Nikita Kniazev

From the root of Boost, where `bootstrap.sh` is located: `b2 libs/config/test` to run the whole test suite of the Boost.Config library and `b2 libs/config/test//name_of_the_test` to run a test named `name_of_the_test`...

Actually, it seems that I am wrong and you need to do `b2 libs/config/test/all` to run the test suit that includes that test, and `b2 libs/config/test/all//BOOST_NO_CTYPE_FUNCTIONS` to run the test...

> Boost 1.80 > > ``` > humpty$ b2 libs/config/test/all//BOOST_NO_CTYPE_FUNCTIONS > Performing configuration checks > > - default address-model : 64-bit (cached) [1] > - default architecture : x86 (cached)...

> The tests under /all/ are useful sometimes, but need to be taken with a pinch of salt also. I personally fixed wrong defines in config that lived for a...

I don't want to think how this should be resolved, feel free to just close the PR.

That's still about a megabyte repository for three lines of code. I believe it is not a large burden to transfer that header, is it?

The change it conflicts with (035c096da942a3fa7f75180ae8b2f2732db0a9f4) is just wrong and should be reverted, then this will land cleanly.

What is `the original purpose of those workarounds`? To slow compilation for everyone because of Boost.LexicalCast? I fixed Boost.LexicalCast and it is not need the workaround in Boost.Config.

I have reproduced the https://svn.boost.org/trac10/ticket/4115 problem by applying https://github.com/boostorg/lexical_cast/pull/31 and https://github.com/boostorg/config/pull/307 on VC9.0/10.0/11.0, the fix for LexicalCast proposed in https://github.com/boostorg/lexical_cast/pull/32.

I did not say it is a bad idea, it is convenient but costly.