APK is 27.7 MB since v1.3.2 (was 5.1 MB): ML Kit barcode scanner ships four ABIs

[DuarteSantos8]

Migrated from GitLab issue #51, opened by the openGym maintainer (project bot) on 2026-09-05. openGym moved back to GitHub on 2026-09-11; the GitLab thread is closed and points here.

The Android build grew from 5.1 MB (v1.3.1) to 27.7 MB (v1.3.2 and v1.3.3). unzip -l on openGym-1.3.3.apk:

file	size
classes.dex	9.5 MB
lib/x86/libbarhopper_v3.so	6.1 MB
lib/x86_64/libbarhopper_v3.so	5.9 MB
lib/arm64-v8a/libbarhopper_v3.so	4.9 MB
lib/armeabi-v7a/libbarhopper_v3.so	3.2 MB

libbarhopper_v3 is Google ML Kit's barcode model, pulled in by @capacitor-mlkit/barcode-scanning for the gym membership QR check-in (a797729, v1.3.2). Nothing is wrong with the feature; the cost is 20 MB of native code for four ABIs plus the ML Kit classes, on a phone that only needs one ABI.

Options, cheapest first:

1. ABI splits / abiFilters in frontend/android/app/build.gradle: ship arm64-v8a (and armeabi-v7a if we still care about 2016 phones); drop x86/x86_64 — no real phone runs them. Saves ~12 MB on its own.
2. Unbundled ML Kit variant (barcode-scanning via Google Play services instead of barcode-scanning-bundled model): the model downloads on first use, the APK loses libbarhopper entirely. Trade-off: first scan needs Play services + network, which breaks on de-Googled phones — so probably not for openGym.
3. ProGuard/R8 minify for release: classes.dex 9.5 MB is unminified. Needs a keep-rule check for Capacitor plugins.

Worth doing before the in-app updater (v1.3.3) makes people download this on every release.

[DuarteSantos8]

Still open. Plan: ABI filters (drop x86/x86_64) first, which alone saves ~12 MB, then look at the unbundled ML Kit variant. Not in v1.3.6.

[DuarteSantos8]

Fixed in v1.3.9: the release APK only ships ARM code (arm64-v8a and armeabi-v7a), which should make it about 12 MB smaller.

Released in v1.3.9: https://github.com/DuarteSantos8/openGym/releases/tag/v1.3.9