trcc-linux 9.8.6__tar.gz → 9.8.8__tar.gz

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: trcc-linux
-Version: 9.8.6
+Version: 9.8.8
 Summary: Linux implementation of Thermalright LCD Control Center
 Project-URL: Homepage, https://github.com/Lexonight1/thermalright-trcc-linux
 Project-URL: Documentation, https://github.com/Lexonight1/thermalright-trcc-linux#readme
@@ -126,7 +126,7 @@ Description-Content-Type: text/markdown
 
 Native Linux port of the Thermalright LCD Control Center (Windows TRCC 2.1.2). Control and customize the LCD displays and LED segment displays on Thermalright CPU coolers, AIO pump heads, and fan hubs — entirely from Linux.
 
-> **This project wouldn't exist without our testers.** I only own one device. Every supported device in this list works because someone plugged it in, ran `trcc report`, and told me what broke. 32 testers helped us go from "SCSI only" to full C# feature parity with 6 USB protocols, 16 FBL resolutions, and 12 LED styles. Open source at its best — see [Contributors](#contributors) below.
+> **This project wouldn't exist without our testers.** I only own one device. Every supported device in this list works because someone plugged it in, ran `trcc report`, and told me what broke. 32 testers helped us go from "SCSI only" to full C# feature parity with 6 USB protocols, 16 FBL resolutions, and 13 LED styles. Open source at its best — see [Contributors](#contributors) below.
 
 > Unofficial community project, not affiliated with Thermalright. Built with [Claude](https://claude.ai) (AI) for protocol reverse engineering and code generation, guided by human architecture decisions and logical assessment.
 
@@ -306,7 +306,7 @@ Set the angle to **90°** (or 270°) in the GUI, then open **Cloud Themes** —
 | **Media** | Video/GIF playback on LCD, video trimmer, image cropper, screen cast (X11 + Wayland), mic audio visualization |
 | **Overlay Editor** | Text, sensors, date/time overlays — font picker, dynamic scaling, color picker |
 | **Hardware Sensors** | 77+ sensors — CPU/GPU temp, fan speed, power, usage — customizable dashboard |
-| **LED Control** | 12 LED styles, zone carousel, breathing/rainbow/static/wave modes, per-zone color |
+| **LED Control** | 13 LED styles, zone carousel, breathing/rainbow/static/wave modes, per-zone color |
 | **Display** | 16 resolutions (240x240 to 1920x462), 0/90/180/270 rotation, 3 brightness levels |
 | **Multi-device** | Per-device config, auto-detect, multi-device with device selection |
 | **Security** | udev rules, polkit policy, SELinux support, no root required after setup |
@@ -364,7 +364,7 @@ Run `lsusb` to find your USB ID (`xxxx:xxxx` after `ID`), then match it below.
 **HID LED devices** — RGB LED control:
 | USB ID | Devices |
 |--------|---------|
-| `0416:8001` | AX120 DIGITAL, PA120 DIGITAL, Peerless Assassin 120 DIGITAL ARGB White, Assassin X 120R Digital ARGB, Phantom Spirit 120 Digital EVO, HR10 2280 PRO Digital, and others (model auto-detected via handshake) |
+| `0416:8001` | AX120 DIGITAL, PA120 DIGITAL, Peerless Assassin 120 DIGITAL ARGB White, Assassin X 120R Digital ARGB, Phantom Spirit 120 Digital EVO, HR10 2280 PRO Digital, Magic Qube, and others (model auto-detected via handshake) |
 
 > See the [full device list with protocol details](doc/REFERENCE_DEVICES.md) and the [Device Testing Guide](doc/GUIDE_DEVICE_TESTING.md) if you have an untested device.
 
@@ -409,20 +409,21 @@ If a probe goes `BAD`, that's the bug — fix in `src/`, re-run, see `PASS`. Eve
 | HID Type 3 | pyusb interrupt | TARAN ARMS |
 | Bulk | pyusb bulk | GrandVision 360, Mjolnir Vision 360, Wonder Vision Pro 360, Frozen Warframe Pro |
 | LY | pyusb bulk (chunked) | Trofeo Vision 9.16 LCD |
-| LED | pyusb HID | All LED segment display devices (12 styles) |
+| LED | pyusb HID | All LED segment display devices (13 styles) |
 
 ## Code Contributors
 
 These folks didn't just report a problem — they opened the editor and sent a fix. On a solo project, a merged pull request is about the most generous thing you can do, and every one of these shipped to every user:
 
+- **[@jphilipb](https://github.com/jphilipb)** — [#215](https://github.com/Lexonight1/thermalright-trcc-linux/pull/215): reverse-engineered and hardware-validated the Thermalright Magic Qube (65-LED segment display) — a device that doesn't even exist in the Windows app — and wired up full support from scratch.
 - **[@Hythera](https://github.com/Hythera)** — [#209](https://github.com/Lexonight1/thermalright-trcc-linux/pull/209): had the flake read its version straight from `pyproject.toml`, killing an entire place I used to have to bump by hand every release.
 - **[@elsiedotcafe](https://github.com/elsiedotcafe)** — [#123](https://github.com/Lexonight1/thermalright-trcc-linux/pull/123): pinned the PySide6 package to a stable source so the build stops breaking.
 
-Thank you both — genuinely. PRs are always welcome, and you'll land right here.
+Thank you all — genuinely. PRs are always welcome, and you'll land right here.
 
 ## Contributors
 
-A huge thank you to every single person who filed an issue, pasted a `trcc report`, or just said "it doesn't work." You are the reason this project supports 6 USB protocols, 16 resolutions, and 12 LED styles from a single developer with one device. Every bug report, every test on hardware I don't own, every "hey this is broken" — that's open source at its best. You're not wasting your time. You're building something together.
+A huge thank you to every single person who filed an issue, pasted a `trcc report`, or just said "it doesn't work." You are the reason this project supports 6 USB protocols, 16 resolutions, and 13 LED styles from a single developer with one device. Every bug report, every test on hardware I don't own, every "hey this is broken" — that's open source at its best. You're not wasting your time. You're building something together.
 
 Special thanks to everyone who has contributed invaluable reports to this project:
 

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/README.md

@@ -53,7 +53,7 @@
 
 Native Linux port of the Thermalright LCD Control Center (Windows TRCC 2.1.2). Control and customize the LCD displays and LED segment displays on Thermalright CPU coolers, AIO pump heads, and fan hubs — entirely from Linux.
 
-> **This project wouldn't exist without our testers.** I only own one device. Every supported device in this list works because someone plugged it in, ran `trcc report`, and told me what broke. 32 testers helped us go from "SCSI only" to full C# feature parity with 6 USB protocols, 16 FBL resolutions, and 12 LED styles. Open source at its best — see [Contributors](#contributors) below.
+> **This project wouldn't exist without our testers.** I only own one device. Every supported device in this list works because someone plugged it in, ran `trcc report`, and told me what broke. 32 testers helped us go from "SCSI only" to full C# feature parity with 6 USB protocols, 16 FBL resolutions, and 13 LED styles. Open source at its best — see [Contributors](#contributors) below.
 
 > Unofficial community project, not affiliated with Thermalright. Built with [Claude](https://claude.ai) (AI) for protocol reverse engineering and code generation, guided by human architecture decisions and logical assessment.
 
@@ -233,7 +233,7 @@ Set the angle to **90°** (or 270°) in the GUI, then open **Cloud Themes** —
 | **Media** | Video/GIF playback on LCD, video trimmer, image cropper, screen cast (X11 + Wayland), mic audio visualization |
 | **Overlay Editor** | Text, sensors, date/time overlays — font picker, dynamic scaling, color picker |
 | **Hardware Sensors** | 77+ sensors — CPU/GPU temp, fan speed, power, usage — customizable dashboard |
-| **LED Control** | 12 LED styles, zone carousel, breathing/rainbow/static/wave modes, per-zone color |
+| **LED Control** | 13 LED styles, zone carousel, breathing/rainbow/static/wave modes, per-zone color |
 | **Display** | 16 resolutions (240x240 to 1920x462), 0/90/180/270 rotation, 3 brightness levels |
 | **Multi-device** | Per-device config, auto-detect, multi-device with device selection |
 | **Security** | udev rules, polkit policy, SELinux support, no root required after setup |
@@ -291,7 +291,7 @@ Run `lsusb` to find your USB ID (`xxxx:xxxx` after `ID`), then match it below.
 **HID LED devices** — RGB LED control:
 | USB ID | Devices |
 |--------|---------|
-| `0416:8001` | AX120 DIGITAL, PA120 DIGITAL, Peerless Assassin 120 DIGITAL ARGB White, Assassin X 120R Digital ARGB, Phantom Spirit 120 Digital EVO, HR10 2280 PRO Digital, and others (model auto-detected via handshake) |
+| `0416:8001` | AX120 DIGITAL, PA120 DIGITAL, Peerless Assassin 120 DIGITAL ARGB White, Assassin X 120R Digital ARGB, Phantom Spirit 120 Digital EVO, HR10 2280 PRO Digital, Magic Qube, and others (model auto-detected via handshake) |
 
 > See the [full device list with protocol details](doc/REFERENCE_DEVICES.md) and the [Device Testing Guide](doc/GUIDE_DEVICE_TESTING.md) if you have an untested device.
 
@@ -336,20 +336,21 @@ If a probe goes `BAD`, that's the bug — fix in `src/`, re-run, see `PASS`. Eve
 | HID Type 3 | pyusb interrupt | TARAN ARMS |
 | Bulk | pyusb bulk | GrandVision 360, Mjolnir Vision 360, Wonder Vision Pro 360, Frozen Warframe Pro |
 | LY | pyusb bulk (chunked) | Trofeo Vision 9.16 LCD |
-| LED | pyusb HID | All LED segment display devices (12 styles) |
+| LED | pyusb HID | All LED segment display devices (13 styles) |
 
 ## Code Contributors
 
 These folks didn't just report a problem — they opened the editor and sent a fix. On a solo project, a merged pull request is about the most generous thing you can do, and every one of these shipped to every user:
 
+- **[@jphilipb](https://github.com/jphilipb)** — [#215](https://github.com/Lexonight1/thermalright-trcc-linux/pull/215): reverse-engineered and hardware-validated the Thermalright Magic Qube (65-LED segment display) — a device that doesn't even exist in the Windows app — and wired up full support from scratch.
 - **[@Hythera](https://github.com/Hythera)** — [#209](https://github.com/Lexonight1/thermalright-trcc-linux/pull/209): had the flake read its version straight from `pyproject.toml`, killing an entire place I used to have to bump by hand every release.
 - **[@elsiedotcafe](https://github.com/elsiedotcafe)** — [#123](https://github.com/Lexonight1/thermalright-trcc-linux/pull/123): pinned the PySide6 package to a stable source so the build stops breaking.
 
-Thank you both — genuinely. PRs are always welcome, and you'll land right here.
+Thank you all — genuinely. PRs are always welcome, and you'll land right here.
 
 ## Contributors
 
-A huge thank you to every single person who filed an issue, pasted a `trcc report`, or just said "it doesn't work." You are the reason this project supports 6 USB protocols, 16 resolutions, and 12 LED styles from a single developer with one device. Every bug report, every test on hardware I don't own, every "hey this is broken" — that's open source at its best. You're not wasting your time. You're building something together.
+A huge thank you to every single person who filed an issue, pasted a `trcc report`, or just said "it doesn't work." You are the reason this project supports 6 USB protocols, 16 resolutions, and 13 LED styles from a single developer with one device. Every bug report, every test on hardware I don't own, every "hey this is broken" — that's open source at its best. You're not wasting your time. You're building something together.
 
 Special thanks to everyone who has contributed invaluable reports to this project:
 

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "trcc-linux"
-version = "9.8.6"
+version = "9.8.8"
 description = "Linux implementation of Thermalright LCD Control Center"
 readme = "README.md"
 license = "GPL-3.0-or-later"

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/__version__.py

@@ -1,4 +1,4 @@
 """TRCC Linux version information. History in doc/CHANGELOG.md."""
 
-__version__ = "9.8.6"
+__version__ = "9.8.8"
 __version_info__ = tuple(int(x) for x in __version__.split("."))

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/adapters/device/__init__.py

@@ -71,7 +71,8 @@ class DeviceFactory:
 
 # Side-effect imports: load each device module so its @register decorator
 # fires and populates the registry above.  Anything that imports
-# ``trcc.adapters.device`` triggers all five.
+# ``trcc.adapters.device`` triggers them all.
+from . import ali_lcd as _ali_lcd  # noqa: E402, F401
 from . import bulk_lcd as _bulk_lcd  # noqa: E402, F401
 from . import hid_lcd as _hid_lcd  # noqa: E402, F401
 from . import led as _led  # noqa: E402, F401

trcc_linux-9.8.8/src/trcc/adapters/device/ali_lcd.py

@@ -0,0 +1,141 @@
+"""AliLcd — Device implementation for USBLCDNew "Ali" variant LCD hardware.
+
+A DISTINCT bulk protocol from BulkLcd (GrandVision).  The Ali device
+(0416:5406, Elite Vision 360 ARGB) is driven by its own C# worker thread
+``ThreadSendDeviceDataALi`` (USBLCDNEW.decompiled.cs:548) — different write
+endpoint, handshake, validation, and a fixed 320x320 RGB565 canvas.  Sharing
+BulkLcd would mean a wire-selecting branch inside one class; the C# itself
+splits the two into separate threads, so we mirror that with a separate wire.
+
+Handshake:   write 16-byte request + 1024 zero pad → EP 0x02 (cs:571-577),
+             read 1024 → validate resp[0] in (101, 102) (cs:610).
+Frame send:  16-byte header + 320x320 RGB565 (204800 B) → EP 0x02,
+             then read a 16-byte ack (cs:646-660).  Fixed 204800-B buffer
+             (cs:628) — no PM/resolution table for this device.
+"""
+from __future__ import annotations
+
+import logging
+
+from ...core.errors import HandshakeError, TransportError
+from ...core.models import HandshakeResult, ProductInfo, Wire
+from ...core.ports import BulkTransport, Device
+from ...core.protocol import DeviceProfile
+from . import DeviceFactory
+
+log = logging.getLogger(__name__)
+
+
+# ── Wire constants (line-cited to USBLCDNEW.decompiled.cs) ──────────────
+
+_EP_WRITE = 0x02          # WriteEndpointID.Ep02 (cs:569) — NOT 0x01 (GrandVision)
+_EP_READ = 0x81           # ReadEndpointID.Ep01 (cs:568)
+
+# 16-byte handshake request + 1024-byte zero pad (cs:571-577).
+_HANDSHAKE_REQUEST = bytes([
+    0xF5, 0x00, 0x01, 0x00, 0xBC, 0xFF, 0xB6, 0xC8,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00,
+]) + bytes(1024)
+
+# 16-byte frame header (cs:646-650); bytes[12:16] LE = 0x032000 = 204800.
+_FRAME_HEADER = bytes([
+    0xF5, 0x01, 0x01, 0x00, 0xBC, 0xFF, 0xB6, 0xC8,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x03, 0x00,
+])
+
+_HANDSHAKE_READ_SIZE = 1024      # byte[1024] read buffer (cs:570)
+_ACK_READ_SIZE = 16              # Read(first, 0, 16, ...) (cs:660)
+_HANDSHAKE_TIMEOUT_MS = 1000
+_WRITE_TIMEOUT_MS = 5000
+_READ_TIMEOUT_MS = 1000
+
+# Fixed 320x320 RGB565 — new byte[204800] = 320*320*2 (cs:628).
+_WIDTH = 320
+_HEIGHT = 320
+_FRAME_BYTES = _WIDTH * _HEIGHT * 2
+_VALID_IDENTITY = (101, 102)     # resp[0] (cs:610)
+
+
+@DeviceFactory.register(Wire.BULK_ALI)
+class AliLcd(Device[BulkTransport]):
+    """USBLCDNew "Ali" bulk LCD (fixed 320x320 RGB565)."""
+
+    def __init__(self, info: ProductInfo, transport: BulkTransport) -> None:
+        super().__init__(info, transport)
+        # Fixed canvas — the C# hardcodes the 204800-B buffer, no PM table.
+        self._profile = DeviceProfile(_WIDTH, _HEIGHT, jpeg=False, big_endian=True)
+
+    @property
+    def profile(self) -> DeviceProfile | None:
+        """Fixed 320x320 RGB565 profile (no handshake-derived geometry)."""
+        return self._profile
+
+    # ── Device ABC ────────────────────────────────────────────────────
+
+    def connect(self) -> HandshakeResult:
+        log.info("AliLcd %s: opening transport", self.info.key)
+        if not self._transport.open():
+            log.error("AliLcd %s: transport open failed", self.info.key)
+            raise HandshakeError(f"Failed to open USB transport for {self.info.key}")
+
+        try:
+            self._transport.write(_EP_WRITE, _HANDSHAKE_REQUEST, _HANDSHAKE_TIMEOUT_MS)
+            resp = self._transport.read(_EP_READ, _HANDSHAKE_READ_SIZE, _HANDSHAKE_TIMEOUT_MS)
+        except TransportError as e:
+            log.error("AliLcd %s: handshake I/O failed: %s", self.info.key, e)
+            raise HandshakeError(f"AliLcd handshake I/O failed: {e}") from e
+
+        if len(resp) < 1 or resp[0] not in _VALID_IDENTITY:
+            log.error(
+                "AliLcd %s: handshake validation failed (len=%d, resp[0]=%s)",
+                self.info.key, len(resp), resp[0] if resp else "N/A",
+            )
+            raise HandshakeError(
+                f"AliLcd handshake validation failed "
+                f"(len={len(resp)}, resp[0]={resp[0] if resp else 'N/A'})"
+            )
+
+        model_id = resp[0] - 1     # obj3[0] = array[0] - 1 (cs:614)
+        result = HandshakeResult(
+            resolution=self._profile.resolution,
+            model_id=model_id,
+            pm_byte=resp[0],
+            fbl=self.info.fbl,
+            raw_response=bytes(resp[:64]),
+        )
+        self._handshake = result
+        log.info("AliLcd handshake OK: resp[0]=%d model_id=%d resolution=%s (RGB565)",
+                 resp[0], model_id, result.resolution)
+        return result
+
+    def send(self, payload: bytes) -> bool:
+        if not self._transport.is_open or self._handshake is None:
+            log.error("AliLcd %s: send() called before connect()", self.info.key)
+            raise TransportError(
+                f"AliLcd {self.info.key} not connected — call connect() first"
+            )
+
+        if len(payload) != _FRAME_BYTES:
+            log.warning("AliLcd %s: payload is %d bytes, expected %d (320x320 RGB565)",
+                        self.info.key, len(payload), _FRAME_BYTES)
+
+        frame = _FRAME_HEADER + payload
+        log.debug("AliLcd %s: sending %d-byte frame (RGB565, 320x320)",
+                  self.info.key, len(frame))
+
+        def _write_frame() -> bool:
+            # C# writes the header+buffer in one transfer, then reads a 16-B ack
+            # (cs:653-660).  One write call; PyUSB splits into USB packets.
+            self._transport.write(_EP_WRITE, frame, _WRITE_TIMEOUT_MS)
+            self._transport.read(_EP_READ, _ACK_READ_SIZE, _READ_TIMEOUT_MS)
+            return True
+
+        # Shared reconnect + recovery policy (base Device): one in-place
+        # close→open→handshake retry absorbs an intermittent NAK / heals a
+        # stale handle after resume, before escalating to the recovery tracker.
+        return self._send_with_recovery(_write_frame)
+
+    def disconnect(self) -> None:
+        log.info("AliLcd %s: disconnecting", self.info.key)
+        self._transport.close()
+        self._handshake = None

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/adapters/device/led.py

@@ -26,7 +26,13 @@ from ...core.errors import (
     UnsupportedOperationError,
 )
 from ...core.led_models import LedPayload
-from ...core.led_protocol import remap_led_colors, resolve_pm
+from ...core.led_protocol import (
+    is_fingerprint_header,
+    remap_led_colors,
+    resolve_handshake,
+    resolve_model_name,
+    resolve_pm,
+)
 from ...core.models import HandshakeResult, LedHandshakeResult, ProductInfo, Wire
 from ...core.ports import BulkTransport, Device
 from . import DeviceFactory
@@ -206,11 +212,15 @@ class Led(Device[BulkTransport]):
                     time.sleep(_HANDSHAKE_RETRY_DELAY_S)
                     continue
 
-                # Windows DeviceDataReceived1 doesn't validate magic/cmd —
-                # we warn on mismatch but accept.
-                if resp[0:4] != _MAGIC:
-                    log.warning("Led handshake: unexpected magic %s", resp[0:4].hex())
-                if len(resp) > 12 and resp[12] != 1:
+                # Windows DeviceDataReceived1 doesn't validate magic/cmd — we
+                # accept regardless.  A recognised fingerprint header (e.g. the
+                # Magic Qube's DC DD AA 01) is expected, so only warn on a truly
+                # unknown magic / cmd byte.
+                header = bytes(resp[0:4])
+                recognised = is_fingerprint_header(header)
+                if header != _MAGIC and not recognised:
+                    log.warning("Led handshake: unexpected magic %s", header.hex())
+                if len(resp) > 12 and resp[12] != 1 and not recognised:
                     log.warning("Led handshake: unexpected cmd byte %d", resp[12])
 
                 # PM and SUB extraction — matches Windows UCDevice.cs offsets:
@@ -218,10 +228,13 @@ class Led(Device[BulkTransport]):
                 self._pm = resp[5]
                 self._sub = resp[4]
 
-                # PM byte → device style + readable model name. Falls back to
-                # the registry's ``led_style`` / ``product`` when the firmware
-                # reports a PM not in PmRegistry (e.g. a new SKU).
-                entry = resolve_pm(self._pm, self._sub)
+                # Handshake → device style + readable model name.  The header
+                # fingerprint wins first (some SKUs reuse a PM byte but ship a
+                # distinct handshake header — e.g. the Magic Qube shares PM=208
+                # with CZ1 but answers with DC DD AA 01), then the PM registry.
+                # Falls back to the registry's ``led_style`` / ``product`` when
+                # the firmware reports a PM not in PmRegistry (e.g. a new SKU).
+                entry = resolve_handshake(header, self._pm, self._sub)
                 if entry is not None:
                     style = entry.style
                     model_name = entry.model_name
@@ -278,7 +291,10 @@ class Led(Device[BulkTransport]):
         cached = _probe_cache_load(self.info.vid, self.info.pid)
         if cached is not None:
             cpm, csub, cname = cached
-            entry = resolve_pm(cpm, csub)
+            # A fingerprint device (e.g. Magic Qube) shares its PM byte with
+            # another product, so recover it by cached model name first — the
+            # cache doesn't persist the handshake header — then by PM byte.
+            entry = resolve_model_name(cname) or resolve_pm(cpm, csub)
             log.warning(
                 "Led handshake: live failed after %d retries — using "
                 "disk cache (PM=%d SUB=%d model=%s, last_err=%s)",

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/adapters/system/_udev.py

@@ -69,6 +69,7 @@ _WIRE_SUBSYSTEMS: dict[Wire, tuple[str, ...]] = {
     Wire.HID:  ("hidraw", "usb"),
     Wire.LED:  ("hidraw", "usb"),
     Wire.BULK: ("usb",),
+    Wire.BULK_ALI: ("usb",),
     Wire.LY:   ("usb",),
 }
 

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/adapters/system/linux.py

@@ -13,6 +13,7 @@ Key pieces:
 from __future__ import annotations
 
 import ctypes
+import errno
 import logging
 import os
 from pathlib import Path
@@ -219,6 +220,20 @@ class LinuxScsiTransport(ScsiTransport):
             return True
         except OSError as e:
             log.error("LinuxScsiTransport: open failed for %s: %s", self._path, e)
+            if e.errno == errno.EACCES:
+                # A resolved-but-unreadable node almost always means setup was
+                # never run: the /dev/sd* block fallback is root-only (the udev
+                # 0666 rule only covers scsi_generic), and even /dev/sg* is
+                # root-only until the setup rule lands.  Name the remediation
+                # instead of a bare EACCES.  (#217)
+                is_block = self._path.startswith("/dev/sd")
+                detail = ("this is a root-only block node (the sg kernel module "
+                          "isn't loaded) — " if is_block else "")
+                log.warning(
+                    "LinuxScsiTransport: permission denied on %s — %srun "
+                    "`trcc system setup` then reboot to load sg and grant "
+                    "0666 access without sudo (#217)", self._path, detail,
+                )
             return False
 
     def close(self) -> None:

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/core/commands/__init__.py

@@ -127,6 +127,7 @@ from .theme import (
     LoadImage,
     LoadTheme,
     LoadVideo,
+    RestoreDeviceState,
     RestoreLastTheme,
     SaveTheme,
     UploadCustomMask,
@@ -190,6 +191,7 @@ __all__ = [
     "RenderDcStandalone",
     "RenderLed",
     "ResetDevice",
+    "RestoreDeviceState",
     "RestoreLastTheme",
     "RunDoctor",
     "RunHealthCheck",

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/core/commands/device.py

@@ -191,18 +191,29 @@ class ConnectDevice(Command[ConnectResult]):
                     override.button_image or "(cutout-only)",
                 )
 
-        # LED style: enrich ``led_style`` from the same handshake fingerprint.
-        # The registry leaves it None ("resolved at runtime from PM byte"), so
-        # without this the LED handler reads None and falls back to a default
-        # style.  resolve_pm lives in core, so this stays inside the hexagon.
+        # LED style: enrich ``led_style`` from the handshake.  The registry
+        # leaves it None ("resolved at runtime"), so without this the LED
+        # handler reads None and falls back to a default style.  Prefer the
+        # LED adapter's already-resolved style — it honours the header
+        # fingerprint, so a Magic Qube (shares PM=208 with CZ1) isn't
+        # mis-tagged as CZ1 — and fall back to the header/PM registry from the
+        # raw handshake bytes.  resolve_* live in core, so this stays inside
+        # the hexagon.
         if device.info.wire is Wire.LED and device.info.led_style is None:
-            from ..led_protocol import resolve_pm
-            led_entry = resolve_pm(handshake.pm_byte, handshake.sub_byte)
-            if led_entry is not None and led_entry.style is not None:
-                device.info = _dc_replace(device.info, led_style=led_entry.style)
+            led_hs = getattr(device, "led_handshake", None)
+            led_style = getattr(led_hs, "style", None)
+            if led_style is None:
+                from ..led_protocol import resolve_handshake
+                led_entry = resolve_handshake(
+                    handshake.raw_response[:4],
+                    handshake.pm_byte, handshake.sub_byte,
+                )
+                led_style = led_entry.style if led_entry is not None else None
+            if led_style is not None:
+                device.info = _dc_replace(device.info, led_style=led_style)
                 log.info(
-                    "ConnectDevice %s: LED style PM=%d → %s",
-                    self.key, handshake.pm_byte, led_entry.style.value,
+                    "ConnectDevice %s: LED style → %s",
+                    self.key, led_style.value,
                 )
 
         # Install theme/cloud/mask data for the HANDSHAKE-resolved resolution.

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/core/commands/theme.py

@@ -1533,6 +1533,85 @@ class RestoreLastTheme(Command[ThemeResult]):
             key=self.key, path=resolved, reset_overrides=False,
         ).execute(app)
 
+@dataclass(frozen=True, slots=True)
+class RestoreDeviceState(Command[ThemeResult]):
+    """Ensure *key* has a renderable display state from persisted settings.
+
+    The shared "make this device show something" path every UI dispatches at
+    its display-start entry — the GUI on connect, the CLI ``display play`` /
+    ``keepalive`` before their loop, the API ``restore-theme`` endpoint.
+    Idempotent: a no-op when a theme is already active (a daemon / GUI already
+    primed it), so it's safe to dispatch unconditionally at each entry.
+
+    Order (mirrors the GUI connect-restore, minus GUI-only widget restore):
+      1. Active theme already loaded → done.
+      2. ``RestoreLastTheme`` — the persisted theme + its mask + overlays +
+         bundled video (preserves the device's overrides).
+      3. No persisted theme → auto-load the first available theme
+         (``ListThemes`` → ``LoadTheme``), matching the GUI first-install
+         auto-load so a fresh device still shows something.
+      4. Replay the persisted ``background_path`` video override
+         (``PlayVideo``) so a cloud / user video resumes on top of the theme.
+
+    Steps 2-3 render + send an immediate first frame, which also leaves a
+    cached frame — so a keepalive resend has something to push (the terminal
+    "No cached frame — render at least once first" gap this closes). (#150)
+    """
+    key: str
+
+    def execute(self, app: App) -> ThemeResult:
+        log.info("RestoreDeviceState: key=%s", self.key)
+        existing = app.active_themes.get(self.key)
+        if existing is not None:
+            log.info("RestoreDeviceState: %s already has an active theme — no-op",
+                     self.key)
+            return ThemeResult(ok=True, key=self.key, theme_name=existing.name,
+                               theme_path=str(existing.path),
+                               message="Display state already active")
+
+        # 2. Persisted theme (reset_overrides=False preserves overlay/mask edits).
+        RestoreLastTheme(key=self.key).execute(app)
+
+        # 3. Nothing persisted → first available theme (GUI first-install parity).
+        #    Resolve via the shared oriented resolver (device profile → native →
+        #    registry) so a rotate panel lists its portrait catalog, matching
+        #    where the GUI browser lists themes.
+        if app.active_themes.get(self.key) is None:
+            resolution = _resolve_oriented_resolution(app, self.key)
+            if resolution is None:
+                log.warning("RestoreDeviceState: %s — cannot resolve resolution "
+                            "to auto-load a theme", self.key)
+            else:
+                listing = ListThemes(resolution=resolution).execute(app)
+                if listing.themes:
+                    first = Path(listing.themes[0].path)
+                    log.info("RestoreDeviceState: %s no persisted theme — "
+                             "auto-loading first theme %s", self.key, first)
+                    LoadTheme(key=self.key, path=first).execute(app)
+
+        theme = app.active_themes.get(self.key)
+        if theme is None:
+            log.warning("RestoreDeviceState: %s — no theme available to restore",
+                        self.key)
+            return ThemeResult(
+                ok=False, key=self.key,
+                message=("No theme available for this device — install themes "
+                         "or load a theme first"),
+            )
+
+        # 4. Resume the persisted cloud / user video background over the theme.
+        bg = app.settings.for_device(self.key).background_path
+        if bg:
+            log.info("RestoreDeviceState: %s replaying persisted background %s",
+                     self.key, bg)
+            PlayVideo(key=self.key, path=Path(bg)).execute(app)
+
+        return ThemeResult(
+            ok=True, key=self.key, theme_name=theme.name,
+            theme_path=str(theme.path),
+            message=f"Display state restored ({theme.name})",
+        )
+
 @dataclass(frozen=True, slots=True)
 class ListCloudThemes(Command[CloudThemesListResult]):
     """List themes available in Thermalright's hosted catalog.

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/core/geometry.py

@@ -116,20 +116,24 @@ def plan_orientation(
     rotate_panel = profile.rotate and w != h and orientation in (90, 270)
     if rotate_panel and not content_is_portrait and not profile.widescreen:
         return OrientationPlan((w, h), False, orientation)
+    if rotate_panel and profile.widescreen:
+        # Widescreen JPEG panels (1600×720 etc.): compose on the transposed
+        # portrait canvas (720×1600), preview upright — but post_rotate=0.  The
+        # WIRE rotation is owned entirely by resolve_encode_angle (via wire_angle),
+        # which the C# ``ImageToJpg`` applies to the whole composite to reach the
+        # device's fixed landscape dims: 0→180, 90→90, 180→0, 270→270.  A
+        # post_rotate here would double-rotate on top of it (the pre-fix code
+        # returned 180 at 270 AND relied on portrait_content=True to suppress the
+        # wire rotate — which silently sent an unrotated 720×1600 portrait frame,
+        # #169).  On-glass handedness still wants a reporter photo (#203/#169).
+        return OrientationPlan((h, w), True, 0)
     if rotate_panel:
-        # Portrait content is composed on the transposed canvas with the
-        # orientation baked into the pixels — but the asset is authored for 90.
-        # 270 is 90 + a 180° flip, and 180° is dimension-preserving, so rotate
-        # the WHOLE composite (bg + mask + text) as one unit and everything
-        # stays true to the physical rotation instead of frozen at 90.
-        # Widescreen JPEG panels keep post_rotate=0: their 90/270 send-unrotated
-        # behaviour is hardware-verified (#169) and must not change.
-        # 270 is 90 + a dimension-preserving 180° flip — applies to EVERY rotate
-        # panel including widescreen (the C# decompile shows widescreen 854×480
-        # rotates 90→270° / 270→90°, a 180° difference, not frozen).  The flip
-        # keeps the portrait-buffer dimensions (#169) and only re-orients the
-        # content, so it's dimension-safe (no clip).  On-glass handedness for
-        # widescreen still wants a reporter photo (#203) to confirm.
+        # Non-widescreen portrait content: composed on the transposed canvas with
+        # the orientation baked into the pixels — but the asset is authored for
+        # 90.  270 is 90 + a dimension-preserving 180° flip, so rotate the WHOLE
+        # composite (bg + mask + text) as one unit and everything stays true to
+        # the physical rotation instead of frozen at 90.  These panels take no
+        # wire rotation (portrait_content=True opts them out of wire_angle).
         post_rotate = 180 if orientation == 270 else 0
         return OrientationPlan((h, w), True, post_rotate)
     return OrientationPlan(oriented_resolution((w, h), orientation), False, 0)

{trcc_linux-9.8.6 → trcc_linux-9.8.8}/src/trcc/core/led_models.py

@@ -200,6 +200,10 @@ LEGACY_STYLE_ID: dict[LedStyle, int] = {
     LedStyle.LF11:  10,
     LedStyle.LF15:  11,
     LedStyle.LF13:  12,
+    # Synthetic id for the Linux-added Magic Qube (no legacy C# style_id);
+    # continues the sequence so every LED_STYLES entry has an id.  It shows
+    # plain sensor gauges (led_panel_for treats any non-LC1/LF11/LC2 id so).
+    LedStyle.MAGIC_QUBE: 13,
 }
 
 # Inverse lookup — legacy int → :class:`LedStyle` enum.  Used by GUI
@@ -268,6 +272,13 @@ LED_STYLES: dict[LedStyle, LedStyleSpec] = {
                                  zone_assets=_ZONE_BTN_MODE56),
     LedStyle.LF13:  LedStyleSpec(62,  62, 0, "LF13",
                                  "led_preview_lf13",  "led_bg_lf13"),
+    # Magic Qube — 65 LEDs: two 7-seg digits (3 LEDs/segment, LED 0-41) + 4
+    # corner metric indicators (42-49) + a 15-LED contour border (50-64).
+    # Same 2-digit / 4-phase design as CZ1.  Reuses the generic segment assets
+    # until dedicated art is added.
+    LedStyle.MAGIC_QUBE: LedStyleSpec(65, 14, 4, "MAGIC_QUBE",
+                                      "led_preview_cz1", "led_bg_segment",
+                                      zone_assets=_ZONE_BTN_N),
 }
 
 
@@ -302,6 +313,7 @@ ZONE_STYLE_ASSETS: dict[LedStyle, tuple[tuple[str, str], ...]] = {
     LedStyle.LF10:  _ZONE_ASSETS_BTNN,
     LedStyle.CZ1:   _ZONE_ASSETS_BTNN,
     LedStyle.LF11:  _ZONE_ASSETS_BTNN,
+    LedStyle.MAGIC_QUBE: _ZONE_ASSETS_BTNN,
 }