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

[Tacvek]

----- Original Message ----- 
From: "Ronald Lamprecht" <address@hidden>
To: "Tacvek" <address@hidden>
Cc: "Daniel Heck" <address@hidden>; "enigma-devel" <address@hidden>
Sent: Wednesday, June 28, 2006 4:23 PM
Subject: Re: [Enigma-devel] Lua 5.1 "luaL_error" problems on Windows


> Hi,
>
> Tacvek wrote:
> > And finally a slightly suspicious one:
> >
> > > printf-args.c: In function `printf_fetchargs':
> > > printf-args.c:83: warning: `wint_t' is promoted to `int' when passed 
> > > through `...'
> > > printf-args.c:83: warning: (so you should pass `int' not `wint_t' to 
> > > `va_arg')
> > > printf-args.c:83: note: if this code is reached, the program will abort
> >
> >
> > I've always had this appear, although Daniel did not seem to be able to 
> > reproduce it.
> > I had assumed I have some bad header somewhere.
> >
> > If this error is also showing up for other people then I would be 
> > suspicious.
> > Aborting does indeed appear to be what is happening, and I think the 
> > error printing code
> > does use *printf.
>
> This warning occurs in GNU gettext compilation. I saw it, too. But the 
> crash messages are not translated. So it is very unlikely to be the 
> reason.
>
> The gettext version included is only compiled for Windows - otherwise we 
> use the OS-installed gettext version.

Indeed. Well actually, the included gettext would also be used if enigma was 
compiled
an a very obscure platform that lacks its own gettext.

Anyway, it is more than just gettext, it looks like a whole libintl. It 
includes implementations
of the *printf functions, and i suspect it is overiding the ones included in 
the C library,
or compiler (the C standard allows those functions to be implemented in 
either location).

I'm still not clrear what exactly a wint_t is, but if it is being used, it 
could explain the error.

It would also explain the consistancy. The quickest way to tell is to 
use --disable-nls.

I tested that. Besides a few small tweaks that are needed to make it compile 
using with that, it worked.
The resulting exe also crashed.

So my next guess is an uncaught C++ exception.

> Eliminating the reason for this warning would be anyway a step forward.
>
> @Tacvek: The gcc 3.4.5 version you tested did use which exception method - 
> DW2 or sjlj exceptions (see 
> http://www.mingw.org/MinGWiki/index.php/build%20a%20Win32%20x-compiler%20for%20Linux)?

I was using whatever is is used by the Debian 3.4.5 build.

Apparently that is compiled using "--enable-sjlj-exceptions".
http://ftp.debian.org/debian/pool/main/m/mingw32/mingw32_3.4.5.20060117.1-1.diff.gz

> I am still wondering why the crashes occur very stable with the calls 
> listed in this thread independent of Lua, minggw and Enigma version, but 
> other luaL_error calls work totally stable independet of the versions at 
> the same time. What is the difference between these luaL_error calls?

There does not appear to be any difference.


Wait a second: The test on line 17 (enigma.GetAttrib(sender, "notexisting"), 
the one not inside the function) does not always crash. Sometimes I manage 
to get an error message instead.

Also, if with an "--disable-optimize" build the version of 
enigma.GetAttrib(sender, "notexisting") found inside the function causes the 
program to crash. Dr watson reports a " 0xC0000096" error, which is 
apparently "Privileged instruction error".

Loading it into the Visual studio debugger tells me:
"Unhandled exception at 0x008e19c0 in enigma.exe: 0xC0000096: Privileged 
instruction."

The instruction appears to be a "hlt" instruction.
It is followed by two
"db ffh" instructions which are apparently illeagal instructions.