Attached is a chopped down autopilot.xml that, for this user, causes fgfs to exit during startup with debug 'backtrace' showing 'no stack'
Breakpoint at autopilotgroup.cmm#189 addAutopilotFromFile
then
Breakpoint at fg_osg_viewer#340 .. F9 continue a few times as previous BP is hit,
When at fg_osg_viewer #340 F5 step into and again and again and ..
gdb seems to lose its stack at: globals.cxx #199 :
inline double get_sim_time_sec () const { return sim_time_sec; }
At that point pressinfg F5 'next' has no effect and the program must be restarted.
Print sim_time_sec yields $1 = 0
This symptom showed, consistently, on both desktop ( Ryzen 1300 / SSD ) and laptop ( I7 / HDD ) many times while I found that removing the one line: <max-rate-of-change> made the problem go away. On the laptop I disabled 'Master Audio' and the symptom stopped and hasn't returned with the short autopilot.xml I posted; the issue still shows with the full length autopilot.xml.
Turning off the master audio on the desktop has made no difference, although I noticed that on a 'good' run the audio startup log messages are the next to show on the console.
I've tried all multithreading options, is there a way of running single thread and being 100% positive there's no persistent configuration getting accessed ? Tks </max-rate-of-change>
I've rebuilt after disabling Audio and AEON Wave by running ccmake on simgear.
Now ( desktop ) when the error occurs I see ' Qt - like' popup dialog:
I can't see any property name matching that but, since the model is yet to be developed, it's possible I'm specifying properties yet to exist. Is that the cause, with the gui message somehow getting obliterated if Audio gets to start up ? Tks
It looks as if the culprit is:
Changing this to a one-liner makes the error go away, ( props.cxx#171)
Once switching off audio allows the error message to show this is a non issue for me, I can open the ticket under Audio if desired or should I look at changing the source to either / or : change to a logging incident / strip whitespace ? Tks
Last edit: Huntley Palmer 2020-12-04
Slighty losing the plot here :) The ' found in property name is fixable, that's a missing call to 'trim' to remove leading and trailing white-space from the <input> element.
The audio thing, please make a seperate ticket for, I reckon.
OK, yep
Note for posterity: I'm adjusting the Autopilot XML loading code, to accept leading/trailing whitespace in strings, so we tolerate the XML issue.