (b9fadb5:docs/IMPLEMENTATION_STATUS.md).

# Verified by tests that ran (simulators « SPNoteNext iPhone » = iPhone 17 Pro, « SPNoteNext iPad » = iPad Pro 13-inch (M5), iOS 26.5, Xcode 27.2)

What Evidence
App unit tests 393 tests in 87 suites passed on both simulators, no known issue (Swift Testing, Scripts/build.sh test)
UI tests, iPhone 33 executed, 0 failures, 8 skipped (5 iPad-only, 3 live-only): 928 s
UI tests, iPad 33 executed, 0 failures, 15 skipped (12 iPhone-only, 3 live-only), 2 expected failures recorded (Escape key event, note window restoration after relaunch: not reproduced by the simulator): 2 299 s
Live scenarios (integration server M4M64c:8486, web 3.0.0, scratch database, never production) Scripts/test-live.sh: 3 tests passed — the mandatory offline scenario, tasks created offline reach the server once and a web completion is pulled, a capture made offline is in the server inbox after the sync. Scripts/test-live.sh ui: 3 UI tests passed on each simulator (hosted screens, below)
Store migration v1 → v2 StoreMigrationTests: a COPY of the database written by the shipped 1.0 code path (Tests/Fixtures/v1.sqlite, Core fixture of commit 267a673) inside a container shared with the widget reader and the share hand-off: every row kept, backup notes.sqlite.backup-v1 written, idempotent; the widget's read-only connection on the v1 file answers « unavailable » and leaves the file byte-identical, then reads it once the app migrated; a share made before the migration is imported after it
Open failures StoreFailureTests: damaged file, newer schema → typed error, blocking StoreFailureScreen, file untouched, never an in-memory substitute
Loading does not dirty a page LoadDoesNotDirtyTests: list item ids assigned at load are written only with the first real edit
Scoped invalidation ScopedInvalidationTests: typing 200 characters (20 autosaves) = 20 store changes; 1.0 rule: 120 query re-runs and 20 widget reloads; 2.0: 1 query re-run (the list, after the pause) and 1 widget reload
3.0 nodes in the editor EditorNodesTests, EditorInputTests, EditorNodesExportTests: insert, byte-identical round trip (unknown nodes and unknown attrs included), cards rendered light / dark / accessibility size, tap → detail, slash menu, Markdown shortcuts with one-step undo, callout kinds, task due / priority, reveal(blockId:), Markdown and plain text byte-equal to Core's fixtures
Ink and Pencil logic InkConversionTests (PK → ink → PK → ink: position 0.20 pt, pressure 0.007, incl. a drawing produced by the web serializer), ShapeRecognizerTests (216 synthetic shapes recognised, 34 impostors rejected, snap = one undo step), DrawingBlockTests (1.0 drawings kept or converted, nothing deleted), PDFAnnotationTests (page ⇄ normalised ⇄ view mapping on rotated pages, PDF file byte-identical after annotating)
Shell routes with block ids, route coordinator on several scenes (one route = one window), single activity handling, Handoff activity content, tab ⇄ split adaptation, reachability of every 1.0 screen, hosted sections visibility, command palette, tasks / inbox / backlinks models on the fake backend (offline outbox, rejected operations, 2.x server)
Capture and voice capture kinds → inbox, offline then sync creates one page, shared items become captures once, recorder state machine + one real AVAudioRecorder run in the simulator, transcript segmentation, server transcription flow on recorded answers, AI summary inserted only on request
System widgets redacted when the app lock is on (mirrored flag), interactive task widget writing through a second connection, quick actions mapping, task / capture / ask intents' model paths, task due notification plan and responses, Spotlight task identifiers
Hosted screens opened on screen (HostedSectionsLiveUITests, Scripts/test-live.sh ui, signed in on the integration server with rows seeded by App/LiveSeed.swift) iPhone and iPad, 3 tests each, 0 failures: the ten sections (brief, ask + model picker, server search, databases → a database → a row, canvas → a canvas, graph, constellation, timeline, health, references) and the three note-menu sheets (outline / semantic zoom, page timeline + time machine, ask about the note) each show content and do not crash; 22 screenshots per device (hosted-*), looked at. Found and fixed on compact width: a second large title above each screen's own title row, and two back chevrons on a database / canvas. Nothing paid was triggered (no question sent, no synopsis, no summary)
SPNotesFeatures hosting ShellFeatureSectionsTests: ten sections registered, none listed until the server is known to offer the feature, router kept alive by the shell, canvas ⇄ PencilKit bridge result

# Implemented, not verified (no test drove it, or the simulator cannot)

  • Shared block views now used for chart, queryView, blockRef, bibliography, runCode: provider selection is tested, the sheets were not opened in a UI test after the swap (the mermaid sheet and the in-app chart sheet were, before it).
  • The PencilKit editor presented from a canvas drawing card (bridge logic tested, presentation never driven).
  • Live drawing, lasso / ruler / erasers, PDF ink capture from the overlay, PDF selection → highlight, handwriting region drag: no touch input in a headless simulator.
  • Fluid Navigation zoom, Living Pages on rows, Intelligent Panels: decisions tested on both Reduce Motion paths, never looked at in motion.
  • On-device speech recognition and dictation (logic tested around fakes); camera and scanner inside Quick Capture.
  • Quick actions, Siri phrases, notification delivery and actions, Spotlight results, widgets on a real Home / Lock Screen.
  • English catalogue: 681 keys; strings with interpolations written by hand may not match their generated keys (they fall back to French).

# Only verifiable on a real device (never done: no device was connected)

Passkey sign-in with Face ID (the only way into the app) · Apple Pencil pressure / tilt / hover / squeeze, palm rejection, real handwriting recognition quality · camera, scanner · microphone quality, call interruptions, AirPods routes, background behaviour of a recording · widgets on the Home / Lock Screen (interactive button latency, locked state) · Stage Manager and several windows (the note window scene was reached by one iPad UI test in the simulator only) · hardware keyboard menus · notifications · Spotlight · Handoff to a Mac · VoiceOver · Instruments.

# Known limits and deviations from the specification

  • Universal links: the server's AASA now lists applinks for /p/* and the app declares applinks:note.spboucher.ai; parsing and activity handling are unit-tested, the real hand-off from Safari needs a provisioned build on a device.
  • Hosted screens: opened and looked at, not exercised further in this app (no drag on the graph or canvas, no database edit beyond opening a row, no Ask question sent — paid); interaction coverage is that of SPNotesCore's own tests.
  • « Aujourd’hui » is the native local screen (pinned, recents, tasks, inbox count); the server's DailyBriefScreen is the hosted section « Brief du jour ».
  • The command palette is the app's own (⌘K; the editor's link shortcut moved — see docs/UI_CONVENTIONS.md), not CommandCenterView; Tasks and Inbox are the app's own screens (offline fallback of the 1.0 triage, rejected operations), not the shared ones.
  • Mermaid, canvasLink, citation, audio, pdf and drawing details stay in-app (audio: on-device speech; pdf / drawing: Pencil).
  • Cards are bitmaps drawn at load and cached; PDF export shows the new blocks as pictures; bibliography entries on the card use a neutral author–year form.
  • Smart shapes are triggered by the « Formes » toggle, not by pause-and-hold. PDF ink has no eraser (delete from the list).
  • Reminders stay the 1.0 per-note reminders (not merged with tasks, by design of the sync engine).

# What an upgrading 1.0.0 user will notice in their data

  • The local database is migrated v1 → v2 at first launch (additive; a backup copy is kept next to it for a while). A 1.0 build cannot open the file afterwards.
  • Until the app has been opened once after the update, widgets show « open the app » instead of notes.
  • Shared items and captures now land in the Inbox — on the server too (the page creation carries inboxAt / captureKind; live test aCaptureMadeOfflineIsInTheServerInboxAfterTheSync) — instead of the notes list; untitled ones take their first line as title.
  • List items receive block ids at the first edit of a note; toggles stored closed show collapsed; new callouts are kind note.
  • Typing - , # , 1. , [] , > , ```, ---, / transforms the paragraph (one undo restores the characters).
  • The 1.0 « new drawing » entry creates a drawing block (ink + preview + PencilKit data); 1.0 drawings stay as they are unless converted on request; nothing is deleted.
  • iPad draws with the Pencil only by default; the automatic whole-drawing OCR alert is replaced by recognition of a region on request.
  • With the app lock on, widgets are redacted. The accent colour is ultramarine.
Files main
Menu

Theme