Re: [Enigma-devel] Lua 5.1 "luaL_error" problems on Windows

[Ronald Lamprecht]

Hi,

Tacvek wrote:
> > I am very confident that the bug is not located in Enigma. But I still 
> > have no idea whether it is Lua or Mingw gcc that causes the trouble.
> >
> > A mingw gcc 3.4.5 - DW2 test would be interesting.
>
> That is what I did compile the one that gives the unusual  termination 
> dialog box with.However I cross compiled, which for some reason seems to 
> make it al but imposible to debug.
>
> Lets see what we do know for sure:
> * It looks like Enigma is throwing the XLevelRuntime, but for some 
> reason Enigma's exception handlers are not getting a chance to catch the 
> exception.
> * Apparently something is causing the program to segfault while
> * With slsj we are getting a segfault. Apparently, SDL's parachute is 
> doing something, as the program terminates rather than truely crashing, 
> but truely crashes when run in wizard mode.
> * It looks like SDL is not wring a parachute deployment notice to 
> stderr.txt, but I'm pretty sure  I've seen such a notice in stderr.txt 
> in other cases.
>
> * With -DW2 exceptions there is a unusual dialog box. This appears to be 
> raised by MSVC++'s runtime library. (More likely it is actually the C 
> runtime library). This unusual dialog box occurs in wizard mode as well 
> as normal mode.
>
> What a mess.

I looked at Lua how it processes luaL_error. luaconf.h offers 3 methods
for LUAI_THROW/LUAI_TRY: C++ exceptions, _longjmp/_setjmp and
longjmp/setjmp. Linux and Mac use _longjmp/_setjmp, Windows uses
longjmp/setjmp independent of the MingW exception method (SJLJ, DW2).

The MingW setjmp.h looks strange: setjmp is mapped to _setjmp, _longjmp
does not exist. The comment to the used buffer is funny:
> /*
>  * The buffer used by setjmp to store the information used by longjmp
>  * to perform it's evil goto-like work. The size of this buffer was
>  * determined through experimentation; it's contents are a mystery.
>  * NOTE: This was determined on an i386 (actually a Pentium). The
>  *       contents could be different on an Alpha or something else.
>  */
> #define _JBLEN 16

One experiment would be to compile Lua as C++ - this would guarantee the
same exception mechanism being used in Enigma and Lua. (Some Enigma
sources would need minor adaptation. No idea about tolua++' exception 
dependency). The effects on SJLJ and DW2 would be interesting.

On the other hand a Visual Studio Express compilation should use
Microsoft SEH. Another chance to bypass tan expection handling problem.

BTW the DW2 unusual dialog box raised by MSVC++'s runtime library may be
caused by an exception passing dll boundaries:
http://thread.gmane.org/gmane.comp.gnu.mingw.user/9193/focus=9232
The following versions seems to be the first that should be able to pass
dll boundaries:
http://thread.gmane.org/gmane.comp.gnu.mingw.user/18860/focus=18860

> I'll be vacationing the next two weeks so I'll likely not be much help 
> in fixing this problem.

- Ronald