Fwd: LNUM patch Good Friday revise

[Asko Kauppi <askok@...>]

A critical bug in the LNUM patch has been revealed by Aladdin Lampé  
(thanks!).

It happens in all LNUM_INT64 compilations of Lua applying the patch,  
and essentially changes certain 64-bit values (s.a. 0x1 00000001) to  
small integers (this one to 1) at code compilation (but not always).  
Anyone using LNUM for 64-bit support is adviced to fix this immediately.

The quick fix is (in ltable.c 'luaH_getint()'):

  if (cast(unsigned int, key-1) < cast(unsigned int, t->sizearray))
    -->
  if (cast(unsigned LUA_INTEGER, key-1) < cast(unsigned int, t- 
>sizearray))

The actual fix on LNUM svn and LuaForge (not uploaded yet) will be a  
bit different, but both essentially do the same.

Sorry about the issue.

-Asko



Edelleenlähetetty viesti alkaa:

> Lähettäjä: Asko Kauppi <askok@dnainternet.net>
> Päiväys: 10. huhtikuuta 2009 klo 20:34.08
> Vastaanottaja: Lua list <lua@bazar2.conectiva.com.br>
> Aihe: LNUM patch Good Friday revise
>
>
> Happy Easter, everyone.
>
> There's been some fixes to LNUM patch, mostly concerning  
> optimizations on x86 and x86_64 platforms. All known cases of  
> crashing are now solved.
>
> For people using LNUM succesfully on embedded arena, there's nothing  
> much new. But having the fixes done makes me feel better since  
> there's no unknown issues now.
>
> Download patch here:
>
> 	http://luaforge.net/frs/?group_id=214
>
> Or check out a pre-applied source base, here:
>
> 	svn co svn://slugak.dyndns.org/public/2009/LNUM2
>
> - asko
>
>
> The LNUM patch allows easy and wide configuration of Lua number  
> types, with the use of simple compiler defines (LNUM_INT32,  
> LNUM_INT64, LNUM_FLOAT, LNUM_DOUBLE, LNUM_LDOUBLE, LNUM_COMPLEX). It  
> provides added integer accuracy (32- or 64-bit), enhances speed on  
> non-FPU systems and allows use of complex numbers (C99 required).