QtProbe tests

Run make test at the repository root, or run make -C qtprobe test-core, test-ui, or test-ioc separately. Tests build in the selected Qt object tree. On Windows, build-windows.bat test 4 initializes the compiler and runtime environment and runs all three suites. Text logs are retained as test-core.txt, test-ui.txt and test-ioc.txt in the object directory. The QtStripTool launch test builds a native recorder helper, so it needs neither a Unix script nor an installed QtStripTool. The two X11 drag interoperability tests are skipped by the offscreen suite on Windows, as on other offscreen platforms. Qt Test is required. The IOC test also requires EPICS Base's softIocPVA (or softIoc for CA-only builds); it creates its own records and sets explicit loopback-only CA/PVA address lists and isolated server ports. It never uses site PV names. It verifies CA/PVA metadata, values, writes, errors, arrays, read-only behavior, reconnect and the headless JSON command, the record explorer, plus precision display, alarm limits, CA/PVA long strings (DESC$ and a CHAR waveform), record processing, and an inspector's write → Undo → Process round trip with its log. UI tests use Qt's offscreen platform, with temporary settings. Missing IOC prerequisites fail.

Artifacts, including IOC output and a UI screenshot, are stored under the Qt object directory's test-artifacts/. IOC processes are terminated by test cleanup. Tests do not install packages, change EPICS configuration or alter production channels.

The offscreen UI suite also checks Help menu/F1 integration, full-guide search and wrapping, topic navigation and window reuse, the offline manual and its site override, and About credits, animation lifetime and Open help behavior. Manual links use a Qt URL handler in tests, so no external browser is launched. Help and About screenshots are saved beside the inspector screenshot.

Workspace coverage includes the welcome view, tab and Next/Previous PV navigation, table filtering and closing, exact Readback staging, write availability while paused, section toggles shared across PVs, window fitting (growing, shrinking, the user-height floor and the empty reset) appearance persistence, and dark-theme popups and disabled menu text (menu-dark.png). workspace-empty.png, workspace-one.png, workspace-sections.png, workspace-light.png, workspace-dark.png, workspace-write.png and workspace-compact.png use synthetic UI fixtures with acquisition disabled. They never contact an IOC.

Table tests cover natural, value and state sorting, quick filters and their counts, click/Ctrl/Shift selection, text/CSV/name copies, keyboard navigation and the age timer stopping when hidden. Other tests cover the range gauge and Trend limit lines, write logging and Clear, confirmation (answered by the test), pasted name lists, zoom of fonts and rows, and Open in QtStripTool through a native recorder helper selected by QTPROBE_STRIPTOOL. Core tests check that history stays within its count and the shared memory limit (including the split between PVs and long values shortened to 1000 characters), and that write-log values are limited. UI tests check that hidden inspectors and closed sections release plot and table copies, that a paused inspector keeps them, and that old QtStripTool configurations are pruned. Text sizing is checked with hinted and unhinted fonts from 6 to 16 pt (remote X displays often render unhinted, with fractional widths): plot axis labels, elided labels and the History timestamp column must show their text in full. Enum trends are checked for state names, step drawing in the inspector and the table, no limit lines or gauge, and time-in-state statistics (core test with fixed receipt times, including a disconnect). Numeric arrays and text waveforms hide Trend, add no numeric history samples and leave the table sparkline empty, while scalars keep Trend. Structure browser tests cover in-place updates (rows, expansion and selection kept), default expansion of NT metadata, change marks on values and on collapsed parents, the search, row and whole-structure copies, the "more elements" row, CA PVs without Structure and release of hidden rows (UI, synthetic trees). Core tests check its text and JSON forms and the dotted Info paths derived from it. The IOC test reads an NTScalar and a QSRV group (an NTEnum member and a plain array) over PVA. Core tests cover record link parsing (PV links, constants, hardware and JSON links), the record-type field lists and every diagnosis rule. The IOC test reads a periodic calc with CP and forward links and a disabled record over CA and PVA, and follows a link by clicking it; a UI test checks that following a link opens or selects the target inspector. Core tests also cover precision display, long-string decoding and text writes, alarm-limit detection, the bounded write log and its CSV, and the QtStripTool configuration.

The multi-PV tests verify simultaneous table rows, exact 64-bit and array readbacks, alarms/disconnects, shared history, independent inspector pause, column alignment and narrow-window columns, filtering, closing, batch preflight and read-only behavior. Text-drop tests cover whitespace-separated QtEDM payloads, provider selection, duplicates, editors/tabs, copy actions, source selection release, cancelled gestures, and rejection of malformed/file/move-only drops. Table previews are saved as workspace-all-pvs-{light,dark,compact}.png.

For a real QtEDM middle-button gesture, with QtEDM already built, run:

make -C qtprobe test-qtedm-drop

This optional Linux test builds a small X11/XTest mouse driver and launches the unmodified sibling medm/bin/Linux-x86_64/qtedm. The target starts its own Xvfb display with xvfb-run (override with XVFB_RUN), so the synthetic pointer input never reaches your desktop. Override QTEDM_TEST_PROGRAM=/path/to/qtedm when needed. It checks drops on the welcome page, name editor, inspector value, staged value editor, a table row and the tab strip, including duplicates and preservation of the staged value without writes. The source inherits loopback-only test discovery settings; no production channels are used. This case is skipped by the normal offscreen suite.

Legacy MEDM drops are checked with a small Motif program that starts a drag the way MEDM does (XmDragStart, COMPOUND_TEXT, copy) and drops on the window. It needs Motif, Xt and XTest development files:

make -C qtprobe test-motif

It covers a new PV, a duplicate, the provider selector and a refused name. Driving the real MEDM binary with synthetic input does not work under Xvfb: MEDM waits in its Motif clipboard copy before the drag starts.

CPU checks are opt-in because timings depend on the host and Qt platform:

make -C qtprobe benchmark-cpu
cd qtprobe
python3 tests/benchmark_ioc_cpu.py O.Linux-x86_64-qt5/benchmark_cpu

The first command measures an empty window, 100 static synthetic PVs, 100 PVs updating at 10 Hz with 10,000 retained events each, and one inspector with a full history, with no section and with Trend open. The helper reports JSON containing process CPU time as a percentage of one core, elapsed time, and delivered update counts. Its options also select History or a hidden window. The Python runner needs Python 3 and softIocPVA; it owns a loopback-only IOC and foreground CA repeater and checks 1/100 static CA PVs, 100 active CA PVs, and 100 static/active PVA PVs. Set EPICS_TEST_BIN for another EPICS installation. No PV writes are issued by the benchmark client. Correctness tests additionally cover deferred rendering, idle timers, scrolling into stale table rows, incremental history rows, peak/gap preservation in dense plots, and shared CA/PVA delivery after peer removal.

The benchmarks also support native Windows builds from an initialized MSVC shell (use the same Make variables and runtime paths as the normal build). Select O.Windows-x86_64-qt6/benchmark_cpu.exe for the Python runner and set EPICS_TEST_BIN to the EPICS bin/windows-x64 directory. On Windows the helper uses process kernel/user CPU time and peak working set; on Unix it uses process CPU time and peak RSS. Both report CPU as a percentage of one core.