[lmi] Replacing boost with C++ standard library [Was: Replacing boost with std C++11]

[Greg Chicares]

On 2017-04-25 14:27, Greg Chicares wrote:
[...remaining boost dependencies...]
> 
> (I) Standard replacements available with C++17:
> 
>   #include <boost/filesystem/*>
[...]
> Requires gcc-5.3 and '-lstdc++fs':
>   
> http://stackoverflow.com/questions/33149878/experimentalfilesystem-linker-error/33159746#33159746

IIRC, the API changed; I haven't tackled this one yet.

>   #include <boost/optional.hpp>
> Used by auxiliary rate-table program only. Apparently in gcc-4.9 with 
> '-std=c++1y':
>   
> http://stackoverflow.com/questions/17013708/implementation-status-of-stdoptional-in-g-libstdc/17013985#17013985

Replaced with std::experimental::optional.

>   #include <boost/any.hpp>
> Used by the product editor only, which is really just an auxiliary program 
> that
> happens to be built into lmi. Apparently in C++17.

Replaced with std::experimental::any.

> (II) Available in C++11, but gcc's implementation is slow:
> 
>   #include <boost/regex.hpp>
[...]
>   wx_test_about_version.cpp
> where presumably it could be replaced with <regex>

Done.

>, and by one auxiliary program:
>   test_coding_rules.cpp
> for performance reasons (because gcc's implementation is slow).

But that's a standalone program, so any special warning flags it
requires don't affect lmi.

> (III) Convenient, but not strictly necessary:
> 
>   #include <boost/operators.hpp>
> Used by lmi's datum*, mc*, and tn* classes as a convenience; the operators 
> could
> instead be written by hand.

Still used in three headers; eliminated in eight others today.

> (IV) Not really needed at all:
[...]
> Used only by unit tests, and only to compare boost to a preferred alternative
> that is actually used in lmi.
> 
> Upgrading to C++17 apparently takes care of category (I). We could throw away
> all of (IV) anytime. Category (II) could be dispensed with by writing the
> operators ourselves. That leaves only category (III), where we could either
> retain boost for 'test_coding_rules' only, or just put up with gcc's slowness.

Ignoring 'filesystem' (in namespace 'fs'), and filtering out comments,
the unit tests (category IV), and 'test_coding_rules':

$grep 'boost::' *.?pp |sed -e'/^[^:]*: *\/\//d' -e'/^[^:]*_test.cpp:/d' 
-e'/^test_coding/d'           
calendar_date.hpp:    :boost::additive<calendar_date,int>
calendar_date.hpp:    ,boost::totally_ordered<calendar_date>
calendar_date.hpp:    ,boost::unit_steppable<calendar_date>
config.hpp:namespace fs = boost::filesystem;
mc_enum.hpp:    ,private boost::equality_comparable<mc_enum<T>,mc_enum<T>>
mc_enum.hpp:    ,private boost::equality_comparable<mc_enum<T>,T>
mc_enum.hpp:    ,private boost::equality_comparable<mc_enum<T>,std::string>
tn_range.hpp:    ,private boost::totally_ordered    <tn_range<Number,Trammel>>
tn_range.hpp:    ,private 
boost::equality_comparable<tn_range<Number,Trammel>,Number>
tn_range.hpp:    ,private 
boost::equality_comparable<tn_range<Number,Trammel>,std::string>

> Anyway, if we upgrade the C++ dialect and write comparison operators 
> ourselves,
> we can remove boost utterly from lmi, with all the problems it brings...and
> enable all the warnings it has forced us to turn off.

The only remaining obstacles are 'filesystem' and a few sets of operators.