Re: [RFA] Move cpplib to the toplevel

[Paolo Bonzini <paolo dot bonzini at polimi dot it>]

> Yes the IA64 C++ ABI says that sizeof(bool) should equal to
> 1.  But Apple did not want to break backward compatibility
> with code written before the ABI was written so they kept
> the sizeof(bool) and sizeof(_Bool) to equal to 4 instead
> of 1 like the ABI.  So powerpc-apple-darwin does not follow
> the IA64 C++ ABI to the wire but instead is modified version
> of it with only the sizeof(bool) being different.

Ah, ok.

But there's a more important reason to discard such a sizeof (bool): the 
bool/_Bool that is detected by ACX_HEADER_STDBOOL *really* must match 
whatever is used as a default by system.h, in this case a char. 
Otherwise, this will cause problems for example with toplevel bootstrap: 
you start with a compiler that has no stdbool.h and no _Bool, so 
stage1/libiberty/libcpp/libbanshee will use a 1-byte bool on Darwin, 
while stage2 will happily use the 4-byte builtin _Bool.

This is the same problem that (with gcc configured just once) occurs 
with runtime libraries, see 
http://gcc.gnu.org/ml/gcc/2004-03/msg01219.html (Darwin's saveFP/restFP).

I think that replacing bool with char is a bad thing because it will 
lead to subtle bugs one day or another.  But if we are to emulate bool, 
well, we should do it well and ensure that we ever use a non-emulated 
bool only if it is the same as the emulated one.

Paolo