binary executable size increase: gcc-9 vs gcc-10

[sheerluck]

I use fmt in some simple toy project.
I noticed increase of binary executable size:

• after compilation with gcc-9 the executable has 427K
• after compilation with gcc-10 the executable has 472K

I used https://github.com/google/bloaty this way:

$ bloaty -n 0 -d symbols exe-9 > txt-9
$ bloaty -n 0 -d symbols exe-10 > txt-10

This is excerpt from txt-9 sorted by name:

    FILE SIZE        VM SIZE    
 --------------  -------------- 
   1.2%  5.30Ki   1.6%  5.08Ki    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::arg_formatter<fmt::v6::buffer_range<char> >, fmt::v6::basic_format_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::arg_formatter<fmt::v6::buffer_range<char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_format_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.3%  1.16Ki   0.3%     984    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::arg_converter<long, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::arg_converter<long, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)

   2.6%  11.2Ki   3.5%  11.0Ki    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char>::format<fmt::v6::printf_arg_formatter<fmt::v6::buffer_range<char> > >() (.constprop.0)
   0.3%  1.14Ki   0.3%     979    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char>::format<fmt::v6::printf_arg_formatter<fmt::v6::buffer_range<char> > >() (.constprop.0.cold)
   0.2%     883   0.2%     722    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::internal::arg_formatter_base<fmt::v6::buffer_range<char>, fmt::v6::internal::error_handler>::operator()<int, 0>(int)

 100.0%   427Ki 100.0%   314Ki    TOTAL

This is excerpt from txt-10 sorted by name:

    FILE SIZE        VM SIZE    
 --------------  -------------- 
   3.1%  14.5Ki   4.0%  14.3Ki    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::arg_formatter<fmt::v6::buffer_range<char> >, fmt::v6::basic_format_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::arg_formatter<fmt::v6::buffer_range<char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_format_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.2%  1.03Ki   0.2%     848    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::arg_converter<long, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::arg_converter<long, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.2%  1.03Ki   0.2%     848    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::arg_converter<long long, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::arg_converter<long long, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.2%  1.13Ki   0.3%     950    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::arg_converter<short, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::arg_converter<short, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.2%  1.13Ki   0.3%     950    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::arg_converter<signed char, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::arg_converter<signed char, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.2%     803   0.2%     591    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::printf_width_handler<char>, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::printf_width_handler<char>&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&)
   0.2%     803   0.2%     586    decltype(fp(0)) fmt::v6::visit_format_arg<fmt::v6::internal::printf_width_handler<char>, fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> >(fmt::v6::internal::printf_width_handler<char>&&, fmt::v6::basic_format_arg<fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char> > const&) (.cold)

   2.8%  13.0Ki   3.6%  12.8Ki    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char>::format<fmt::v6::printf_arg_formatter<fmt::v6::buffer_range<char> > >() (.constprop.0)
   0.1%     573   0.1%     381    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::basic_printf_context<std::back_insert_iterator<fmt::v6::internal::buffer<char> >, char>::format<fmt::v6::printf_arg_formatter<fmt::v6::buffer_range<char> > >() (.constprop.0.cold)
   0.1%     334   0.0%     173    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::internal::arg_formatter_base<fmt::v6::buffer_range<char>, fmt::v6::internal::error_handler>::operator()<double, 0>(double)
   0.1%     334   0.0%     173    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::internal::arg_formatter_base<fmt::v6::buffer_range<char>, fmt::v6::internal::error_handler>::operator()<float, 0>(float)
   0.9%  4.36Ki   1.2%  4.20Ki    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::internal::arg_formatter_base<fmt::v6::buffer_range<char>, fmt::v6::internal::error_handler>::operator()<int, 0>(int)
   0.1%     354   0.1%     193    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::internal::arg_formatter_base<fmt::v6::buffer_range<char>, fmt::v6::internal::error_handler>::operator()<long double, 0>(long double)
   0.7%  3.13Ki   0.8%  2.97Ki    std::back_insert_iterator<fmt::v6::internal::buffer<char> > fmt::v6::internal::arg_formatter_base<fmt::v6::buffer_range<char>, fmt::v6::internal::error_handler>::operator()<unsigned int, 0>(unsigned int)

 100.0%   472Ki 100.0%   356Ki    TOTAL

I don't know if anything can to be done in fmt -- this may be gcc's fault

[vitaut]

What {fmt} version and exact compile/link flags did you use?

[sheerluck]

{fmt} version 6.2.0
CMakeLists.txt:

set(CMAKE_CXX_STANDARD 20)
set(CMAKE_CXX_FLAGS_RELEASE "-march=native -mtune=native -O3 -mmmx -msse -msse2 -msse3 -pedantic -Wextra -Wshadow -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -fgraphite-identity -floop-nest-optimize -fdevirtualize-at-ltrans -fipa-pta -fno-semantic-interposition -flto=9 -fuse-linker-plugin -pipe -falign-functions=32")

[vitaut]

I recommend using the master branch which has some major library size optimizations. Also make sure to disable assertions by defining NDEBUG. As for difference between gcc 10 and 9, I'm not sure - probably some inlining heuristic changed.

[vitaut]

See also http://www.zverovich.net/2020/05/21/reducing-library-size.html