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>
`UDPSCOPE_PROBE_SIDE_TAB_MS` is the one reading for the design's §5 rule -- the statistics
accumulator keeps running on the hidden page -- which nothing tested before. It switches to
Measure, forces one pass so there is a baseline, switches away to Signals and reads again two
seconds later: `advanced=4` is the assertion, and `advanced=0` is the defect where the
accumulation is gated on the visible tab, which would leave a gap in min/max/avg for exactly as
long as the trigger was on screen.
Two things the probe has to do for itself, and getting either wrong makes `advanced=0` a lie
about the tab:
- it **enables its own signal**, because a signal starts disabled (enabling is what starts
acquisition) and a probe that waited for one to be on would be reading another probe's side
effects -- and `UDPSCOPE_ENABLE_ALL_MS` is not part of this run;
- it **assigns that channel to a plot**, because a channel with no trace leaves the live window
*still*, and a still window is precisely the state `update_measures()` is supposed to skip.
Measured the hard way: without the assignment the probe read `advanced=0 min_n=0` with the view
pinned and only four passes in twelve seconds, none of them the tab's doing.
Docs: `probe-readings.md` gains the probe's entry and the signals panel's new shape
(`rows == signals + 1 + math`, `divider=0` a list that lost its `+`), loses the whole `math
panel:` entry, and follows `pane_sensitive=`, the gone `revealed=` field, the six toolbar
controls and the tab-reading `m`/`g` key pairs. `verification.md` gains the probe's bullet and
the re-measured warning figures. The invariants record the sidebar as a switcher above a stack
with a scroller per page, the hidden-page trap that replaces the unrevealed-revealer one for it,
that a feature's on/off is sensitivity rather than visibility, that a math channel is a row of
the signal list, and that a cursors-off readout blanks to `--` with the panel left on screen.
`shortcuts.ui` and the help page's window and sidebar bullets stop describing five revealers.
The README records the deviation from `SPECS.md` that names the panes as separate collapsible
sections.
Verified: `[side-tab-probe] page=signals rows=4 accums=4 advanced=4 min_n=16`;
`signals panel: labels=2 list=1 rows=9 signals=4 math=4 divider=1`; `panels: … (page_current=1)
(sensitive=0) switcher=1`; `toolbar icons:` six controls with every name `(ok)`, with the
Preferences dialog opened in the same run; zero `Gtk-CRITICAL` and no missing icon.
README gains the three controls where each lives -- the trigger's `band`/`reject`, the
measurement's `noise`/`hf`, and the math channel's `de-noise` node -- each with its unit, what
"off" is, and what it costs, plus the two deliberate deviations from the design spec next to the
features they belong to: the band applies to `edge` and `window` only (a pulse's width is measured
from a crossing to its return, so a band there changes the measured width rather than rejecting
noise; a dropout is an absence with no level to compare), and the band is verified through
`BufferSearch` hit counts rather than `[condition-probe]` latch times, because probe-readings.md
states that a `t0` is not an assertion.
The invariant docs gain the rules a later change must not break: `EventCondition`'s two new fields
and the once-per-packet `configure(dt)` that must not reset; the crossing rule that a change is
between two *decided* sides; the measurement path's exception to "a readout quotes a stored
sample", why the band moves the edge decision only, why `measure_accumulate` loops samples outer,
and why a filtered window disables both bucket fast paths; that `MathOp` is append-only (the index
*is* the serialisation), that the median window is pooled in the runtime rather than held per node,
why the pole reuses `b0`/`z1`, and that median-then-pole is the design; the session's named
`noise`/`hf` keys and its thirteen-field node string that still reads twelve; and the two GTK
traps -- positionally brace-initialised dialogs, and a control that cannot apply being hidden
rather than disabled.
The probe gains a staged chain rather than one phase, and the schedule moves with it:
`UDPSCOPE_PROBE_NOISE_MS` now times the *setup* stage, which installs the de-noise graph through
`apply_math_channels()` -- a structural rebuild that replaces the buffer and drops its history --
five seconds before anything scans it. The legs and the five measurements then run at about
10.5 s, and the reading that says the median rejected every impulse at 15.5 s, so the documented
run needs a 20 s mock and an 18 s exit.
The de-noise channel is installed **disabled** and enabled only at 13 s, for its own reading. An
acquired channel is a share of the ring, so a fifth one shortens every channel's retained span --
and the six `BufferSearch` leg counts and `pkpk bare` are all peaks and counts *of that window*.
Installing it enabled (as the plan had it) moved `bare` from 49 to 21, `ratio` from 7.00 to 3.00
and `pkpk bare` from 2.988 to 2.488, i.e. it would have rewritten the older feature's documented
proof to record a regression that is not one. With it enabled only at the end, every reading from
Tasks 4 and 8 is bit-for-bit what the doc records, and the new line reads
`denoise raw=2.987696/1 denoised=2.000000/1 ratio=0.669 channels=1` -- `denoised` exactly the
clean sine's 2.0, which is the claim the node makes.
Two other findings, both folded in: a derived channel whose source is switched off through the
panel's checkbox follows it off and does *not* come back when the source returns (the enable-all
probe's last act switches `sine` off at ~3 s), and the probes reload leftover math channels from
`$XDG_CONFIG_HOME/udpscope`, which is why a `setup channels=` count above 1 means an earlier run's
session, not a defect.
`math_probe` gains a third channel with a `DeNoise` node so the editor's median row is built and
synced on a baseline run: `3 math channels on 'sine' and 'noise'`, every other line as before,
`math cost: 103.3 ns/slot` (one reading of the same path).
verification.md: the recipe, and why the two stages may not be collapsed -- a probe that rebuilt
the buffer under its own legs reads `bare=0` and looks like a broken comparator. Also that the
noise and math probes may not share a run: both write `app.math`.
A derived channel cannot be computed from a signal nobody acquires: `Source` resolves to a
*channel*, and a disabled signal has none -- so a channel that reads one silently computed
nothing, which is what "all math signals depending on it do not compute anymore" looked like
from outside.
Rather than leave it enabled and blank, the channels that read the signal follow it off:
`disable_dependent_math_channels()` walks the graphs to a **fixpoint**, because one math
channel may name another, and the switch goes through the same path the pane's own toggle
uses -- `MathChannel::enabled` then `apply_math_channels()`, the single writer -- so the
pane's switches follow. Their traces leave the plots with the signal's, and the notice says
both halves.
The channels that do *not* read it are untouched, which is the assertion's other half:
[enable-all-probe] switched 'sine' off: traces_left=0 (want 0)
math_off=[math x2,math3] math_on=[math lp]
notice=''sine' is off: 1 trace left the plots; math off: math x2 and 1 more'
`math x2` and `math3` both source `sine`; `math lp` sources `noise` and stays on. The probe's
switch step now runs late enough that a math probe armed behind it has installed its graphs,
which is why the earlier reading was empty -- `app.math` had nothing in it yet, not
"no dependants".
Re-enabling the signal does not switch them back on, for the same reason it does not restore
the traces: that is the user's decision, and the rule is written down in
invariants-math-channels.md.
`set_math_channels()` rebuilt the store for any change, and a rebuild replaces the buffer:
every channel in the window started again, which is the "modifying one expression invalidates
all the others" report.
The plan is a *partition over the channels*, so it only moves when the set moves. When the
set is unchanged -- same channels, names and enables -- nothing is rebuilt now: the store is
untouched, every other channel keeps its samples and its runtimes, and the surgical path
prepares and recomputes only the affected graphs. The affected set is the edited channels plus
whatever *names* them, because one channel's output is another's input, and the recompute
computes the whole list in order (the inputs have to be there) while writing only those in
the mask -- which is why the step-loop is now `math_apply_locked(sample, write_mask, timed)`.
Adding, removing or renaming a channel still moves the plan, and still rebuilds.
[math-probe] graph edit: stored_slots 800 -> 900 (want the second at least the first, and non-zero)
The count grows because the stream keeps filling while the edit runs; the point is that it
never falls to 0, and that every channel afterwards still reads `N samples vs M time slots`
with N == M.
Three tightenings, all measured with the reloads now reported on stderr.
`set_math_channels()` treats an identical list as no change at all. It rebuilt
unconditionally, and a rebuild replaces the buffer and drops its history -- which a file
view then has to re-read from disk. The comparison is the graphs', and it leaves the nodes'
`x`/`y` out deliberately: those are the canvas's own, no op reads them, and a graph that
merely *moved* should not cost the buffer.
`reload_file_view()` fires only when the store is actually empty. A change that really
happened leaves it so in a file view (the rebuild replaces it, and there is no producer to
refill it); one that found nothing to do leaves it full, and reading a 200 MB file again for
that is the waste. `ScopeModel::stored_slots()` is the one-line accessor it needed.
Every reload now says so, because a second read of a file is worth accounting for:
[udpscope] reloaded <path> after a structural change: 402 samples back
Measured, `--open` a file:
session with no graphs 0 reloads
the math probe (3 real changes) 3 reloads, and all the readings stay equal
(`math x2 402 samples vs 402 time slots`)
I had offered this as "one extra read at `--open` startup", and that was wrong: with no
session there is no `apply_math_channels()` call at all, so there was never an extra read
to remove. What the tightenings buy is that a no-op application is now free, and that a
reload cannot happen unless the buffer was really lost. Each *genuine* change still costs
one re-read -- a rebuild drops the history by design, and preserving it across a rebuild is
a model change rather than this one.
Two defects, both reached by adding a math channel to an open waveform or a replayed
event, and both measured before and after.
**A file view has no producer, so a structural change left it empty.** `set_math_channels()`
rebuilds the channels *and* the buffer, and a rebuild drops the history: right for a stream,
whose producer refills it from the next packet, and wrong for a file, which has nothing to
refill it with. The `--open` dump read `0 samples vs 0 time slots` for *every* channel,
the file's own `sine`/`square`/`ramp`/`noise` included, though the load had reported 402 --
that is what "adding a math node breaks the plotter" was. `reload_file_view()` re-reads the
file after such a change, which is also what computes the derived channels over it: the
load's own pushes run the math once the set is in place, so `--open` with a session that
defines graphs works for the first time.
**The file path pushed into the derived rings, and so did the engine.** `push_loaded_slot()`
did not honour the rule `ChannelInfo::math` states for the packet path ("the ingest must skip
it"), so each loaded slot put a value in a derived ring and `push_math_sample()` then put the
computed one: two samples per slot, a trace at double rate -- the derived channels read
`804 samples vs 402 time slots`. Skipping them there is the whole fix.
Readings now, all equal, which is the assertion this file's own invariant makes:
--open the file: math x2 402 samples vs 402 time slots, y=[-1.9021, 0.0063] (2x the sine)
math lp 402 vs 402, mean=-0.4900 (a real filter output, not zero)
live: math x2 25100 vs 25100
replayed event: math x2 4000 vs 4000
I first tried filtering the file's *derived columns* out by name, since an export writes
every plotted channel. It removed the double count and zeroed `math lp`, because the column
it dropped was the source a session graph reads -- so it is reverted: those columns are
legitimate data, and the fix belongs in the push, not the loader.
Moving the editor's frame into math-editor.ui left `math_body` and `math_actions` as
<child> elements of the *window* rather than of `math_outer`, because those edits were
anchored on the file's tail and closed the outer box two elements early. A GtkWindow
takes exactly one default <child>, so nothing failed loudly: the body -- with the canvas
and the parameter form inside it -- was left parentless, visible but never mapped, 0x0,
drawing nothing. Two symptoms, one cause: the canvas is blank, and the window is sized
around the head row and the buttons alone.
Measured against the pre-move revision in a worktree, same probe, same environment:
before this fix `canvas=0x0 mapped=0 window=756x940`; after it `canvas=420x836 mapped=1
window=938x940`, which is the old revision's own reading exactly.
It survived because that slice's probe read the editor's *values* (nodes, gain, output,
name, unit) and nothing about its structure. Every other panel moved that day has a
structural reading -- `controls_inside=13/13`, `nested=1`, `labels=2 list=1` -- and none
of them was affected. `[math-probe] editor frame:` is that reading now: the canvas's
allocation, its mapped state, and whether the canvas and the name entry are inside the
dialog at all, delayed 500 ms because an allocation only exists after a layout. In the
broken build it read `0x0 mapped=0 in_dialog=0`.
check_ui passes both files: the XML was valid and every property type-checked.
The second half of this plan row, and the half of that dialog which is template.
src/ui/math-editor.ui holds the window (its size and modality), the box around it,
the identity row -- the name and the unit -- and the body with the canvas in it.
What stays in code is the drawing: the canvas is a GtkDrawingArea the file *declares*
and the code draws, with the gestures that make it a graph editor, and the parameter
form follows which node is selected. The palette and the buttons go next.
The builder is held on `App` and cleared on the *next* opening rather than with the
window: the window belongs to the builder's object list, so freeing the builder under a
live window frees the window with it. That is the source form's pattern, and the probe's
close-then-reopen path is what exercises it -- `reopened: dialog=ok, nodes=2`.
The identity row is read back too, since a `.ui` can hold the widget but not the value:
`dialog build: ... name='math3' unit='' ...`, the entries having been filled from the
channel being edited. check_ui is at 12 ok, 0 failed.
`math_dialog_sync_form()` wrote the label and then moved a node whose source
names an unacquired channel -- a mutation that changes what
`math_dialog_problem()` reports. With the sync re-entry now guarded, the nested
call that used to repair the label is suppressed, so the line kept describing a
source that had just been fixed and left `Apply` insensitive until the next
mutation. The reconcile moves above the write and the dropdown reuses its list.
A review of the whole node editor. The hit test measured the distance to the
24 curve samples rather than to the segments between them, so a wire spanning
several rows was unpickable between two samples -- the probe now places its
boxes 280 units apart and prints the nearest sample's own 8.4 against a 3.0
tolerance, which the point test would have missed. A refused connection said
nothing: the message is now carried in `MathDialog::refusal` and shown once by
the single writer of the label, and a self-wire is refused through the real
drag-end path instead of only through `math_connect()`.
Two defects were found only by the new assertions. `math_dialog_sync()`
re-entered itself -- `gtk_drop_down_set_selected()` emits `notify::selected`
for a programmatic set too, and the handler answers by syncing again, which
consumed the refusal before it could be drawn -- so the body is now guarded
one level deep. And a deleted node left `selected_wire` naming an index that
afterwards belongs to a different wire, so `Delete` cleared an input on the
wrong node.
A node is placed in the first free grid cell rather than at the end of the
list's own cell, so a node added after a delete no longer lands on top of a
survivor; the label ellipsizes with a tooltip rather than wrapping, because a
wrapped line makes the dialog's height depend on the text in it.
Each invariant records what went wrong rather than only the rule: a port hit
that lost its node, a hit test at the identity view that asserts nothing, a
`Delete` that must not fall through to the selected node, and a column pitch
too narrow for a wire's own tread. The warning baseline drops to four.
The file was 1967 lines and is auto-loaded into every context, so every
session paid for the whole of it to reach one invariant. It now carries the
project's goal, the three facts that are always relevant, and a table of
contents; the body is fourteen files under docs/claude/ reached by link
rather than by import, so nothing is loaded until it is asked for.
The content is moved verbatim -- the only edits are each file's own header
and the cross-references that became cross-file ("see the invariant below"
is now a link to the file that owns it).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
