Re: [ANN] Lua 5.2.0 (rc2) now available

[Georg Lehner <jorge-lua@...>]

Hello!

Visual Studio 2008 Professional with Win32 and ARMV4I compilers here

Using Warning Level 4:
----------------------

return (exit(status), 0);

-> warning C4702: unreachable code

if (L) exit(status);
return 0;

-> no warning


exit(status);
return 0;

-> warning C4702: unreachable code


Using Warning Level 3:
----------------------

return (exit(status), 0);

-> no warning

if (L) exit(status);
return 0;

-> no warning


exit(status);
return 0;

-> no warning

- - -
exit() is annotated as "noreturning" in Microsofts stdlib headers, so 
the compiler expects not to return
after exit().
- - -

When compiling the complete lua5.2 DLL with warning level 4, I get 
several reasonable "possible loss of data"
warnings which could be avoided by proper explicit casting.

Then I get some "unreachable" code warnings in places where 
luaG_runerror() is followed by "return 0;".
This is understandable, since luaG_runerror() is annotated as 
"noreturning" via  "l_noret".

To be consistent, either don't use l_noret in Lua code, use "return 0" 
after not-returning function calls
and use compiler warning settings to filter the "unreachable code" 
warning messages with "smart" compilers --
or use l_noret wherever applicable, remove the "return 0" after 
not-returning function calls and use
compiler warning settings to filter the "function must return a value" 
warning messages with the "dumb" compilers.

Another idea: Is it feasible to redeclare exit() this way:

  extern l_noret exit(int code);

inside some appropriate Lua header to help "not so dumb" compilers which 
have stdlib.h headers without annotation?

Best Regards,

    Jorge-León


On 11/29/11 13:19, Roberto Ierusalimschy wrote:
> > > >    return (exit(status), 0);
> > > >          
> >      
> > > >    if (L) exit(status);  /* test to avoid warnings of 'exit' x 'return' */
> > > >    return 0;
> > > >          
> > I have tested the previous two patches on the bunch of compilers
> > currently installed on my computer.
> > None of them produced a warning for both tricks.
> > The first one is probably better because it avoids an additional
> > useless test-and-jump.
> >      
> According to [1], something similar to the first option gave warnings
> in MSVC2005.
>
> [1] http://lua-users.org/lists/lua-l/2008-01/msg00375.html
>
> It would be nice if other people could try these options in their
> systems.
>
> -- Roberto
>
>