frametop-hand-recorder (a separate repo, with this one as a submodule) ships hands/ with its own
dashboard window and binaries built for the SteamOS host. A standalone.json at the top of its tree
marks it (takes.standalone()):
- session.py starts ft-hands directly instead of through distrobox, for recording and tracking
- session.json's and the manifest's tool name its build ("ft-handrec <describe> (<name> <version>)")
- the repair hints say to run its install command again (fix_hint)
ft_handrec.py takes --qml and --style for that window, and has toolVersion and standalone for it.
The Makefile passes LDFLAGS to the ft-hands links (its release links them statically), and
test_qml_backend.py checks another window with FT_HANDREC_QML.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
reference.md's ft-screens keys paragraph, the relay's top docstring and
hazards.md's key releases section now say that a mouse button passed
through as a key in pointer mode goes to the screen the pointer is on,
not where typing goes, only while one shows and Frametop isn't paused,
that ft-screens lets its release through and releases it itself when
the pointer leaves the screens, its screen hides, or Frametop pauses,
and that a relay that starts releases mouse buttons too. hazards.md
adds that the button still reaches the virtual mouse, which gamescope
reads, and that a keyboard's media keys come from its Consumer Control
node, which stays ungrabbed, so they reach gamescope as well as the
desktop even while typing goes to the desktop. design.md says why the
buttons follow the pointer: as keys, a release was lost whenever typing
moved, and wlroots counts presses per button. reference.md names
scripts/test-relay-buttons.sh, and keys-test's wider scope.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
keys-test gains two fake nodes with volume keys, a keyboard's Consumer
Control node (USB) and the headset's gpio-keys (BUS_HOST), both as if
their volume keys were remapped. Its virtual devices now record what
they're sent, and the relay's volume handling is a recorder, so no
wpctl runs.
New checks: a relay that starts releases the modifiers and mouse
buttons on the desktop. A media key reaches the desktop and no virtual
device, and one held over the once-a-second check stays down until its
node lets go. A volume stand-in goes to the volume handling only; an
original volume key a remap missed, and the headset's KEY_SELECT, reach
nothing. Clicks with POINTER=0, clicks while paused, and a pass-through
mouse's clicks don't reach the desktop; a plain click in pointer mode
is the helper's alone. A side button mapped to "key" in pointer mode
reaches the desktop and the virtual mouse, and its release still
reaches the desktop after a pause starts.
51 checks, all passing. Against the relay at the PR merge (82979e4), 9
of the new ones fail.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
screens/tests/relay-buttons-test.c checks relay-buttons.h, built and
run by scripts/test-relay-buttons.sh in the dev container like the
controller click test (-Wall -Wextra -Werror). Which codes go to the
keyboard, the pointer, or nowhere (0x100-0x10f, 0x118-0x15f and
BTN_TRIGGER_HAPPY up dropped, KEY_OK up to BTN_TRIGGER_HAPPY keys); a
press without the pointer on a visible screen dropped along with its
release; a release without a press ignored; a release that still goes
once typing or the pointer moved elsewhere; and ft-screens' own
releases, after which the relay's find nothing held.
It fails against a header whose release needs the pointer, and against
one that makes every code from BTN_MISC up a button, as PR #41 did.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR #41 sent the relay's codes from BTN_MISC up to the seat's pointer
through send_key and handle_key, which lets a release through only if
the wlr keyboard holds that key, and buttons never go through it. So
the release was dropped whenever typing stopped going to the desktop
while a button was held (the dashboard closing in dashboard mode,
looking away in gesture mode, the hide hotkey, a click on another
panel, a pause), and wlroots 0.20 counts presses per button: one lost
release swallowed that button for good, the laser's BTN_LEFT clicks
included, and KWin kept it held. The laser leaving the screens lost it
too, since vr.cpp holds FocusLeave only for its own presses.
The relay's mouse buttons (BTN_MOUSE..BTN_TASK) now follow the pointer
rather than typing, tracked in their own bitmask (relay-buttons.h). A
press needs the pointer on a screen that shows, nothing paused, and
that button not held; a release goes whenever its bit is set. ft-screens
releases the held ones itself when the pointer leaves the screens
(before it clears the focus), when the pointer's screen closes, and on
the first tick after that screen hides or everything pauses
(ft_vr_screen_visible, new in vr.cpp). Each button gets its pointer
frame, and the button state is wl_pointer_button_state, not the
keyboard's enum (-Wextra warned about it).
send_key is keyboard-only again, so relay buttons don't count as typing
for the frame rates (typed_ms). Codes from KEY_OK up to
BTN_TRIGGER_HAPPY take the keyboard path (xkeyboard-config names
keycodes above 255); 0x100-0x10f, 0x118-0x15f and BTN_TRIGGER_HAPPY up
are dropped.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The copy is frozen at install time, and SteamOS's entry and steamos-nested-desktop come from
steamdeck-kde-presets, so that package gets a retest hint too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The summary always listed it, and when it was the only thing left (step 1 done by an
uninstall.sh from before this entry) step 1 was skipped and the copy stayed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without /usr/share/applications/deckard-nested-desktop.desktop, the sed failed inside the
set -e install script: install.sh stopped at step 7 and left an empty entry, so the get.sh
one-liner broke. Now the copy is skipped with a note. It's written to a temporary file and
moved into place, Name is replaced whole (localized names dropped), and X-Steam-Special
stays on Frametop's entry only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A relay that went away with a key-mapped mouse button down left it down
in ft-screens, and ft-screens takes no second press of a button it
holds. The start's release loop, which let go of only the modifiers,
releases BTN_MOUSE..BTN_TASK as well. ft-screens ignores the release of
a button it doesn't hold.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR #41 sent ft-screens every key and button a device has. So with
POINTER=0 every physical click went to ft-screens as well as to the
virtual mouse, clicks were sent while paused, a pass-through mouse's
clicks landed wherever a laser last was, and gamepad, joystick,
digitizer and BTN_TRIGGER_HAPPY codes became desktop buttons.
to_screens() sends keys below BTN_MISC as before, mouse buttons
(BTN_MOUSE..BTN_TASK) only when asked to, and nothing else, but the
release of anything the desktop has down. The pointer branch asks for
it only in pointer mode, where reaching it means the button's map says
"key": the PR's case, a side button passed through for Back. Clicks for
the virtual controller or the virtual mouse don't go to ft-screens,
and a key-mapped button held into a pause still has its release sent.
Keys from KEY_OK (0x160) up stay in the relay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR #41 let volume-role nodes fall into the pointer-device branch. Those
nodes are never grabbed: the headset's gpio-keys (its KEY_SELECT click
button and volume), keyboards' and mice's Consumer Control nodes, and
pmic_resin. So media keys were re-emitted on the virtual keyboard and
gamescope got them twice, the headset's click button went to ft-screens,
the device button maps applied to them, and after a partial volume
remap failure an original volume key went out on the virtual keyboard,
where gamescope reads it (a volume key with nothing focused aborts it).
A volume node's keys stop there again. Only a USB or Bluetooth node's
non-volume keys (the media keys) go on, to ft-screens, like a
pass-through keyboard's: ft-screens types them only while typing goes
to the desktop, and reconcile_desktop_keys releases one whose node goes
away while it's held (physically_down reads volume nodes too).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ft-gaze's "eye-server.mmap has eye data, but its layout is not recognized"
reached only the journal. The Gaze page said "the eye tracker isn't
sending", or "no eyes seen (is the headset on?)" for our tracker's first
calibration, which waits for SteamVR to see an eye, so nothing pointed at
the real cause after a SteamOS update.
ft-gazed now notes that line (cleared by ft-gaze's "layout:" line, and when
ft-gaze starts or stops), and the problem under the Gaze pointer switch
says SteamVR's eye data has a layout Frametop doesn't know, so SteamVR's
eye tracker can't be used, or with our tracker, that its calibration can't
open. A calibrated own tracker needs nothing from that file, so it gets no
problem. ft-gaze logs the line only after 30 s of eye data with no layout
fitting, so the tracker's warm-up after the headset goes on doesn't show it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With PR #26 ft-gaze reads the SteamOS 0.4.x beta's layout too, but
update-check.py still knew only stable's: on the beta it reported FAIL (warn
without gaze) and said to "update the offsets in gaze/ft-gaze.cpp", as in
issue #15's report, though the gaze worked.
It now tries both layouts with ft-gaze's own test (EyeFile::Detect: the
timestamp within 2 s of the clock and moving on, both set-1 directions unit
vectors), says which one it found, and fails only when neither fits, for
1.5 s so a blink or a moment with an eye lost doesn't fail it. The failure
says to add the new layout to EyeFile::Detect, and to check again after the
tracker's warm-up when the headset just went on. EYE_NEED is ft-gaze's
kNeed now (it was 0x1d3 against ft-gaze's 0x1f3, now 0x1f8). design.md's
update-check sentence and the offset table in gaze/tracker/findings.md say
there are two layouts.
Checked with made-up files: stable and beta ok, the fields moved by 3
bytes, and a direction always zero, FAIL, an eye lost for two samples ok,
an idle counter skip.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gaze/test/mmap-layout-test.sh builds mmap-layout-test.cpp against
ft-gaze.cpp itself in the dev container and plays the loop's passes on a
made-up file with a made-up clock (Detect takes the time), so it needs no
SteamVR and no eye tracker. It checks that both layouts are found from the
next sample, at 90 and at 15 samples a second (as seen on the beta), and at
3; that a writer too slow to confirm, a stopped one, one that stopped just
now, a direction that isn't a unit vector, an empty file and the same fields
moved by any other amount from 1 to 16 bytes are all refused; that a
timestamp read mid-write doesn't end the wait; that a sample is read from
the beta's moved fields once that layout is found; and that the counter
reads 0 without the file instead of touching memory.
Checked against three broken copies of ft-gaze.cpp: without the "timestamp
moved on" check (6 failures), with a 60 ms window (2), and with the
counter read unguarded (a segfault).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
design.md's notes on KWin's script engine get the quirk the watchdog from #35
is for: callDBus never calls back when a call fails (an error reply, the name
gone from the bus, or the 25-second timeout), and KWin only logs it. The long
poll's description says how the watchdog calls again (after 30 seconds, then
longer each time while no reply comes) and why the 30 seconds have to stay
above KWin's timeout. floating-windows.md's line on the poll says the script
also calls again after 30 seconds without an answer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
While ft-floatd isn't running, the watchdog fired every 30 s for as long as
the desktop ran, and each fire was two lines in the journal: the script's own
and KWin's "Received D-Bus message is error" for the call that failed. That
was about 5800 lines a day, for nothing: only starting ft-floatd again helps,
and it reloads the script when it starts. ft-floatd starts from the session's
autostart, and nothing restarts it.
The script now says it once per outage, and the next reply clears that. Each
fire after the first waits twice as long, up to 5 minutes, which also cuts
KWin's line. A reply starts over at 30 s, so a lost reply is still polled
again after 30 s.
In the poll model: ft-floatd gone for 10 minutes went from 20 fires and 20
prints to 4 fires (after 30, 60, 120 and 240 s) and 1 print. A lost reply
still recovers at 30 s. Two lost in a row, with no reply between them, now
wait 30 and then 60 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The watchdog from #35 starts a new NextCommand call when the old one hasn't
answered. If the old call's reply still came after that, its callback started
another poll, and two polls would answer each other for good: ft-floatd answers
a waiting poll empty whenever the next one arrives, so KWin and ft-floatd would
trade D-Bus calls in a loop until the script reloads. Each call now has a
serial. A reply to a call the watchdog gave up on still runs its commands
(ft-floatd sends each one once), but only the current call polls again.
With the period at 30 s that reply can only come if KWin stalls with it
queued: KWin's callDBus ends every call by its 25 s D-Bus timeout. The period
was ft-floatd's POLL_SECONDS plus 10 s, though, so lowering POLL_SECONDS
below 15 would have made the loop easy to reach. The period is now a plain
30 s, and the comment says it has to stay above KWin's timeout.
In a model of the poll (the script's code run in Qt's JS engine against
ft-floatd's wait() and QtDBus's timeout): with the watchdog at 15 s and
ft-floatd blocked for 15 s, the old code looped (5000 calls in 10 s), and
this one makes 2. The same with KWin stalled and the timer run before the
queued reply. In normal use nothing changes: the same calls, no fires.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
frametop.conf.example said HANDS_SWAP_SIDES=0, and desktops.sh copies it on a
fresh install, so every new install forced ft-camd's side camera names and
turned the hand tracker's own side check off. Some SteamVR restarts swap those
names, and then hands land beside their cutouts and recordings are mislabelled.
The example now says auto. scripts/conf-migrate.sh, run by install.sh and
hands/rec/install.sh, replaces the old line only where it's still exactly as
the example wrote it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With HANDS_SWAP_SIDES set to 0 or 1, ft-hands kept the forced naming as the
published truth even after the hands showed it was backwards. The hand recorder
took that as the session's decision, so the export labelled slam_left and
slam_right the wrong way round (dataset PR #4: every take swapped).
ft-hands now publishes the hands' answer once they disagree (state "forced,
disagrees"); tracking keeps the forced names. sides.read_live corrects the same
case from an ft-hands built before this, and the recorder's log says to use auto.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The pause section of reference.md now says the relay releases a click
still held as it lets go of the 3D mouse, and lists
input/test/pause-buttons-test.py with the other pause test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
pause-buttons-test checked the pointer's driver_down set and filtered
the sent commands down to "btn" lines. On the code before the fix it
stopped at a missing attribute, before any behavior was checked, and it
couldn't see where "hide" went. The order matters: a release after
"hide" wakes a helper that wakes on any btn.
It now compares the exact commands stand_down sends (["btn trigger 0",
"hide"] for a held left button) and adds the cases that were missing:
the pointer already off with a button held (the idle timeout,
pointer_toggle), a mapped controller button and a key combination, and
gaze holds, which are the helper's to end. Its labels now say what each
case is: they called a plain mouse press a gaze drag. Against the relay
before the fix, 5 of its 11 checks fail; against the contributor's
commit, the 2 pointer-off ones.
keys-test gains a pause check through the relay's main(): in pointer
mode, a left button held into a pause comes up, then the pointer hides,
its release during the pause reaches no one, and clicks go to the helper
again after resume.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two installers give ft-camd its capabilities: hands/run.sh install and the
Hand Recorder's (through hands/run.sh caps). #30 made hands/run.sh uninstall
take them back unconditionally, before it removed the units. With the
recorder installed too, its next session then stopped at "ft-camd needs its
capabilities", and so did ft-cutouts. The recorder's own uninstall, the one
hand-recorder.md points to, still left them on the binary.
One rule now: hands/run.sh uncaps takes them back only while neither the
Hand Recorder's menu entry nor the frametop-camd unit is installed, and says
which one keeps them otherwise. It checks through on_frame, in one call, so
it works from a PC too. hands/run.sh uninstall runs it after removing the
units, and hands/rec/install.sh uninstall after removing the entry, so each
stops counting itself. uninstall.sh removes both and always takes them back.
The docs say which commands ask for the password, and AGENTS.md says
uninstall.sh also has to undo file capabilities.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR #26's detection slept 60 ms per layout inside ft-gaze's 4 ms loop, so no
head poses and no SteamVR events in that time, and it needed a new sample
inside those 60 ms: at 15 samples a second (seen on the beta) about 1 try in
10 missed, at 10 a second half of them. It also ran once at startup and
again on the loop's first pass, and again after every pause of 2 s or more,
where a failed try (SteamVR's tracker warming up after the headset went on)
threw away a layout that was known. That can't help: the layout changes
only with a new eye server, which comes with a SteamVR restart, and the gaze
service stops and starts with SteamVR.
Detect is now a step per pass: it notes the layouts that fit (timestamp near
the clock, unit set-1 directions, as before) and takes one once its
timestamp has moved on, within 0.5 s, so down to 2 samples a second. It runs
only while no layout is known, starting when the counter moves, and the
counter is read at startup so a file left from a stopped server doesn't
start it. "Not recognized" goes to the journal only after 30 s of writing
with no layout fitting, past the tracker's warm-up (about 20 s), so the
warm-up no longer logs it. The startup line with the layout is one of the
probe's startup lines now, not an error worth repeating when ft-gaze stops.
TimePlausible is static, and the size check Open's kNeed already covers is
gone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
In gaze mode a mouse press is held back while you aim, and its release
clicks. A release of the right button sets pressRight with the click, so
the click goes out as a right one. When the pointer goes off in the same
loop (the relay's stand_down sends the release, then "hide", as Frametop
pauses), the click is dropped, but pressRight stayed set. The next press
after resume, from the mouse, Meta+J or a pinch, then went out as a
right click.
Dropping the click now forgets pressRight with it. A "hide" read a loop
after the release still comes too late to stop the click; that needs the
helper's socket to be full as the pause starts, and is noted in the
relay's stand_down.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With PR #26, ft-gaze printed nothing at all while the mmap's layout wasn't
known: not on a layout it doesn't recognize (the next SteamOS update that
moves the fields), and not while SteamVR's tracker warms up after the
headset goes on, when the counter already ticks but the timestamp and the
directions aren't usable yet. Our own tracker, the default, needs nothing
from that file, but its gaze rides on ft-gaze's lines, so it stopped too.
Without a usable mmap (none, or its layout not known yet) ft-gaze now prints
at 90 Hz, as it always did without the file, and the mmap's sources print
as {"ok":0} and "eye" as null. They have to stay out: the unread sample's
zeros would print an "unc" of 0, which gazecheck takes as SteamVR seeing
both eyes, and checks would start on nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Any "btn" or "scroll" from the relay woke a pointer that was off. When
Frametop pauses for a game, the relay's stand_down sends the releases of
the clicks held into the pause, and with its pointer already off (the
idle timeout or pointer_toggle during a press) it used to send them with
no "hide" after. The helper woke on them, connected the virtual
controller and took the right hand role during the game.
Now a button release or a scroll back to 0 0 doesn't wake it. The
release is still handled as before (it ends its press, or goes to the
driver), so the driver doesn't reconnect later with the button down.
Outside a pause this changes one case: a mouse button let go after a
moving controller released the pointer mid-press. That release used to
reconnect the controller, recenter it and release there; now it updates
the disconnected controller, and ft-screens still gets its "up".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since b61b4d6, <file>.ft-bak is written only before the first repair. That
keeps the first config, but no repair after it had a backup, and the log
still said "(backup: <file>.ft-bak)" for each one. The repair writes 0 over
a moved panel's old screen number, so a backup is the only record of where
the panel was. Restoring the file the log named after a later repair also
threw away every Plasma change since the first one: panels, widgets, pinned
apps, wallpapers. Repeated repairs are the case b61b4d6 was written for.
Tried on a copy: after a first repair, a new top panel on screen 2 with
pinned apps, and a drop to one screen, the second repair moved that panel
and named a backup that had neither.
Before each repair the config now also goes to <file>.ft-bak.last (whole or
not at all, like <file>.ft-bak). <file>.ft-bak stays write-once. The log
names both, and the docstring and docs/design.md describe both. Both live
in ~/.config/frametop, which uninstall.sh already deletes with the
settings.
The new test repairs twice, with a widget changed between the runs, and
checks that .last holds the changed config, .ft-bak the first one, and the
log names both. test_repair checks that the first repair writes both and a
run with nothing to repair writes neither.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR #26's layout check read the sample counter on every pass of the loop,
also when /dev/shm/eye-server.mmap wasn't mapped. Without the file the
mapping is a null pointer, so ft-gaze died with SIGSEGV on its first pass,
and the gaze service started it again every 3 s: another container-up.sh,
distrobox enter and SteamVR client each time, while the path that prints at
90 Hz without the file (our own tracker needs nothing from it) could no
longer be reached.
The counter is now read only through the mapping (EyeFile::Counter, 0
without one). And since SteamVR's eye tracker creates the file a second or
two after SteamVR starts, an ft-gaze that started first now looks for it
again every 2 s, one open each time, instead of going without it until the
next restart.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stand_down sent "hide" only while the relay's pointer was on. It can be
off with a button still held: the idle timeout or pointer_toggle during
a press. Then the pause sent "btn trigger 0" on its own, the helper woke
on it (any btn or scroll wakes it), connected the virtual controller and
took the right hand role during the game, and nothing hid it again: the
relay doesn't tick a paused pointer.
Now every release stand_down sends (held buttons, the system and claim
pulses, a scroll pulse) is followed by "hide". The helper also stops
waking on releases (next commit), but the relay and the helper update
separately, so the relay can't count on that.
The comments now say what driver_down covers: the left, right, middle
and back actions from a mouse, a mapped controller button or a key
combination. Gaze holds go to the helper as gazekey, gazedrag and
precision, and it ends them itself when the pointer hides. The docstring
also records the one case left: a gaze-mode press held back for aiming
clicks once if the helper reads "hide" a loop after the release.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
shutil.copy2 wrote <file>.ft-bak in place, so a copy cut off partway (a
full disk, a crash, the battery running out) left a short file. Since
b61b4d6 the backup is written only once, so that short file stayed for
good: os.path.exists finds it, the next repair skips the backup and goes
ahead. Before, the next repair at least wrote it again. With a 1 KiB limit
on file size, the first run left 1024 of the config's 2536 bytes, and the
second kept them and repaired.
The copy now goes to <file>.ft-bak.tmp, is fsynced, and only then takes its
name with os.replace; a failed copy removes the temporary file. If the
backup can't be written, the panels stay as they are, the reason goes to
the log, and the script exits 1 (the session goes on: it runs fix-panels
with || true). The next start tries again. That was already the outcome
of a failed copy, but through a traceback.
The new test runs fix-panels under that 1 KiB limit and checks that only
the config is left in its directory, unchanged, and that the next run
without the limit backs it up and repairs it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step 1 ran only while something it lists was left, and ft-camd's capabilities
weren't part of that test. When they were the only thing left, a re-run
skipped step 1 and never removed them. That happens to everyone who ran step 1
with the uninstaller from before #30, and to a re-run after the password
failed. Now they count like the rest.
The folded sudo call passed ft-camd as FTCAMD=$(...), an unquoted word that a
repo path with a space split in two: sudo then ran the second half as the
command, and the system files stayed. ft-camd now goes to the root script as
its first argument, quoted, which also stops depending on sudoers letting
VAR=value through. The capabilities-only warning says to run it again, which
works now.
The README and the script's header say step 1 also takes the capabilities.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
test_backup_keeps_the_original ran fix-panels.py against a 24-line shell and
awk stand-in for kwriteconfig6, and its last check (lastScreen is 0) tested
that stand-in's edit, not fix-panels. The stand-in also ignored --key, so it
would go wrong without a sound the day fix-panels writes another key. The
file's test_repair already needs the real tool, which the Frame and the dev
container both have, and the old test passes unchanged without the stub.
The new test puts a backup in place first and checks that a repair leaves it
as it was and still moves the panel and its tray. It fails on the code
before b61b4d6 (the backup gets overwritten) and passes on it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
posix_spawn also fails when the log can't be opened: with 9c66475's
O_NOFOLLOW, a symlink or a directory where the /tmp fallback's log goes
makes it fail with ELOOP or EISDIR, and ft-layout doesn't run. The message
said only "can't run .../ft-layout", which reads as a missing ft-layout.
It now names the log and the error.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
After 9c66475 and the previous commit, ft-layout's log is
$XDG_RUNTIME_DIR/frametop-layout.log, but report.sh still collected
/tmp/frametop-layout.log, so bug reports lost it, and reference.md still
named the /tmp path.
report.sh reads the new path (it already sets XDG_RUNTIME_DIR to
/run/user/<uid>). reference.md and the comment on RunLayout name the host's
runtime directory, which ft-screens and the desktop start share. The nested
desktop has its own, /run/user/<uid>/frametop, removed at every desktop start
and stop, and the log isn't there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
9c66475 moved the log of ft-screens' ft-layout runs to $XDG_RUNTIME_DIR, but
the desktop start still wrote /tmp/frametop-layout.log with a shell redirect,
which truncates and follows a symlink planted there. The fixed /tmp path stayed
in use at every start, and the layout's output was split over two files, one
of them never truncated.
The start now writes $XDG_RUNTIME_DIR/frametop-layout.log, the host runtime
directory ft-screens gets from this script, so one file has the run from the
last desktop start and ft-screens' runs after it. The script already needs
XDG_RUNTIME_DIR here (set -u; ft-screens' socket check reads it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
