Re: [ANN] Lua 5.2.0 (work1) now available

[David Burgess <dabsoft@...>]

I am not a guru, but let me relate my experiences:

1) if you are going to use windows.h it should be the first header, this 
is because the windows headers modify macros used by the RTL, I have 
read MS documentation on this (but cant remember where).
This means that loadlib.c *needs* an in the module #ifdef to include 
windows.h as the first header. I vaguely remember specific problems with 
modules that include <io.h>, so in writing a Lua library for windows the 
header order is impoortant.
i.e.
<windows.h>
<lua.h>

2) One can argue that VC_EXTRALEAN WIN32LEAN_AND_MEAN STRICT should be 
on the compiler command line invocation not in the headers e/g

/DSTRICT=1 /DWIN32_LEAN_AND_MEAN=1 /DVC_EXTRALEAN=1 
/D_CRT_SECURE_NO_DEPRECATE

and these should be considered depending on what you are building
/D_CONSOLE or /D_WINDOWS\
/D_WIN32_WINNT=0x0500 \
/D_MBCS  or not \


DB


Alexander Gladysh wrote:
> > > > Or just include windows.h where it is needed in loadlib.c under #elif
> > > > defined(LUA_DL_DLL) and not in the luaconf.h file.
> > > > Works without problems for me.
>
> > > Much better. windows.h is a serious kitchen-sink header.
>
> > Agreed.  #include <windows.h> has no business out in my public header chain.
>
> Indeed.
>
> I've expected that, at least, it would be framed by #if
> defined(LUA_CORE) checks, but no, it goes right to public. Nasty.
>
> Furthermore, I remember that including windows.h without extra defines
> is a bad idea.
>
> When I coded for Windows, I always used to do at least this:
>
> #define WIN32_LEAN_AND_MEAN
> #define NOMINMAX
> #include <windows.h>
>
> Much less hassle. (Especially without min and max defined as macroses.)
>
> Also, this (outdated) article suggests there can be even more useful defines:
>
> http://support.microsoft.com/kb/166474
>
> Perhaps some Win32 guru would enlighten us more?
>
> Alexander.