Repository navigation
Update to latest version of maplibre-compose to reduce memory leaks - #1180
Merged
Merged
Conversation
…up again The monkey can leave the app where LaunchHome couldn't get back to Home: in Street Preview, whose Home has no search bar and isn't left by Back; in the guided tutorial, whose dialogs only Continue closes; or with its window shrunk to a pop-up or left blank by the phone's freeform modes. Restore then failed and the phone was set up again, wiping the app, about once an hour on each phone, so no process ran long enough to show whether its memory levels off. LaunchHome now leaves Street Preview, and continues the tutorial's dialogs and cancels it from the menu. If Restore still fails, the farm restarts the app and runs it once more before setting the phone up again. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ex6Fdp5ftuhE1ZEyhZb7WQ
…live uptime and memory - local_apk tests a locally built APK in place of the CI releases, on every phone or one. It's usually signed differently, so whatever is on the phone is uninstalled first; unsetting it goes back to the releases the same way. - When Restore fails after the monkey soak, the app is first brought back to the front and Restore retried, keeping its process running: Home and am start (MainActivity is singleTask), with a floating task made briefly unresizeable so it leaves freeform, and dark mode switched over and back, which brought back the Moto's blank grey window, whose Compose content had stopped drawing. Only then is the app restarted. - pause_on_restore_failure leaves a phone as it is before the farm would restart its app, until resume-<serial> exists or pause_minutes pass. - LaunchHome leaves Street Preview from inside its Back loop too, as backing out of another screen can land on it; Restore starts by cancelling a guided tutorial left running without a dialog showing (CancelTour.yaml). - The report's devices table shows how long the app has been running and its PSS now, read every minute with the battery, else as at the newest cycle that recorded them: cycles the farm's own restarts stop record neither. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ex6Fdp5ftuhE1ZEyhZb7WQ
… 1.12 On the bench the new map engine used about 50 MB less than 0.13.1 over the same walk, giving memory back as the map moved on where the old engine kept it. - Kotlin 2.3.10 -> 2.4.21, Compose Multiplatform 1.11.1 -> 1.12.2 (the Compose plugin's version now follows composeMultiplatform; it was pinned separately, which stopped the Res class being generated), KSP 2.3.12, and the forced kotlin-daemon-client follows the Kotlin version. - maplibre-compose 0.13.1 -> 0.20.0, on MapLibre Native FFI with the OpenGL runtime (the Cubot has no Vulkan; the merged manifest declares no Vulkan feature). The MapLibre Android SDK, its OpenGL substitution and the unused annotation plugin are gone, and with them the SDK's prefab package, so the prefab comments are now history. - MapContainerLibre rewritten for rememberMapState, MapInteractions (pan only full screen, no rotate or tilt), animateCamera/animateCameraToBounds and no overlay. Its parameters are unchanged. While a beacon is being edited its marker is drawn over the centre of the map rather than as a layer following the camera, so it doesn't lag a pan; it's laid out to match the layer: the pin's point at the centre, at 1.2x, with its number. - DestroyMapWhenDisposed is gone: the map's state closes itself when it leaves the composition. - iOS: MapLibre Native now comes in through the Shared framework, so the MapLibre SwiftPM package is dropped and the system libraries it needs are linked, as maplibre-compose's iOS setup lists them. Checked on a Pixel: Home, full screen map (draws straight away), panning only when full screen, editing a marker's location, and an offline map's extract outline and fit. Not checked: the iOS build, route waypoint markers, and pinch-to-zoom inside a scrolling screen. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tile source was resolved again for every new map centre, which on Home is every location update: listing the extracts and opening and reading each of them, on the main thread, about once a second while walking. Which extract covers the map only changes when the centre moves into another z14 tile, so it's now remembered against that tile. On a phone standing still, 40 seconds of location updates resolved it once instead of about 17 times. On iOS the whole style JSON was also rebuilt for every location, and then compared in full by maplibre-compose against the one it had. It's now built only when the tile source or the colours change, as on Android, and the layer overrides are made inside that on both. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The red arrow, a symbol layer that jumped once a second, is replaced by maplibre-compose's location indicator: a dot with a wedge pointing the way the user faces, which animates between fixes and turns the short way past north. It's the same red in every theme, as the arrow was, and the arrow's drawable, now unused, is removed. It's given the location and heading the rest of the app is using - the head tracker's heading when one is paired, otherwise the compass - rather than maplibre-compose's own providers, which would make their own platform requests and know nothing of the head tracker, GPX replay or a simulated location. On Home it also shows how accurate those are: a circle for the fix's horizontal accuracy, which HomeState already had, and a sector either side of the heading for its accuracy, which HomeState now keeps too. That's the fused orientation's heading error, iOS's heading accuracy, or the head tracker's; the plain Android sensor provider reports none, and so shows no sector. They reach the map through new optional userLocationAccuracy and userHeadingAccuracy parameters on PlatformMapContainer, MapContainerLibre and rememberFullScreenableMap. The other screens don't pass them. Unlike the symbol layer, the indicator takes no part in label placement, so labels are no longer pushed out from under it, and it draws above the beacon marker. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
davecraig
had a problem deploying
to
development
October 23, 2026 17:13 — with 
GitHub Actions
Failure
GitHub Actions
Failure
DocumentationScreens.getHelp makes a ResourceEnvironment for each language through its internal constructor, which Compose Resources 1.12 gave a script qualifier, between the language and the region. Looking for the old four-parameter constructor found none, and the test failed. It now uses the five-parameter one, with no script, as none of the values folders has one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This branch was successfully deployed
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.
No description provided.