Skip to content

Merge upstream 2026-11-20 - #44

Merged
tiper merged 9 commits into
mainfrom
upstream-merge-2026-11-20
Nov 20, 2026
Merged

tiper merged 9 commits into
mainfrom
upstream-merge-2026-11-20

Conversation

@tiper

@tiper tiper commented Nov 20, 2026

Copy link
Copy Markdown
Owner

Merges PlxNative main through 70cbda9 (plxnative#1478, "URI drivers report their own display evidence"): 7 upstream commits (plxnative#1348, plxnative#1462, plxnative#1470, plxnative#1471, plxnative#1474, plxnative#1477, plxnative#1478). No upstream release is pending, so PLXNATIVE_BASE stays 0.11.0-dev.

Conflicts resolved

  • Rename the bottom layer crate plx_platform to plx_os plxnative/plx-native#1474, plx_platform renamed plx_os (25 files): kept FinNative's side and renamed it (AC3 conversion in ff.rs/native_executor.rs/transcoder.rs, Jellyfin ids in poster keys, AddUser, AudioConversion, the Jellyfin login form, PACK_APP_VERSION, FinNative's app ids in storage tests, PLXNATIVE_BASE in About, finstate DB8 keys). tools/rename-os-layer.py then renamed the FinNative-only crates and files (plx_jellyfin, */finnative.rs, jellyfin.yml); --check reports 0 old-name sites. jellyfin keeps its place in ci/module-layers.ini and the staticlib test lists.
  • Open Watch Party: the protocol crate, its checks and eight conformance tables plxnative/plx-native#1462 (Apache-2.0 plx_owp): README and CONTRIBUTING keep FinNative's wording and name the Apache-2.0 crate.
  • ci/proofs regenerated; ci/readonly_generated.py is upstream's with FinNative's os/src/i18n/mod.rs hash re-added.
  • Version files keep FinNative's version.

Gates (committed tree)

  • make check-nodef: ok.
  • cargo test --lib -p plx_jellyfin: 162 passed, 0 failed.
  • make check: 14049 passed, 39 failed, all in tests that rely on unwritable directories or the network (plex session persistence, media prefs characterisation, app session, telemetry persistence, curlio), which fail when the suite runs as root in the build container. CI is the gate for these.

Generated by Claude Code

GLinnik21 and others added 9 commits November 19, 2026 15:02
…plxnative#1348)

* Detail and Home hero: name HDR and Dolby Vision beside the resolution

The identity line said "4K" for an HDR or Dolby Vision film and nothing more: the poster caption's Video quality part was the only place the app worded the dynamic range. The detail page now draws one more keyline chip right after the resolution, "Dolby Vision" for a file that carries it and "HDR" for any other HDR video (the facts the page already holds, of the same default version the resolution chip describes). Home's hero draws "HDR" in the same place when its row says so; a listing row knows only that its video is HDR, so a Dolby Vision title reads "HDR" there, as it does under a poster.

Both go through the shared identity row, so the chip takes part in the text run's budget and elision like the others.

Signed-off-by: Tiago Pereira <1698241+tiper@users.noreply.github.com>

* Home hero: a Dolby Vision title says so, as its detail page does

The hero read "HDR" for a Dolby Vision title because a listing row knew only that its video is HDR. The row now also carries whether that stream is Dolby Vision, along the lane the HDR fact already travels (ListingFacts, PmsMovie), and the hero words both through fmt::dynamic_range.

Signed-off-by: Tiago Pereira <1698241+tiper@users.noreply.github.com>

* Hero identity: its doc names the dynamic-range chip

Signed-off-by: Tiago Pereira <1698241+tiper@users.noreply.github.com>

---------

Signed-off-by: Tiago Pereira <1698241+tiper@users.noreply.github.com>
…lxnative#1471)

Adds the review-pr skill: a local review of someone's pull request, posted as
a formal GitHub review from the owner's account after the owner has read the
draft.

It follows umputun's review workflow and writing rules (cc-thingz,
plugins/review) around this repo's own analysis. It reads the conversation,
bot reviews and CI first, fetches the PR into a detached worktree, and runs
pr-reviewer (and fw-compat-reviewer) on origin/main...head. Tests run only for
authors with write access; a fork's PR gets CI's results unless the owner says
yes after the build inputs are read. The draft is lgtm or a short numbered
list, with findings inline. One review is posted on the owner's click, pinned
to the reviewed SHA and recorded with pr-land.py reviewed; approving never
merges.
The bottom layer crate, the one that says what the television is and keeps, is renamed so the platform crates the plan in discussion plxnative#1460 adds (webOS, the desktop simulators, Samsung Tizen) can take the word "platform":

- `plx_platform` becomes `plx_os`, `rust-modules/platform/` becomes `rust-modules/os/`, the layer `[platform]` becomes `[os]`, the fence keyword `[port X]` becomes `[platform X]` and `SRC_PLATFORM` becomes `SRC_OS`. The inner `tv` module keeps its name. Nothing behaves differently.
- `tools/rename-os-layer.py` (stdlib, `-I`) makes the rewrite and stays for the open PRs and the release lines. It runs forward (idempotent), `--reverse` on a tree, `--reverse-patch FILE` on a backport's diff, `--check` (every old-name site in three spellings plus prose, separator-list and table-cell uses, classified or failing) and `--equivalence BASE` (HEAD against the script applied to BASE, and back). `ci/test_rename_os_layer.py` (20 cases) runs in `ci/check-python-steps.txt`.
- An open PR picks it up by merging `main`, running the script and `--check`, and committing.
- Both reviewed certificates are re-pinned: `ci/readonly_generated.py` 1b3b34fa… to 55e6b3ac… (its current i18n snapshot 2c978e3a… to 867bb410…) and `ci/readonly_dependencies.py` 915ab64b… to 47228bf8…. Every changed hash follows from the rename: names, paths, manifest and lock hashes, and certified sources whose only edits are comments or the package name. `Cargo.lock` has the same lines with `plx_os` re-sorted; no version moves.

Measured before and after on one Mac: `RUST_INPUTS` (1500 files) and `STORAGE_INPUTS` (16) are identical after the rename, the make query targets print the same bytes, and no build-bench row is slower beyond its spread. Three boots on the dev set under `--guest --mock` differ only in the image-cache cap, by half the space the shorter crate name saves in the binary. The cargo seed key changes once. Crash reports now name `plx_os::` frames, so a Sentry issue whose top frames are in this crate can regroup after the next release.
…store mode (plxnative#1477)

Samsung's hosted app has no credential store it can trust across launches, so the session layer gains a mode that keeps no credential, for the Tizen platform to choose (step W3a of the platform plan, discussion plxnative#1460). Nothing selects it yet; webOS and desktop behave and write exactly as before.

- `secure::StoreMode { Durable, MemoryOnly }` is published once with `publish_mode` (default `Durable`); tests use `set_mode_for_test` under the serial test lock, reset on drop.
- Under `MemoryOnly` the session cache cell is the only authority for reads, guarded edits, revocation and clear: no canonical commit, no sealed or plaintext legacy fallback, no revocation marker. The gate sits in `read_live_locked`, `save_locked_with_authority` and `clear_for_erase`, above the test-file and target-arch branches; the logic is in `session/memory_only.rs`.
- A sign-in there completes as `CompletionOutcome::Ephemeral` (`PersistOutcome::Memory`, `DiskOutcome::Memory`): the auth owner releases the fresh-login handoff and goes Home, with no saved-login claim and no warning. `LiveWrite::classify()` stays the one decision site.
- `install_preferences` v2 adds `client_id` beside language and interface resolution and is written only under `MemoryOnly`; load accepts v1 and v2, and webOS and desktop keep writing a byte-identical v1. Sign-out keeps the file, "Delete all local data" removes it, and every other setting is forgotten each launch. A credential file an earlier build left is never read, written or deleted.

Tests cover a whole-data-directory scan that finds no byte of the mock tokens after a memory-only sign-in (a `Durable` control does write them), edit, sign-out and erase, a fresh process over planted credentials, a forced canonical failure that reaches the legacy fallback only under `Durable`, v1 byte-identity, the owner's handling of `Ephemeral`, and an end-to-end owner flow through the real session layer.
…o the LG driver (plxnative#1478)

The URI route no longer names LG's display binding, so a URI driver on another platform can report its own display instead of faking an ACB-shaped answer (step W3b of the platform plan, discussion plxnative#1460). webOS behaves as before.

- `UriDriver::display()` returns a `DisplayEvidence`: whether the display is ready, whether the driver can place a preview rectangle, and an opaque tag the boot log prints. `ProductEvidence` carries it in place of `display_exported`, `binding` and `preview_rectangle`, joined to the driver's `display()` at install time.
- `Binding` and the `VP_EXPORTED` comparison move into the LG driver (`plx_play_uri::backend`), which derives its evidence from them. `UriNotReady` is `DisplayUnavailable` or `Driver(&'static str)`; `AcbHandle` is gone, and the ACB decline wording is unchanged.
- `OriginalRequest.binding` is deleted. `ProductQualifier` (LG's multi-audio rule on the exported window and the opt-in check) moves from `plx_media` to `app/adapters/backends.rs`, built from the binding the composition ends with, after the exported-window override; `plx_media` keeps a test-only `CarrierQualifier`.
- The composition builds the driver first, reads its `display()` and joins it to the product's callable and certified capabilities (`UriProduct`, `install_uri_product_with_ports`). `ready_to_start` stays an on-thread check asked at the offer.
- The boot log gains `uri: display path=<tag> ready|unavailable (<reason>)`; on the dev set it reads `uri: display path=acb ready`, and the rest of the boot and the exit status match main.

New tests pin today's `Offer::rank` order across a Feed and a URI driver (Forced PMS-only staying `Unqualified`), a driver certified for everyone being admitted with no opt-in, the qualifier's decline reasons and its following the exported override's binding, the production product certifying nothing, and the boot line.
…-11-20

# Conflicts:
#	CONTRIBUTING.md
#	Makefile
#	README.md
#	ci/module-layers.ini
#	ci/proofs/platform-generated-enums.json
#	ci/proofs/serde-json-immutable.json
#	ci/readonly_dependencies.py
#	ci/readonly_generated.py
#	ci/test_no_host_staticlib.py
#	rust-modules/os/src/storage/state.rs
#	rust-modules/play-feed/src/ff.rs
#	rust-modules/play-feed/src/native_executor.rs
#	rust-modules/plex/src/plex/session/migration_tests.rs
#	rust-modules/plex/src/plex/session/prefs_characterisation_tests.rs
#	rust-modules/plex/src/plex/session/tier1_characterisation_tests.rs
#	rust-modules/plex/src/plex/transcoder.rs
#	rust-modules/screens/src/account_menu.rs
#	rust-modules/screens/src/legal.rs
#	rust-modules/screens/src/login.rs
#	rust-modules/screens/src/preferences.rs
#	rust-modules/src/app/adapters/poster.rs
#	rust-modules/src/app/bridge.rs
#	rust-modules/src/app/run.rs
#	rust-modules/telemetry/src/telemetry/persistence.rs
#	rust-modules/ui/src/decision_alert.rs
@tiper
tiper merged commit 191123a into main Nov 20, 2026
14 checks passed
@tiper
tiper deleted the upstream-merge-2026-11-20 branch November 20, 2026 06:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants