WIP: UI smoke tests for axis, touchy, gmoccapy, qtdragon

debian/control.top.in

@@ -48,6 +48,23 @@ Build-Depends:
     tk@TCLTK_VERSION@-dev,
     xvfb <!nocheck>,
     x11-xserver-utils <!nocheck>,
+    python3-opengl <!nocheck>,
+    python3-pyqt5 <!nocheck>,
+    python3-pyqt5.qsci <!nocheck>,
+    python3-pyqt5.qtsvg <!nocheck>,
+    python3-pyqt5.qtopengl <!nocheck>,
+    python3-pyqt5.qtwebengine <!nocheck>,
+    pyqt5-dev-tools <!nocheck>,
+    python3-dbus <!nocheck>,
+    python3-dbus.mainloop.pyqt5 <!nocheck>,
+    python3-qtpy <!nocheck>,
+    python3-cairo <!nocheck>,
+    python3-gi <!nocheck>,
+    python3-gi-cairo <!nocheck>,
+    gir1.2-gtk-3.0 <!nocheck>,
+    gir1.2-gtksource-4 <!nocheck>,
+    python3-numpy <!nocheck>,
+    python3-configobj <!nocheck>,
     libfmt-dev,
     yapps2
 Build-Depends-Indep:

tests/ui-smoke/.gitignore

@@ -0,0 +1,9 @@
+# Runtime artifacts left by launch.sh; per Bertho's clean-tree rule,
+# generated build/test outputs must be gitignored or committed.
+linuxcnc.out
+linuxcnc.err
+linuxcnc.pid
+ui-smoke.out
+ui-smoke.err
+result
+stderr

tests/ui-smoke/README

@@ -0,0 +1,26 @@
+UI smoke tests
+~~~~~~~~~~~~~~
+
+These tests launch each GUI (axis, touchy, gmoccapy, qtdragon) under
+xvfb-run against an existing sim config and verify the GUI starts and
+the NML task is reachable. Phase 1 ("does it start"); functional
+checks (load G-code, jog, MDI) belong in tests/ui-functional/.
+
+Each test directory contains:
+  test.sh       launches the GUI under xvfb-run, runs drive.py
+  checkresult   examines the captured output for crash markers
+  skip          skips this test only when xvfb-run is not on the host
+
+Shared helpers live in _lib/:
+  drive.py            common NML driver, prints UI_SMOKE_OK on success
+  launch.sh           xvfb-run wrapper, signal escalation for shutdown
+  cleanup-runtime.sh  belt-and-suspenders: kill stray daemons, ipcrm
+                      shared memory, drop /tmp/linuxcnc.lock
+  checkresult.sh      shared pass/fail predicate
+  skip-if-missing.sh  shared skip predicate
+
+Skip vs fail policy: the only condition we skip on is xvfb-run absence
+(rare local dev env). Python and gi typelib deps the GUIs need are
+declared in debian/control under !nocheck so apt-get build-dep
+installs them on CI; if they are missing the test should fail loudly
+rather than silently skip, so missing deps surface during review.

tests/ui-smoke/_lib/checkresult.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# Shared result check for UI smoke tests.
+#
+# Pass if the driver printed UI_SMOKE_OK and did not print UI_SMOKE_FAIL.
+# The driver only emits UI_SMOKE_OK after a successful NML round-trip
+# (linuxcnc task ready and stat.poll() still alive after settle), so
+# this is sufficient evidence that the GUI booted. We do not grep for
+# generic crash markers like "Segmentation fault" or "Traceback":
+#   - linuxcnc's own scripts/linuxcnc Cleanup may emit shutdown-side
+#     segfaults (Qt/GTK teardown races) that are out of scope for a
+#     startup smoke test;
+#   - launch.sh interleaves linuxcnc.{out,err} before ui-smoke.{out,err}
+#     so an "in-log ordering" check would still catch shutdown noise.
+set -u
+LOG="${1:-/dev/stdin}"
+content=$(cat "$LOG")
+
+if grep -q '^UI_SMOKE_FAIL' <<< "$content"; then
+    echo "FAIL: driver reported UI_SMOKE_FAIL" >&2
+    exit 1
+fi
+
+if ! grep -q '^UI_SMOKE_OK$' <<< "$content"; then
+    echo "FAIL: driver did not report UI_SMOKE_OK" >&2
+    exit 1
+fi
+
+exit 0

tests/ui-smoke/_lib/cleanup-runtime.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+# Reset linuxcnc-related runtime state so the next ui-smoke test starts
+# from a clean environment. Used as both a pre-launch belt-and-braces
+# cleanup and as a post-shutdown last-resort if scripts/linuxcnc's own
+# SIGTERM trap could not reap everything in time.
+#
+# Single source of truth for the daemon list and shared memory keys
+# (BsAtHome review, PR #3999) so they are not duplicated across
+# launch.sh.
+
+set -u
+
+DAEMONS=(linuxcncsvr milltask halui rtapi_app)
+SHM_KEYS=(0x48414c32 0x48484c34 0x00000064)
+
+for proc in "${DAEMONS[@]}"; do
+    pkill -KILL -x "$proc" 2>/dev/null || true
+done
+
+rm -f /tmp/linuxcnc.lock
+halrun -U 2>/dev/null || true
+
+for key in "${SHM_KEYS[@]}"; do
+    shmid=$(ipcs -m | awk -v k="$key" 'tolower($1)==k {print $2}')
+    [ -n "$shmid" ] && ipcrm -m "$shmid" 2>/dev/null || true
+done
+
+exit 0

tests/ui-smoke/_lib/drive.py

@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+# Minimal UI smoke driver: confirm linuxcnc task came up and the GUI
+# did not crash. The smoke layer answers Bertho's "does it start"
+# question only; functional behaviour (home, run a file, verify
+# position) belongs in tests/ui-functional/ (Phase 2).
+
+import linuxcnc
+import sys
+import time
+
+CONNECT_TIMEOUT_S = 60.0
+SETTLE_S = 3.0
+
+
+def connect_and_wait_ready(timeout):
+    """Wait until linuxcnc.stat().poll() returns without error and
+    reports a non-negative echo_serial_number. The NML status buffer
+    can be 'invalid err=3' for the first ~30s while linuxcncsvr is
+    still initialising; recreate the stat object on every iteration so
+    a stale invalid buffer does not stick after linuxcncsvr is ready."""
+    deadline = time.monotonic() + timeout
+    last_err = None
+    while time.monotonic() < deadline:
+        try:
+            stat = linuxcnc.stat()
+            stat.poll()
+            if stat.echo_serial_number >= 0:
+                return linuxcnc.command(), stat
+        except linuxcnc.error as e:
+            last_err = e
+        time.sleep(0.5)
+    sys.stderr.write(
+        f"UI_SMOKE_FAIL: task not ready within {timeout}s "
+        f"(last NML error: {last_err})\n")
+    return None, None
+
+
+def main():
+    cmd, stat = connect_and_wait_ready(CONNECT_TIMEOUT_S)
+    if cmd is None:
+        return 1
+
+    # Give the GUI process enough time to finish constructing itself
+    # (load .ui files, compile resources.py if needed, etc.) and
+    # settle. If the GUI was going to crash on startup it has crashed
+    # by now.
+    time.sleep(SETTLE_S)
+
+    # Re-check task is still alive; a GUI crash may have torn linuxcnc
+    # down via Cleanup.
+    try:
+        stat.poll()
+    except linuxcnc.error as e:
+        sys.stderr.write(f"UI_SMOKE_FAIL: task disappeared after GUI startup: {e}\n")
+        return 1
+
+    print("UI_SMOKE_OK")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

tests/ui-smoke/_lib/launch.sh

@@ -0,0 +1,105 @@
+#!/bin/bash
+# Shared launcher for UI smoke tests.
+# Usage: launch.sh <sim-config-ini>
+#
+# Spawns linuxcnc -r <ini> under xvfb-run, then runs the common driver
+# script against it via NML. Captures stdout/stderr to per-test files.
+#
+# Skip vs fail (BsAtHome / hdiethelm review, PR #3999): the only thing
+# we skip on is xvfb-run absence, which can happen on a developer's
+# minimal machine. CI is expected to have all required deps; if any
+# python module the GUI needs is missing the test should fail loudly
+# rather than silently skip, so missing deps surface during review.
+# Per-GUI deps are declared in debian/control under !nocheck, so
+# apt-get build-dep installs them on CI.
+
+set -u
+
+CONFIG_INI="$1"
+TEST_DIR="${TEST_DIR:-$(cd "$(dirname "$0")" && pwd)}"
+LIB_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+if ! command -v xvfb-run >/dev/null 2>&1; then
+    echo "SKIP: xvfb-run not installed" >&2
+    exit 77
+fi
+
+cd "$TEST_DIR" || exit 1
+rm -f ui-smoke.out ui-smoke.err linuxcnc.pid
+
+# Pre-launch cleanup: a previous ui-smoke test in the same job may
+# have left a daemon listening on the NML TCP port or HAL shared
+# memory still attached. Run the shared cleanup once before we start.
+bash "$LIB_DIR/cleanup-runtime.sh"
+
+# Launch linuxcnc inside xvfb-run. The outer timeout is a safety net
+# so a wedged GUI cannot hang CI.
+LINUXCNC_TIMEOUT=240
+DRIVER_TIMEOUT=90
+
+# Force software OpenGL (Mesa llvmpipe). CI runners have no GPU and
+# Qt/GL widgets segfault under hardware GL with no display. The Qt-
+# specific knobs cover qtdragon's QtQuick + RHI paths.
+export LIBGL_ALWAYS_SOFTWARE=1
+export GALLIUM_DRIVER=llvmpipe
+export QT_QUICK_BACKEND=software
+export QSG_RHI_BACKEND=software
+export QT_OPENGL=software
+
+# touchy's filechooser reads $HOME/linuxcnc/nc_files at startup and
+# crashes if the dir is missing (no try/except in filechooser.py).
+# CI runners have a clean $HOME with no linuxcnc/ tree; create the
+# minimum set of dirs the GUIs poke at on boot.
+mkdir -p "$HOME/linuxcnc/nc_files"
+
+xvfb-run -a --server-args="-screen 0 1024x768x24" \
+    timeout "$LINUXCNC_TIMEOUT" \
+    bash -c "
+        # Run linuxcnc in its own process group so we can signal the
+        # whole group cleanly (linuxcnc forks task, motion, GUI, halrun).
+        setsid linuxcnc -r '$CONFIG_INI' >linuxcnc.out 2>linuxcnc.err &
+        LINUXCNC_PGID=\$!
+        echo \$LINUXCNC_PGID >linuxcnc.pid
+
+        # The driver polls NML readiness itself (BsAtHome review:
+        # avoid real-clock waits where status polling will do).
+        timeout $DRIVER_TIMEOUT python3 '$LIB_DIR/drive.py' >ui-smoke.out 2>ui-smoke.err
+        DRIVE_RC=\$?
+
+        # Clean shutdown: GUI-specific quit first (lets linuxcnc's
+        # own SIGTERM trap run Cleanup which unloads halrun and reaps
+        # shared memory). axis-remote works only for axis but is
+        # harmless otherwise. Then SIGTERM to the script PID so the
+        # trap runs in-process. Wait up to 60s for Cleanup to finish
+        # before falling back to SIGKILL + cleanup-runtime.sh.
+        if command -v axis-remote >/dev/null 2>&1; then
+            axis-remote --quit 2>/dev/null || true
+        fi
+
+        kill -TERM \$LINUXCNC_PGID 2>/dev/null || true
+        for i in \$(seq 60); do
+            kill -0 \$LINUXCNC_PGID 2>/dev/null || break
+            sleep 1
+        done
+        if kill -0 \$LINUXCNC_PGID 2>/dev/null; then
+            echo 'WARN: linuxcnc did not exit on SIGTERM, escalating to KILL' >&2
+            kill -KILL -\$LINUXCNC_PGID 2>/dev/null || true
+            sleep 2
+            bash '$LIB_DIR/cleanup-runtime.sh'
+        fi
+
+        exit \$DRIVE_RC
+    "
+RC=$?
+
+# Surface logs so checkresult and CI artifact upload can see them.
+echo "=== linuxcnc.out ==="
+[ -f linuxcnc.out ] && cat linuxcnc.out
+echo "=== linuxcnc.err ==="
+[ -f linuxcnc.err ] && cat linuxcnc.err
+echo "=== ui-smoke.out ==="
+[ -f ui-smoke.out ] && cat ui-smoke.out
+echo "=== ui-smoke.err ==="
+[ -f ui-smoke.err ] && cat ui-smoke.err
+
+exit "$RC"

tests/ui-smoke/_lib/skip-if-missing.sh

@@ -0,0 +1,17 @@
+#!/bin/bash
+# Shared "skip" predicate for ui-smoke tests.
+# runtests semantics: a `skip` script that returns non-zero causes the
+# test to be skipped. Per-test skip files invoke this.
+#
+# We only skip on xvfb-run absence (rare local dev env). Python /
+# typelib deps are declared in debian/control under !nocheck so CI
+# always has them; missing deps should fail the test loudly rather
+# than silently skip (BsAtHome / hdiethelm review, PR #3999).
+set -u
+
+if ! command -v xvfb-run >/dev/null 2>&1; then
+    echo "skip: xvfb-run not installed" >&2
+    exit 1
+fi
+
+exit 0

tests/ui-smoke/axis/checkresult

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/checkresult.sh" "$@"

tests/ui-smoke/axis/skip

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/skip-if-missing.sh"

tests/ui-smoke/axis/test.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+TEST_DIR="$(cd "$(dirname "$0")" && pwd)" \
+exec "$(dirname "$0")/../_lib/launch.sh" \
+    "$(cd "$(dirname "$0")"/../../../configs/sim/axis && pwd)/axis.ini"

tests/ui-smoke/gmoccapy/checkresult

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/checkresult.sh" "$@"

tests/ui-smoke/gmoccapy/skip

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/skip-if-missing.sh"

tests/ui-smoke/gmoccapy/test.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+TEST_DIR="$(cd "$(dirname "$0")" && pwd)" \
+exec "$(dirname "$0")/../_lib/launch.sh" \
+    "$(cd "$(dirname "$0")"/../../../configs/sim/gmoccapy && pwd)/gmoccapy.ini"

tests/ui-smoke/qtdragon/checkresult

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/checkresult.sh" "$@"

tests/ui-smoke/qtdragon/skip

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/skip-if-missing.sh"

tests/ui-smoke/qtdragon/test.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+TEST_DIR="$(cd "$(dirname "$0")" && pwd)" \
+exec "$(dirname "$0")/../_lib/launch.sh" \
+    "$(cd "$(dirname "$0")"/../../../configs/sim/qtdragon/qtdragon_xyz && pwd)/qtdragon_metric.ini"

tests/ui-smoke/touchy/checkresult

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/checkresult.sh" "$@"

tests/ui-smoke/touchy/skip

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec "$(dirname "$0")/../_lib/skip-if-missing.sh"

tests/ui-smoke/touchy/test.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+TEST_DIR="$(cd "$(dirname "$0")" && pwd)" \
+exec "$(dirname "$0")/../_lib/launch.sh" \
+    "$(cd "$(dirname "$0")"/../../../configs/sim/touchy && pwd)/touchy.ini"