src/ui/main-window.ui holds the window, its two panes, the plot column with the status line under it,
and the sidebar's scroller. What stays in code is what depends on the stream: the events pane, the two
plot areas, the sidebar's panels and the status popover, each put into a slot by id.
The header bar is deliberately not in the file. It is the window's own titlebar, every button on it is
wired to an action, and an AdwApplicationWindow refuses a titlebar outright -- a fatal Adwaita-ERROR
rather than a warning that leaves the window merely ugly -- so the window is a GtkApplicationWindow and
the header bar stays where it was.
gtk_paned_set_position() is the other thing that had to stay in code, and the reason is order rather
than shape: it clamps against its children *now*, so stating it before the sidebar's panels exist moved
the plot column by ~180 px.
The layout, events and box-zoom probes read what they read before this change, and the header reads
window_button_w=130 bar_w=756 as it always has -- checked by building the previous commit and running
the same probe against it, because the layout numbers in the notes turned out to be stale from an
earlier session. Five UI files now.
src/ui/source-form.ui is the form a source is added or edited with: an AdwDialog of
AdwPreferencesGroup rows (AdwEntryRow, AdwSpinRow, AdwSwitchRow) with its action in the header bar,
where the label says which mode the form is in -- Connect for a new host, Save for an existing
configuration. The grid of labels and bare entries it replaces had its hints in placeholders, which
vanish as soon as there is a value; they are the groups' descriptions now.
src/ui/help.ui is the help window's chrome: an AdwWindow with a header bar and an
AdwNavigationSplitView whose sidebar is the page list. The rows stay in C++ because the fourteen
pages are data, not template.
Four UI files now, 653 lines of template that used to be widget construction. Two libadwaita traps
cost a run each and are in docs/claude/invariants-gtk-and-chrome.md: an AdwDialog has no content
property (its child is a plain <child>, unlike an AdwToolbarView), and AdwEntryRow has no subtitle
(it is an AdwPreferencesRow, not an AdwActionRow). The unused form_field() helper went with them.
Every probe reading is unchanged -- [sources] form:, [sources] edit form:, [help] -- which is what
refactoring a shape rather than a behaviour should look like.
src/ui/preferences.ui is an AdwPreferencesDialog whose rows are libadwaita's own -- AdwEntryRow,
AdwSpinRow, AdwSwitchRow, AdwComboRow, AdwButtonRow -- instead of a label and a bare GtkEntry in a
GtkBox. There is no Apply button: a row applies as it changes, which is what those widgets are
designed around, and [prefs-probe] reads the settings file back after every change.
The folder chooser is cancellable now, so a picker still open when the dialog closes cannot write
into a row that is gone; the cancellable is also how the probe takes the cancel path without a
mouse.
Two traps cost a run each and are recorded in docs/claude/invariants-gtk-and-chrome.md: an
AdwDialog is not a GtkWindow (it cannot parent a GtkFileDialog, and in floating mode it adds a
native surface to gtk_window_get_toplevels() that is neither a widget nor a window), and a .ui
that is not listed in the GResource is a Gtk-ERROR at runtime rather than a build failure -- which
is exactly what the probe opening every window is for.
GTK4 and libadwaita. It receives one UDP stream through the vendored
udpsclient, buffers it in a large ring, and draws it with the vendored
plotter: a trigger with a pre/post event window, a pause that freezes the
window, cursors, measurements, math channels in a node editor, an event cache,
named configurations, preferences, and an export of the visible window to
.npy/.csv. Waveform files can be opened and analysed in place.
The interface is two threads -- a producer that owns the socket and the GTK
main thread that owns every widget and the GL context -- sharing one model
under one mutex, with samples crossing through lock-free rings.
deps/plotter is a submodule pinned to the commit this tree builds against;
deps/udpsclient is vendored in place. docs/claude/ documents the rules that
cost a defect to learn, and SPECS.md is the requirement document.
Re-measured all four recorded cases and added a fifth for the re-plan path,
which needs 100 kS/s: at 10 kS/s the ring already covers 838 s, so no window
the control can reach moves time_stride and the window probe re-plans nothing.
Also notes that `stale` is 0 by construction while resequencing is on, and that
the reservation cap is counted in packets, so its sample cost scales with the
batch.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The lookup compared raw uint32 counters with >= and <, so a hole spanning the
2^32 wrap matched nothing and every late packet belonging to it was dropped. A
signed delta, the idiom deps/udpsclient already uses throughout.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Four latches were process-lifetime -- two of them function-local statics, so
process-wide rather than per-object -- and a condition that warned once stayed
silent about every later occurrence in a long-lived session. A diagnostic that
goes quiet after the first occurrence is the failure mode worth fixing in a
feature whose job is diagnosis.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
rebuild_locked() replaced the store, so produced() restarted at 0 while the
producer's tick_ kept counting: a hole's slot and tick addressing then described
a buffer that was gone, forget_holes_before() could not retire it, and a
backfill's amend fell outside the ring -- repairing nothing while the hole was
still marked repaired, which is the band reporting the opposite of the truth. A
re-plan generation is the signal rather than the stride/channel-count guard,
which a reconnect with an identical CONFIG leaves unchanged. slot_base_ is left
alone: it is relative mode's clock origin and must not walk backwards when the
user retypes the window.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
A packet that needed a slot and found none was neither held nor delivered: no
counter recorded it and the consumer reserved a hole nothing could ever backfill.
It now commits the gap to free a slot and takes the ordering decision once more
against the sequence that commit left behind, bounded to one retry; past that it
is delivered in order with the counters it skipped as its `lost`, never flagged
late, because a packet ahead of the sequence has no hole to be looked up by. A
duplicate can now replace its slot at a full ring, as hold_packet's own comment
already promised, and the 20 ms default window is a named constant.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fix the false trailing clause in the udps_stats_t::late_resequenced comment,
move udps_held_t from the tunable-limits section to internal-state next to
udps_slot_t, and add braces to three braceless if bodies to match house style.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
.superpowers/sdd/ is scratch and gitignored; that one report was tracked
by accident in an earlier phase.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The four original items are done. What replaces them: the invisible
riser, relative timing, named configurations, frame storage for
forensics, opening exported waveforms, the plot context menu, drag
resize, an event cache with replay, and a preferences dialog.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
--drop-every/--drop-burst withhold datagrams the counter and stream clock
have already advanced over, and --reorder-every inverts a pair on the
wire; both are deterministic for the reason --glitch-at is, since a probe
asserts the spacing of what it finds and a random count is not a figure.
A `noise` signal gives the dense-trace and intensity-floor paths a input
that is not piecewise smooth.
verification.md records the measured readings, including the one that is
a finding rather than a baseline: a burst past kGlitchPackets (256) is
read as a producer clock glitch, each bridge advances the anchor by one
packet span against a timeline 300 ahead, and four consecutive failures
then throw the stream onto the arrival clock.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
notify::selected passes a GParamSpec* ahead of the user data and
GtkSpinButton::value-changed does not, so on_math_param_changed read the
pspec as its MathDialog* and faulted on the first field it touched --
which is to say on opening the math editor and picking a source.
The probe missed it because filling the form sets the dropdown to the
value it already holds, and GObject emits no notify for that: a channel
sourcing channels().front() picks index 0 on a dropdown already at 0.
math_probe now opens on a channel sourcing the *last* acquired signal,
which does move the selection, and reads back the source the handler
wrote into the node.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two complaints, not one: the glitch is a row dropped before any style
knob is consulted, and the thinness is a choice udpscope made and never
revisited -- it asks deps/plotter for a trace thinner than the library's
own default.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Twelve tasks in five phases, so relative timing ships before any of
the client-side reordering work lands.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The two were asked for together and are one design: the packet counter is what
makes a hole addressable, so relative timing is the coordinate system in which
resequencing is arithmetic rather than inference.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The earlier correction moved the claim rather than fixing it: four places still
said the float-equality test "never fires on a real-valued signal", and a +/-1
square is a real-valued signal -- a standalone check against a band of [1, 1]
fired 19 times, because `inside` is true at +1 and false at -1, so the signal
changes side on every transition.
All four now state the measured behaviour: an exact float-equality test against
the level, which an ordinary signal never satisfies and a signal sitting at the
level satisfies on every transition. Neither answer is a band, which is the
reason the second spin is clamped an increment away rather than the detector
special-cased -- the same conclusion, now resting on a true premise.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
A finite trigger_time() only says the edge was accepted; the mode is Capturing
until the post window fills. Every field the report prints is read from
display_snapshot(), whose stamps span x = 0 only once the event is Held, so
reporting on the latch alone raced the capture: leg 2 printed
`measured=nan at0=nan ... mode=1` on one run and the documented
`at0=-0.500907 earlier=-0.498185 mode=2` on the next. The report now waits for
Held under the same poll budget, and says `latched but never held` when it runs
out -- a different defect from `no latch`, so the line names which.
The readings are corrected against three consecutive leg-2 runs and one leg-3
run: `t0` is a rebased stream stamp and varies every run, so it was never an
assertion; `hits=` likewise tracks the arm phase (12 at one arm, 11 at another).
The mode ordinals were wrong -- `mode=2` is `Held`, not `Capturing`. Shapes 2
and 3 are now measured rather than read from the arm code; shape 1 still is not.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
--glitch-width defaulted to 0.0, so --glitch-at alone produced no glitch and no
warning — the same class of silent no-op that cost a debugging cycle earlier on this
branch. The default is now 0.002 s (2 ms), matching the standalone probe run.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- 08-trigger.txt: names all four condition shapes and the buffer search; fixes the first
paragraph and 'How to reach it' to match what the panel actually contains (fix 3).
- 10-measurements.txt: adds a Statistics bullet — one observation per recompute, what
resets the history, what does not (fix 3).
- probe-readings.md: re-quotes condition-probe lines with level2= and mode= fields;
corrects leg-2 window geometry (0.2 s / 10 % pre, not 1.0 s / 50 %); moves the
wide-window reason to legs 1 and 3; fixes the search-probe scheduling note
(search_probe calls condition_probe directly; do not also set CONDITION_MS) (fix 6).
- verification.md: corrects shape-2 window geometry to match the code (fix 7).
- invariants-session-and-export.md, README.md: document the measure_stats session key —
position, type, default false (fix 8).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- The comment in `feed_window()` and three companion sites claimed a zero-width band
"never fires, because no sample can be on a different side of it than its neighbour".
That is false: `inside = value >= lo && value <= hi` with `lo == hi == 1.0` is true at
+1 and false at -1, so a ±1 square fires 19 times in 20 samples. The correct reason
the panel refuses one is that it reduces the condition to a float-equality test on a
real-valued signal, which in practice never fires (fix 4).
- `EventDetector::condition()`, `BufferSearch::channel()` and `BufferSearch::condition_`
were confirmed unused by grep and deleted (fix 10).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- A buffer rebuild mid-scan (CONFIG or window change) now calls `search.clear()` beside
`reset_measure_stats()` at both sites; without it the scan kept stale at_/end_ and could
report Done over fabricated hits (fix 1).
- `search_marks_label` now carries `PANGO_ELLIPSIZE_END` so the side pane's minimum width
is bounded; the longest text (~58 chars) was driving the pane width (fix 2).
- `update_measures()` gains a third `accumulate` parameter (no default) so every call
states its intent; the Statistics toggle now passes `false`, stopping the same window
being counted twice and deflating sd on repeated toggles (fix 5).
- `search_jump()` records whether `publish_pause_view()` returned an empty snapshot and
`update_search_panel()` appends "(no longer in the buffer)" to the position label when
it did (fix 9).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
place_search_marks() places the selected hit unconditionally to keep it marked when stepping
to it, but the label must distinguish this case from hits that are actually visible on screen,
so it does not lie about what is in view. add sel_off_view to track when the selected hit is
off screen, introduce a shown count that excludes it, and emit a third label case when only
the selected hit is off screen.
The plan's Task 8 named the search probe as its test but never extended it, and leaned on the
exit dump instead -- which stops the tick from Live, where no hit is in view, so it reads
placed=0 for a correct implementation and a broken one alike.
- The statistics label in measures_row() now carries PANGO_ELLIPSIZE_END,
matching the kind label's own treatment, so a long reading cannot widen
the pane beyond its measured minimum.
- The toggled handler on measures_stats_check now returns early when
updating_panels is set; apply_layout() sets updating_panels around its
gtk_check_button_set_active() call and then sets the Reset button's
sensitivity directly, so the loaded state is fully applied without a
spurious schedule_layout_save().
- The measures_row() header comment now names both labels that
update_measures() rewrites and explains the two-line layout.
- gtk_widget_set_sensitive on measures_stats_reset is null-checked in the
toggled handler, matching every neighbouring handler in the file.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Leg 1: widen to 1.0 s / 50% pre so the square's opening edge is in the snapshot
- Leg 2: use sine (not square) for the window channel, edge = falling so the band exit
is deterministic, and read earlier from vals[end-2] rather than half the snapshot
- Leg 3: raise timeout to 0.3 s (> half-period), accept either crossing direction,
widen to 1.0 s / 50% pre to reach the crossing that sits 0.3 s before x=0
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The old `phase=` field compared `trigger_time()` against the mock's absolute
`glitch_at + glitch_width` target. The app's stream epoch is the first packet
it received, not the mock's t = 0; the two clocks differ by however many
packets the CONNECT handshake cost, making the comparison meaningless.
Replace it with a snapshot-relative measurement: scan backwards from x = 0 in
the held event snapshot for the most recent upward crossing of `condition.level`
(the same `prev < level && val >= level` test the detector uses) and negate the
crossing's stamp to get the measured pulse width. A `measured` near 0.002 s is
the glitch caught exactly; near 0.200 s means an ordinary edge was latched.
Remove the now-dead `glitch_at`, `glitch_width`, `square_period` fields from
`ConditionProbe`. No other files changed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Nine tasks with literal code: the mock's injected faults, the shared
EventDetector the trigger and the search both run, the four condition
shapes and their panel rows, the Welford statistics and what resets
them, the chunked buffer search, its marks, and the documentation.
Every probe's expected reading is written as a shape with the measured
figure left to be recorded from a run, never assumed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
format_measure_value() does not exist; the shared SI-prefix formatter is
format_measure(double, const std::string&) at src/ui.cpp:1004.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The spec described the new condition fields as appended session keys. LayoutState
carries no trigger field at all -- the trigger is built from --holdoff and the
panel's defaults on every launch -- so there was nothing to append to. Persisting it
is a separate feature nobody asked for, and a session that silently restored an armed
pulse-width condition would surprise rather than help. The statistics toggle stays as
the one session key this design adds.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The spec claimed a window condition with equal levels degrades to an edge trigger on
that level. It does not: the band test is a change of side, so equal levels make the
signal "inside" only on exact float equality and the condition silently never fires.
Special-casing the detector into an edge trigger would conflate two state machines
that answer different questions -- the divergence the shared evaluator exists to
prevent. The panel refuses the degenerate band instead.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Qualified trigger conditions (pulse width with glitch as a comparison, window,
dropout), per-measurement statistics with a reset, and a search over the retained
ring. The first and third are the same question asked of live and of recorded data,
so they share one allocation-free EventDetector extracted out of TriggerEngine --
and the probe that proves the two paths agree on t0 is part of the scope.
Reference waveforms, PNG export, FFT and a byte-bounded event history are out, each
for a stated reason.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two defects behind one report ("a trigger during a capture overrides the first,
and the hold-off does not apply"); the second is also what made a spike in one
channel look like the acquisition resetting.
The trigger engine keyed its invalidation on the trigger channel's *index*, which
every rebuild of the channel list renumbers -- enabling a signal that sorts before
it is enough. Any rebuild therefore abandoned an open capture and set no hold-off,
so the next edge inside the post window became t0. It keys on the channel's name
now, which changes only when the source genuinely does.
A packet whose stamps jump more than 256 packet spans ahead is now rejected as a
fault and *bridged* at the declared cadence, instead of poisoning the monotonic
mark (which pinned every later packet flat at the glitched value) and condemning
the next good packet. Only four consecutive failures abandon the declared time
signal, and that fallback no longer wipes the rings: the arrival clock is anchored
to continue from the monotonic mark, so the history is still on one timeline --
the wipe is what made one glitching packet look like the stream restarting from 0.
The declared unit is resolved before those checks, since they weigh the stamps
against a cadence in seconds.
Guards: UDPSCOPE_PROBE_TRIGRENUM_MS prints t0_held=1 through the real widgets,
and the mock's --time-glitch-at/--time-glitch-by injects the clock fault.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Clicking the coloured chip in the signal list opens a GtkColorDialog and the
colour it returns overrides that signal's palette entry. Held by name, like
every other identity here, and persisted as a session `[[color]]` table.
RowSet::signal_color() becomes the one resolver: install() builds the trace
through it and all three chip sites read it, because a chip exists to say which
trace a row is about and an override is exactly the state in which two routes to
a colour would disagree. A grey chip names no trace and so is not clickable, and
the way back is `Default colour` in the signal's menu -- a setting with no undo
is a trap.
The gesture sits on the existing label rather than in a GtkButton: padding on
this row is what made the signal list twice as tall once. It claims its sequence
so a click on the chip cannot become the start of a row drag.
UDPSCOPE_PROBE_COLOR_MS reads the result out of plot::Plot::trace() and out of
the chip's own markup, with an override in no palette slot, so neither leg can
pass by asking the resolver what it already said.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
`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.
A single forward pass -- the list is already in evaluation order, so every
input's depth is in hand by the time the reader is reached -- then a fit. The
column pitch is `kNodeW + 48` rather than the add-node grid's 160, which left
28 units for a wire's 40-unit tread and made the bezier double back.
\`add node\` made a gain and nothing else, so every other op was two steps. It
is a menu of the twelve now, each carrying its index through
\`math_op_index()\`'s space rather than a cast of the enum -- which is also
what \`math_dialog_sync()\` now uses to drive the op dropdown, since that
order is the dialog's and not the declaration's.
One `math_dialog_problem()` drives both the line and Apply's sensitivity, so
they cannot disagree, and the two modal alerts are gone -- there was nothing
in either message a user could not have been told while still building the
graph. Written above `math_dialog_sync()`'s early return, since an
out-of-range selection is exactly when a stale line misleads most.
The source dropdown now lists the acquired channels, which is what Apply
validates against: listing every signal let a user pick a source that could
never be made valid, and put the stream's timestamp signal first.
Deleted `const int output_built = ed->output` from `math_probe()` -- it was
dead code, redundant with the `output_after_add` already printed on the same
line, and its removal is what drops the warning baseline from five to four.
The probe's `bad_text` pointer is copied to `std::string` before the second
`math_dialog_sync()` overwrites the label's internal storage.
The evaluator walks the node list in order, which used to make the order the
user's problem -- an input of an earlier node was not even drawn as
connectable. `math_connect()` runs Kahn's algorithm over the trial edge set,
with ties broken by the current index so a graph already in order is not
permuted, and remaps every index the dialog holds in one block. A self wire
and a cycle are refused with a message and change nothing.
\`math_wire_at()\` samples the same cubic the draw lays down -- 24 segments,
well under a pixel of chord error at the scales this canvas allows -- and
\`Delete\` removes the picked one. It deliberately does not fall through to
deleting the selected node: a key whose target depends on an invisible
selection is how a graph gets destroyed by accident.
Record the press in canvas coordinates at drag-begin; use press+total-offset
in both the Wire update and drag-end, replacing the output-port round-trip
that placed the release a column away on a detach. Run the gesture probes at a
non-identity view (scale=2, x=50, y=20) so a missing math_to_graph call causes
failures rather than passing silently. Add a fifth probe: detach and release in
place must reconnect. Update the stale comment that described the deleted
two-click protocol.
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>
One state machine decided at `drag-begin` -- a port starts a wire, a box
moves it, the background pans -- so the two-click protocol and its armed-port
highlight are gone. A drag from an input empties it immediately, which makes
a release over nothing the detach gesture; `Escape` puts it back.
The old assertion called math_canvas_scrolled(nullptr, ...) which resolves the
anchor itself from the event controller — with no controller it fell back to the
canvas centre (or a hardcoded 420x300 size when the widget was not yet allocated).
The probe then measured whether the graph point under (180, 140) was kept. Those
are two different points, so the difference printed was noise.
Extract math_zoom_at(d, px, py, dy) from the handler body: it takes an explicit
canvas anchor and guarantees the graph point under it is still under it after the
zoom. math_canvas_scrolled() resolves the anchor (from the event, or the widget's
own centre with no magic fallback dimensions) and delegates to it. The probe calls
math_zoom_at() with its chosen anchor directly, so dx and dy are an assertion on
the code rather than on the difference between two unrelated points.
Broken-anchor reading: dx=-180.0000 dy=0.0000. Fixed reading: dx=0.0000 dy=0.0000.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
One view transform applied to the cairo context, with `math_to_graph()` as
the inverse that every hit test runs -- a tolerance is a screen distance,
so it is divided by the scale rather than left in graph units. A wheel
zooms about the pointer, a background drag pans (next commit), `f` fits and
Ctrl+0 resets. The drawing area is made focusable, without which a key
controller on it never fires.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
\`math_port_at()\` returned a sentinel for an output hit and the caller
recovered the node with a box test -- but the output square straddles the
box's right edge, so every pixel past it disarmed the gesture silently.
That is the whole of "edges cannot be added".
Cancel also closed with \`gtk_window_destroy()\`, which emits no
\`close-request\`, so the dialog pointer was left dangling and no channel
could be edited twice.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Nine tasks, each ending in a commit: the silent disarm in the port hit test,
the view transform, the drag-based wiring, wire deletion, the topological
re-sort that makes a backwards wire legal, the live validation line, the op
picker, tidy, and the docs with the full verification run.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The reported "cannot add edges" has a specific cause: the output port's hit
radius reaches past the box it belongs to, and the press handler re-derives the
owning node with a box test, so clicking the visible dot's outer half sets
pending_out to -1 and silently disarms. Edge deletion has no code path at all.
The design replaces the two-click protocol with a drag gesture, makes the list
order the editor's own invariant via a topological re-sort rather than a rule the
user must obey, adds a view transform with every hit test routed through its
inverse, and trades the modal alert on Apply for a live validation line. It also
records that the Cancel button still calls gtk_window_destroy -- the dangling
dialog pointer CLAUDE.md documents as fixed was only fixed on the Apply path.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
An A/B of the same fast-mock run with and without a pause reads gaps=0 both
ways, the paused one having published in 86 ms -- so the gaps=73 recorded
yesterday as "a pause costs packet gaps, unexplained" was incidental load on
one run. It never scaled with the publish length and has not reproduced.
Record the publish figures as the range they actually span across runs
(event 69-115 ms, pause 22-86 ms) rather than as single readings, and name
`frame path:` across a pause as the check that the paused trace is drawn like
a live one -- raw=1 lod=555 against raw=0 lod=373, where the defect read
raw=21 lod=118.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
publish_pause_view() filled stamps, values and y and left pyramids empty, so
level_for() descended to raw samples and a frozen pause was re-gathered from
every sample it held on every frame. It cost draw time (12.34 -> 18.18 ms mean)
and, because stage_trace_density() normalises each column against its own peak,
it also changed how the trace looked -- which is how it was noticed.
Both publishes are now timed and reported as `publish cost:`, counted apart
because they stall opposite threads: the event publish runs on the producer
under mutex() and blocks the UI, the pause publish runs on the UI thread and
blocks the producer. That is the number a second live ring has to be judged
against.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
period, frequency and pulse duration counted every sign change about a single
level, which is unusable on any real signal: the mid level is exactly where a
waveform moves fastest, so an interferer faster than the fundamental recrosses
it several times per true crossing. A 5 Hz sine with a 500 Hz tone at a tenth of
the amplitude read 79 crossings for 9, period=0.025622 against 0.2.
MeasureLevel gains a hysteresis band sized once in measure_set_level(): a tenth
of the window's peak-to-peak, clamped to half the room to the nearer extreme so
a threshold set near one stays usable. The band moves the edge decision only --
the sum/count split stays at the bare level, since that split is top/bottom and
has a documented exact reading. The bucket fast path tests the band and the
level separately rather than folding them, or it would refuse exactly the
buckets a crossing passes through: clean cost is unchanged at raw=0.1% and
1.41 ms a recompute.
tools/mock_udpstreamer.py gains --noise, defaulting to 0 so the exact probe
assertions still hold.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Stage 1 cured the moire by collapsing a dense gathered window onto one vertical
segment per pixel column; this is the information that collapse throws away. A
column's min and max are the same whether its samples are spread evenly or piled
at its two ends, so the flat parts of a square wave and the edges between them
drew as one ink. Style::intensity_grading, which the library leaves off and this
app turns on at the first tick (the renderer does not exist until the area is
realized), grades each column by the time the signal spent at each row -- the
digital-phosphor look.
frame path: gains a graded= field, which must equal columns=: grading is a way of
drawing the reduction, not a third path. UDPSCOPE_NO_PHOSPHOR=1 puts the style
back to the library's default and is the A/B, since a frame mean cannot attribute
its own cost -- measured on the fast mock, draw 3.24 ms graded against 2.30 ms,
with segments 1340/4157 against 131/4306 over the same lod=453 frames. That A/B
is also what found the level-floor regression now fixed upstream.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
At a high rate, a high frequency or on noise the plots banded visibly and the bands
crawled under a pan. Three things compound to produce it, none a bug alone: a
pyramid bucket spans one to two pixel columns, Envelope emits two points per bucket
so each contributes a near-vertical hairline and a near-vertical connector, and
every one lands at a fractional x because a trace is deliberately not snapped --
so a column's brightness beats against the pixel grid, and a pan sweeps the beat.
The cure is upstream (deps/plotter, pin moved): draw_traces reduces such a window
onto the pixel grid, one snapped vertical segment per column from that column's
lowest to its highest value. One sub-pixel phase for the whole trace, so there is
nothing left to beat, and the geometry is bounded by axes.w rather than by the
buckets.
frame path: gains a columns= field, which is what separates "the segment count
collapsed because the reduction ran" from "the window got narrower". On the fast
mock at 3 kHz over 20 s: columns=1200, segments 2041/2237 over 1201 painted frames,
against columns=0, segments 10960/216093 over 263 before -- the same run at 4.6x
the frame rate, with trace ink: clean either way.
Also corrects the documented mock invocation: --batch 20000 exceeds the datagram
limit and the mock dies with Errno 90.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
With the axis titles gone the left gutter was 200 px of an 870 px viewport and the right
side had no margin at all, so the newest tick's label was drawn half outside the plot and
the newest sample's ink landed on the frame.
Two causes on the left. The reservation covered `format_si()`'s `%.3g` fallback
(`-1.23e-45`), which no signal on an `Si` axis produces -- the widest real label for a
unitless axis is `-1.23m` -- and `note_y_gutter()` added 6 px of slack to a figure that is
already the maximum of the labels and last frame's floor, making the gutter a ratchet that
climbed to its 200 px cap in ~19 frames. It settles at 104 px now, with `axes.w changes: 1`
where it read 17.
The right side did not exist: a plot's `axes` rect reached the viewport's right edge by
construction. `deps/plotter` gains `Plot::set_right_gutter()` and `Figure::set_right_gutter()`
-- a want rather than a measurement, since nothing is drawn there, imposed figure-wide so
the rows end on one column as they already start on one. udpscope asks for half the widest
time label plus the label padding (59 px), x labels being centred on their tick.
Separately: the exit dump asked two destroyed widgets for their size, which is where the
"exactly two Gtk-CRITICALs" baseline came from. The pointer is freed memory by then, so a
run that had destroyed other windows first (UDPSCOPE_PROBE_HELP_MS opens three) segfaulted
in the middle of the dump instead. Both sizes are captured on the tick; a clean run now
carries no critical.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
A stream that answered CONNECT but has sent no CONFIG has no signal list, so
nothing can be enabled and no DATA frame is owed. Calling that "connected, but
no data is arriving: another reader may hold the source" blamed a second reader
that does not exist, and the badge blinked green over it -- a working pipeline
is exactly what it did not have.
The glance word is `connecting` (ten characters, no wider than `event held`,
so the fixed-set layout rule still holds) and the badge takes a fourth, steady
yellow phase, both from the first frame rather than after the three-second
stall test. The stall notice itself now splits on `have_config`: the unicast
contention wording where a signal list exists, "trying to connect" where it
does not, with `[udpscope] connecting:` on stderr beside the old `starved:`.
Three asked-for changes, one commit because they are one coordinate system.
The live axis is a fixed [-window, 0] with the newest sample at 0, so x reads
as an age and the labels stand still while the trace rolls under them. That
makes the display clock distinct from the stream clock for the first time, so
`ScopeModel::stream_at_x0()` becomes the single conversion between them
(`display_x = stream_t - stream_at_x0()` in all four modes) and a snapshot
carries its own origin in `EventSnapshot::x_offset`. The pause view is now
rebased like an event's, reversing an earlier deliberate choice: a frozen
absolute-stamped picture would sit a whole uptime right of a newest-relative
axis. That also fixed a latent bug, the export adding `trigger_time()` to a
*paused* view's stamps.
The status bar says how long you have been paused -- wall time, since the
producer never stops, through `format_elapsed()` so the field cannot widen.
The export writes two time columns: `time`, the axis as displayed, and
`abs_seconds`, the stream's own clock for the same sample. f64 seconds and
not the wire's integer, because `udps_frame_t::values` is a `const double*`
and the integer never exists in this process.
Cursors and measurement marks move in opposite directions under the rolling
axis, deliberately: a cursor is a position on screen and stays put, a mark
names a sample and travels with it. `App::measure_window` is stream-clocked
for a sharper reason -- a display-coordinate live window never moves, so the
"the window moved" recompute would never fire again.
Moves the deps/plotter pin to 58473db, which stops a time label's mantissa
reaching its own radix: a one-second tick differenced off a live axis read
`1e+3ms`.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
gather() staged x as f32 relative to the series' own x0() -- the start of
the whole retained window -- so the quantisation was a fraction of the
ring rather than of the view: 1.9 us at 20 s of retained data, which at a
10 us view across a 716 px axis is over 100 px between where a sample is
drawn and the x a readout quotes for it (crosshair_point() reads x_at()
in exact f64). It halved on every zoom out, which is what made it read as
"zooming in makes the crosshair select a point beside the data".
The origin is now the caller's (plotter c2c8c59): the renderer passes the
visible range's minimum, so the error is a fixed 2^-24 of what is on
screen -- about 4e-5 px over a 700 px axis -- at any zoom. The parameter
sits before out_xy so every call site is a compile error rather than a
silent change of behaviour.
The renderer no longer reads series.x0() at paint time at all, which
removes two of the three symptoms of the torn window_x0_ read by
construction; the ink probe's now-vacuous kept_fixed_x0 walk goes with
them, replaced by a drawn-vs-quoted pixel figure. neighbours() still
tests the visible range against x_at(0), so the cache stays load-bearing.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The caption that names the relative x labels' anchor was an `Annotation`, whose anchor
is a data coordinate -- so it crawled across the plot under a pan and drifted under a
zoom, which is how it was reported. It is now `Plot::set_x_note()`: the bottom row's
caption, drawn by the renderer in the axes' bottom-left corner in pixels, on the
legend's own backplate.
The text stays the host's, because only the host knows that label + caption has to be
the exact absolute time (`%.6g`, where `format_time()` would round to three figures).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
A `t0 = ...` caption appeared on every sub-second window, including a
triggered one whose axis is already relative to the edge -- so the ticks
were re-anchored away from the one position a triggered axis exists to
show, and the caption had nothing to explain.
The form is now decided by the tick step against `TickFormat::Time`'s own
resolution at the axis' magnitude (three significant figures below a
minute, whole seconds through `m:ss` to a day, three figures of days
above). The span said nothing about that and was wrong both ways: a 2 s
window at twenty minutes of uptime reads `20:00` on every tick and stayed
absolute.
The step is read back from the last layout's majors (a tick value depends
on the range and the axis width, not on the label form); with no ticks yet
the form is left alone rather than guessed, since the hysteresis would
latch a guess. Entering the relative form carries a ULP of slack, because
`step == res` is readable and a bare comparison of two differently
computed doubles flipped the form between frames.
`[timelabels-probe]` and the dump gain `distinct=`, which asserts that no
two major labels read the same -- the thing the relative form is for, and
the only reading that holds whatever the stream's uptime is.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The SI prefix ladder in format_measure() was gated on a non-empty unit, so two
decimals of a small unitless value printed `0.00` -- the reading gone, not merely
coarse. That is a whole class of channels rather than a corner: a CONFIG need not
declare a unit, and a math channel never has one. Below 1e-12 the same loss
happened even with a unit, so that range falls back to three significant digits.
With no unit the prefix goes tight against the number (`585.24m`), since the space
exists to separate the number from its unit and a spaced `m` alone reads as metres.
The mock declared `unit="V"` for every data signal, which is why no probe could see
this; it now takes `--unit ''`.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
