Faster ui-smoke and workplane-retain: wait on the GUI, and two shutdown waits that never ended

lib/python/common/hal_glib.py

@@ -615,8 +615,9 @@ def update(self):
         state_new = self.old['state']
         if state_new != state_old:
 
-            # set machine estop/clear
-            if state_old == linuxcnc.STATE_ESTOP and state_new == linuxcnc.STATE_ESTOP_RESET:
+            # set machine estop/clear; a reset and machine on between two
+            # polls go from ESTOP straight to ON and still clear the estop
+            if state_old == linuxcnc.STATE_ESTOP:
                 self.emit('state-estop-reset')
                 self.emit('interp-idle')
             elif state_new == linuxcnc.STATE_ESTOP:

scripts/linuxcnc.in

@@ -694,9 +694,21 @@ function Cleanup() {
     # Unused variable 'i', just a counter
     # shellcheck disable=SC2034
     for i in $(seq 10); do
-        # (the one component is the halcmd itself; a thread is listed as
-        # a __<name> pseudo component, which stays until realtime stops)
-        if [ "$($HALCMD list comp | tr -s ' ' '\n' | grep -c -v -e '^__' -e '^$')" = 1 ]; then break; fi
+        # Wait only for components that can still leave: a thread is
+        # listed as a __<name> pseudo component, which stays until
+        # realtime stops, and a user component whose process is gone
+        # (a GUI that exited without hal_exit) never leaves. The halcmd
+        # that listed them has exited by the time its pid is checked.
+        COMPS=$($HALCMD -s show comp)
+        BUSY=0
+        while read -r _ TYPE NAME PID _; do
+            case "$TYPE:$NAME" in
+                RT:__*) ;;
+                RT:*) BUSY=1 ;;
+                User:*) if "$KILL" -0 "$PID" 2>/dev/null; then BUSY=1; fi ;;
+            esac
+        done <<< "$COMPS"
+        if [ "$BUSY" = 0 ]; then break; fi
         sleep .2
     done
 

src/hal/user_comps/gladevcp.py

@@ -391,6 +391,11 @@ def main():
         LOG.debug("Register callback '%s' for SIGINT and SIGTERM" %(signal_func))
         signal.signal(signal.SIGTERM, handlers[signal_func])
         signal.signal(signal.SIGINT,  handlers[signal_func])
+    else:
+        # halcmd unload sends SIGTERM. The KeyboardInterrupt hal makes of
+        # it is lost when it lands in a GLib callback (the status polls),
+        # so quit from the main loop instead.
+        GLib.unix_signal_add(GLib.PRIORITY_HIGH, signal.SIGTERM, Gtk.main_quit)
 
     try:
         Gtk.main()

tests/mdi-while-queuebuster-waitflag/README

@@ -1,3 +1,7 @@
-This test tries to hit a race condition where an MDI command is
-executed when a queue buster is waiting to be restarted after
-INTERP_EXEC_FINISH.
+MDI lines sent behind a remapped M400 that yields INTERP_EXECUTE_FINISH
+wait for it and run in order.  Task sets its plan wait while the remap is
+suspended, and an MDI arriving then is queued instead of being handed to
+the interpreter, which would drop it (7e640f7b08).  The test checks the
+order the lines and the remap report in, through the operator messages,
+and that nothing crashed.  It cannot hit the one task cycle the wait flag
+guards on its own: linuxcnc.command waits for task to take each line.

tests/mdi-while-queuebuster-waitflag/checkresult

@@ -1,2 +1,3 @@
 #!/bin/sh
-exit 0 # test failure is indicated by test.sh exit value
+# the MDI lines and the remap ran, in the order they were sent
+grep -q "^in order$" "$1"

tests/mdi-while-queuebuster-waitflag/remap.py

@@ -9,12 +9,9 @@ def init_stdglue(self):
 def M400(self,**words):
 
     self.execute("M66 E0 L0")
-    # self.execute("M66 P0 L1 Q1")
-    print("self.input_flag pre = %s" % self.input_flag)
+    emccanon.MESSAGE("M400 before the yield")
     yield interpreter.INTERP_EXECUTE_FINISH
 
-    print("self.input_flag post = %s" % self.input_flag)
-    # optionally,
-    # yield interpreter.INTERP_OK
+    emccanon.MESSAGE("M400 after the yield")
 
 

tests/mdi-while-queuebuster-waitflag/test-ui.py

@@ -1,26 +1,39 @@
 #!/usr/bin/env python3
+# MDI lines queued behind a remap that yields INTERP_EXECUTE_FINISH run
+# after it, in the order they were sent: see README.
 
-import linuxcnc, hal
+import linuxcnc
 import sys
+import time
 
-# Initialization
 c = linuxcnc.command()
 s = linuxcnc.stat()
+e = linuxcnc.error_channel()
+
 c.state(linuxcnc.STATE_ESTOP_RESET)
 c.state(linuxcnc.STATE_ON)
 c.mode(linuxcnc.MODE_MDI)
 
-c.mdi('(print,pre 1)')
-c.mdi('(print,pre 2)')
-c.mdi('M400')
-c.mdi('(print,post 1)')
-c.mdi('(print,post 2)')
-c.mdi('(print,post 3)')
-c.mdi('(print,post 4)')
-c.mdi('(print,post 5)')
-c.mdi('(print,post 6)')
+# each line reports through the operator messages, which task sends in the
+# order it runs them; the remap reports before and after its yield
+posts = ["post %d" % n for n in range(1, 7)]
+for cmd in ["(debug,pre 1)", "(debug,pre 2)", "M400"] + ["(debug,%s)" % p for p in posts]:
+    c.mdi(cmd)
+want = ["pre 1", "pre 2", "M400 before the yield", "M400 after the yield"] + posts
 
-# Shutdown
+got = []
+deadline = time.time() + 30
+while len(got) < len(want) and time.time() < deadline:
+    m = e.poll()
+    if not m:
+        time.sleep(0.01)
+        continue
+    print("channel:", m)
+    if m[0] in (linuxcnc.NML_ERROR, linuxcnc.OPERATOR_ERROR):
+        break
+    got.append(m[1].strip())
 c.wait_complete()
-sys.exit(0)
 
+print("ran:", ", ".join(got))
+print("in order" if got == want else "not in order, wanted: " + ", ".join(want))
+sys.exit(0)

tests/mdi-while-queuebuster-waitflag/test.sh

@@ -1,10 +1,5 @@
 #!/bin/bash -e
-
-# Variable 'i' is unused, just a loop count
-# shellcheck disable=SC2034
-for i in $(seq 20); do
-    linuxcnc -r test.ini
-    if grep -q '^[^+].*Segmentation fault' stderr; then
-	exit 1
-    fi
-done
+linuxcnc -r test.ini
+if grep -q '^[^+].*Segmentation fault' stderr; then
+    exit 1
+fi

tests/ui-smoke/_lib/checkresult.sh

@@ -3,7 +3,7 @@
 #
 # 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
+# (linuxcnc task ready and stat.poll() still alive once the GUI is up), 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