FG segfault when changing LOD ranges

[AR-1]

@SurferTim
1) This seems correct to me
2) Indeed I have seen "no metar available" for a couple of seconds

[SurferTim]

I don't know what else to tell you. I figured after almost 2 years, that bug would have been fixed. Maybe it isn't the same bug.

Bad news: Yours (and I imagine others) crashes a lot.
Good news: Mine never crashes.

[polly]

I'm sorry, can you possibly add some background to this patch ? I've gone bach trough the thread, which references other threads and all seem to be talking about Canvas, Multi displays, Threading, Shiva with a lot of exposition but no direct reference to any problems with a plain PLib menu causing a crash.
Was there an issue with the LOD menu that was never mentioned in the threads but which cleared while applying a supposed multi-threading patch ? Thanks.

[SurferTim]

Let me look. It was a 2 line change to a SimGear file.

Edit: Here is the post with the fix. Thanks to Hooray!
viewtopic.php?f=4&t=39475&start=75#p390955

[polly]

Thanks, I got that fix, it's whether or not there was a problem with a PLib menu at the time that a Canvas fix somehow affected. The lead developer has requested some detail since the original threads are not easy to follow.

[SurferTim]

You are correct. It is hard to follow. I tried so many things that hadn't worked and was becoming frustrated. That was around page 7.
Basically, any canvas display caused my system to crash, like this:

Initially with one display, it took a little less than an hour to crash.
Finally, with 3 displays, the fails happened quickly. I couldn't even get to the runway before a crash.
All this time, I had to wait for the MP server or FG would crash back to my desktop.

Then Hooray talked me into that 2 line addition to the SimGear file, and both the problems went away, never to return.
...and they all lived happily ever after.
The End.

[polly]

OK Thanks, but I'm trying to understand how this is related to a reported segfault in the View-LOD menu; I'd run the debug and traced it to a spot in the PUI code when I understood your posts to be telling me this patch fixed this PLib issue.
I've flagged this patch as a fix for the View-LOD PLib menu.
What I'm reading now is, your issue was Canvas related and the patch is not confirmed to fix a segfault on a PLib menu, is that correct ? Thanks

[SurferTim]

I don't know about PLib. Is View-LOD a canvas display?

[polly]

No, it's the top-of screen menu, accessed by F10-View-Adjust LOD Ranges; the old menu system predates Canvas.

[SurferTim]

It appears to be down to a dilemma.
1) Keep waiting for MP server before selecting menu items.
2) try the 2 line patch.

The question is: Are you willing to try the patch?

Edit: What old menu system> My main screen menu under "equipment" has a canvas map.

[wkitty42]

Edit: What old menu system>

the entire PLIB menu system accessed via F10... the canvas map was added later as have been all canvas items in FG now...

[Hooray]

I don't know what else to tell you. I figured after almost 2 years, that bug would have been fixed. Maybe it isn't the same bug.
[...]
Let me look. It was a 2 line change to a SimGear file.

[...]
Then Hooray talked me into that 2 line addition to the SimGear file, and both the problems went away, never to return.
...and they all lived happily ever after.
[...]
Edit: Here is the post with the fix. Thanks to Hooray!
viewtopic.php?f=4&t=39475&start=75#p390955

Hi Tim,

as far I as I understand, the folks here are talking about a multiplayer related crash, I don't currently see if/how this should be related to said patch.

to be honest, I could hardly remember that topic - but frankly, now that I do recall the patch, I am also surprised that this got never fixed properly.

It's possible that the nature of my patch was not "proper enough" to make the review, I really don't know - I am not actively involved these days, I do remember talking to Jules back then who appreciated the troubleshooting that we did in your topic - and I do think that I moved some of the relevant bits over to the wiki.

If the issue only happens when [not] using MP, that would indicate a threading path that my patch didn't address - I really don't know, I never used MP at all.
Based on my understanding, MP models are paged in/out using PagedLOD, and I believe that work was done by Stuart (?) - so if there are side effects relating to MP being on/off, that's what I'd personally check next, assuming that PagedLOD may be using worker threads that are not properly synchronized with the ShivaVG, but I am really talking out of my .... here

(I do realize that this must sound like Japanese to you, but looking at the thread, it seems that wkitty42 did provide help and backtraces back then, maybe we can recruit him once again ?)

It's best to file a proper bug report with sufficient details so that people can reproduce the issue - it isn't unlikely that your original issue and my patch are unrelated to what MP users are now seeing here, unless it's also affected by the OSG threading mode ?

EDIT: The Canvas.Path issue is summed up here: https://wiki.flightgear.org/FlightGear_ ... 8OpenVG.29

Given that it's been two years, I believe a proper bug report would make sense at this point - and you should get Stuart, Fernando and James involved to get a proper guesstimate as to what might be going on here...And if it's still Shiva related, you guys should also get Erik and Scott involved/CC'ed since they both worked on Shiva in the past.

I suppose, and hope, with these 5+ folks the critical mass should be sufficient to gather some really useful feedback to help us better understand/fix the issue once and for all.

[SurferTim]

Hi Hooray! Like I said earlier, I don't know what else to tell you, except I'm just telling you the facts.
Before the patch, my FG2020.4.0 d_n_c ALWAYS crashed if I didn't wait a minute before selecting "Fly" from the launcher.
After the patch, it has NEVER crashed, even selecting "Fly" within moments of starting the launcher.
The patch was the only change. It was your recommendation. If it hasn't been fixed, that would be up to you to push for a change.
Good news: Mine always works.
Thanks!

Edit: Just one more thing. My problem crash ended with a Segfault in the log. The segfault in the subject is what caught my attention.

[Hooray]

I believe by now it must have been over a year ago since I built/ran FG, so I am not in a particularly good position to judge what's going on.

While I don't quite see how these issues could be related, I will enjoy the credit you're giving my patch while it lasts

Seriously though, the folks still running into segfaults should definitely consider filing a proper bug report.
If this issue is still the problem and if this patch fixes it, it needs to go up the chain to be discussed among core devs.

Personally, I am unlikely to have any significant spare time in the next couple of months to help with this, and if I had a little more time, I would be more interested in looking at some of the Canvas/Compositor issues that Fernando mentioned - which is to say, I am basically out ... sorry folks

If any of the folks mentioned above are currently still around, I would suggest to get them involved - even if just to discuss the original patch/problem (as in, Stuart, Jules, Fernando, James and Scott that come to mind) - to get the process started, someone who uses the d&c script (hello wkitty42 ??) could open a merge request to get the patch discussed/reviewed and integrated (linking back to the original topic), which is probably going to do more in a single week than sitting idle on the forum for another 2 yrs

[polly]

A segfault happens on a memory access out of bounds and so a fix for one segfault may be unrelated to another's.

In this case, when the High Detail slider is clicked, the pointer for the name of the parent dialog is invalid. This happens if both MP server is not connected and AI traffic is not enabled ( at least here ) ; if either MP or AI are active the Parent Dialog has a name, 'static-lod'. Perhaps an AI/MP dialog shouldn't be accessible if AI and MP are inactive, perhaps those menu entries shouldn't be clicked.

It's not clear the CanvasPath patch has a certain effect, I've seen the segfault after applying the patch, maybe less frequently, so I cannot say for sure if the segfault is caused by uninitialised properties or by a thread timing issue.