Error Message When Closing Dark Radiant

[Fidcal]

Dark Radiant 0.8.1 of 12 Apr 2007

 

I get an error message when I close Dark Radiant if I have selected anything. If I just open DR then close it then no error. If I open, it, select an item, then close DR, then I get the error. If I open, select an item then deselect it and close DR I still get the error. Tried it in two maps and its the same. Doesn't even need a map - just any entity to select.

 

The error produces a very long report but I can't copy and paste it. It also refers to a text file output with a ton of data in it but it doesn't look like the same data. I can upload that if it is of any use.

 

The errror does not seem to affect actually using DR as far as I can tell but just keeps appearing each time I exit.

[greebo]

Without seeing the error I've no idea what this could be. The radiant.log or any screenshots/text dumps would be nice.

[Fidcal]

The error report screenie probably won't help much HERE because it only shows the start of a lengthy scrollable report.

 

The radiant log is HERE

 

The other text output referred to in the screenie I couldn't find and I think it had been removed from the temp folder before I got to it. So I re-ran DR and got another HERE - so it might not exactly match (though I just clicked a light in both cases then exited) but I can do the whole thing again tomorrow if that's crucial.

[greebo]

So this is an ordinary crash? Looking at the logfile I can't see anything suspicious, the logfile gets closed regularly without being interrupted. Any GTK errors visible after it crashed when you run it from a DOS box?

[Fidcal]

OK, the output in a DOS window is HERE

[greebo]

Ok, angua could reproduce it on her machine, whereas I just get some GTK assertions (after opening DarkRadiant, creating a light, deselecting it and closing the app).

 

Maybe this is the GTK library throwing up here? My machine just shows some text errors in the console and quits without "crash", angua's doesn't show the text errors but "crashes". If the GTK DLLs are involved I'm fairly sure that these are related to the double destroy commands in the Entity Inspector we recently talked about. I suggest we revisit this topic after the other issue has been resolved, maybe this problem is gone as well then.

[Fidcal]

Yes, that was my instinct since the error does not happen for you it will probably go away when I put in a later install. it's no problem for me.

[OrbWeaver]

I am seeing a segfault on shutdown on Windows, which seems to exhibit equivalent behaviour (triggered by selecting an entity, does not occur when nothing has been selected). I wonder if this could be related to the fix for the GTK issue on shutdown?

 

EDIT: Fortunately it appears on Linux as well, meaning it can be debugged.

[greebo]

OrbWeaver, have you been able to find out where the crash is happening? If you point me in the right direction I can take a look at it. Otherwise I'd have to go the usual way of perforating the code with std::cout statements to find it (after an educated guess where to start).

[OrbWeaver]

Yes, I know what the problem is. Unfortunately my suggestion to add a conditional GTK_IS_WIDGET() test before destroying the PropertyEditor's widget was not the correct solution, because GTK has already shut down when this happens meaning that the _widget pointer refers to invalid memory, and even GTK_IS_WIDGET() is not acceptable.

 

I think what I will do is add a GTK quit callback (with gtk_quit_add()) that will set the PropertyEditor's widget pointer to NULL when GTK exits. Then in the destructor, the gtk_widget_destroy will only be called if the pointer is not NULL.

 

(If you want to implement this yourself now, feel free -- I won't get to it until this evening).

[greebo]

I will have quick look at this, it's probably sufficient to add such a callback in the base class, wouldn't it?

[greebo]

This should be resolved now. Changes are committed to SVN.