[Maintenance] Reduce code and mobile bundle size with measured budgets

[0xSero]

Goal

Reduce Litter's maintained code, release bundle size, build/runtime cost, and duplicated platform behavior without weakening iOS/Android parity or deleting code on textual-reference guesses.

This builds on merged PR #192, which removed dead Rust crates and folded the mobile bridge. The next phase must be measured and split into small, independently reversible PRs.

Current source/package baseline (2026-08-04 main)

• 1,009 tracked files; roughly 79k Swift, 48k Kotlin, and 73k Rust lines before generated/submodule exclusions are refined.
• 51 direct Android dependency declarations, 7 iOS package declarations, and 5 in-repository Rust manifests across the mobile/kittylitter build.
• 53.5 MB of tracked content. The largest release inputs include two iOS animated home-cat PNGs (~10.5 MB total) and two Android animated WebPs (~8.7 MB total), plus duplicated cross-platform tip/transmission art.

These are repository measurements, not installed IPA/AAB proof. The first PR must establish authoritative packaged-artifact baselines.

Guardrails

• Preserve behavior and all named mobile parity surfaces.
• Shared session/thread/streaming/hydration/discovery/auth/voice policy stays in Rust; do not trade Rust deletion for duplicate Swift/Kotlin state machines.
• Generated bindings, third-party submodules, debug-only harnesses, SwiftUI type inference, and package/release inputs require compiler/build evidence before deletion.
• Report compressed and installed/uncompressed sizes separately. Source LOC is diagnostic, not a success metric by itself.
• Each optimization PR must include before/after measurements, tests, and rollback scope.

Workstreams

1. Establish budgets and regression reporting

• Record IPA/AAB/APK, per-ABI JNI, iOS staticlib/framework, asset, and dependency contribution sizes in CI.
• Add warning/failure budgets with an explicit update process and a human-readable diff artifact.
• Add cold launch, reconnect, thread hydration, first render, and realtime negotiation timing probes where not already covered by perf: instrument the mobile↔host paths so latency claims are measurable #189.

2. Remove package weight

• Re-encode oversized animated assets with visual/alpha/frame-timing comparison and device rendering proof.
• Ensure release packages exclude screenshots, debug harnesses, unused architectures, stale frameworks, symbols, and development-only resources.
• Audit direct dependencies for unused or replaceable libraries; remove one dependency per focused PR with platform builds.
• Use platform release optimizations deliberately: dead stripping and resource rules on iOS; R8/resource shrinking and ABI/app-bundle splits on Android, with startup/JNI tests.

3. Simplify maintained code

• Find duplicated Swift/Kotlin normalization/reducer/cache logic and move only truly shared policy into the existing Rust crate.
• Delete superseded compatibility paths only after current protocol/version usage is proven absent.
• Reduce bridge surface area and generated bindings by moving direct operations to AppClient and keeping AppStore composite-only.
• Split oversized files/modules by coherent responsibility when it reduces change coupling; do not churn files for line-count aesthetics.

4. Improve runtime/build efficiency

• Profile before optimizing: hydration/reconnect, markdown/stream rendering, asset decode, database/index work, UniFFI crossings, and build hot spots.
• Bound unbounded work and remove repeated parsing/projection only when traces show it matters.
• Keep fast local lanes and CI cache behavior reproducible.

Acceptance criteria

• CI publishes authoritative package/component size reports for iOS and Android and rejects unexplained regressions over agreed budgets.
• Every cleanup PR includes before/after bytes and relevant behavior/build/device proof.
• No user-facing parity regression; shared behavior remains one Rust implementation with thin platform projection.
• Release IPA/AAB installed size and cold-start/reconnect timings improve from the recorded baseline, with targets set after measurement.
• Deleted code/assets/dependencies have compiler, build graph, package inspection, and where relevant device evidence.
• The final report lists net maintained LOC, dependency count, package size, launch/reconnect latency, and all intentionally retained complexity.

Initial PR sequence

1. Size-report script + CI artifacts/budgets (measurement only).
2. Animated home asset compression with frame/alpha/visual and device proof.
3. Release-package content audit/removal.
4. Dependency removals and shared-state simplifications, one behavior area per PR.

Keep this issue open only while it is an active measured program; close or split it once the budgets and remaining work are owned by narrower issues.

[0xSero]

First measured implementation is open as #208. It re-encodes the two animated home assets once per platform at quality 90, preserving alpha/frame count/timing, and removes 9,885,936 tracked bytes (iOS -5,822,830 bytes; Android -4,063,106 bytes). Local validation includes Android resource merge, full iOS simulator build/install/launch, visible cat rendering, and two-frame motion proof. Physical Android rendering remains an explicit acceptance item before the draft is ready to merge.

[0xSero]

Second measured reduction is in draft PR #214: four Android tip-cat assets are losslessly 2,350,607 bytes smaller and the clean debug APK is 2,350,592 bytes (3.98%) smaller. Pixel equivalence, packaging, unit tests, and assemble pass; installed Android rendering remains the acceptance gate.

[0xSero]

test

[0xSero]

Agreed direction, with the audit's caveat: measure before cutting. The current footprint is dominated by the Rust static lib (libwebrtc + russh + codex crates) and the Ghostty renderer; Swift/Kotlin source is a small share, so source-level pruning moves the needle less than feature flags on parked paths (nearby-Mac pairing, legacy alleycat markers). Proposal: add measured budgets (per-binary + per-feature) to the release lane first, then act on what the numbers say. Keeping open as a maintenance epic.

[0xSero]

Status: groundwork landed — the render-stall and startup-cost work in #319 removed several hot-path allocations, and the realtime rework deleted the retired codex-ios-audio crate + AecBridge on both platforms (measured budgets still needed). What this issue still needs: (1) agreed budget numbers (app binary, xcframework size, cold-start) wired into CI as failing thresholds rather than a dashboard, (2) a delete pass over parked/unreachable code using the risk register as the source of truth rather than textual-reference guesses, (3) Android equivalent of the iOS size report. No date committed yet — leaving open as the tracking issue.