[boost] [result_of][borland-5.9.2] BOOST_NO_RESULT_OF

From: Jonathan Turkanis (turkanis_at_[hidden])
Date: 2008-01-15 14:12:35

Hi,

I make one fairly modest use of result_of in Boost.Iostreams.

My initial implementation included result_of.hpp and guarded the use of 
result_of with the macro BOOST_NO_RESULT_OF. I then discovered that 
Borland 5.9.2 and Borland 5.9.3 don't support result_of well enough for 
my needs, but that result_of.hpp header doesn't set BOOST_NO_RESULT_OF 
for these toolsets (since they supposedly support SFINAE). So I added 
checks for Borland version to the guard around result_of.

This worked until yesterday. Now it seems that simply including the 
result_of header fails on Borland (see std_test_result_of_header: 
http://tinyurl.com/26umcr).

I'd like to patch Iostreams as soon as possible, so I'm wondering 
whether this regression will be resolved by making the header includable 
by Borland or by marking std_test_result_of_header as an expected 
failure. It seems to me that result_of.hpp either has to be 
unconditionally includable, or BOOST_NO_RESULT_OF has to be defined in 
boost/config.hpp so I can tell whether to include it.

If result_of is now really unusable on Borland (and if late model 
Borland really supports SFINAE), then it's no longer true that 
BOOST_NO_RESULT_OF is simply the disjunction of two existing config 
macros, and so it seems natural to define it in config.hpp instead of in 
result_of.hpp. (It could be defined in config/compiler/borland.hpp and 
also in config/suffix.hpp if BOOST_NO_SFINAE or 
BOOST_NO_TEMPLATE_PARTIAL_SPECIALIZATION is defined.)

Thanks,

-- 
Jonathan Turkanis
CodeRage
http://www.coderage.com