Re: [Boost-build] Python port development

Subject: Re: [Boost-build] Python port development
From: Artyom (artyomtnk_at_[hidden])
Date: 2010-07-09 15:24:56

> >> I can  confirm this. Yesterday I did just that.  And used it for 
>configuration of 
>
> >> wxWidgets with Boost.Build, for defining  SIZEOF_WCHAR_T macro  
>appropriately (other sizeofs weren't required for  now).
> >> Very  funny coincidence that this talk and my implementation happened at  
>nearly 
>
> >> the same, I must say ;-)
> >>    
> > 
> > How did you do this? Using negative array size trick?
> >  
> 
> I would also be interested in knowing exactly how you did this.
> The  way I would handle it would be to build a program
> that prints the size and  read the output.

That is exactly the way it should not be done, and that is exactly the reason 
you
should give it to build system:

Why it can't be done using printing the sizeof: It does not work when you cross 
compile,
for example I compile on Windows Intel to Linux ARM because compiling on arm is 
very slow. You can't run
the compilation test because it is ARM executable, and if you do compile and run 
native (and not target)
you can actually get wrong result.

So the correct solution (that is done in autotools for years and only recently 
in CMake)
is following.

How can we check for example if sizeof(wchar_t) >= 4?

   char test[sizeof(wchar_t)*2 - 7]

If sizeof(wchar_t) == 4 then 4*2-7 == 1 and the code is legal and compilation 
succeeds,
If sizeof(wchar_t) ==  2 then 2*2 - 7 == -3 and this code is not legal and 
compilation fails.

It is very tricky to implement this correctly. This is why it shouldn't be done 
by users
as most of them would use the solution you had suggested, and it is wrong.

Just to make it clear, even CMake had done this right only recently, before this 
it was implemented
incorrectly. This solution was discovered very long time ago by autotools.

Artyom

P.S.: BTW sizeof(wchar_t)==2 only if you work on Windows, rest platforms have 
sizeof(wchar_t)==4