Bug fix: carb dots overlapping SMB dots on the chart#650
Merged
marionbarker merged 1 commit intoJul 20, 2026
Conversation
aug0211
commented
May 9, 2026
Contributor
- Restores pre-PR Differentiate the display of Automated and Manual boluses #305 behavior; no other dot positioning changes
- Carb dots entered alongside an SMB were rendering at the same Y position as the SMB triangle, visually overlapping
- The carb-vs-bolus collision logic was supposed to lift the carb dot above the bolus to BG+70, but only checked for Boluses (not SMB)
- Regression comes from PR Differentiate the display of Automated and Manual boluses #305 (SMB display differentiated from boluses), which split SMBs into their own smbData array. - After the split, findNearestBolusbyTime in processNSCarbs no longer saw SMBs as "nearby treatments," so carb-near-SMB fell through to the default BG+20 offset: the same Y as the SMB itself.
- processNSCarbs now also looks up smbData.
- A carb within 300s of either a bolus OR SMB lifts to BG+70; otherwise BG+20.
- Restores pre-PR loopandlearn#305 behavior; no other dot positioning changes - Carb dots entered alongside an SMB were rendering at the same Y position as the SMB triangle, visually overlapping - The carb-vs-bolus collision logic was supposed to lift the carb dot above the bolus to BG+70, but only checked for Boluses (not SMB) - Regression comes from PR loopandlearn#305 (SMB display differentiated from boluses), which split SMBs into their own smbData array. - After the split, findNearestBolusbyTime in processNSCarbs no longer saw SMBs as "nearby treatments," so carb-near-SMB fell through to the default BG+20 offset: the same Y as the SMB itself. - processNSCarbs now also looks up smbData. - A carb within 300s of either a bolus OR SMB lifts to BG+70; otherwise BG+20.
bjorkert
approved these changes
Jul 19, 2026
bjorkert
left a comment
Member
There was a problem hiding this comment.
Verified on top of the Swift Charts migration: with a carb next to an SMB the carb now lifts to BG+70 as intended, carbs next to manual boluses still lift, and lone carbs stay at BG+20. Merges cleanly into current dev. Thanks for the fix and the clear regression analysis!
marionbarker
approved these changes
Jul 20, 2026
marionbarker
left a comment
Collaborator
There was a problem hiding this comment.
Approved by test using Trio URL.
Applied patch to dev v6.2.25. added carb, observed smb. the two labels are offset and readable.
marionbarker
added a commit
that referenced
this pull request
Jul 25, 2026
* Move release flow to PR-based and add CODEOWNERS for branch protection (#649)
* Move release flow to PR-based and add CODEOWNERS for branch protection
- Add .github/CODEOWNERS
- Cut a release branch in release.sh and open two PRs (sync to dev,
release to main) instead of merging dev into main locally
- Move tag creation into a workflow that fires on push to main
- Skip auto_version_dev when Config.xcconfig was changed in this push,
so merging the release sync-PR into dev does not double-bump
* Make workflow guards portable via fork check
Replace the hardcoded 'loopandlearn' owner check in workflows with a
fork check, so the workflows run on any non-fork repository (including
a test org) while still skipping on contributor forks.
* Allow skipping sister repo updates in release.sh
Set SKIP_SISTER_REPOS=1 to bypass the LoopFollow_Second / LoopFollow_Third
update_follower steps. Default behavior is unchanged: both sister repos
are updated as today and missing directories still cause a hard error,
so a forgotten clone in production fails fast.
* Revert "Allow skipping sister repo updates in release.sh"
This reverts commit c2792b8363d6057e02700663b411ca9c71aadb3d.
* Skip patch hunks for files missing from sister repos
Sister repos (LoopFollow_Second / LoopFollow_Third) are intentionally
stripped of dev-only files like release.sh, auto_version_dev.yml,
lint.yml, and warn_main_pr.yml. Any release patch that touches one of
those files used to abort the sister-repo update with 'No such file or
directory'. Now update_follower runs git apply --check first, parses
the missing-file errors, and re-applies with --exclude for each, so the
sister patch covers the files that actually exist.
* Revert "Skip patch hunks for files missing from sister repos"
This reverts commit 45b9871af675f51ee49037a812c66fad8bd5717d.
* CI: Bump dev version to 6.1.1 [skip ci]
* Diagnose and skip rogue Nightscout profile records (#635)
* Diagnose and skip rogue Nightscout profile records
Profile fetch now uses /api/v1/profiles?count=1 with find[startDate][$lte]=now,
so future-dated records can no longer block the active profile. Adds a
"Run diagnostics" button in the Remote Settings Debug section that fetches
14 days of profile history and surfaces three failure modes:
- Bundle ID mismatch when Loop and Trio share a Nightscout
- Alternating device tokens from multiple installations
- Future-dated profile records left over from a wrong-clock uploader
The bouncing-tokens check compresses consecutive same-token runs and only
warns on actual token alternation, not normal token rotation.
* Widen bouncing-token check and surface shift history
Three changes to the profile diagnostics:
- Drop the 14-day find[startDate][$gte] filter. A slow A→B→A pattern
spread across months only registers as one transition inside a 14-day
window, so the bouncing-tokens check would silently miss it on servers
that honor the filter. The existing 1000-record cap now defines the
scope, which goes back as far as upload frequency allows.
- Fall back to created_at when sorting profile records, so uploaders
that omit startDate don't cluster at .distantPast and corrupt the
run-length compression.
- Include the chronological list of token shifts in the bouncing-tokens
warning. Each row shows when the shift happened and the abbreviated
from→to tokens, so users can see at a glance which devices are
competing instead of just "3 tokens involved across N records".
* CI: Bump dev version to 6.1.2 [skip ci]
* Prevent stale foreground-restart latch from tearing down freshly renewed LA (#656)
pendingForegroundRestart could outlive the condition that triggered it:
a brief foreground entry while the renewal overlay was up latched the
intent, the user backgrounded before didBecomeActive ran, the background
renewal then replaced the LA, and the next foreground entry minutes later
fired the deferred restart against an already-fresh LA.
- adoptPushToStartActivity clears pendingForegroundRestart on every
adoption — a freshly-adopted LA resolves the renewal-window condition
that latched the intent.
- performForegroundRestart re-checks renewalFailed / overlayShowing /
pushToStartLooksStuck before tearing down; bails if none still hold.
- Deferred-foreground-restart push-to-start is tagged
reason="deferred-foreground-restart" via a single-shot
nextStartReasonOverride, so the stale-latch event is no longer
indistinguishable from a real user start in logs.
* CI: Bump dev version to 6.1.3 [skip ci]
* Recover Live Activity after APNs 410 token expiry (#657)
A 6.1.0 user reported the Live Activity vanishing and refusing to come
back without a manual Restart. Trace: APNs returned 410 on the per-
activity push token at 04:42; handleExpiredToken ended the activity but
the eventual iOS .dismissed (4 h later, under the default dismissal
policy) was classified as a user swipe and locked dismissedByUser=true.
Root cause is two cooperating bugs around an app-initiated end():
- end() nulls `current` and clears laRenewBy. handleExpiredToken's
comment said "Activity will restart on next BG refresh via
refreshFromCurrentState()", but renewIfNeeded short-circuits when
current is nil and performRefresh's bind-existing path rebinds to the
just-ended activity. bind() then clears endingForRestart, so the late
.dismissed reads as renewBy=0 / renewalFailed=false / endingForRestart=
false — branch (c) "USER" in the classifier.
- The classifier had no way to recognize a stale observer firing for an
activity the app no longer tracks.
Fixes:
- handleExpiredToken drives the restart synchronously on iOS 17.2+
(attemptPushToStartCreate "expired-token"), so the orphaned post-410
state is short-lived and adoption of the fresh activity cancels the
old observer.
- performRefresh / update bind-existing only to activities in
.active state. Binding to an .ended/.dismissed corpse would clear
endingForRestart and re-attach an observer that only ever delivers
.dismissed.
- .dismissed classifier gains branch (d): if the dismissed activity is
not the one we currently track, log and take no action — only the
foreground LA can be user-swiped, so a stale-observer delivery for an
already-replaced activity must not latch dismissedByUser=true.
* CI: Bump dev version to 6.1.4 [skip ci]
* Auto-restart Live Activity when iOS sends .ended (#659)
When iOS reaches the Live Activity lifetime cap before renewal fires it
delivers .ended, not .dismissed. The state observer only ran restart
logic on .dismissed, so handleForeground saw renewalFailed=false and
renewBy still in the future and returned "no action needed", leaving
the LA dark until manual force-restart.
Mark laRenewalFailed=true on the .ended path (gated on wasCurrent and
!endingForRestart) so the next foreground entry triggers
performForegroundRestart, which sweeps the corpse activity and pushes a
fresh one.
* CI: Bump dev version to 6.1.5 [skip ci]
* Deduplicate Dexcom Share readings to fix delta always showing zero (#661)
Dexcom Share returns each reading twice when both the iPhone Dexcom
app and the Apple Watch app upload to the same account (~9-10 s apart,
same SGV). Without deduplication the two most recent entries in bgData
were always identical, producing delta = 0.
The NS fetch path already had inline deduplication. Extract it into a
shared helper (deduplicateBGReadings) and apply it to the Dexcom-only
path as well.
* CI: Bump dev version to 6.1.6 [skip ci]
* Remove storyboard, migrate to SwiftUI layout (#608)
* Remove Main.storyboard and migrate to SwiftUI app lifecycle
Replace UIKit storyboard/SceneDelegate architecture with SwiftUI App
entry point (LoopFollowApp.swift) and TabView (MainTabView.swift).
Convert MoreMenuViewController to SwiftUI (MoreMenuView.swift). Add
SwiftUI wrappers for Remote and Nightscout tabs. Remove 6 obsolete
UIKit wrapper view controllers and ~300 lines of tab management code
from MainViewController.
* Migrate info table from UITableView to SwiftUI List
Replace UITableView with SwiftUI InfoTableView hosted in
MainViewController. Make InfoManager an ObservableObject so data
updates trigger SwiftUI rebuilds automatically. Remove
UITableViewDataSource conformance and table delegate methods.
No changes needed to the 10 Nightscout controller files that
populate the table data.
* Migrate statistics and pie chart from UIKit to SwiftUI
Replace 7 UILabel properties and DGCharts PieChartView with a
StatsDisplayModel ObservableObject and hosted StatsDisplayView.
The pie chart uses a UIViewRepresentable wrapper for DGCharts
since the Charts pod name shadows Swift Charts. Remove ~60 lines
of UIKit stack layout code from MainViewController setupUI().
* Migrate BG display area from UIKit labels to SwiftUI
Replace BGText, DirectionText, DeltaText, MinAgoText, serverText,
LoopStatusLabel, and PredictionLabel with a SwiftUI BGDisplayView.
Add pull-to-refresh via .refreshable modifier. Move loop status and
prediction text updates to Observable values across DeviceStatus,
DeviceStatusLoop, DeviceStatusOpenAPS, and BGData. Remove
UIScrollView overlay and UIScrollViewDelegate conformance.
* Migrate main layout to SwiftUI with UIKit charts embedded
Replace UIStackView layout with MainHomeView SwiftUI view that composes
BGDisplayView, InfoTableView, LineChartWrapper (UIViewRepresentable for
DGCharts), and StatsDisplayView. MainViewController now hosts a single
UIHostingController instead of managing individual UIView containers.
Visibility of info table, small graph, and stats is now reactive via
Storage observables in SwiftUI, removing several Combine subscriptions.
BG text uses lineLimit + minimumScaleFactor instead of manual font sizing.
* Clean up migration artifacts and fix post-migration bugs
- Fix AVSpeechSynthesizer temporary in AppDelegate that would be
deallocated before speech completes; use stored property instead
- Fix appMovedToBackground tab switching to use Observable instead of
dead UIKit tabBarController reference
- Remove dead code: rebuildTabsIfNeeded(), updateNightscoutTabState(),
traitCollectionDidChange notification relay, UIViewExtension.addBorder
- Remove unused imports (Charts, UIKit, Combine) from migrated files
- Remove unused synthesizer from LoopFollowApp
- Remove redundant .appearanceDidChange subscription from NightscoutVC
- Add missing super calls in viewWillAppear/viewDidAppear
* Replace view hierarchy walking with MainViewController.shared
The getMainViewController() methods in TreatmentsView, SettingsMenuView,
and BackgroundRefreshManager tried to find MainViewController by casting
rootViewController as UITabBarController, which always fails with the
SwiftUI lifecycle. Add a weak static shared reference set during
viewDidLoad and use it everywhere instead.
* Fix MainViewController.shared references for stats and treatments
Pass MainViewController.shared instead of nil when creating
AggregatedStatsContentView in MainTabView and MoreMenuView. Replace
view-hierarchy-walking getMainViewController() in TreatmentsViewModel
with MainViewController.shared.
* Fix info table font size to match storyboard
The storyboard used system 17pt for both title and detail labels.
The SwiftUI migration used .subheadline (~15pt) making text smaller.
* Fix Share Logs sheet rendering blank
Present UIActivityViewController via UIApplication.topMost instead of
wrapping it in a SwiftUI .sheet, which rendered an empty view.
* Fix back navigation from Settings sub-pages
SettingsMenuView declared its own NavigationStack(path:) while already
being pushed onto the outer NavigationStack from MainTabView, so sub-page
back buttons popped the outer stack and jumped past Settings to Menu.
Drop the nested NavigationStack and route Settings entries through the
ambient stack: a single SettingsRoute enum drives a navigationDestination
attached at the MoreMenuView root. The Settings entry itself becomes a
NavigationLink(value:) so it doesn't compete with a navigationDestination
(isPresented:) modifier, which was re-asserting Settings as the top of
stack whenever a sub-page was pushed.
* Harden post-storyboard migration
* MainViewController is now a strong static singleton bootstrapped from
LoopFollowApp.init(). Lifecycle work in viewDidLoad (Combine sinks,
observers, scheduleAllTasks, migrations) runs at launch regardless of
whether the Home tab is rendered, and HomeContentView reuses the
singleton instead of instantiating a fresh VC each time.
* MoreMenuView's eight .navigationDestination(isPresented:) modifiers are
collapsed to a single MenuRoute enum routed through one
.navigationDestination(for:), preventing the same destination-slot
contention that previously caused Settings → Graph back navigation to
jump past Settings.
* MainTabView observes Storage.shared.appearanceMode so theme changes
propagate; the orphaned .appearanceDidChange notification name is
removed.
* OPEN_APP_ACTION notification taps now dismiss any presented modal
before switching to Home, matching prior SceneDelegate behavior.
* Drop the unused Core Data stack (NSPersistentCloudKitContainer,
saveContext) from AppDelegate, the dead AppDelegate.window property,
and the legacy UIRequiredDeviceCapabilities=armv7 /
UIStatusBarTintParameters keys from Info.plist. Switch
AlarmSound's keyWindow access to the connected-scenes API and
generalize UIApplication.topMost likewise so it works on Mac
Catalyst.
* Strip redundant inner NavigationView wrappers from settings sub-views
pushed onto the outer NavigationStack: Graph, General, Advanced,
Calendar, Contact, Dexcom, Nightscout, BackgroundRefresh,
InfoDisplay, ImportExport. Drop unused onBack parameters from
AlarmsContainerView and SettingsMenuView, the unused
isPresentedAsModal flag from MainViewController, and the leftover
debug print in ObservableValue.set.
* LineChartWrapper.updateUIView now flushes the chart on SwiftUI
re-render. MainViewController.deinit removes all observers, not
just the custom "refresh" one. MoreMenuView caches the app version
in @State instead of constructing AppVersionManager on every body
re-render. HomeModalView uses NavigationStack (not deprecated
NavigationView).
* MoreMenuView: render tab-switch buttons in primary color
Buttons in a List inherit the accent tint, so the Features rows that switch tabs appeared blue while the NavigationLink rows that push appeared white. Use .buttonStyle(.plain) to suppress the tint and drop the now-redundant .foregroundStyle(.primary) calls.
* Revert MainViewController singleton bootstrap
Constructing MainViewController.shared from LoopFollowApp.init() — and
reusing the same VC across HomeContentView re-creations — caused tapping
the BG chart to crash with `-[__NSArrayM insertObject:atIndex:]: object
cannot be nil`. Bisected to the singleton+bootstrap piece of the
post-storyboard hardening; the rest of that commit (programmatic UI,
MoreMenuView routing, NavigationView strip-out) is retained.
Restore the prior behavior: shared is a weak static set in viewDidLoad,
HomeContentView constructs a fresh MainViewController each time, and the
LoopFollowApp.init() bootstrap is removed.
Known follow-up: lifecycle work in viewDidLoad (Combine sinks,
scheduleAllTasks, migrations) again only runs when the Home view is
first rendered, so a user who has moved Home off the tab bar gets
degraded behavior until they navigate to it.
* MoreMenuView: make tab-switch rows full-row tappable
Wrap Button labels in an HStack with a trailing Spacer and contentShape
so the entire row is tappable, matching the hit area of NavigationLink
rows. Extract the pattern into a small FullRowButton helper, used for
both tab-switch rows and Share Logs.
* Align units-selection conflict resolution with integration branch
Move the Diagnostics section out of Section("Speak BG") (was nested at
the wrong indent), match StatsDisplayModel field order, and add the
spacing line in updateStats.
* MoreMenuView: fix cross-row tap routing in Features section
Mixing Button and NavigationLink rows in the same List ForEach caused
taps on a NavigationLink row to fire a sibling Button row's action —
e.g. tapping Alarms with Stats in the tab bar would switch to the Stats
tab instead of pushing the alarms detail.
Make every row in the menu's List a uniform FullRowButton and drive
pushes from state via .navigationDestination(isPresented:). Add an
opt-in chevron to FullRowButton so navigating rows render the standard
disclosure indicator.
* MoreMenuView: keep Settings as a value-based NavigationLink
Mixing .navigationDestination(isPresented:) with
.navigationDestination(for:) on the same view shadowed the value-based
SettingsRoute registration once SettingsMenuView was on the stack, so
sub-rows like Units and Metrics couldn't push.
Settings sits alone in its section, so it doesn't need the uniform-Button
treatment used in Features and Logging — restore it to a NavigationLink
and route it through the existing .navigationDestination(for:) channel.
* Fix navigation between alarms and menu
* Drive Before-First-Unlock recovery from AppDelegate
Move BFU recovery (Storage.reloadAll) out of MainViewController and into
AppDelegate so it runs even when the home tab's UIHostingController has
not yet materialized — necessary under the SwiftUI App lifecycle, where a
BG-only launch (BGAppRefreshTask, BLE wake, prewarming) may complete and
the device may unlock without MainViewController ever being created.
AppDelegate observes protectedDataDidBecomeAvailable (authoritative signal)
and willEnterForeground (fallback), with a race-guard re-check immediately
after observer registration. Recovery is idempotent via needsBFUReload.
MainViewController now reacts to a new .bfuReloadCompleted notification by
showing the loading overlay and rescheduling tasks; if it is not alive when
the notification fires, its viewDidLoad will later see the already-reloaded
Storage values and schedule tasks correctly on first load.
* Keep MainViewController alive regardless of tab layout
Make MainViewController.shared a strong, long-lived singleton created
once via bootstrap() on first foreground, so the data pipeline, alarms
and background audio run even when Home is moved into the Menu rather
than a tab. Home views reuse the single instance instead of creating
new ones, so the singleton is never displaced.
Defer the one-shot BG graph zoom until the chart has a real frame and
re-render the graph on every appearance, so the curve stays visible
when Home is reached from the Menu or moved between tab bar and Menu.
Restore the one-time telemetry consent prompt that was lost when
SceneDelegate was removed, presenting it from MainTabView on first
appearance for undecided installs.
* Fix issues found in post-migration review
- Speak BG quick action: under the SwiftUI scene lifecycle UIKit
delivers Home Screen quick actions to the window scene delegate and
never calls application(_:performActionFor:). Install a scene
delegate via configurationForConnecting that handles warm taps,
cold-launch shortcut delivery, and mirrors the Live Activity la-tap
URL handling that moves with it.
- Nightscout tab: show a hint instead of a blank page when no URL is
configured, and recreate the web view when the URL or token changes
(the page was loaded once in viewDidLoad and stayed stale until app
restart).
- Stats: resolve MainViewController lazily with a fallback to the
shared instance, so stats work when the tab is built before
bootstrap() has run (cold launch with Statistics as selected tab).
- Remove the unwired NightscoutSettingsViewModel delegate chain.
- Make LineChartWrapper.updateUIView a no-op; MainViewController
already notifies the charts whenever it mutates their data.
* Fix squished Home layout caused by phantom keyboard inset
iOS sometimes replays a stale keyboard frame when the app returns to the
foreground, compressing the Home screen by a keyboard's height until a
rotation forces the safe area to recompute. Home has no text input, so
opt out of keyboard avoidance at both hosting layers.
* Suppress silent passive notifications in willPresent handler (#655)
UNUserNotificationCenterDelegate.willPresent returned
[.banner, .sound, .badge] unconditionally, which meant any notification
iOS routed through this handler while the app was foregrounded produced
sound — including the Live Activity push-to-start payload, which is
intentionally silent (interruption-level: passive, empty title/body).
Now returns [] for passive notifications and for ones with empty
title/body. The four intentional alerts (renewal-failed, APNs
credentials missing, push-to-start token missing, alarms) all use
non-empty title/body and the default .active interruption level, so
they continue to surface.
Also expanded the willPresent log line with interruptionLevel and
title/body presence so future reports can confirm whether iOS routed
a given payload here.
* Bump fastlane to 2.235.0 and jwt to 3.2.0 (CVE-2026-45363) (#663)
* Bump fastlane to 2.235.0 and jwt to 3.2.0 (CVE-2026-45363)
jwt < 3.2.0 accepts attacker-forged tokens when an empty or nil key is
used with HMAC algorithms (GHSA-c32j-vqhx-rx3x). fastlane 2.233.1 pinned
jwt < 3, blocking the fix. fastlane 2.235.0 relaxes that to jwt < 4,
allowing the upgrade to 3.2.0.
* Remove json and addressable pins no longer needed with fastlane 2.235.0
* Fix BG history truncation caused by duplicate readings from multiple uploaders (#662)
* Increase Nightscout BG entry count to handle multiple uploaders
Some users have multiple sources uploading BG entries to Nightscout for
the same sensor (e.g. a closed-loop system plus two Dexcom platform apps),
resulting in up to 3 entries per 5-minute slot. The previous count cap of
days × 2 × 288 was too low in those cases, silently truncating the oldest
portion of the requested time window before deduplication.
Raise the multiplier from 2 to 4 in both BGData.swift and
StatsDataFetcher.swift, giving headroom for up to 4 uploaders. The
date[$gte] filter still bounds the window, so no extra data is returned
when fewer entries exist.
* Supplement Dex Share with NS data when Dex doesn't cover the full window
When Dexcom credentials are present, LoopFollow fetches from Dex Share
first. If the Dex data doesn't reach back to the start of the requested
window (e.g. Dex Share's 24h cap when more days are configured, or any
other gap), fall through to webLoadNSBGData so Nightscout fills in the
older portion. The existing merge logic already handles stitching the two
sources together correctly.
* Deduplicate Dex Share readings before use
Multiple Dexcom uploaders (e.g. G7 iPhone app + Apple Watch) each write
readings to Dexcom's cloud, causing the Share API to return 2+ entries
per 5-minute slot. With the API's hard cap of 288 readings, duplicates
consume the budget and the returned data covers only ~15h instead of 24h.
Dedup Dex Share data with the same 30-second window used for Nightscout,
so both the NS-supplement path and the direct ProcessDexBGData path
receive clean, deduplicated readings.
* Extract BG entry-count multiplier into named globalVariables.maxExpectedUploaders
* Reuse deduplicateBGReadings helper for Dexcom Share dedup
* Match Trio's override/temp-target colors for non-Loop backends (#668)
* Quick-Pick Boluses and Meals for remote commands (#603)
* Common remote commands
* Help texts
* Rename Common to Quick-Pick throughout
Renames all files, classes, types, and UI labels introduced by the
quick-pick feature from "Common" to "Quick-Pick":
- CommonBoluses/ → QuickPickBoluses/
- CommonMeals/ → QuickPickMeals/
- CommonBolusesManager → QuickPickBolusesManager
- CommonMealsManager → QuickPickMealsManager
- CommonSectionHeader → QuickPickSectionHeader
- UI labels: "Common Boluses/Meals" → "Quick-Pick Boluses/Meals"
* Add Nightscout WebSocket support for real-time data updates (#606)
* Add Nightscout WebSocket (Socket.IO) support for real-time data updates
Connect to Nightscout's Socket.IO endpoint to receive push notifications
when new data arrives, instead of waiting for the next poll cycle. The
WebSocket acts as a smart trigger: when a dataUpdate event arrives, only
the relevant data types (BG, treatments, device status, profile) are
fetched based on which keys are present in the delta payload.
When WebSocket is connected and authenticated, polling intervals are
extended (BG 3x, device status 3x, treatments 2→10 min, profile 10→30
min) so HTTP polling becomes a safety net. On disconnect, polling
immediately reverts to normal intervals.
The feature is always-on when Nightscout is configured — no user setting
needed. A read-only connection status is shown in Nightscout settings.
- Add Socket.IO-Client-Swift 16.1.1 via SPM
- Add NightscoutSocketManager for connection lifecycle
- Add NightscoutSocketDataHandler for selective push-trigger logic
- Extend polling intervals when WebSocket is authenticated
- Show WebSocket status in Nightscout settings
- Wire up lifecycle in MainViewController (init, foreground, refresh)
- Add staleness detection (10 min fallback to polling)
* Add opt-in WebSocket toggle with info sheet and refresh on disconnect
- Add webSocketEnabled storage property (default off) so users opt in
- Replace read-only status with toggle + info button in Nightscout settings
- Info sheet explains real-time updates, polling fallback, and battery impact
- On toggle off: disconnect socket and trigger full refresh to restore
normal polling intervals immediately
- On unexpected socket disconnect: post refresh notification so extended
polling intervals revert to normal without waiting for them to expire
* Fix stale WebSocket session on config change, remove redundant staleness timer
- Disconnect WebSocket when Nightscout URL/token validation fails,
preventing the old session from streaming data from a previous server
while polling has switched to the new config
- Reorder removeAllHandlers() before disconnect() so intentional
disconnects don't fire the event handler that would reconnect to
an invalid URL
- Remove staleness timer — the extended polling intervals (3x/5x)
already serve as the safety net when WebSocket data stops flowing
* Limit WebSocket to foreground and default-on
Disconnect the Nightscout WebSocket when LoopFollow moves to the
background and re-establish it when returning to the foreground.
Polling continues to handle background updates, so the persistent
connection no longer holds the cellular radio out of idle while the
app is not in use. With the battery cost bounded to foreground time,
the toggle now defaults to on. Inline documentation updated to
describe the foreground/background behavior.
* Resume Nightscout polling promptly when the WebSocket disconnects
While the socket is authenticated, each REST poll is rescheduled with a
multiplier on the assumption WS will publish the next reading before the
poll fires. Without an explicit catch-up, those long delays carry over
when the socket goes away — most visibly on background entry, where the
user could sit on stale data for 10–15 minutes after the screen locked.
Fire each Nightscout poll immediately on disconnect. Their actions then
reschedule on the normal un-multiplied cadence, since connectionState is
no longer .authenticated. Gate on the previous state so the reconnect
dance in connectIfNeeded() and other no-op disconnect paths don't trigger
spurious REST round-trips.
* Default debug logging on and prompt for description when sharing logs (#627)
Flip the debugLogLevel default to true and add a migration step so
existing users with it stored as false also get it enabled, ensuring
shared logs contain useful detail when reporting problems.
When the user taps Share Logs, present a sheet asking for a short
description of the issue. The description is written to a
ShareNotice_<timestamp>.txt file (date, app version, branch+sha,
user description) and included alongside the log files in the iOS
share sheet.
* Add optional yesterday BG comparison line to main graph (#665)
Adds a Graph Settings toggle (Nightscout only) that overlays yesterday's
BG curve on the main graph, time-shifted by 24h so it aligns with the same
clock time today. Drawn as a thin dimmed gray line with no dots, purely for
comparison. When enabled, one extra day of history is fetched and the overlay
is capped to now plus the configured hours of prediction so it never extends
further into the future than the prediction line.
* Scale info data table text with Dynamic Type (#667)
* Remove Main.storyboard and migrate to SwiftUI app lifecycle
Replace UIKit storyboard/SceneDelegate architecture with SwiftUI App
entry point (LoopFollowApp.swift) and TabView (MainTabView.swift).
Convert MoreMenuViewController to SwiftUI (MoreMenuView.swift). Add
SwiftUI wrappers for Remote and Nightscout tabs. Remove 6 obsolete
UIKit wrapper view controllers and ~300 lines of tab management code
from MainViewController.
* Migrate info table from UITableView to SwiftUI List
Replace UITableView with SwiftUI InfoTableView hosted in
MainViewController. Make InfoManager an ObservableObject so data
updates trigger SwiftUI rebuilds automatically. Remove
UITableViewDataSource conformance and table delegate methods.
No changes needed to the 10 Nightscout controller files that
populate the table data.
* Migrate statistics and pie chart from UIKit to SwiftUI
Replace 7 UILabel properties and DGCharts PieChartView with a
StatsDisplayModel ObservableObject and hosted StatsDisplayView.
The pie chart uses a UIViewRepresentable wrapper for DGCharts
since the Charts pod name shadows Swift Charts. Remove ~60 lines
of UIKit stack layout code from MainViewController setupUI().
* Migrate BG display area from UIKit labels to SwiftUI
Replace BGText, DirectionText, DeltaText, MinAgoText, serverText,
LoopStatusLabel, and PredictionLabel with a SwiftUI BGDisplayView.
Add pull-to-refresh via .refreshable modifier. Move loop status and
prediction text updates to Observable values across DeviceStatus,
DeviceStatusLoop, DeviceStatusOpenAPS, and BGData. Remove
UIScrollView overlay and UIScrollViewDelegate conformance.
* Migrate main layout to SwiftUI with UIKit charts embedded
Replace UIStackView layout with MainHomeView SwiftUI view that composes
BGDisplayView, InfoTableView, LineChartWrapper (UIViewRepresentable for
DGCharts), and StatsDisplayView. MainViewController now hosts a single
UIHostingController instead of managing individual UIView containers.
Visibility of info table, small graph, and stats is now reactive via
Storage observables in SwiftUI, removing several Combine subscriptions.
BG text uses lineLimit + minimumScaleFactor instead of manual font sizing.
* Clean up migration artifacts and fix post-migration bugs
- Fix AVSpeechSynthesizer temporary in AppDelegate that would be
deallocated before speech completes; use stored property instead
- Fix appMovedToBackground tab switching to use Observable instead of
dead UIKit tabBarController reference
- Remove dead code: rebuildTabsIfNeeded(), updateNightscoutTabState(),
traitCollectionDidChange notification relay, UIViewExtension.addBorder
- Remove unused imports (Charts, UIKit, Combine) from migrated files
- Remove unused synthesizer from LoopFollowApp
- Remove redundant .appearanceDidChange subscription from NightscoutVC
- Add missing super calls in viewWillAppear/viewDidAppear
* Replace view hierarchy walking with MainViewController.shared
The getMainViewController() methods in TreatmentsView, SettingsMenuView,
and BackgroundRefreshManager tried to find MainViewController by casting
rootViewController as UITabBarController, which always fails with the
SwiftUI lifecycle. Add a weak static shared reference set during
viewDidLoad and use it everywhere instead.
* Fix MainViewController.shared references for stats and treatments
Pass MainViewController.shared instead of nil when creating
AggregatedStatsContentView in MainTabView and MoreMenuView. Replace
view-hierarchy-walking getMainViewController() in TreatmentsViewModel
with MainViewController.shared.
* Fix info table font size to match storyboard
The storyboard used system 17pt for both title and detail labels.
The SwiftUI migration used .subheadline (~15pt) making text smaller.
* Fix Share Logs sheet rendering blank
Present UIActivityViewController via UIApplication.topMost instead of
wrapping it in a SwiftUI .sheet, which rendered an empty view.
* Fix back navigation from Settings sub-pages
SettingsMenuView declared its own NavigationStack(path:) while already
being pushed onto the outer NavigationStack from MainTabView, so sub-page
back buttons popped the outer stack and jumped past Settings to Menu.
Drop the nested NavigationStack and route Settings entries through the
ambient stack: a single SettingsRoute enum drives a navigationDestination
attached at the MoreMenuView root. The Settings entry itself becomes a
NavigationLink(value:) so it doesn't compete with a navigationDestination
(isPresented:) modifier, which was re-asserting Settings as the top of
stack whenever a sub-page was pushed.
* Harden post-storyboard migration
* MainViewController is now a strong static singleton bootstrapped from
LoopFollowApp.init(). Lifecycle work in viewDidLoad (Combine sinks,
observers, scheduleAllTasks, migrations) runs at launch regardless of
whether the Home tab is rendered, and HomeContentView reuses the
singleton instead of instantiating a fresh VC each time.
* MoreMenuView's eight .navigationDestination(isPresented:) modifiers are
collapsed to a single MenuRoute enum routed through one
.navigationDestination(for:), preventing the same destination-slot
contention that previously caused Settings → Graph back navigation to
jump past Settings.
* MainTabView observes Storage.shared.appearanceMode so theme changes
propagate; the orphaned .appearanceDidChange notification name is
removed.
* OPEN_APP_ACTION notification taps now dismiss any presented modal
before switching to Home, matching prior SceneDelegate behavior.
* Drop the unused Core Data stack (NSPersistentCloudKitContainer,
saveContext) from AppDelegate, the dead AppDelegate.window property,
and the legacy UIRequiredDeviceCapabilities=armv7 /
UIStatusBarTintParameters keys from Info.plist. Switch
AlarmSound's keyWindow access to the connected-scenes API and
generalize UIApplication.topMost likewise so it works on Mac
Catalyst.
* Strip redundant inner NavigationView wrappers from settings sub-views
pushed onto the outer NavigationStack: Graph, General, Advanced,
Calendar, Contact, Dexcom, Nightscout, BackgroundRefresh,
InfoDisplay, ImportExport. Drop unused onBack parameters from
AlarmsContainerView and SettingsMenuView, the unused
isPresentedAsModal flag from MainViewController, and the leftover
debug print in ObservableValue.set.
* LineChartWrapper.updateUIView now flushes the chart on SwiftUI
re-render. MainViewController.deinit removes all observers, not
just the custom "refresh" one. MoreMenuView caches the app version
in @State instead of constructing AppVersionManager on every body
re-render. HomeModalView uses NavigationStack (not deprecated
NavigationView).
* MoreMenuView: render tab-switch buttons in primary color
Buttons in a List inherit the accent tint, so the Features rows that switch tabs appeared blue while the NavigationLink rows that push appeared white. Use .buttonStyle(.plain) to suppress the tint and drop the now-redundant .foregroundStyle(.primary) calls.
* Revert MainViewController singleton bootstrap
Constructing MainViewController.shared from LoopFollowApp.init() — and
reusing the same VC across HomeContentView re-creations — caused tapping
the BG chart to crash with `-[__NSArrayM insertObject:atIndex:]: object
cannot be nil`. Bisected to the singleton+bootstrap piece of the
post-storyboard hardening; the rest of that commit (programmatic UI,
MoreMenuView routing, NavigationView strip-out) is retained.
Restore the prior behavior: shared is a weak static set in viewDidLoad,
HomeContentView constructs a fresh MainViewController each time, and the
LoopFollowApp.init() bootstrap is removed.
Known follow-up: lifecycle work in viewDidLoad (Combine sinks,
scheduleAllTasks, migrations) again only runs when the Home view is
first rendered, so a user who has moved Home off the tab bar gets
degraded behavior until they navigate to it.
* MoreMenuView: make tab-switch rows full-row tappable
Wrap Button labels in an HStack with a trailing Spacer and contentShape
so the entire row is tappable, matching the hit area of NavigationLink
rows. Extract the pattern into a small FullRowButton helper, used for
both tab-switch rows and Share Logs.
* Align units-selection conflict resolution with integration branch
Move the Diagnostics section out of Section("Speak BG") (was nested at
the wrong indent), match StatsDisplayModel field order, and add the
spacing line in updateStats.
* MoreMenuView: fix cross-row tap routing in Features section
Mixing Button and NavigationLink rows in the same List ForEach caused
taps on a NavigationLink row to fire a sibling Button row's action —
e.g. tapping Alarms with Stats in the tab bar would switch to the Stats
tab instead of pushing the alarms detail.
Make every row in the menu's List a uniform FullRowButton and drive
pushes from state via .navigationDestination(isPresented:). Add an
opt-in chevron to FullRowButton so navigating rows render the standard
disclosure indicator.
* MoreMenuView: keep Settings as a value-based NavigationLink
Mixing .navigationDestination(isPresented:) with
.navigationDestination(for:) on the same view shadowed the value-based
SettingsRoute registration once SettingsMenuView was on the stack, so
sub-rows like Units and Metrics couldn't push.
Settings sits alone in its section, so it doesn't need the uniform-Button
treatment used in Features and Logging — restore it to a NavigationLink
and route it through the existing .navigationDestination(for:) channel.
* Fix navigation between alarms and menu
* Drive Before-First-Unlock recovery from AppDelegate
Move BFU recovery (Storage.reloadAll) out of MainViewController and into
AppDelegate so it runs even when the home tab's UIHostingController has
not yet materialized — necessary under the SwiftUI App lifecycle, where a
BG-only launch (BGAppRefreshTask, BLE wake, prewarming) may complete and
the device may unlock without MainViewController ever being created.
AppDelegate observes protectedDataDidBecomeAvailable (authoritative signal)
and willEnterForeground (fallback), with a race-guard re-check immediately
after observer registration. Recovery is idempotent via needsBFUReload.
MainViewController now reacts to a new .bfuReloadCompleted notification by
showing the loading overlay and rescheduling tasks; if it is not alive when
the notification fires, its viewDidLoad will later see the already-reloaded
Storage values and schedule tasks correctly on first load.
* Keep MainViewController alive regardless of tab layout
Make MainViewController.shared a strong, long-lived singleton created
once via bootstrap() on first foreground, so the data pipeline, alarms
and background audio run even when Home is moved into the Menu rather
than a tab. Home views reuse the single instance instead of creating
new ones, so the singleton is never displaced.
Defer the one-shot BG graph zoom until the chart has a real frame and
re-render the graph on every appearance, so the curve stays visible
when Home is reached from the Menu or moved between tab bar and Menu.
Restore the one-time telemetry consent prompt that was lost when
SceneDelegate was removed, presenting it from MainTabView on first
appearance for undecided installs.
* Scale info data table text with Dynamic Type
Drive the info table font size and row height from @ScaledMetric so the
top-right info data grows and shrinks with the iPhone's text-size setting,
keeping the existing 17pt/21pt look at the default size. Cap the scaling at
accessibility1 so the compact top strip stays within its layout.
* Use two rows if needed
* Info table adjustments
* CI: Bump dev version to 6.1.7 [skip ci]
* ITMS-91061 - privacy manifest for Charts framework (#666)
* CI: Bump dev version to 6.1.8 [skip ci]
* Remove the Nightscout-based remote command path (#618)
Trio dev's APNS-based remote command support has merged to main, so the
Nightscout treatment-posting variant of remote commands is no longer
needed. Drop the .nightscout case from RemoteType, the Nightscout
remote view and controller, the unused NoRemoteView fallback, and all
.nightscout switch arms in the remote settings, import/export, and
device validation paths. Storage keys are reused by the TRC and
LoopAPNS variants and remain untouched.
Existing users with remoteType set to "Nightscout" automatically fall
back to .none on next launch: StorageValue's JSONDecoder fails on the
removed enum case and returns the default value.
* CI: Bump dev version to 6.1.9 [skip ci]
* Cap prediction cone to shortest predBG array length (#653)
Fixes #637.
The cone of uncertainty in updateOpenAPSPredictionDisplay() was capped
at the longest predBG array (.max()), which let the band visibly deform
at the tail as shorter arrays dropped out one by one.
Cap at the shortest array length instead so every cone point is computed
from the same set of contributing arrays. Matches Trio's ForecastSetup
(Trio/Sources/Modules/Home/HomeStateModel+Setup/ForecastSetup.swift),
which uses allForecastValues.map(\.count).min() and then iterates
0 ..< localMinCount.
Renamed maxLength to coneLength since the variable no longer represents
a max.
* CI: Bump dev version to 6.1.10 [skip ci]
* CI: Re-run lint when a PR's base branch changes (#676)
Add 'edited' to the lint workflow's pull_request trigger, guarded so the
SwiftFormat job only re-runs when the base branch actually changed.
Retargeting a PR (e.g. main -> dev) fires pull_request: edited, which the
default trigger types (opened, synchronize, reopened) ignore. When the new
base requires the SwiftFormat status check, GitHub leaves it 'Expected -
waiting for status to be reported' with no run behind it, blocking the PR
indefinitely. Re-running on base change ensures the required check actually
reports against the new base.
* CI: Bump dev version to 6.1.11 [skip ci]
* Remove empty Nightscout group from project file (#675)
The Nightscout remote command path was removed in #618, deleting the only
file under LoopFollow/Remote/Nightscout/. The now-empty PBXGroup was left
behind, still pointing at a folder with no contents. Remove it.
* CI: Bump dev version to 6.1.12 [skip ci]
* update version to 6.2.0 [skip ci]
* Sanitize Nightscout token to prevent WebSocket crash on launch (#688)
* Sanitize Nightscout token to prevent WebSocket crash on launch
Strip whitespace, newlines, and control characters from the token
before storing it and before opening the WebSocket. A stray character
(typically pasted in) produced an invalid percent-encoded query in
Socket.IO's URL builder, which traps on iOS 26 and crashed the app on
startup. Existing saved tokens are sanitized defensively at connect time.
* Sanitize URL entry path, not just token, to prevent WebSocket crash
* CI: Bump dev version to 6.2.1 [skip ci]
* Add search to the alarms list (#694)
* CI: Bump dev version to 6.2.2 [skip ci]
* Add search filter to the Add Alarm sheet (#681)
* CI: Bump dev version to 6.2.3 [skip ci]
* Fix stale snooze state in alarm list (#678)
The alarm list categorized snoozed vs. active alarms against Date()
sampled at render time, but nothing re-rendered the view when a snooze
expired. A lapsed snooze could stay in the Snoozed section for hours
while the editor correctly showed it as not snoozed.
Refresh a now snapshot on appear and on returning to the foreground,
and drive all snooze comparisons through it.
* CI: Bump dev version to 6.2.4 [skip ci]
* Prefer structured ISF/CR from device status over reason string (#677)
Ignore a placeholder 0 in the structured ISF field so the info row
falls back to the profile value instead of showing a 0 enacted ISF.
Read the carb ratio from the structured CR field when present, keeping
the reason-string regex only as a fallback for uploaders that don't
expose it.
* CI: Bump dev version to 6.2.5 [skip ci]
* Bump fastlane to 2.237.0 (#701)
Update the pinned fastlane version from 2.235.0 to 2.237.0 and refresh
the transitive dependencies in Gemfile.lock.
Notable upstream changes:
- addressable minimum raised to 2.9.0 (CVE-2026-35611)
- excon upper bound relaxed, resolving to 1.5.0
- match: keychain import path escaping fix, fail fast on SSH auth
prompts in non-interactive shells
- gym: cleaner xcpretty error handling, xcbeautify preferred
* CI: Bump dev version to 6.2.6 [skip ci]
* Only clear the forecast when the data source changes (#707)
* Only clear the forecast when the data source changes
The refresh handler used to wipe the Loop and Trio forecasts on every refresh, including every websocket disconnect, which made the Loop forecast disappear and only come back on the next loop cycle. Move the clearing so it only happens when the active system actually changes: when a device status flips between Loop and Trio, and when Nightscout is turned off so there is no device status source at all, such as Dexcom only.
* Rebuild the Loop forecast when it is empty
The forecast was only rebuilt when the loop cycle advanced (previousLastLoopTime < lastLoopTime), so any time the prediction was cleared within the same cycle it stayed blank until the next loop cycle. Also rebuild when predictionData is empty, so the forecast is restored on the next device-status poll regardless of cycle timing.
* Fall back to the loop timestamp when the pump reports no clock
A pump that omits pump.clock (e.g. some Omnipod configurations) left alertLastLoopTime at 0, so the Loop forecast anchored to epoch 0 and stretched the graph across decades. Use the loop cycle timestamp as the loop time when no pump clock is present, so the forecast anchors correctly and refreshes each cycle.
* CI: Bump dev version to 6.2.7 [skip ci]
* Remove migration step 1 and the legacy UserDefaults framework it kept alive (#698)
Step 1 shipped in v3.0.0 (2025-07-07) and its one-year window has passed.
It was already unreachable: migrationStep defaults to 9, so any user without
a stored migrationStep skips every step, and users who have one are at >= 1.
Removing migrateStep1() also retires its exclusive dependencies: the
UserDefaultsValue / ObservableUserDefaultsValue stack, UserDefaultsValueGroups,
ObservableUserDefaults, ObservationToken, AnyConvertible, and the orphaned
HKQuantity+AnyConvertible. Nothing in Swift reads the app-group UserDefaults
suite anymore, so AppConstants.APP_GROUP_ID goes too.
* CI: Bump dev version to 6.2.8 [skip ci]
* Add Nightscout database size info row and alarm (#697)
Nightscout's dbsize plugin exposes the Mongo data + index size at
/api/v2/properties/dbsize. Poll it on a new 6 hour scheduler task and
surface it two ways:
- An optional "DB Size" info row showing used MiB and the percentage of
the site's configured DBSIZE_MAX.
- A new "Nightscout Database Size" alarm that fires at or above a
percentage threshold, defaulting to 75% to match the urgent level of
the plugin itself.
The database fills over weeks, so the alarm defaults to day-only and
never repeats at night.
Nightscout still reports the property with a zero size when db.stats()
fails, which would read as an empty database rather than a missing
reading. Treat that as no reading, mirroring how the plugin's own pill
hides itself in that case.
* CI: Bump dev version to 6.2.9 [skip ci]
* Migrate charts to Swift Charts (#704)
* Migrate charts to Swift Charts
Replaces the danielgindi Charts and SwiftAlgorithms CocoaPods with native
Swift Charts.
- Rebuilds the BG graph, treatments, and prediction cone on Swift Charts
with a custom pan/zoom/inspect interaction layer.
- Ports the stats views (AGP, TIR, GRI) off the old Charts dependency.
- Drops the Charts and SwiftAlgorithms pods and their generated support files.
Chart appearance and behaviour are kept at parity with the previous
implementation.
* Raise app and Live Activity extension deployment target to iOS 18.0
* Show every reading and treatment under the chart scrubber
When the detail selector is dragged across the chart, the pill now lists
every item near the finger — the nearest glucose reading plus any treatments
whose symbols fall within a screen-space capture band — instead of a single
winner chosen by time. Every glucose reading stays inspectable, including
where a bolus, carb entry, or an on-trace treatment (BG check, note, suspend,
resume, sensor start) shares its time.
* Enhance AGP, GRI, and TIR graph views with threshold indicators and improved padding
* Refactor threshold label handling in AGPGraphView to use Localizer for unit display
* Add label support to BandRect, enhance band label rendering in BGChartView and fix upper domain for Y label
* Show Trio override names on the graph
Trio uploads overrides as Exercise events with the name in the notes
field and no reason field, so the graph band label fell back to the
generic "Override". Resolve the name notes-first (falling back to
reason for Loop) when building the override graph data, matching the
info table and treatments list.
---------
Co-authored-by: codebymini <daniel@codebymini.se>
Co-authored-by: Daniel Mini Johansson <42831533+codebymini@users.noreply.github.com>
* CI: Bump dev version to 6.2.10 [skip ci]
* release.sh: verify latest script before running (#682)
Guard the release flow against running an outdated copy of release.sh. Before any git operations, fetch origin/dev and compare the running script against origin/dev:release.sh, aborting if they differ.
* CI: Bump dev version to 6.2.11 [skip ci]
* release.sh: set gh pr repo explicitly (#685)
Derive owner/repo from the origin remote and pass --repo to both gh pr create calls so the release flow does not depend on gh's inferred default repo, which fails in clones with multiple remotes.
* CI: Bump dev version to 6.2.12 [skip ci]
* release.sh: drop [skip ci] from version bump; allow manual tag_on_main (#686)
The [skip ci] on the version-bump commit skipped the required SwiftFormat lint check (blocking the release PRs) and the tag_on_main workflow (no auto-tag). Remove it; auto_version_dev already prevents re-bumping via its Config.xcconfig guard. Also add a workflow_dispatch trigger to tag_on_main so the tag can be created manually if a run is ever skipped.
* Add directional/charging skip toggles to Low, High and Battery alarms (#692)
* Add option to skip low BG alarm while rising
The low alarm can now stay silent while BG is climbing, firing only
when the latest reading is flat or still falling. Off by default, so
existing alarms are unaffected.
* Add skip-if-falling for high alarm and skip-while-charging for battery
The high alarm can now stay silent while BG is falling (mirror of the
low alarm's skip-if-rising). The phone-battery alarm can stay silent
while the phone is charging, using the uploader's charging status from
Nightscout. Both are off by default.
* CI: Bump dev version to 6.2.13 [skip ci]
* Mark scheduled runs with nothing to build as cancelled instead of successful (#705)
* Mark scheduled runs with nothing to build as cancelled instead of successful
* update testflight.md, remove alive references, update loopdocs links
---------
Co-authored-by: marionbarker <marionbarker@earthlink.net>
* CI: Bump dev version to 6.2.14 [skip ci]
* Use NightscoutUtils.constructURL for the Nightscout web viewer URL (#712)
* CI: Bump dev version to 6.2.15 [skip ci]
* Add fallbacks for missing TDD and timestamp in devicestatus (#709)
* Fix TDD and Updated dropping out on Trio SMB follow-up uploads
* Remove comments
* CI: Bump dev version to 6.2.16 [skip ci]
* Format Rec. Bolus with the pump's bolus increment (#700)
* Format Loop's recommended bolus with InsulinMetric
The Loop device status parser formatted Rec. Bolus with a hardcoded
String(format: "%.2fU"), while the OpenAPS/Trio parser routed the value
through InsulinMetric. The same recommendation rendered as "0.00U" on a
Loop URL and "0" on a Trio URL. Use InsulinMetric on both paths.
* Format Rec. Bolus with the pump's bolus increment
Rec. Bolus was formatted two different ways: the Loop parser used a
hardcoded String(format: "%.2fU"), while the OpenAPS/Trio parser went
through InsulinMetric, which drops to one fraction digit below 10 U. The
same recommendation rendered as "0.00U" on a Loop URL and "0" on a Trio
URL.
Both parsers now use InsulinFormatter, which derives its fraction digits
from the pump's reported bolusIncrement, so the value is shown at the
precision the pump can actually deliver. The unit suffix is dropped to
match the other insulin rows in the info table.
Clear the info row and deviceRecBolus when the device status carries no
recommendation, so a stale value can't linger in the info table, the
Rec. Bolus alarm condition, or the Live Activity.
* CI: Bump dev version to 6.2.17 [skip ci]
* Remove pre-iOS 18 Live Activity code paths (#711)
* Remove pre-iOS 18 Live Activity code paths
With the deployment target now at iOS 18, delete the legacy fallbacks and
availability guards that only ran on older systems:
- LiveActivityManager: remove startIfNeededLegacy and attemptLegacyRenewal
(the Activity.request() start/renewal paths), unwrap the iOS 17.2
push-to-start guards, and drop the now-redundant @available attributes.
- LoopFollowLiveActivity: collapse the dual ActivityConfiguration into the
family-adaptive layout and inline contentMargins.
- StorageCurrentGlucoseStateProvider: the renewal overlay is driven solely
by laRenewalFailed.
- EKEventStore: request full calendar access directly.
- RestartLiveActivityIntent: drop the redundant iOS 16.4 availability
attributes.
Also delete the orphaned duplicate LoopFollow/LiveActivity/RestartLiveActivityIntent.swift,
which was not referenced by any target and shadowed the maintained root copy.
* Tighten Live Activity comments to describe end state
* CI: Bump dev version to 6.2.18 [skip ci]
* Gate startup on storage readiness for Before-First-Unlock recovery (#710)
* CI: Bump dev version to 6.2.19 [skip ci]
* Add custom alarm sounds (#699)
* Add custom alarm sounds
Users can now import their own audio for alarm tones via the tone picker
(document picker, `.audio` UTType) or by dropping files into the app's
Documents folder from the Files app. Imported files are copied into
`Documents/CustomSounds/`, validated with AVAudioPlayer, and capped at
2 MB / 30 s.
SoundFile is now an enum with `.builtin(String)` and `.custom(UUID)`
cases. Built-in values encode as a bare string so existing stored alarms
keep loading; custom values encode as a keyed object. Static aliases
(`.indeed`, `.wrongAnswer`, …) preserve all existing call sites.
AlarmSound.setSoundFile now resolves built-in via the bundle and custom
via CustomSoundStore, and falls back to the default built-in with a log
when a file is missing (deleted custom sound, etc.) instead of crashing.
* Harden custom alarm sound imports
Validate files dropped into Documents in place before moving them into the
store, applying the same size, duration, and format checks as the in-app
importer. This prevents a user's file from being deleted when it fails the
audio check and stops oversized files from bypassing the caps.
Reset the alarm's sound to the fallback built-in when the selected custom
sound is deleted, and avoid rewriting the index when nothing changed.
* Prevent duplicate custom sound when importing from shared folder
Move the picked file into the store instead of copying when it already lives
in the app's shared Documents folder, so absorbDroppedFiles() doesn't re-ingest
the leftover original as a second entry. Validate before the file operation so a
rejected import never deletes the user's original.
* CI: Bump dev version to 6.2.20 [skip ci]
* Add dismissable top banner for Nightscout, Dexcom, and heartbeat issues (#696)
* Add dismissable top banner for Nightscout and Dexcom errors
Failed BG downloads were only visible in the log viewer, and a failed
Dexcom Share login silently fell back to Nightscout. Users with a bad
token, mistyped URL, or wrong Dexcom credentials saw no data with no
explanation.
Add a banner above the tab view (visible on every tab, pushes content
down) that surfaces these failures with a clear cause:
- Nightscout fetch failures trigger a rate-limited status.json probe
to classify the error (invalid token, token required, site not
found, no network) using the existing NightscoutError messages.
- Dexcom Share errors are mapped to friendly text (account not found,
incorrect username or password, too many attempts), including the
current AccountPasswordInvalid code alongside the legacy SSO_ ones.
The previously silent Nightscout fallback is now shown as a warning.
- Banners clear automatically on a successful fetch or when the
source is deconfigured. Dismissing a banner suppresses that exact
error for 30 minutes; a different error shows immediately.
- Repeated failing retries reuse the same message so the banner does
not re-animate every cycle.
Also fix verifyURLAndToken transport-error classification, which
reported DNS/connection failures as siteNotFound and everything else
as networkError; it now distinguishes them by URLError code, which
also improves the settings status label.
* Add banner warning for frequent Bluetooth heartbeat dropouts
* CI: Bump dev version to 6.2.21 [skip ci]
* Add search to the Menu (#695)
* Add search to the Menu covering Settings sub-items
Make SettingsRoute the single source of truth for the Settings list by
moving each row's title, icon and search keywords onto the route, and
grouping them via menuSections(nightscoutConfigured:). SettingsMenuView
now renders from that data.
Add .searchable to the Menu. Search results are assembled from the same
sources that build the menu — the Settings routes, the tab features and
the static rows — so they stay in sync automatically. A settings result
deep-links straight into its sub-screen; features, log and support-link
results reuse the menu's existing navigation.
* Make bottom-level settings findable in Menu search
Add a leaves list to SettingsRoute describing the individual settings
inside each sub-screen (row titles plus search synonyms), and include
them in the Menu search index after the screen-level items. A leaf
result shows a Settings → <screen> subtitle and opens the screen that
contains the setting, e.g. searching "basal" finds Graph Basal under
Advanced. Info Display leaves are derived from InfoType so they follow
the app's info rows automatically. NavigationRow gains an optional
subtitle to render these results.
* CI: Bump dev version to 6.2.22 [skip ci]
* Add color thresholds to the Info Display (#693)
* Add color thresholds to the Info Display
Individual info rows (IOB, COB, battery, pump battery, TDD, rec bolus,
carbs, and sensor/cannula/insulin ages) can now turn yellow or red when
their value crosses user-set thresholds — a visual cue only, with no
alarm. The concerning direction is fixed per metric (battery colors when
low, ages when high) and in-range values read green.
Replaces the parallel infoSort/infoVisible arrays with a single ordered
[InfoDisplayItem] store (migrated via migrateStep10), and reworks the
Info Display settings into drill-in rows that reorder and open a per-row
detail at the same time.
* Use steppers for Info Display color thresholds, and color the pump row
Threshold entry now uses steppers with a per-type range and step, mirroring
the alarm editors, so values that need decimals (IOB, rec. bolus, the ages)
accept them. The old text fields re-parsed on every keystroke and swallowed
the decimal separator, making fractional thresholds impossible to enter.
Direction, unit, bounds and defaults move into a single InfoColorConfig per
type, so isColorable is simply "has a config".
The pump reservoir row is now colorable, low-is-concerning. Pumps that only
report "50+" count as exactly 50, matching what the reservoir alarm already
does with latestPumpVolume.
* CI: Bump dev version to 6.2.23 [skip ci]
* Add new section to export APNS setting (#714)
* CI: Bump dev version to 6.2.24 [skip ci]
* release.sh: use merge-commit for release PRs to preserve release notes (#687)
* release.sh: use merge-commit for release PRs to preserve notes
The PR-based flow recommended rebase-merge, which rewrites the feature-PR commits to new SHAs on main. That breaks GitHub's auto-generated release notes (only the release author is credited, no PR list) and diverges main from dev. Switch the guidance to 'Create a merge commit' so main retains the original commits and the release notes generate correctly.
* release.sh: trim rationale from release PR body
* CI: Bump dev version to 6.2.25 [skip ci]
* Bug fix: carb dots overlapping SMB dots on the chart (#650)
- Restores pre-PR #305 behavior; no other dot positioning changes
- Carb dots entered alongside an SMB were rendering at the same Y position as the SMB triangle, visually overlapping
- The carb-vs-bolus collision logic was supposed to lift the carb dot above the bolus to BG+70, but only checked for Boluses (not SMB)
- Regression comes from PR #305 (SMB display differentiated from boluses), which split SMBs into their own smbData array. - After the split, findNearestBolusbyTime in processNSCarbs no longer saw SMBs as "nearby treatments," so carb-near-SMB fell through to the default BG+20 offset: the same Y as the SMB itself.
- processNSCarbs now also looks up smbData.
- A carb within 300s of either a bolus OR SMB lifts to BG+70; otherwise BG+20.
* Migrate main LoopFollow target to a synchronized folder group (#702)
Convert the main LoopFollow/ group from explicit file enumeration to a
PBXFileSystemSynchronizedRootGroup so files are picked up from disk. This
shrinks project.pbxproj from ~2900 to ~880 lines and removes it as a source
of merge conflicts when files are added, moved, or deleted.
Shared LiveActivity/GlucoseConversion sources that also build into the Live
Activity extension are moved to a new Shared/ folder referenced by both
targets. Info.plist is kept out of the copy phase via a membership exception.
Also remove dead code that was on disk but never in the build: a stale
duplicate RestartLiveActivityIntent, four unreferenced orphan files, and a
leftover Core Data model unused since the initial commit.
* CI: Bump dev version to 6.2.26 [skip ci]
* Fix Mac Catalyst build: guard liveActivity keywords case (#715)
* CI: Bump dev version to 6.2.27 [skip ci]
* Remove unneeded UIKit imports (#713)
* Remove unneeded UIKit imports
* Remove duplicate UIKit import in SettingsMenuView
* CI: Bump dev version to 6.2.28 [skip ci]
* Parameterize bundle ID and build workflow for identical sister-repo code (#690)
* Derive Fastlane bundle identifiers from app_suffix
Read app_suffix from LoopFollowDisplayNameConfig.xcconfig in the Fastfile and
build the app and Live Activity extension identifiers from it, replacing the
hardcoded com.<team>.LoopFollow strings. The same Fastfile now produces the
correct bundle identifiers in LoopFollow, LoopFollow_Second and
LoopFollow_Third without per-repo edits.
Update release.sh so the conflict notice no longer lists the Fastfile, which is
now identical across the sister repositories.
* Make swiftformat and build workflow identical across sister repos
Accept any LoopFollow* directory in Scripts/swiftformat.sh instead of only the
exact 'LoopFollow' folder, so formatting runs consistently in LoopFollow,
LoopFollow_Second and LoopFollow_Third.
Drive the variant name in build_LoopFollow.yml's not-configured summary from
${{ github.event.repository.name }} / ${{ env.UPSTREAM_REPO }} and drop the
variant name from two comments, so the workflow no longer hardcodes 'LoopFollow'
in cosmetic text. The only intentional per-repo differences left are the cron
minute and UPSTREAM_REPO.
* Keep swiftformat folder guard intact
The exact-'LoopFollow' folder check is intentional: it prevents swiftformat from
reformatting source files when the tree is checked out as a sister repo
(LoopFollow_Second / LoopFollow_Third), which would create spurious diffs. The
script stays identical across all three repos and skips by folder name at
runtime.
* Sync sister repos by mirroring instead of patching
Replace the git diff / git apply patch flow in release.sh's update_follower
with an rsync --delete mirror of the release tree into each sister repo,
followed by re-applying that instance's overlay (app_suffix, display_name,
UPSTREAM_REPO, cron) and renaming the workspace. The per-instance values are
derived from the sister directory name; only the staggered cron minute is
passed in.
Mirroring makes each sister an exact copy of the release tree except for the
overla…
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.