We recently branched, so it's a good time to think about compiler baselines.
Currently our Windows and macOS baselines are leading the pack in terms
of latest C++ features. Our Linux distro baseline with RHEL/CentOS 7
provides relatively fresh GCC and Clang versions through extra repositories.
GCC 11 in SCL:
https://centos.pkgs.org/7/centos-sclo-rh-x86_64/devtoolset-11-gcc-11.2.1-1.2.el7.x86_64.rpm.html
llvm 13 in EPEL 7:
https://centos.pkgs.org/7/epel-x86_64/llvm13-13.0.1-1.el7.x86_64.rpm.html
In the future, probably the biggest pressure for bumpage will be on
Clang as it is lagging behind the competition in features.
I'm not advocating for any particular bump target, but it would be
interesting to hear if people have any wishes regarding C++20 features.
What has so far come up:
- template-parameter-list for generic lambdas
https://en.cppreference.com/w/cpp/language/lambda#Syntax (available by
bumping GCC to 8 and Clang to 9)
- std::endian https://en.cppreference.com/w/cpp/types/endian (available
by bumping GCC to 8)
I got the idea to look at what Boost library features we use and what
could be replaced by standard features. I don't mean we should aim for
some wholesale banishment of Boost, but for sure minimising its use has
various benefits like reduced build times. Cleaner code, easier
usability and better performance are good reasons to stick with Boost libs.
Command used to discover Boost stuff:
git grep -Eho 'boost::[a-z_:]+' | sort | uniq
Irrelevant stuff like patches in external/boost ignored.
Boost libraries we might want to replace:
- boost::bind http://www.boost.org/libs/bind
https://en.cppreference.com/w/cpp/utility/functional/bind Already in
C++11. Has some subtle differences to std::bind, but is only used in
sdext/source/pdfimport/pdfparse/pdfparse.cxx
- boost::enable_shared_ptr_from_this Only mentioned in obsolete comments
in slideshow/source/inc/soundplayer.hxx
- boost::filesystem https://www.boost.org/libs/filesystem/doc/index.htm
https://en.cppreference.com/w/cpp/filesystem Would require GCC 8 (after
bumping macOS runtime to 10.15 in Jan 2023)
- boost::lexical_cast
https://www.boost.org/doc/html/boost_lexical_cast.html
https://en.cppreference.com/w/cpp/utility#Elementary_string_conversions
Would require GCC 8. Full floating-point support not in any Clang
version yet, but we use Dragonbox for float-to-string anyway
- boost::optional (::make_optional only used in libetonyek)
https://www.boost.org/libs/optional
https://en.cppreference.com/w/cpp/utility/optional
https://www.cppstories.com/2018/05/using-optional/#migration-from-boostoptional
C++17, no need to bump anything
- boost::shared_ptr (related ::dynamic_pointer_cast)
https://www.boost.org/libs/smart_ptr
https://en.cppreference.com/w/cpp/memory/shared_ptr Already in C++11.
Used mostly because it is used in libcmis
- boost::starts_with
https://www.boost.org/doc/libs/release/doc/html/boost/algorithm/starts_with.html
https://en.cppreference.com/w/cpp/string/basic_string/starts_with Would
require GCC 9. Used in libcmis
- boost::variant https://www.boost.org/libs/variant
https://en.cppreference.com/w/cpp/utility/variant
https://www.cppstories.com/2018/06/variant/#migration-from-boostvariant
C++17, no compiler bump needed. We just bumped our macOS runtime to
10.14, so all features related to variant are safe to use
Boost libraries we can revisit when C++23 features become available:
- boost::gregorian and boost::posix_time
https://www.boost.org/libs/date_time
https://en.cppreference.com/w/cpp/header/chrono std::chrono might have
enough features starting from C++20. However, a design issue related to
formatting is only fixed in C++23 (approved):
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p2372r3.html
- Range (::adaptors, ::make_iterator_range)
https://www.boost.org/libs/range
https://en.cppreference.com/w/cpp/ranges More fully featured in C++23
- boost::split (used with boost::is_any_of)
https://www.boost.org/doc/libs/1_79_0/doc/html/boost/algorithm/split.html
https://en.cppreference.com/w/cpp/ranges/split_view Becomes more robust
in C++23 (approved):
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p2210r2.html
Boost libraries we probably don't want to replace (happy to be proven
wrong):
- boost::circular_buffer https://www.boost.org/libs/circular_buffer A
special container with good performance
- boost::container::flat_map Special container for special purpose
- boost::hash https://www.boost.org/libs/container_hash
https://en.cppreference.com/w/cpp/utility/hash Maybe used for some good
reason?
- boost::intrusive::circular_list_algorithms
https://www.boost.org/libs/intrusive Special container for special
purpose. Used only in sw/inc/ring.hxx
- boost::intrusive_ptr https://www.boost.org/libs/smart_ptr Special
pointer with no equivalent in STL
- boost::iterator_facade https://www.boost.org/libs/iterator
https://vector-of-bool.github.io/2020/06
