Changelog

All notable changes to the plugins in this repository will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Entries up to and including 0.2.1 describe TriggeredAvg, which was developed in its own repository and merged into this one at 0.3.0. Its history and this changelog came with it. Entries from 0.3.0 onwards cover all three plugins.

[0.4.1] - 2026-08-25

Added

• Switchable display units on the Bar Mapper — degrees, millimetres or screen pixels. A stimulus is measured on a screen, and until now everything the mapper showed or took was degrees of visual angle: speed, sweep start, resolution, map centre, the map axes and the RF readout. Show units in at the top of ANALYSIS now switches all of them at once, given a Viewing distance and, for screen pixels, a Screen resolution (px/mm); both are saved with the signal chain, so the rig is typed in once rather than converted by hand at the bench. Degrees remain the unit everything is computed, stored and exported in — the conversion happens in the parameter fields and the panel readouts, and the three display parameters are the only ones in this plugin that do not ask for a recompute, so switching unit is incapable of moving a receptive field or changing a map pixel. What is typed is converted back to degrees and clamped against the parameter's own range, so the range does not change with the unit either: 2000 mm/s comes back as 1990 mm/s, which is the 200 deg/s the parameter has always allowed. The factor is the small-angle one, mm/deg = distance × π/180 (9.95 mm/deg at 570 mm), which under-reports position by about 4% at 20° eccentricity — one factor for positions and extents alike keeps a map pixel the same size in millimetres wherever it sits. Saved sessions keep every *_deg array and attribute unchanged and now also carry viewing_distance_mm, screen_px_per_mm and screen_mm_per_deg, so an offline analysis can convert without being told the rig separately

• Degree axes on the Bar Mapper's maps — ticks and numbers along the bottom and down the left of every panel, in visual-field degrees, so the centre of a mapped receptive field can be read off the picture. Until now a map showed position only relative to its own centre pixel, and the actual coordinates of a peak had to be counted out in pixels from the middle. Ticks land on round multiples in visual-field coordinates rather than on fractions of the map, so a map covering the origin always has a tick there; the numbers thin out on a small panel while the ticks stay, and the axes step aside entirely when the panel is too small for them to be worth the map pixels they cost

• Triggered Average's axis limits moved into an AXES popout. Two labels, two AUTO/MANUAL toggles and four editors — some 570 pixels of options bar — for settings that are adjusted once and then left alone, crowding out the controls that are used constantly. The popout shows the captured window the X range is clamped to, so a value that comes back changed is explained; a range typed with its ends the wrong way round is now rejected instead of silently reset, and both ranges are remembered while an axis is on AUTO. Saved layouts keep the same attribute names, so an existing signal chain restores unchanged

• SAVE and LOAD buttons on the Triggered Average canvas, alongside CLEAR, so a run can be resumed after a break and the data read outside the GUI. They are the same SessionControls component and the same session bundle the Bar Mapper already used — one XML file plus one .npy per array — so the two plugins cannot drift into two formats that look alike and are not interchangeable. SAVE works during acquisition (the accumulators are copied under their own lock and written on a background thread); LOAD is refused while acquiring, and a session that does not match the current sample rate, trial window, channel count or trigger conditions is refused with the reasons rather than half-applied

• Triggered Average's session also carries what the canvas draws: averages, standard_deviations and a time_ms axis with the trigger at sample pre_samples. Derived from the accumulators and ignored on load — they exist so that reading a session in Python or MATLAB hands over the traces rather than the accumulator state behind them. Conditions with no trials are written as zeros, never NaN. The format is documented in the README

• SAVE and LOAD buttons in the trigger table, in all four plugins. A rig usually runs several of these plugins off the same conditions, and the trigger table — names, TTL lines, colours and the three message patterns per condition — had to be retyped identically into each of them. SAVE writes the table to a small XML file; LOAD replaces the table with one from such a file, or from a saved signal chain, whose CUSTOM_PARAMETERS block carries exactly the same TRIGGERSOURCE elements. In the Bar Mapper the sweep angles and the direction generator's settings travel with the table, so a direction set arrives meaning what it meant where it was saved. LOAD is disabled during acquisition, and refuses a file with no trigger sources in it rather than emptying the table

• The Receptive Field Bar Mapper's direction generator is configurable. Besides the number of directions it now takes the TTL line (fixed, or one line per direction counting up from a base) and the arm message it writes — text before the number, the number for the first direction, and text after it — so VSTIM: TRIALTYPE, 200, TIMESEQUENCE generates VSTIM: TRIALTYPE 200 TIMESEQUENCE, … 201 …, and so on. A preview line shows the first and last pattern before REPLACE writes them, and the settings are saved with the signal chain

Changed

• The trigger source table now has one serialiser rather than one per place that stores it. The signal chain, a saved session and the new trigger-settings file all go through writeTriggerSourcesToXml / readTriggerSourcesFromXml in TriggerCore, so a field added to a trigger source cannot reach one of them and not the others

• pendingTimeoutMs now defaults to 5000 ms, up from 2000. The old default was shorter than the gap between a sweep's TTL edge and the trial-end message that commits it in a typical mapping run, so the last trials of a block were dropped. Sources in an existing signal chain keep whatever they were saved with

Fixed

• The Receptive Field Bar Mapper was never released. Every artifact list in both workflows named the three plugins that existed when they were written, so the mapper was built on every CI run and then dropped on the floor: it appeared in no release archive and in no CI artifact, and the only way to obtain it was to build it. if-no-files-found: error was already set on all four lists, which is what will catch the next one -- it fails the job rather than shipping an archive that is quietly short a plugin

• Triggered Average's options bar could place its right-hand buttons where the scrollbar could not reach them. The bar was laid out at least 775 px wide while its controls needed nearly twice that, and the holding viewport scrolls only as far as the bar's own bounds, so on a narrow window CLEAR was drawn past a boundary nothing could scroll to. The width is now computed from the same layout that is performed

• The Receptive Field Bar Mapper recomputed the whole map every time a trial was parked awaiting its commit message, producing exactly the map already on screen. It was the only one of the four plugins to report a parked capture as a display change; the other three already returned "nothing to show", which is what that return value means

[0.3.0]

TriggeredAvg merged into the Triggered Spectra repository, which becomes Event-Triggered Analysis and now builds three plugins over two shared cores. The averaging plugin was the structural template for the spectral core and had since been overtaken by it; rather than copy the improvements back, the shared layer was split out and all three plugins now sit on it.

This release does not preserve compatibility with saved TriggeredAvg signal chains. Parameter defaults, the trigger-type model and the channel selection all changed; see Removed and Changed.

Added

• trigger_core: the ring buffer, work queue, capture worker, trigger sources, broadcast-message matching, the trigger configuration window and the trigger monitor. Contains no FFTW and no DSP, so a plugin that only needs trial windows does not inherit a numerical dependency
• TriggeredCaptureNode: shared GenericProcessor base owning everything about getting a trial window. TriggeredPower, TriggeredCoherence and TriggeredAverage all derive from it
• Triggered Average gains a MONITOR window: live per-source counters through each stage (edges → queued → captured → committed), TTL and message totals, the text of the last broadcast message, a console-echo toggle and a one-line diagnosis of the commonest failures
• Triggered Average gains a channel selector. Everything downstream is linear in the number of selected channels, so this is the main performance lever
• Triggered Power and Triggered Coherence gain undo/redo for trigger-source edits, which they never had
• Pending captures are backed by a shared, separately tested PendingCaptureStore<Payload>
• average_tests binary. 272 tests now run across three binaries, up from 209

Changed

• Repository renamed to event-triggered-analysis; namespace is now EventTriggered throughout
• Arming is derived from the arm pattern rather than declared by a trigger type. Setting an arm pattern is what makes a source gated, mirroring the commit pattern, which already decided whether a capture was provisional
• Triggered Average's window defaults widen: pre_ms 250 → 500 ms, post_ms 750 → 1000 ms, both ranges 5 s → 10 s
• FFTW is discovered and installed by the spectral layer alone, so the repository configures without it for a trigger_core-only build
• SpectralWorker renamed to CaptureWorker: it never did any spectral work
• Every plugin binary now installs from Build/<config>/

Removed

• TriggerType::TTL_AND_MSG_TRIGGER. Whether a source was gated was stated twice — as a type and as whether it had an arm pattern — and the two could disagree; a TTL_AND_MSG source with an empty arm pattern could never fire at all. Saved sources of this type load as plain TTL, with their gating carried by the arm pattern they already had
• The trigger-type column and its dropdown, which now had nothing to choose
• Triggered Average's own ring buffer, trigger sources, capture queue, DataCollector thread and configuration window, all replaced by the shared ones

Fixed

• Broadcast messages no longer take a lock on the audio thread. handleBroadcastMessage is dispatched from checkForEvents() inside process(), and Triggered Average took its DataStore mutex there — the same one the message thread holds while repainting — and copied an AudioBuffer. Commit, discard and expiry now run on the capture worker
• Removing a trigger source no longer leaves its buffers behind. Nothing erased the DataStore entries keyed by the freed pointer, so a later source allocated at the same address inherited the dead one's average
• ctest no longer reports success without running anything. The test binary landed in a per-config directory its runtime libraries never reached, so test discovery found nothing and ctest exited 0 printing "No tests were found!!!"
• Triggered Average's editor: the channel selector had no control at all, the Max Trials row was drawn below the visible area, and every inline control stretched across the stream-selector drawer when it was opened. The last two also affected the spectral plugins
• With auto-scale on, the average and the individual trials were normalised to different ranges and drawn on the same axes, so the average appeared to swing wider than the traces it was computed from
• Ring-buffer reads now report Overrun when the writer laps the reader mid-copy, instead of silently returning a mixture of old and new samples
• TriggeredCoherence honours commit patterns (#8). It implements the same three pending-capture hooks the other two plugins do, so a trigger source with a commit pattern parks its trial until a commit message folds it in, a cancel discards it, or the timeout expires. Trigger sources are shared configuration, and until now the same source accumulated immediately in Coherence while Triggered Power waited — two answers from one configuration, with nothing in the UI to say so. A parked trial is also kept out of the shift predictor: its null pairs consecutive kept trials, and a trial that is later discarded must not become the partner of the one after it
• Triggered Average's monitor no longer double-counts (#21). CaptureWorker bumps trialsCaptured for every window it extracts and pendingCommitted for every commit that finds one parked, for all three plugins; TriggeredAvgNode counted both again on top, so CAPTURED and COMMITTED read twice the real numbers there while Triggered Power and Triggered Coherence read correctly. The accumulated average itself was never affected
• CI builds again after the repository rename. The cached Open Ephys GUI build tree carries the absolute path it was configured at, and the workspace path contains the repository name, so every job restored a tree configured under .../TriggeredSpectra/ and failed at Configure Open Ephys GUI. The cache key and its restore-key are now scoped to github.repository, and a guard step discards a restored tree whose CMakeCache.txt points somewhere else

Known gaps

Tracked as issues rather than listed exhaustively here:

• Message-only triggering is declared but never fires (#11)
• Spike-triggered averaging (#9), which would need per-stream analysis (#10)
• The display layer has no automated coverage; every UI fix above was found by manual testing (#12)

[0.2.1] - 2026-08-04

Changed

• CI: Windows builds now use the Visual Studio 2026 toolchain (Visual Studio 18 2026 generator); the retired Visual Studio 17 2022 generator is gone from the GitHub-hosted images
• CI: Windows jobs install a current CMake via lukka/get-cmake, since the runner image ships CMake 3.31.x which predates the VS 2026 generator
• CI: dropped the ilammy/msvc-dev-cmd step, which is unnecessary when configuring with a Visual Studio generator

[0.2.0] - 2026-08-04

Added

• Message pattern matching per trigger source: each condition now has independent armPattern, cancelPattern, and commitPattern fields (case-insensitive substring match; empty = disabled)
• Pending capture workflow: when a commitPattern is set, captured data is held in a per-source pending slot until a commit or cancel message arrives, rather than being immediately added to the average
• pendingTimeoutMs (default 2000 ms) auto-discards uncommitted pending captures when the commit message never arrives
• DataStore methods: storePendingCapture, commitPendingCapture, discardPendingCapture, discardExpiredPendingCaptures
• SetTriggerSourcePattern undoable action for persisting pattern edits through the GUI undo stack
• Three new editable columns (Arm, Cancel, Commit) in the trigger-source configuration popup table
• Tests for pending capture lifecycle, pattern matching, and XML round-trip

Changed

• Arm matching changed from exact equalsIgnoreCase(source->name) to containsIgnoreCase(source->armPattern), decoupling the condition display label from message matching
• Cancel is evaluated before commit when a message matches both patterns (always cancels rather than commits)
• TriggerSource is now a plain data struct — the TriggeredAvgNode* back-pointer has been removed; popup UI components receive the node pointer at construction time instead
• Configuration popup window widened to 840 px to accommodate the new pattern columns

[0.1.1] - 2026-06-18

Fixed

• Null dereference in handleAsyncUpdate when the canvas is not open
• Static local lastSampleNumber was shared across processor instances, causing incorrect trigger detection with multiple instances
• Data race in cached plot-path rebuild — DataStore lock is now held for the full rebuild
• Custom X-axis limits not clamped to the data collection window ([-pre_ms, post_ms]), silently producing a blank plot for out-of-range values (#8)
• x_min, x_max, and use_custom_x_limits parameters not handled in parameterValueChanged, so programmatic changes (XML load, config messages) had no effect on the display

Changed

• Replaced assert(false) in the MSG_TRIGGER broadcast path with a log warning so the plugin degrades gracefully on unexpected message types
• Removed the dead 60 Hz polling timer in TriggeredAvgCanvas (display updates are now purely event-driven via AsyncUpdater)
• Removed unused post_ms member from GridDisplay
• Moved SampleNumber type alias to a dedicated Types.h to remove header coupling

[0.1.0] - 2026-02-10

Added

• Initial release of TriggeredAvg plugin
• Event-triggered averaging functionality for continuous data
• Real-time visualization of averaged waveforms
• Configurable pre-trigger and post-trigger windows
• Compatible with Open Ephys GUI API v10