Document Linux PC-host feasibility blocker on lxso2

Co-authored-by: SAPHID <saphid@users.noreply.github.com>

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 ## Unreleased
 
+- Distinguish an unreadable gamescope panel list from a confirmed empty list in
+  Desktop Capture and Control (issue #73). Failed discovery now asks users to
+  check SteamVR and report the exact message and SteamOS version, rather than
+  claiming no panel is open. This does not establish a new capture fix.
+
 - Add an opt-in desktop low-battery alert without starting a family session.
   Existing status reads confirm low charge and drain; recovery hysteresis and a
   ten-minute cooldown limit repeat alerts. Off on page open/headset switch;

app/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "frame-control",
-  "version": "0.4.16",
+  "version": "0.4.17",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "frame-control",
-      "version": "0.4.16",
+      "version": "0.4.17",
       "license": "MIT",
       "devDependencies": {
         "electron": "^44.4.5",

app/package.json

@@ -1,7 +1,7 @@
 {
   "name": "frame-control",
   "productName": "Frame Control",
-  "version": "0.4.16",
+  "version": "0.4.17",
   "description": "Desktop app for managing a Valve Steam Frame over SSH",
   "private": true,
   "main": "main.js",

docs/evidence/clip-long-2026-10-24.json

@@ -0,0 +1,29 @@
+{
+  "date_utc": "2026-10-24",
+  "version": "v0.4.17",
+  "steam_os_version": "0.5.6",
+  "steam_os_build": "20261015.6141207",
+  "result": "PASS: the real Steam-saved clip exported beyond 60 seconds",
+  "method": "Steam manual recording of a reserved test title, then Frame Control's production frame_clips.export_clip path; no packaged GUI button click",
+  "source": {
+    "kind": "saved Steam clip",
+    "duration_seconds": 100.801,
+    "chunks": 68,
+    "gaps": 0
+  },
+  "export": {
+    "format": "MP4",
+    "bytes": 1377849,
+    "sha256": "82ef31f65e6933d122bc326d664beb2eb2e3ba6969f26c0032552e80fbe8d823",
+    "duration_seconds_ffprobe": 100.833335,
+    "video": "H.264, 1920x1080, 6049 frames",
+    "audio": "AAC, 48000 Hz, stereo",
+    "full_ffmpeg_decode": "passed without errors"
+  },
+  "cleanup": "Restored Steam recording mode to its prior off setting; stopped and removed only the reserved title and its processes; no recording or running Steam apps remained; Game Mode active, battery 100%, shared test lock released",
+  "limits": [
+    "This exercised the production export code path, not a packaged GUI Export click.",
+    "Visible gameplay quality, perceptual audio/video sync, trims and longer recordings were not checked.",
+    "The newly saved Steam source clip was retained; no user recording was deleted."
+  ]
+}

docs/evidence/linux-pc-host-lxso2-2026-10-24.md

@@ -0,0 +1,114 @@
+# Linux PC host feasibility on lxso2, 2026-10-24
+
+**Blocked on this host's current desktop environment. No feature enabled and
+no feature PR opened.** Investigation for [issue #23](https://github.com/saphid/frame-control/issues/23),
+starting at `4e45648` (main after #154). This concerns flat desktop panels,
+not Linux PC VR streaming.
+
+## Decision
+
+The smallest useful candidate is one explicitly selected **whole screen**,
+captured through the ScreenCast portal's restricted PipeWire connection,
+encoded locally with GStreamer/x264, and sent to the existing viewer through
+the shared SSH controller. Input, individual windows and virtual displays
+would remain unsupported until separately verified. This candidate was **not
+implemented**: lxso2 currently cannot exercise the real capture permission path.
+
+Development packages could unblock compilation, but would not create a desktop,
+portal backend or consented screen. Synthetic encoding, an Xvfb screen grab or
+using the Frame as its own host would not verify Linux PC desktop capture.
+No host packages or services were changed.
+
+## Evidence
+
+Host: `lxso2`, Linux x86-64, kernel `6.8.0-139-generic`.
+
+- `DISPLAY`, `WAYLAND_DISPLAY` and `XDG_SESSION_TYPE` were unset. The user bus
+  was reachable at `/run/user/1000/bus`.
+- `/run/user/1000/pipewire-0` and `/run/user/1000/wayland-0` did not exist.
+  No processes matched Xorg, Xwayland, Xvfb, GNOME Shell, KWin Wayland, Weston,
+  PipeWire, WirePlumber or xdg-desktop-portal at the process check.
+- D-Bus `ListNames` returned D-Bus, systemd and anonymous connections, with no
+  `org.freedesktop.portal.Desktop`. `ListActivatableNames` also had no desktop
+  portal. These calls queried D-Bus itself without activating capture services.
+- `dpkg-query` found no `xdg-desktop-portal*` or `pipewire*` packages.
+  `libgstreamer1.0-0:amd64` was installed at `1.24.2-1ubuntu0.1`, but
+  `gstreamer1.0-tools` was uninstalled. `gst-launch-1.0` and `gst-inspect-1.0`
+  were absent from PATH. Xvfb was installed, but supplies no portal by itself.
+- `pkg-config --modversion` failed for `gstreamer-1.0`,
+  `gstreamer-app-1.0`, `gstreamer-video-1.0` and `libpipewire-0.3`.
+  `gio-unix-2.0` reported `2.80.0`.
+- `desktop/bundle/pc-host.so` did not exist.
+  `python3 desktop/build.py --with-x264 --out /tmp/frame-linux-pc-feasibility`
+  exited **1** before compilation: required GStreamer pkg-config modules were
+  missing. No native build or encoder runtime success is claimed.
+
+Reproduce the non-activating portal checks in the intended desktop user's
+session, using that session's actual runtime directory and bus:
+
+```sh
+gdbus call --session --dest org.freedesktop.DBus \
+  --object-path /org/freedesktop/DBus --method org.freedesktop.DBus.ListNames
+gdbus call --session --dest org.freedesktop.DBus \
+  --object-path /org/freedesktop/DBus --method org.freedesktop.DBus.ListActivatableNames
+pkg-config --modversion gstreamer-1.0 gstreamer-app-1.0 gstreamer-video-1.0 gio-unix-2.0
+```
+
+## Existing implementation reviewed
+
+- `ui/frame_pcview.py` selects PCView only on Windows and rejects Linux even
+  when enabled. Linux installers do not carry its native bundle.
+- `desktop/portal.c` creates a **RemoteDesktop** session, requests pointer and
+  keyboard devices, and selects monitors and windows (`types=3`). It is not
+  ScreenCast-only or screen-only. Its restricted fd/node and fresh-connection
+  handling are reusable.
+- `ui/frame_pc_agent.py` calls the selection a shared window or screen and
+  picks VA-API when the element is present, falling back to x264. A bounded
+  x264 slice needs explicit encoder policy and unsupported-input behavior;
+  exposing the existing Linux path alone would not meet that scope.
+- `app/main.js` starts Python with the inherited environment; it does not
+  supply a compositor, portal backend or capture permission.
+- `.github/workflows/pc-host.yml` installs native dependencies and tests
+  synthetic x264/protocol behavior with fake capture consent/input. This does
+  not establish interactive Linux desktop capture.
+- Issue #23 and its six existing comments distinguish prior synthetic
+  helper/viewer results from untested Linux desktop capture and consent.
+
+## Verification and limits
+
+`python3 -m unittest discover -s tests -p 'test_pc*.py' -v` exited **0**:
+27 tests ran successfully, with one additional native-class setup skip
+(`native PC libraries not built on this host`). Coverage includes the compiled
+C controller trace, Python adapters, grants and default-off/Windows-only gates.
+It does not prove native GStreamer capture or local-client streaming.
+
+No fake/local native stream, consent prompt, real desktop capture, Frame viewer
+run, benchmark, input or window mode was tested in this investigation. No SSH
+connection to the Frame was needed; no lock was acquired and no headset install,
+launch, stop or settings change was attempted. Issue #113 was left alone.
+No issue comments or other public messages were sent.
+
+## Concrete unblock and next slice
+
+1. Supply an interactive Linux desktop, physical or VM, with a working
+   ScreenCast portal/backend, PipeWire and accessible consent dialog. Installing
+   the portal frontend alone is insufficient. A VM can establish software
+   capture/x264 behavior while physical-host behavior remains unverified.
+2. Install the native dependencies listed in `pc-host.yml` in that environment
+   and build a local `--with-x264` bundle. Resolve x264 distribution
+   licensing/source obligations before shipping it: the current build labels
+   that option test-only and the Windows installer omits it.
+3. Implement a ScreenCast-only session with monitor-only selection, no
+   RemoteDesktop/device request, disabled input in protocol and viewer, hidden
+   window/virtual-display modes, and explicit x264 selection. Preserve the
+   default-off gate and existing Mac/Windows paths.
+4. Verify consent acceptance/refusal/cancellation, Stop/reopen, connection loss
+   and portal closure with a local client decoding actual captured content.
+   Run the shared benchmark and record results without inferring hardware or
+   physical laser behavior.
+5. Only after local capture works, acquire the shared Frame lock once and
+   follow [the headset procedure](../testing.md#headset-smoke-test) for a brief
+   viewer test. If occupied, leave the lock alone and defer device testing.
+
+These are prerequisites and proposed scope, not a ready feature. The Linux
+host request in #23 remains open.

docs/evidence/panel-discovery-2026-10-24.md

@@ -0,0 +1,65 @@
+# Panel discovery investigation, 2026-10-24
+
+Base: v0.4.16 (`040581a`). Scope: issue #73, Desktop Capture and Control.
+
+## Evidence and change
+
+Issue #73 reports SteamOS 0.4.5 with its desktop visibly running, without a
+Frame Control version or steps. The maintainer's 2026-10-12 comment identifies
+the duplicate `loadPanels` UI function fixed in 0.4.3 and requests the exact
+remaining message and SteamOS build if it persists. There is no reporter
+follow-up. At this base, the functions are distinct and the existing Node
+regression checks their uniqueness and the Desktop GET request.
+
+Read-only source inspection found a remaining diagnostic ambiguity:
+`read_props()` returns `{}` for failed/timed-out X queries and missing root
+properties. `panels()` treated that like a confirmed empty focusable-window
+list. With no detected Plasma process it claimed no panel was open; with Plasma
+it claimed SteamOS was not listing the desktop. Neither claim follows from an
+unreadable list.
+
+The change checks for the presence of `GAMESCOPE_FOCUSABLE_WINDOWS` before
+those claims. An absent list now reports unknown panel state, suggests checking
+SteamVR and retrying/Headset view, and asks for the exact message and SteamOS
+version if persistent. Missing-tool and whole-run deadline explanations retain
+precedence. No new probes, retries, timeouts, input or capture routes are added.
+
+## Real-device boundary
+
+A preliminary read-only SSH inspection returned SteamOS 0.5.6, BUILD_ID
+`20261015.6141207`, battery 99%, discharging, and X0/X1 socket names. It made
+no headset changes. The shared lock was absent at that moment.
+
+The subsequent atomic `mkdir /tmp/frame-test.lock` failed with `File exists`.
+The guarded command exited with status 3 before the session, X-property or
+panel-helper probes ran. The other task's lock was left alone. No application
+was installed, started, stopped or focused; no headset settings changed.
+
+That implementation probe was blocked by the occupied shared device. The
+coordinator later supplied a separate narrow read-only observation on SteamOS
+0.5.6, BUILD_ID `20261015.6141207`, at battery 99%, with the shared lock absent
+and both `xprop` and `xwininfo` installed:
+
+- In **Game Mode**, `DISPLAY=:0 xprop -root GAMESCOPE_FOCUSABLE_WINDOWS
+  GAMESCOPE_FOCUSED_WINDOW` returned both properties **present but empty**.
+- On `DISPLAY=:1`, both properties were reported **not found**.
+
+This observation distinguishes a confirmed empty list on `:0` from absent
+properties on `:1`. It does **not** reproduce #73's Desktop Mode report on
+SteamOS 0.4.5, and it does **not** test this PR's code on the headset. It is a
+coordinator-supplied observation, separate from the blocked implementation probe
+above; no new headset probe was run for this follow-up.
+
+Remaining limit: there is no reproduction establishing a further
+capture/discovery fix for the reported Desktop Mode/build combination. This PR
+improves a demonstrated diagnostic error; it must not close #73 as a verified
+new capture fix.
+
+## Offline verification
+
+Focused Python regressions distinguish absent, focus-only, failed/timed-out X
+queries, confirmed empty and desktop-running lists. Existing tests cover
+missing tools, alternate displays, successful discovery and the 12-second
+deadline. The production page's Node behavior test verifies that both Capture
+and Control preserve the unreadable-list advice. These use fake X/HTTP/DOM
+boundaries and do not prove real SteamOS 0.4.5 capture or physical input.

docs/pc-in-headset.md

@@ -107,6 +107,12 @@ The Linux adapter from #38 (desktop portal, PipeWire, VA-API) is in the tree,
 because it shares `capture.c` and the controller, and CI still builds and tests
 it. The app never offers it and the Linux installers don't include it.
 
+The [2026-10-24 lxso2 feasibility check](evidence/linux-pc-host-lxso2-2026-10-24.md)
+found no running or activatable desktop portal, no PipeWire socket and missing
+native build dependencies. Linux support remains disabled. The evidence records
+the build failure and prerequisites for a screen-only, view-only x264 slice;
+Linux capture, input and window modes have not been verified there.
+
 ## Measuring
 
 The Mac benchmark accepts the PC host:

docs/recording.md

@@ -170,14 +170,26 @@ The temporary server stopped, the shared device lock was released, and the
 headset remained in Game Mode at 99% battery. See
 [`docs/evidence/clip-gui-export-2026-10-24.json`](evidence/clip-gui-export-2026-10-24.json)
 for the narrow proof. This did not exercise the packaged desktop installer.
+
+On 2026-10-24, the exact v0.4.17 tag exported a second real Steam-saved clip
+through `frame_clips.export_clip`, the same production path used by **Export**.
+Steam listed the clip at **100.801 seconds**, with 68 chunks and no gaps. The
+copied MP4 was **100.833 seconds** by `ffprobe` and decoded completely with
+`ffmpeg`; it was not cut at 60 seconds. The test restored Steam's recording
+mode, removed its reserved title and processes, and left Game Mode active at
+100% battery. See [the long-clip evidence](evidence/clip-long-2026-10-24.json).
+This did not click **Export** in the packaged desktop app or verify visible
+gameplay quality or perceptual audio/video sync.
+
 Remaining checks include:
 
 - Perceptual audio/video sync and audible content. Frame Control's separate
   **Record** path stamps camera and audio inputs with the wall clock because
   they otherwise count from different clocks.
-- Long saved clips, trims starting partway through a Steam recording, and the
-  packaged desktop app's **Export** button. One real saved clip's directory
-  layout was checked; the parser remains deliberately loose for other layouts.
+- Clips longer than the 100-second test, trims starting partway through a Steam
+  recording, and the packaged desktop app's **Export** button. Two real saved
+  clips' directory layouts were checked; the parser remains deliberately loose
+  for other layouts.
 - How much x264 at 1080p30 costs the Frame while a game is running.
 - Automatic copying: that 30 seconds without a write is a safe sign Steam has
   finished saving a clip, and how a once-a-minute listing feels on the Frame.

docs/streaming.md

@@ -7,6 +7,9 @@ This covers three directions, plus input:
 - **C. iPhone → Frame**: mirror the phone inside the headset.
 - **PC VR from Linux**: [feasibility and options](linux-vr-streaming.md),
   including Valve's streaming and USB support. No Linux host tested yet.
+- **Linux desktop → Frame panels**: [PC-host feasibility on lxso2](evidence/linux-pc-host-lxso2-2026-10-24.md).
+  Blocked on missing desktop portal/PipeWire and native build dependencies;
+  Linux desktop capture remains disabled and unverified.
 - **Input**: type and point in the Frame from the Mac or iPhone.
 - **Live view and Control**: watch a panel flat and tap on it to use it.
 
@@ -285,6 +288,13 @@ How (**verified 2026-09-29**, SteamOS 0.4.1, build 20260925.6191901):
   [#73](https://github.com/saphid/frame-control/issues/73) (SteamOS 0.4.5, desktop
   open but not found). The likely cause there was the page itself: a second
   `loadPanels` (the panel switcher's) replaced the Desktop view's, fixed in #84.
+  If the `GAMESCOPE_FOCUSABLE_WINDOWS` property cannot be read, discovery now
+  reports that the panel list is unreadable and whether a panel is open is
+  unknown. A failed X connection or missing property is not evidence that the
+  desktop is closed. Check that SteamVR is running, retry or use Headset view;
+  if it persists, report the exact message and SteamOS version. The confirmed
+  empty-list advice is retained when the property is present but empty.
+  [2026-10-24 investigation and verification limits](evidence/panel-discovery-2026-10-24.md).
 - Tested from the iPhone app (Simulator): a tap on the Desktop view focused a
   text box in the panel and the text field typed into it; a trackpad move went
   exactly (+40, +25).

tests/page/desk_panels.mjs

@@ -40,6 +40,15 @@ function page(answer) {
   if (!p.noPanelWhy().startsWith("The headset's desktop is running")) fail("why for capture: " + p.noPanelWhy());
   if (!p.ctrlTarget().why.startsWith("The headset's desktop is running")) fail("why for control: " + p.ctrlTarget().why);
 }
+// An unreadable list is unknown: both Capture and Control retain that distinction.
+{
+  const why = "Frame Control couldn't read gamescope's panel list, so it can't tell whether an app panel is open. " +
+    "Check that SteamVR is running in the headset, then try again or use the Headset view. " +
+    "If this persists, report your SteamOS version and this exact message.";
+  const p = page(() => ({ panels: [], focus: null, why }));
+  await p.loadPanels();
+  if (p.noPanelWhy() !== why || p.ctrlTarget().why !== why) fail("unreadable list reason lost");
+}
 // An older headset script without "why": the generic advice.
 {
   const p = page(() => ({ panels: [], focus: null }));

tests/test_touch.py

@@ -355,6 +355,8 @@ class FindingPanels(unittest.TestCase):
         t.subprocess.run = lambda *a, **k: subprocess.CompletedProcess(a, 0, out, "")
         self.assertEqual(self.real_read_props(":0"), {"GAMESCOPE_FOCUSABLE_WINDOWS": [1, 0, 2, 3, 769, 4],
                                                       "GAMESCOPE_FOCUSED_WINDOW": [0]})
+        t.subprocess.run = lambda *a, **k: subprocess.CompletedProcess(a, 0, "GAMESCOPE_FOCUSABLE_WINDOWS(CARDINAL) = \n", "")
+        self.assertEqual(self.real_read_props(":0"), {"GAMESCOPE_FOCUSABLE_WINDOWS": []})
         t.subprocess.run = lambda *a, **k: subprocess.CompletedProcess(a, 0, "GAMESCOPE_FOCUSED_WINDOW:  not found.\n", "")
         self.assertEqual(self.real_read_props(":0"), {})
 
@@ -392,17 +394,52 @@ class FindingPanels(unittest.TestCase):
         self.assertIn("Headset view", out["why"])
 
     def test_desktop_running_but_not_listed(self):
+        self.props[(":0", "GAMESCOPE_FOCUSABLE_WINDOWS")] = []
         self.t.desktop_running = lambda: True
         out = self.t.panels()
         self.assertEqual(out["panels"], [])
         self.assertIn("desktop is running", out["why"])
 
     def test_nothing_open(self):
+        self.props[(":0", "GAMESCOPE_FOCUSABLE_WINDOWS")] = []
         out = self.t.panels()
         self.assertEqual(out, {"focus": None, "focus_display": None, "panels": [],
                                "why": "No app panel is open in the headset. Open the desktop or an app in "
                                       "the headset, then try again."})
 
+    def test_absent_list_is_unknown_even_when_desktop_is_running(self):
+        for desktop in (False, True):
+            with self.subTest(desktop=desktop):
+                self.t.desktop_running = lambda: desktop
+                self.t._ROOT.clear()
+                out = self.t.panels()
+                self.assertEqual(out["panels"], [])
+                self.assertIn("couldn't read gamescope's panel list", out["why"])
+                self.assertIn("can't tell whether", out["why"])
+                self.assertIn("Headset view", out["why"])
+                self.assertNotIn("No app panel is open", out["why"])
+
+    def test_focus_property_without_panel_list_is_unknown(self):
+        self.props[(":0", "GAMESCOPE_FOCUSED_WINDOW")] = [0]
+        self.assertIn("couldn't read gamescope's panel list", self.t.panels()["why"])
+
+    def test_failed_xprop_does_not_claim_no_panels_are_open(self):
+        t = self.t
+        t.read_props = self.real_read_props
+        saved = t.subprocess.run
+        self.addCleanup(setattr, t.subprocess, "run", saved)
+        for failure in ("refused", "timeout"):
+            with self.subTest(failure=failure):
+                def run(*args, **kwargs):
+                    if failure == "timeout":
+                        raise subprocess.TimeoutExpired(args[0], kwargs["timeout"])
+                    return subprocess.CompletedProcess(args[0], 1, "", "Can't open display")
+                t.subprocess.run = run
+                t._ROOT.clear()
+                out = t.panels()
+                self.assertIn("couldn't read gamescope's panel list", out["why"])
+                self.assertNotIn("No app panel is open", out["why"])
+
     def test_missing_x_tools_are_named(self):
         self.plasma_on(":0")
         self.t.shutil.which = lambda tool: None if tool == "xwininfo" else "/usr/bin/" + tool

ui/frame_touch.py

@@ -271,7 +271,7 @@ def panels(limit=None):
         out["why"] = ("The headset took too long to list its panels (its X tools didn't answer in time). "
                       + ("Some may be missing. " if found else "") + "Try again, or use the Headset view.")
     elif not found:
-        out["why"] = why_none(unfound)
+        out["why"] = why_none(unfound, props)
     return out
 
 
@@ -295,13 +295,18 @@ def desktop_running():
     return False
 
 
-def why_none(listed):
+def why_none(listed, props):
     """Why panels() found nothing, in words the person at the computer can act on.
-    listed: how many of the windows gamescope lists aren't X11 windows on any display."""
+    listed: how many of the windows gamescope lists aren't X11 windows on any display.
+    props: the root properties actually read; an absent list is not an empty list."""
     missing = [tool for tool in ("xprop", "xwininfo") if not shutil.which(tool)]
     if missing:
         return (f"This SteamOS build has no {' or '.join(missing)}, which Frame Control uses to find the "
                 "headset's panels. Use the Headset view instead, and please report your SteamOS version.")
+    if "GAMESCOPE_FOCUSABLE_WINDOWS" not in props:
+        return ("Frame Control couldn't read gamescope's panel list, so it can't tell whether an app panel is open. "
+                "Check that SteamVR is running in the headset, then try again or use the Headset view. "
+                "If this persists, report your SteamOS version and this exact message.")
     if listed:
         # gamescope lists windows, but none is an X11 window on any display: a Wayland
         # window (a desktop drawn straight to gamescope, say) can't be found or captured this way.