[boost] [Boost.Config] Using <cstddef> instead of <utility> for std lib detection

Subject: [boost] [Boost.Config] Using <cstddef> instead of <utility> for std lib detection
From: David Deakins (ddeakins_at_[hidden])
Date: 2009-03-27 17:06:04

I am in the process of migrating our VeecoFTC regression tests from VC8 
and STLport 5.1 to VC9 and STLport 5.2.  While working on this, I have 
experimented with turning on STLport's _STLP_USE_BOOST_SUPPORT macro. 
When this macro is defined, STLport will pull in the Boost.TypeTraits 
header files instead of its own type traits implementations.  It also 
suppliments a few of the standard headers (memory and functional) with 
Boost-based TR1 functionality much like Boost.TR1 does.  However, 
similar to some of what Boost.TR1 had to deal with, some caution is 
required to avoid circular dependencies in the headers.  One area where 
this tends to come up is in detecting the standard library in use in 
Boost.Config.  Currently config/select_stdlib_config.hpp and 
config/stdlib/stlport.hpp include <utility> to determine what std 
library is provided.  However, <utility> can tend to pull in some code 
that wants type traits and thus may circularly pull in Boost.  One 
convenient way to resolve this would be to use the much simpler 
<cstddef> header instead of <utlity> to determine the std lib.  Is this 
a reasonable change to make or is there something special about using 
<utility> in the select_stdlib_config code?

-Dave