Multicast was verified on a loopback group (239.255.42.99 on 127.0.0.1); a real network was not exercised.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The E-1608 source marks every restart as lost and logs each overrun; its form stops touching
siblings after the chooser destroys them; the link skips command id 0, retries a full-buffer
discover send and ignores a zero-length scan read. Session files escape \n and \r; get_int and
get_double reject out-of-range values; the UDPS port spin reports changes; a skipped source panel
says why on stderr. The E-1608 probe now checks the signal names and the reconnect itself.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Over SSH with X11 forwarding the app does not appear: no usable Vulkan device means GTK cannot
create a swapchain, and `--renderer cairo` fixes that half only -- the plot is a GtkGLArea and asks
for its own GL context, so indirect GLX is the second wall. A virtual X server on the machine that
runs the app sidesteps both, because the GL context then lives next door to the samples.
`tools/udpscope-remote.sh` starts Xvfb on a free display, runs the app there with
`--renderer cairo`, and serves the screen through x11vnc bound to localhost only -- the SSH tunnel
is the authentication, so `-nopw` is deliberate and `-localhost` is what keeps a mistake from being
a remotely reachable screen. It prints the tunnel and viewer commands rather than assuming a
client, mentions the websockify/noVNC route for a browser, and takes the X and VNC servers with it
when the app exits. `--no-vnc` leaves just the virtual display, for another viewer; the preflight
names the install command instead of a bare "command not found".
Verified: the preflight refuses with exit 1 and the install line when x11vnc is absent; with
`--no-vnc` the app runs against the virtual display and reads `buffer allocated: 4 channels` with
no Vulkan warning and no critical, and no Xvfb is left behind when the timeout ends it. That run
also exercises what the earlier log showed by omission: nothing is enabled, so a working app
plots nothing.
The README's Limitations gains the bullet a person actually needs -- the exact warning string, that
both halves need the software preset, the script itself, and that a window which appears but stays
empty usually means no signal is enabled.
GTK4 defaults to Vulkan, and an X11-forwarded display has no device to make a
swapchain from; the plot is a GtkGLArea, so it then asks for a GL context that
is not there either. --renderer cairo is the software preset for that display:
it exports GSK_RENDERER=cairo and LIBGL_ALWAYS_SOFTWARE=1 in main() before
gtk_init()/adw_init(), because both are read when the renderer and the context
are created. --renderer gl is the faster choice when indirect GL works, and
--renderer vulkan makes the default explicit. Nothing is exported, and GTK's
own choice stands, when the option is absent.
The once-a-second header line is now behind UDPSCOPE_PROBE_HEADER, following
the [notrace] rule: its widths are still sampled for the exit dump, but the
line is detail behind that reading rather than output every run prints.
UDPSCOPE_PROBE_RENDERER prints the exported pair before GTK starts and the
GskRenderer GTK actually built.
README, the manual's overview and docs/claude record the two symptoms and the
cure.
The plotter caps a density grid at kDensityCellCap now (deps/plotter), so a 2 s
window with persistence on no longer scales the image with the window: the
probe's four one-signal rows read 1 044 714 cells / 4 178 856 bytes where they
read 2 226 150 / 8 904 600, and a moved frame uploads 1 044 714 bytes where it
uploaded 2 226 150. The probe's own legs are unchanged (on peak=1.0000,
step-acquisitions peak=1.5000, step-time ratio=0.367879).
The paint probe no longer counts a frame the density is drawing as *traceless*.
`Renderer::draw_traces()` returns while the ink is the image, so every frame of
a persistence display stages no trace: the dump read `traceless paints: 653
(637 after the first bind)` for a run whose real figure is 4 (0), and the
`[notrace]` line printed for each of those frames. It is now behind
UDPSCOPE_PROBE_NOTRACE -- a per-paint line is a log flood and not a reading --
while the counters it details stay in every build, and the line stays reachable
for the state it was written for (a bound row that draws nothing).
Docs corrected where they stated the old rule: the README, the trigger
invariants, the density probe's readings, the verification list (the new
variable) and the manual page behind the panel's tooltip.
The plotter now deposits every acquisition -- the newest included -- into one grid per trace and
draws no solid trace over it, each grid in that trace's colour and with a one-sweep ceiling (its own
commit, **1b8b55d**). This side needed almost no change of its own, because the colour was already
the trace's and the trace's colour is `RowSet::signal_color()` -- the invariants' one route -- but the
figures it reads back are now per trace:
- **The ceiling is left at the library's default of one sweep**, no longer derived from the fade.
Deriving it from the steady state is the gain offset the user reported between the latest
acquisition and the persistent image; a signal crossing the same place every acquisition now settles
at the trace's colour, at full strength, and the newest sweep reads as the trace it is a picture of.
- **`cells`/`grid_bytes` are the sum over the row's traces** -- one grid each, a trace with no series
holding none -- and both the probe and the exit dump carry `traces=` beside `rows=`. That is what
makes the per-signal memory figure read: **556 537 cells and 2.2 MB per signal**, at 8904600 bytes
over the four one-signal rows of the probe's run.
The probes, re-read on this build (the `verification.md` command, `--rate 500000 --batch 1000`):
[density-probe] off enabled=0 ... rows=0 traces=0 cells=0 ink=0.0 peak=0.0000
[density-probe] on enabled=1 ... rows=4 traces=4 cells=2226150 ink=226786.0 peak=1.0000
[density-probe] step-acquisitions ... peak=1.5000
[density-probe] time-before ... peak=2.1359
[density-probe] step-time ... ratio=0.367879
[density-probe] frame cost: tick 0.62 -> 6.96 ms (184/23 frames), draw 2.02 -> 6.16 ms (177/18)
[udpscope] density: frames=748 rows=4 traces=4 cells=2226150 grid_bytes=8904600 ink=489400.2 peak=2.2 upload_bytes=2226150
`on` `peak=1.0000`, `step-acquisitions` `peak=1.5000` and `step-time` `ratio=0.367879` are the legs
that were expected, unchanged; the *draw* mean is now **below** the run before the split (2.02 -> 6.16
against 2.09 -> 20.91), because the solid traces are no longer drawn over the image.
Docs updated with it: the trigger manual page (`persistence` is the trace, one image per signal, the
newest included), the README's persistence section, `invariants-trigger-and-pinning.md` (five rules,
the ceiling and the no-stroke rule among them), `verification.md` and `probe-readings.md` (the lines
above), the trigger panel's comment and tooltips, and the plotter submodule.
`cmake --build build --clean-first -j4`: **0 errors, the documented three warnings** (two
`MathNode::source`, one `-Woverlength-strings` at 12352); `make -C build test-ui` **13 ok**;
`make -C build test-help` **green** (15 pages); the self-check **0 `Gtk-WARNING`/`Gtk-CRITICAL`** and
`gtk warnings: 0`, two `buffer allocated:` lines.
The library's `full_scale` now defaults to the default fade's steady state and `gamma` to 1.0 (its
own commit); this side states the ceiling for the fade *it* configured, because that is what the
ceiling means: the ink the display settles at, so the newest acquisition is a fraction of full ink
and white is reached only where acquisitions agree.
- **acquisitions**: `1/(1 - factor)` -- the geometric sum, 2.0 at the panel's 0.5, and 1.0 at 0
(only the newest sweep, at full ink).
- **time**: the library's default of two sweeps. A time fade has no fixed steady state -- it
settles where the trigger's rate and the constant meet, which is the *point* of it -- so a fast
trigger reads brighter than a slow one the way a phosphor's does.
- `gamma` is left at the library's 1.0: a cell is a count of sweeps, so half as many must read as
half the ink. With the ceiling at one sweep and gamma at 0.45 (the library's first defaults) one
sweep was already full white and every display looked the same -- which is exactly what the user
reported seeing, and why the assertion for it is a framebuffer one in the library's case
(`empty=21 full=83 quarter=54`) rather than anything `ink`/`peak` can say.
The probe's own figures are unchanged in kind (they are the grid's): `on` `peak=1.0000`,
`step-acquisitions` `peak=1.5000`, `step-time` `ratio=0.367879`, `off` all zeros -- re-read on this
build, with the frame cost `tick 0.53 -> 7.49 ms`, `draw 1.89 -> 16.78 ms` and `gtk warnings: 0` in
the same run. The trigger/pinning invariants carry the ceiling rule and why the eye found it.
The trigger pane gains the display the user asked for and the library now provides, and the
persistence it already had is renamed to what it is.
**Persistence is one image; the last-N-traces display is "multiple events".** Both live in a new
`Advanced` expander in the trigger panel, together with the settings that are set up once and left
alone. `persistence` (a switch) turns on the plotter's `Plot::set_density()` for every row:
every triggered acquisition is accumulated into a screen-sized intensity grid drawn behind the
traces, brightness being how many acquisitions covered a cell -- no depth to choose, no memory
that grows with the acquisitions (2.2 MB over four 573x412 rows in the probe's run). `fades by`
picks the fade the user asked for: **acquisitions** (each keeps a factor of what was before it,
0.5 by default -- the phosphor afterglow) or **time** (a time constant in stream seconds, so the
display fades while you watch whatever the trigger rate).
- **Trigger mode only, as asked.** `apply_density()` (the tick) states the display on for `Held`
and `Capturing` and off for `Live`, `Waiting`, `Paused` and a file, and turning it off drops the
grid at once rather than at an acquisition a quiet trigger never sends. It is unpickable and
reserves no layout space, exactly as the invalid-span bands are: a density can never be in the
way of a cursor or a drop, and it cannot move the framing.
- **One deposit per acquisition, one frame after the frame that frames it.** `Plot::accumulate_density()`
deposits what the plot holds through the transforms of its *last* layout -- the frame's own is
computed inside the paint that follows the tick -- so `density_pending` is set beside
`record_pending` and consumed by the next tick. In the same tick it was right on a Held-to-Held
transition and wrong on the first latch, whose axis was still the live ring's. Switching the
display on under an acquisition already on screen deposits it too, so the switch does not appear
to do nothing until the next latch (`single` mode may never send one).
- **The session carries it**: `trigger_density`, `trigger_density_decay` ("acquisitions"/"time",
the words the panel shows), `trigger_density_tau` and `trigger_density_factor`, written whether
or not it is on -- and `trigger_persistence` still means the multiple-event depth, so an older
file is unchanged.
- **The library changed twice under this**: the deposit now counts a cell **once per trace per
sweep** (counting the hits put one dense sweep at the ceiling of the brightness scale and hid the
sweeps behind it -- the probe's `on` leg read `peak=18` for a single sweep and reads `peak=1`
now), and a frame's upload is every row's rather than the last row's. Both are in the plotter's
own commit.
`UDPSCOPE_PROBE_DENSITY_MS` reads it all back off the plots -- the plotter's own `ink()`/`peak()` --
with the panel's own widgets driving the switch and the fade, in four legs: **off** (`rows=0
ink=0.0 peak=0.0000`), **on** (`rows=4`, `peak=1.0000`), **step-acquisitions** (`peak=1.0000 ->
1.5000`, the factor then the new sweep) and **step-time** (`ratio=0.367879`, `1/e` for `dt = tau`).
It also reads the frame cost as an A/B **inside one run** (the accumulators at the switch against
the frames after it): `tick 0.74 -> 9.99 ms`, `draw 2.09 -> 20.91 ms` at four rows and 2.2 M cells
on llvmpipe.
Acceptance: clean build 0 errors and exactly the three documented warnings (12352-byte GResource
blob, two `MathNode::source`), `test-ui` 13/13, `test-help` green, a plain run with **0
Gtk-CRITICAL / Gtk-WARNING** and `gtk warnings: 0`, and the panel's own frame reading
`boxes=16 controls_inside=15/15 advanced_widgets=5/5 advanced_inside=5/5` (the block's controls
counted from the expander's child, which is not in the widget-parent chain -- and the search
block's walk now names *its* expander instead of "the first one", which the second expander would
have stolen).
Docs: the trigger/pinning invariants carry the rule and its five consequences, `probe-readings.md`
the four legs and the dump line, `verification.md` the probe's sequence and assertions, and the
manual's trigger page explains both displays to a user.
Two changes in the trigger pane, landed together.
**Persistence.** Trigger mode keeps the last N published acquisitions and draws them fading toward
the background, N being a panel control (1..8, default 1) with N=1 meaning today's behaviour
exactly, so the knob is the switch. The layers are the accumulate path's own -- one layer per
(signal, age) in the row's free trace slots, named with the accumulate vocabulary so
`layer_base_name()` strips the suffix and `signal_color()` resolves the base signal -- so a layer is
not a signal: the signal list, the colours and the measurements see one row per signal.
`ScopeModel` retains the outgoing snapshot in seven atomic `shared_ptr` slots (no lock, no copy; the
snapshot already exists) and a layer's channel is resolved **by name inside the snapshot**, because
the channel list is renumbered by any structural change since that acquisition was cut.
`RowSet::sync_layers()` re-plans from the rows and the depth alone, so a latch costs no rebuild, and
`event_data_window()` skips layer slots so an older acquisition can never re-pin the axis. The
schedule is 0.5^age -- the acquisition on screen at full brightness and each older one half the
previous, independent of N, so raising the depth adds dimmer acquisitions rather than rescaling the
ones already on screen. Memory is bounded by the ring's own `--buffer-mb` rather than by the depth,
which is the bound that cannot be driven into an OOM, and the panel says when either bound bites.
This also wired up `set_trace_alpha()`, which the accumulate path was setting and nothing was
reading -- so the documented 0.35 translucency of an accumulation was opaque in fact.
**Force.** In the trigger pane beside Arm/Disarm, shown only while trigger mode is on: one
acquisition now, with the configured window. It is a synthetic edge rather than a second publisher
-- the engine opens one capture at the present moment and everything else is the ordinary latch, so
a forced acquisition is cut, pinned, framed, recorded and cached exactly like a latched one, and
the button owns no window and no publish. The pre-trigger gate is deliberately skipped (capture now
is a request for the history that exists, and the snapshot reports its own shorter pre_s), a
capture already filling is refused rather than restarted, and the state afterwards is the mode's:
Normal stays armed, Single disarms. The request is one atomic flag written under the model mutex
with `armed` -- the producer reads `armed` at the top of a packet and the request at the bottom, so
a request written between the two would be dropped -- and consumed at the end of a packet, after
the channel loop and before the pending publish.
**A defect fixed in passing, in the function Force was modelled on:** `on_arm_clicked()` passed
`TriggerConfig::channel` (an index into `channels()`) to `set_signal_enabled()` (an index into
`signals()`). With one signal acquired, arming switched on the stream's clock signal and left the
watched signal untouched -- and since a signal the panel was not showing is a change, it rebuilt
the buffer under the click, dropping the very pre-trigger window the arming, and now the Force, was
about to capture. Measured before: `0.0100 s of history behind it`; after: `2.5299 s`.
`trigger_signal_index()` is now the one resolution, used by both buttons.
A pane is recomputed over the whole visible window on a cadence, and at a 15 s window
(11.8 M samples, 184 k level-0 level buckets, `--buffer-mb 4000` so the stride stays 1)
that is priced per *measured channel* and per *distinct share*. Measured with the new
`[levels-probe]` sweep, per recompute: one window with one level 6.58 ms, four shares
10.28 ms, sixteen shares 24.55 ms, and four measurements one share each over four
channels **35.69 ms** -- against a tick mean of 1.34 ms with an empty pane and 4.58 ms
with that one. That is the reported "very slow with multiple measurements": the pass
runs inside `tick()`, so it is a *frame* cost, and not an ingest one -- `gaps` and
`packets` stay clean even though each `MeasureWindowReader` holds the model mutex for
its whole life, because a socket buffer absorbs a few tens of milliseconds of producer.
The largest single term inside a pass was a call. `measure_accumulate_bucket()` folds
one bucket into one level, and it is the innermost loop of a walk whose iteration count
is the window's buckets *times* the pane's levels; it lived out of line in
`measure.cpp`, so every one of those iterations was a cross-translation-unit call.
Moving it, and the crossing state it updates, into `measure.hpp` as `inline` takes
sixteen shares on one channel from 24.55 to **17.25 ms** a recompute -- with every value
in the dump's `measurements:` unchanged. Two levers remain and both are measured: the
fetch's three out-of-line pyramid accessors plus a 48-byte copy per bucket (3.3 ms of an
11.8 M-sample window's 11.6 ms pass) needs a combined accessor upstream, and `raw` is
the *signal's* share (4.6 % on the mock's noisy channel against 0.1 % on its sine),
because a level inside a bucket's own `[lo, hi]` has to be walked sample by sample.
Bisected with the pane populated (the earlier bisect had it empty, which cannot see
this): the parent of the recent range and HEAD read **8.29** and **7.92 ms** a
recompute for the same four-type pane at the same window, and neither of the two
commits between them touches `ui_measure.cpp` or `measure.cpp` at all. So the cost is
the scale and not a regression, and the fix makes a pass cheaper rather than removing
one -- the plot's own per-frame cost was checked the same way (four commits, equal
within 3 %, and unchanged by the window, the ring size and the visible sample count).
`UDPSCOPE_PROBE_LEVELS_MS` with `UDPSCOPE_PROBE_LEVELS` and
`UDPSCOPE_PROBE_LEVELS_CHANNELS` is the sweep; `docs/claude/invariants-measurements.md`
records the curve and the inline rule, `verification.md` and `probe-readings.md` the
probe and what its numbers must read, and the README the pane's cost at this scale.
A measurement is a statistic of the window the pane is *about*, and on a
pinned view -- a held event, a paused stream, an opened file -- that is the
frame, not the part of it on screen. The pane followed the viewport
instead, so a wheel on a latched event moved `average` and `top`: the
reported "when measuring a still frame, moving the zoom makes measurements
such as average or top change". Nothing about the data had changed; the
user was only looking closer.
The window is `RowSet::pinned_range()` while the pin frames the view and
the viewport otherwise, and the choice is made in `update_measures()` --
the call site -- rather than inside `MeasureWindowReader`, because it is
*booked* as well as read: that window is what the "it moved" test compares
against `App::measure_window`, so a gesture inside the pin spends no
recompute either, and a reader that chose its own would leave a pinned
view recomputing every gesture frame for a number that cannot change.
`pinned_range()` is now the one expression of the pin's extent, shared
with the framing (`hold_pinned_window()`, `fit_x()`), and the reader is
handed the window instead of reading the plot. Live is unchanged: there
the window really moves and the statistic follows it, and the
accumulator's rule about what resets the history is about neither.
`UDPSCOPE_PROBE_MEASUREPIN_MS` is the guard, three legs of before/after a
gesture on the row's own axis. On a latched event zoomed to a quarter
(`view=[-1..1]` -> `[-0.25..0.25]`, `pin` unchanged): `same=4/4
for_same=1 pass_delta=0`, every value bit-identical -- `average=0.000000
top=0.957975 peak to peak=2.000000 amplitude=1.915950` both sides. The
same on a pause taken from it. Since the key `for=` follows `pin=` now,
`[measure-trace]` gained the field, so a held run reads `for==pin` with
`swin` free to sit inside it. On a live view a zoom to an eighth reads
`same=0/4 for_same=0 pass_delta=3 span_after=0.125000 follows=1` with the
values moving -- the case the rule must not swallow.
Before the fix the same probe reads `same=1/4 for_same=0 pass_delta=1
span_after=0.500000` on the held event (`top` `0.957975` -> `0.957966`)
and `same=0/4 span_after=0.125000` on the pause (`top` `0.957944` ->
`0.358614`). Live reads the same either way.
Also fixed, found by the pause leg: pausing while an event is *held*
froze an **empty** picture. `on_pause_toggled()` read `stream_at_x0()`
after setting the pause state, and that accessor follows the display --
`paused` with no pause view published yet answers `-display_epoch()`,
which on a held event is 0, so the cut searched the ring at edge-relative
x, found nothing, and the frame bound an empty snapshot: every
measurement `--`, every ruler gone (`ev=650001` on the probe's line now,
`ev=0` before). The offset and the window are read first, which is the
offset the axis is actually displaying. The live case is unaffected, which
is why the pause probe, which pauses a live view, never saw it.
Docs corrected where they asserted the old half-truth: the invariants now
separate the accumulator's rule (unchanged) from the window a value is
computed over (this change), the trigger/pinning file names the pin as
load-bearing for both the framing and the numbers and records the pause
ordering rule, `verification.md`/`probe-readings.md` describe the new
probe and `pin=`, and the README and the manual say what a still frame
measures.
Checked: `cmake --build build --clean-first -j4` 0 errors and the
documented three warnings; `make -C build test-ui` 13 ok; `make -C build
test-help` green; an exit-dump run 0 Gtk-WARNING/Gtk-CRITICAL and `gtk
warnings: 0`; `[measurepin-probe]`, `[pan-probe]` (`held=1`,
`allowed=1`, `limited=1`), `[pause-probe]` (`event_size` non-zero exactly
while paused, `axis` inside `frozen_x`, `skips` 0 live / 30 paused),
`[stats-probe]` (`dn=1` across the pause), `[zoom-probe]` (`gap_px=0.0`
at every step) and `[measure-probe]` + the dump's `measures
window:` (all-types run `marks_in_window=4/4`, `pin=off` live) on the fast
mock. `measure cost:` is unchanged in kind: `raw=37298 (0.0%)` of
`covered=121704537`, 1.04 ms a pass -- the runs that read higher are the
probe's own zoomed-to-a-slice windows, where the window is legitimately
shorter than a bucket.
Three defects, all of them a window describing data that is not there.
A math edit refilled the derived ring with `clear()`, which renumbers it
from 0, leaving `produced()` short by the inputs' retained base. Since
`live_window_locked()` clamps against every channel's `produced()`, that
one lagging channel answered `n = 0` for the live view, every pause view
and every trigger capture until a whole ring of fresh data had gone by --
the reported freeze, invisible until the ring first wraps, hence its
intermittence. It now restarts with `clear_to(base)`, and
`ChannelBuffer::republish_from(base)` puts the published window back:
`acquire()` is incremental, so a rewrite in place published nothing and
the pyramid went on describing the samples the edit replaced.
The trigger's channel is an index, re-resolved by name after a renumbering
-- but only in `rebuild_enabled_locked()`, so a math add or delete went
through `set_math_channels()` with no restore, the index was clamped to
-1, and `process()` returned at once while the panel still named the
channel. The restore moves into `rebuild_channels()`, which *is* the
renumbering, so no caller can forget it.
And `update_measures()` suppressed its "the window moved" test while an
event was held -- the one mode where a gesture is the only thing that can
move it. Zooming a latched event left the pane describing the whole
captured event. A pinned window needs no exemption: it lands on the range
it already measured and fails the 2 % test by itself.
The checks that missed all three: `UDPSCOPE_PROBE_MATHWRAP_MS` edits a
graph whose ring has already wrapped, which `math_probe` structurally
cannot reach; `[measure-trace]` now prints the mode, the window on screen
and the window the numbers describe; and the dump's `marks_in_window`
counts against the window on screen instead of against the one the marks
were computed over, where it was a tautology.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
**The clock.** Trigger mode showed a blank plot and never latched when the session defined a math
channel, because such a session gives the model a plan *before* the stream's own signals are in it:
one derived channel, declared rate 0, so `plan_partition()`'s 1 kHz fallback is the interval those
early frames are stamped at -- 1300x too fast for a 1.3 MS/s stream -- and they have nowhere to go,
since the value loop skips derived channels. Relative mode then stamped a *position* times the plan's
*current* interval, so when the CONFIG's channels landed and the plan took the real rate (769 ns) the
same positions produced times 1300x *smaller* than what was already in the ring, and the monotonic
clamp held the clock at the old value for the rest of the run: one repeated stamp, `history=0`, no
latch, and a window the renderer cannot draw. The same family as the restarted-client poisoning
fixed in 620473a, through a different door. The filter's own output was finite and correct
throughout, and disabling the math channel only appeared to cure it because that is a structural
change which rebuilds the store -- the third time that channel has looked guilty while being a
bystander.
Relative mode now accumulates *seconds* (`rel_base_s_ += n * dt`), so a re-plan is a slope change and
the clock cannot jump backwards by construction; a frame is only stamped into a model that has a
*streamed* channel; and the epoch is adopted once per connection, with "unset" a state rather than
the value zero. A one-shot warning names the family if it returns: `the clock stalled at ...: N
samples stamped [a..b] are all behind it`.
**The dialog.** The Configurations window's Close button and the window-manager close both did
nothing: `close-request` was connected to a bare `void` lambda, and that signal is `gboolean` -- TRUE
means "handled" and suppresses the default handler that destroys the window, while `G_CALLBACK` is a
cast, so GTK read whatever the compiler left in the return register. A standalone GTK program of that
shape reproduces it: `void` handler -> `window_alive=1` after `gtk_window_close()`, `gboolean FALSE`
-> 0. The same audit found `gtk_window_present(GTK_WINDOW(...))` on an `AdwDialog`, a type assertion
and a no-op; it is `adw_dialog_present()` now. The menu row stays: it is the only way to switch or
save a configuration without restarting, and what made it look useless was that it could not close.
Verified: trigger mode with that session's graph reads `ring=[0.0058..0.2446] -> [0.0058..1.9969]`,
`history 0.237 -> 2.0 s`, `edge at 1.120 s, latched at 2.120 s`, `display=event(frozen) latches=1`,
`traceless paints 266` against 1519 before the fix. Baselines hold: 0 errors, the documented three
warnings, test-ui 13 ok, test-help green, `gtk warnings: 0`.
The user's real host reported "connected and live" with one missing-data band over the whole window
and no trace in it. The counter is the only thing that claims a packet existed, and a reservation
was booked on its word alone: a source whose DATA counter counts samples, or frames of an inner
layer, advances by more than 1 per datagram, the client cannot check that claim (it has no clock of
its own to compare against), so it holds each packet, saturates its reorder ring, delivers it in
order and reports the skipped counters as lost. Every frame therefore reported a phantom gap, and
`reserve_hole()` wrote `lost * batch_` NaN samples with no bound but the 256-PACKET cap -- the
sample form of that rule ("the number to weigh against the ring is 256 * batch_, not the 256") was
never implemented. With relative time on -- which that session carries -- each phantom gap is
materialised as NaN slots by construction: the window filled with reservations, there was never
anything missing to repair them, and the gaps also stretched the relative clock, which is why the
status bar read live at ten times the true rate.
`Acquisition::gap_plausible()` now gates every reservation on two independent tests: the arrival
clock must corroborate the claim (that many missing packets would take that many frame intervals;
deliberately permissive, because a false refusal costs a shading and never data), and the ring must
be able to hold the hole. The verdict is taken before the stamps, and relative mode's slot advance
is skipped with it, since the advance and the reservation are the same claim. Refusals are counted
in `StreamStats::gaps_refused` and said once per connection with the numbers, naming the wire shape
they describe.
tools/mock_udpstreamer.py gains `--counter-step N`: a counter that is not a packet counter on an
intact link, which is the one shape the mock could not produce and the shape the report came from.
Verified in both directions. A counter stepping 1000 per datagram reads `holes=0 hole_slots=0
backfilled=0` with the traces present (3 traces, 1.4 M samples) against 7.9 M claimed gaps, and the
warning above names the cause. Every other packet withheld -- a real loss -- still reads
`holes=3579`, so genuine gaps are still reserved: a fix that stopped believing real losses would
have traded one wrong picture for another. The user's control read as "the math channel is the
problem" and was confounded -- disabling it is a structural change, which rebuilds the buffer and
clears the accumulated NaN.
`top`, `bottom` and `amplitude` became the *flat levels* in e83ad86 -- the level the
signal dwells at, read from the density of the window's buckets -- and that change kept
the split's own two means as its internal fallback for a window the pyramid cannot
summarize. Those means are now three types of their own, so the two different questions
about one split can be asked side by side: `above %` and `below %` are the plain mean of
the samples on one side of a share -- the reading the old `top`/`bottom` gave -- and
`between % %` is the mean of the band between two shares, which no single split can
answer, because a band's two edges are two thresholds.
Nothing about the arithmetic is new, and nothing is a new pass. `Above`/`Below` are the
`sum[]`/`count[]` split `measure_accumulate()` already folds -- the very figures
`Top`/`Bottom` fall back to -- and `Between` is the **difference of two levels' upper
sides**, both of which come out of one pass two over one window. It rides the bucket
fast path unchanged: a bucket wholly inside the band is strictly above the lower level
and strictly below the upper one, so both accumulators fold it whole, and only the
buckets that straddle either edge are opened. `measure_level_pcts()` is the one place the
shares are read, and it clamps both and puts `between`'s pair **in order**, so "between
80 % and 20 %" is the band between 20 % and 80 % -- normalized rather than refused,
because the shares name two edges rather than a direction. Nothing to measure reads `--`,
as for every other type: a share at 100 % has no sample above it, and a band can hold
none at all.
Measured, `[bandavg-probe]`, fast mock at a 2 s window (`--rate 1300000 --batch 1000
--freq 1`):
sine `above 50` 0.636620 (2/pi), `below 50` -0.636620, `between 50 75` 0.255873
square `above 50` 0.986496 (30 harmonics) / 1.000000 (--square-harmonics 0),
`between 50 75` 0.302744 / --
ramp `between 50 75` 0.250000 and `above 50` 0.500000, `above 100` --
The first three are the figures the old `top`/`bottom` documented; the ramp's are what a
uniform density predicts (a band's mean is its own centre); the ideal square's empty band
is the nothing-to-measure rule. The pane offers the types and edits their percentages
through the dialog's second share -- which only `between` shows -- and a measurement's
pair survives add/edit/delete and the session file (`threshold2`), where a file written
before it existed loads at the default.
No raw pass: with the measurements probe adding one of every type at a 2 s window, `raw`
went from 113966 of 240719031 covered (0.0%) to 134978 of 255372033 (0.1%), 0.8 -> 1.1
ns a covered sample.
Docs: the band-average invariant (and the two assertions the flat-level change moved),
the probe readings for the new probe, the verification recipe, the manual's measurements
page and the README.
A bench scope separates Vtop/Vbase from Vmax/Vmin because a band-limited
edge overshoots and rings: the mock's `square` at 30 harmonics reaches
±1.179 where its plateau is ±1. On the rippled square the old readings were
a property of the edges rather than of the waveform -- `top` 0.986496,
`amplitude` 1.179350 -- and `amplitude` reported the overshoot.
`top`, `bottom` and `amplitude` are now the flat levels and their delta.
A flat level is a density question -- the plateau holds most of the
samples and the ringing a few percent -- so `measure_scan_buckets()` folds
a 64-bin histogram of the whole buckets' midpoints while it has the list in
hand (one observation per bucket, each bin carrying its buckets' exact sum
and finite count), and `measure_flat_levels()` takes, either side of the
split, the densest bin holding at least half the densest one's buckets --
the outermost of those, which is what makes a waveform with no plateau read
its peak -- and the exact mean of the samples of the buckets within 5 % of
the window's peak-to-peak of it. 5 % is deliberately under the ~9 %-of-swing
Gibbs overshoot. `amplitude` is `top - bottom` (a scope's Vamplitude), not
the half-swing it was.
Measured, fast mock, 2 s window, harmonics 0 -> 30: top 1.000000 ->
0.999983, bottom -1.000000 -> -0.999983, amplitude 2.000000 -> 1.999966,
while `peak to peak` keeps the extremes (2.000000 -> 2.358701). On the 1 Hz
sine top is 0.957988 (it was 0.636620, the half-wave mean; a sine has no
plateau, so the densest band is at its peak) with sd 0.0014 over a scrolling
window, and `peak to peak` is still exactly 2.
No raw pass and no new fetch: the histogram is the same fetched bucket list
read a second time, gated on a pane that holds one of the three types.
`raw` is unchanged (0.1% clean, 2.1% with the --noise 0.1 interferer,
against 0.1%/2.1% before) for 0.44 -> 0.50 ms and 0.52 -> 0.65 ms a pass.
A window the pyramid cannot summarize -- an hf-filtered walk, or one shorter
than a bucket -- keeps the old readings: the mean of each side of the split,
and the window's peak to peak for amplitude.
Docs: the flat-level invariant (and the two 2/pi assertions it replaces),
the probe readings, the manual's measurements page and the README.
