Skip to content

final-1.2.6.md

Latest commit

 

History

History
105 lines (82 loc) · 4.65 KB

final-1.2.6.md

File metadata and controls

105 lines (82 loc) · 4.65 KB
Raw

Version 1.2.6 Performance And Footprint Results

This closeout compares version 1.2.6 at main commit d532d52 with the pre-refactor baseline at e38ffba. Measurements use the same build commands, supported ABIs, and two physical Android acceptance devices described in baseline-e38ffba.md.

Source health

Metric Baseline 1.2.6
First-party files checked 180 451
Files above 500 lines 38 0
Classes above 500 lines 5 0
File allowlist entries 38 0
Class allowlist entries 5 0

The final count includes the new scale and list-window tests. Run npm run check:source-size to enforce the empty allowlists.

Web build

Command: npm run web:build

Artifact Baseline bytes 1.2.6 bytes Change
Initial application JavaScript 252,434 206,594 -18.2%
Vue/vendor JavaScript 140,548 140,604 +0.0%
Lazy MapLibre JavaScript 803,071 803,071 0.0%
All files under dist/assets 1,744,131 1,710,181 -1.9%

The initial-bundle target of at least 15% is met. Platform-specific entrypoints keep native and mock client implementations out of production web builds.

Android release build

The signed 1.2.6 APK and AAB contain all supported ABIs and use 16 KB-aligned native libraries.

Artifact Baseline bytes 1.2.6 bytes Change
arm64-v8a libreticulum_mobile.so 13,379,216 10,283,472 -23.1%
armeabi-v7a libreticulum_mobile.so 9,955,548 7,509,632 -24.6%
x86_64 libreticulum_mobile.so 13,646,704 11,441,624 -16.2%
Signed universal APK 33,127,536 33,677,303 +1.7%
Release AAB 32,016,205 32,632,288 +1.9%

The native-library target of at least 10% is exceeded for every ABI. The universal APK/AAB target is intentionally not claimed: preserving Android 16 KB page compatibility changes native packaging and offsets the raw library reduction. Removing that compatibility or a supported ABI solely to meet the container-size target would remove supported functionality.

Scale and responsiveness

Executable tests cover:

  • 1,000 peer projections;
  • 10,000 message projections;
  • 1,000 event and 1,000 telemetry projections;
  • 100 checklists with 200 tasks each;
  • a 1,000-command priority burst;
  • normal-queue saturation while local, SOS, acknowledgement, and lifecycle capacity remains available;
  • dense and sparse 1,000-position map clustering below the 50 ms long-task ceiling.
  • shared 200-row render windows for peers, conversations, messages, events, action messages, checklists, and checklist tasks, with controls that retain access to every record.

The projection scale test enforces p95 below 500 ms. Bounded queues reject excess normal work with a typed timeout, and network work remains asynchronous. The list-window unit tests verify that 1,000-row collections never render more than 200 rows at once and that final partial pages remain reachable. Message-thread reads use one additive composite SQLite index covering their conversation filter and stable timestamp/message ordering. An EXPLAIN QUERY PLAN regression test verifies that SQLite selects the index.

Physical-device sample

The signed release APK was installed with adb install -r on both devices.

Device Cold launches Average Warm launches Average PSS Jank Critical logs
Pixel 7, Android 17 / SDK 37 502, 529, 498 ms 509.7 ms 126, 23, 30 ms 59.7 ms 456,070 KB 4/254 (1.57%) 0
Samsung SM-G950W, Android 9 / SDK 28 920, 551, 516 ms 662.3 ms 96, 75, 66 ms 79.0 ms 287,763 KB 68/195 (34.87%) 0

The first Samsung cold launch is an outlier after replacement install; its next two launches are close to baseline. Frame jank improved on both devices, including 62.50% to 34.87% on the Android 9 device. Memory is a point-in-time sample with retained production history and active Reticulum traffic, so it is not treated as a controlled leak comparison.

Paired-device acceptance

The Pixel and Samsung both reached READY and completed retained-history and fresh-workflow checks for discovery/announces, direct LXMF messaging, delivery acknowledgement, event command acceptance, checklist history, MapLibre, background/foreground restoration, manual sync, offline transport failure, automatic reconnect, and SOS activation/cancellation. The controlled Samsung offline cycle kept the core runtime ready, reported the failed TCP interface, and recovered after restoring its original Wi-Fi and mobile-data settings.

No crash, ANR, JNI failure, or uncaught exception was found in the acceptance log samples.