straw/docs/sulkta/RECON.md

NewPipe recon — 2026-05-23

Clone: /root/build/newpipe-recon/ (shallow clone of upstream master 94005cf).

TL;DR

NewPipe is mid-migration to Kotlin Multiplatform + Compose Multiplatform, but the new app is at literal day-zero. Of 4 modules, only :app (the legacy Android app) actually works. :shared has 892 LOC of scaffolding but zero screens. :desktopApp is an 18-line stub. :iosApp has 0 lines of Kotlin.

The "fork the legacy app and modernize it" angle is dead — they're modernizing. The actual opening is: pick up the abandoned-looking KMP scaffold and execute the next year of work in a day's worth of vertical-slice ambition. Bake in SponsorBlock + RYD as first-class while we're at it.

Module breakdown

Module	LOC	Kt files	Java files	Status
:app (legacy NEWPIPE_APPLICATION_ID_OLD)	82,427	173	295	What users run as v0.28.6
:shared (KMP, NEWPIPE_APPLICATION_ID_NEW)	892	21	0	Hello-world stage, zero screens
:desktopApp	18	1	0	Main.kt stub only
:iosApp	0	0	0	Xcode shell only

:shared current state

shared/src/commonMain/kotlin/net/newpipe/app/ contains:

• App.kt — @Composable fun App(...) that sets up Koin + AppTheme {} with an empty body. The KMP app literally renders nothing.
• Constants.kt — empty file-level constants.
• navigation/NavDisplay.kt — wraps androidx-nav3's NavDisplay, but entryProvider { } block is empty. No screens registered.
• navigation/Screen.kt — sealed interface with zero subclasses.
• di/KoinApp.kt — bare @KoinApplication object KoinApp with no modules.
• composable/TopAppBar.kt — basic top bar.
• theme/ — Color, ServiceTheme, Theme. Solid Material 3 scaffolding.
• preview/ThemePreviewProvider.kt — Compose preview helper.
• di/settings/SettingsModule.kt — multiplatform settings DI (separate commonMain interface + androidMain/jvmMain/iosMain impls).

That's it. The entire new app is theme + navigation primitive + Koin stub.

:shared/androidMain

• ComposeActivity.kt — Android entry that hosts the Compose UI.
• Constants.kt — Android-specific constants.
• extensions/Context.kt — extension utilities.

Stack (already chosen by NewPipe, modern as of 2026)

Layer	Pick	Version	Notes
Language	Kotlin	2.3.21	Bleeding edge
UI	Jetpack Compose + Compose Multiplatform	1.11.1	Material3 1.11.0-alpha07
Build	AGP	9.2.1	Latest
Nav	androidx-navigation3	1.1.1	Type-safe, KMP-friendly
DI	Koin	4.2.1	Annotation-driven via koin-plugin
Async	kotlinx.coroutines	1.11.0	Plus kotlinx-coroutines-rx3 bridge to legacy
Serialization	kotlinx.serialization	1.11.0	JSON, parcelable replacement
Image loading	Coil 3	3.4.0	KMP-native via coil-network-okhttp
HTTP	OkHttp	5.3.2	Android, used by extractor
DB	Room	2.8.4	Android-only, KMP-incompatible (legacy app)
Player	ExoPlayer 2.19.1	—	Stale. Should be androidx.media3 ≥ 1.4
Crash	ACRA	5.13.1	Send-to-self crash reporter
Leak detection	LeakCanary	2.14	Debug builds only

Critical dependency: NewPipeExtractor

• teamnewpipe-newpipe-extractor = "v0.26.2" pulled from JitPack
• JVM-only. Uses Jsoup + nanojson + Java HTTP.
• Cannot live in commonMain for KMP. Lives in :app and would need androidMain / jvmMain source sets in :shared.
• This is why their KMP migration is slow — porting the extractor to KMP is the actual big yak. They haven't started it.

For our fork: same constraint applies. Android-only target for day-1; KMP extractor port is a separate (multi-week) project.

What's actively being deprecated

NewPipe's own codebase has TODOs and migrations in flight:

• RxJava 3 → coroutines (bridges via kotlinx-coroutines-rx3 show partial)
• ExoPlayer 2 → Media3 (still on legacy)
• Fragment + ViewBinding → Compose (the LeakScope findings live in this layer)
• Material 1 → Material 3 (mid-migration; material = "1.11.0" # TODO: update)
• KSP2 incompatibility flagged on statesaver
• Apache commons in app/src/main/java/org/apache/commons/ — vendored, would go away in KMP world

Where to hook SponsorBlock + Return YouTube Dislike

For a from-scratch new app:

• SponsorBlock: ExoPlayer/Media3 player listener that consumes a list of segments fetched at video resolution time. Skip via player.seekTo() on onPositionDiscontinuity/period-update callbacks. API: SHA-256 prefix lookup at sponsor.ajay.app/api/skipSegments/<prefix>.
• Return YouTube Dislike: HTTP GET to returnyoutubedislike.com/votes?videoId=<id>. Render in video detail screen next to the like count. No player integration needed.

Both clients can live in :shared/commonMain using Ktor client (KMP HTTP). That makes them iOS/desktop-ready once an extractor port exists.

Existing forks doing the same thing

Surfaced from NewPipe issue thread #13512:

• PipePipe — fork with SponsorBlock + RYD. Active.
• Tubular — fork with SponsorBlock + RYD. Active.
• YouPipe — fork with extra features.
• SkyTube — separate Android YouTube app, not NewPipe-based.

So this isn't novel territory. We're doing it for the build experience (element-x energy), not to fill a market gap.

Differentiators we could pursue

If we want to be more than "another fork":

1. Compose-native from day 1 (vs. PipePipe/Tubular which are still on Fragment+XML legacy). Smaller, faster, easier to maintain.
2. Media3 ExoPlayer (vs. NewPipe's legacy ExoPlayer 2).
3. Ktor-based clients for SB/RYD in commonMain — KMP-ready for the day someone ports the extractor.
4. First-class settings sync via our existing Sulkta auth (Authentik OIDC) for cross-device subscriptions/history. None of the existing forks do this.
5. Sulkta-Coop signing cert — joins the rest of our app family with consistent signature.