- Boost-users - lists.preview.boost.org

Very slow destructors of binary_oarchive and filtering_streambuf
by boost.5nyej＠simplelogin.com 03 Oct '22

03 Oct '22

Hello, I just posted a question on StackOverflow, but I realised that this mailing list might be a better place. The link to the question is: https://stackoverflow.com/questions/73938656/very-slow-destructors-of-binar… Tl;dr; I have a code that serializes a big object by using binary_oarchive and filtering_bufstream. The writing on its own takes around 28 minutes, but the deconstructors of the mentioned classes take 37 and 36 minutes, respectively. It is very strange to me that they would take so much time to run, as the first deconstructor adds only 2KiB of data, and the second deconstructor doesn't change anything. Could you help me find the root cause of those execution times?

1 0

[serialisation] Strange "direct access" warning on OSX for basic_text_oprimitive
by Marco Craveiro 29 Sep '22

29 Sep '22

Hi Boost Users, First, a great thanks to the authors of the boost serialisation library, which I have used successfully across projects and has proven invaluable. Now to my problem. I have been investigating for a little while a strange warning I get when linking on OSX; the results of my investigation can be found here [1]. The TL; DR of it is as follows: my OSX debug build has a warning (indented for ease of reading): ld: warning: direct access in function 'boost::archive::basic_text_oprimitive< std::__1::basic_ostream<char, std::__1::char_traits<char> > >::~basic_text_oprimitive()' from file 'vcpkg_installed/x64-osx/debug/lib/libboost_serialization.a(basic_text_oprimitive.o)' to global weak symbol 'std::__1::basic_ostream<char, std::__1::char_traits<char> >& std::__1::endl<char, std::__1::char_traits<char> > (std::__1::basic_ostream<char, std::__1::char_traits<char> >&)' This warning is caused by a mismatch in visibility settings between my project and vcpkg's build of boost. However, as far as I can see, both projects are using "-fvisibility=default", meaning there should be no mismatch. Also, no other warnings show up but if I change the project settings I then get hundreds of warnings (again, for details see ticket). What I find even more puzzling is that I am using all three types of archives (Text, XML and Binary). Of these three, only Text produces this warning. As the problem is with the destructors, I did a quick inventory of the destructors to see how they differ: basic_xml_oarchive.hpp: [2] BOOST_ARCHIVE_OR_WARCHIVE_DECL ~basic_xml_oarchive() BOOST_OVERRIDE; basic_text_oarchive.hpp: [3] ~basic_text_oarchive() BOOST_OVERRIDE {} basic_binary_oarchive.hpp: [4] no destructor. So my question is, should text archive also have a BOOST_ARCHIVE_OR_WARCHIVE_DECL macro, and would that fix my warning? If so I can submit a PR. Many thanks for your time. Marco [1] https://github.com/Microsoft/vcpkg/issues/4497 [2] https://github.com/boostorg/serialization/blob/develop/include/boost/archiv… [3] https://github.com/boostorg/serialization/blob/develop/include/boost/archiv… [4] https://github.com/boostorg/serialization/blob/develop/include/boost/archiv… -- Marco Craveiro MD, Domain Driven Consulting blog: https://mcraveiro.github.io/index.html twitter: https://twitter.com/MarcoCraveiro That the Ideas are themselves manifestations (of the Idea-Idea) and that the Idea-Idea is a-kind-of Manifestation-Idea—which is a-kind-of itself, so that the system is completely self-describing— would have been appreciated by Plato as an extremely practical joke [Plato]. -- Alan Key

1 0

Re: [Boost-users] [boost] [signals2][variant] inline warning with MSVC 2022
by Frank Mori Hess 22 Sep '22

22 Sep '22

On Thu, Sep 22, 2022 at 8:57 AM Peter Dimov <pdimov(a)gmail.com> wrote: > > That's complete nonsense. First, Microsoft invented __forceinline, so > making BOOST_FORCEINLINE be a no-op on the very compiler for which > the feature has been added would be ridiculous. Second, it does force > inlining, that's why it's called __forceinline. The whole point of the warning > is to warn about the rare cases where the compiler does not honor the > command; it exists because people want to know whether their "force > inline" orders are ignored. But no boost user wants to be warned about some internal implementation detail of a boost library. So every boost usage of BOOST_FORCEINLINE needs to add a #pragma warning(disable: 4714), awesome! Who is the maintainer of variant anyways? -- Frank

1 0

TSan error with boost::interprocess::message_queue on macOS ARM
by vl01＠jens-carl.de 22 Sep '22

22 Sep '22

Hello List, I'm using the boost::interprocess::message_queue in a project and to verify everything is working I created a test case. In that test case I'm sending and receiving data via threads. The test passes on Linux, Windows with no sanitizer and also with AddressSanitizer and ThreadSanitizer. I recently added Apple Silicon (M1) to the test matrix and now the test run with ThreadSanitizer enabled fails. Is boost::interprocess::message_queue thread-safe? Can the reported TSan error be an false-positive? If it that case, what test can I run to verify that? I looked at code of atomic_cas32() and it uses the legacy built-in __sync_val_compare_and_swap(). Can it be that that built-in isn't supported on ARM? Is there a newer built-in which should be used? The reported TSan error: -- snip ================== WARNING: ThreadSanitizer: data race (pid=4146) Atomic write of size 4 at 0x000104af4028 by thread T2: #0 __tsan_atomic32_compare_exchange_val <null>:81245684 (libclang_rt.tsan_osx_dynamic.dylib:arm64e+0x57854) #1 boost::interprocess::ipcdetail::atomic_cas32(unsigned int volatile*, unsigned int, unsigned int) atomic.hpp:581 (tsan_boost:arm64+0x100005b30) Previous read of size 4 at 0x000104af4028 by thread T1: #0 boost::interprocess::ipcdetail::spin_mutex::try_lock() mutex.hpp:100 (tsan_boost:arm64+0x100006eb0) #1 void boost::interprocess::ipcdetail::try_based_lock<boost::interprocess::ipcdetail::spin_mutex>(boost::interprocess::ipcdetail::spin_mutex&) common_algorithms.hpp:62 (tsan_boost:arm64+0x100006e28) SUMMARY: ThreadSanitizer: data race atomic.hpp:581 in boost::interprocess::ipcdetail::atomic_cas32(unsigned int volatile*, unsigned int, unsigned int) -- snap Here is a code example which demonstrates the test case -- snip #include <boost/interprocess/ipc/message_queue.hpp> #include <iostream> #include <vector> #include <thread> #include <memory> struct MessageQueueImpl { MessageQueueImpl() { boost::interprocess::permissions file_permissions; file_permissions.set_permissions(0660); queue = std::make_unique<boost::interprocess::message_queue>(boost::interprocess::create_only, "tsanqueue", 10, 15, file_permissions); } std::unique_ptr<boost::interprocess::message_queue> queue; }; class MessageQueue { public: MessageQueue() : m_impl(std::make_unique<MessageQueueImpl>()) {} void send(const std::vector<uint8_t>& buffer) { m_impl->queue->send(buffer.data(), buffer.size(), 0); } void receive(std::vector<uint8_t>& buffer) { boost::interprocess::message_queue::size_type received_size = 0; buffer.resize(m_impl->queue->get_max_msg_size()); uint32_t prio; m_impl->queue->receive(buffer.data(), buffer.size(), received_size, prio); buffer.resize(received_size); } private: std::unique_ptr<MessageQueueImpl> m_impl; }; int main() { auto uut = std::make_unique<MessageQueue>(); std::thread sender([&uut]{ for(uint8_t i = 0; i < 3; ++i) { const std::vector<uint8_t> send_buffer = {i}; uut->send(send_buffer); } }); std::thread receiver([&uut]{ for(uint8_t i = 0; i < 3; ++i) { std::vector<uint8_t> receive_buffer; uint32_t prio; uut->receive(receive_buffer); std::cout << "buffer received:\n"; for (const auto i : receive_buffer) { std::cout << i << '\n'; } } }); if(sender.joinable()) { sender.join(); } if(receiver.joinable()) { receiver.join(); } return 0; } -- snap Cheers, Jens

3 3

Re: [Boost-users] [boost] [signals2][variant] inline warning with MSVC 2022
by Frank Mori Hess 22 Sep '22

22 Sep '22

On Wed, Sep 21, 2022 at 12:32 PM Peter Dimov <pdimov(a)gmail.com> wrote: > > That's not very user friendly; maybe the user _wants_ the warning in his code, > and the above would disable it even if no Boost header that uses > BOOST_FORCEINLINE (and generates a warning) has been included. > Okay, then maybe BOOST_FORCEINLINE shouldn't be defined as __force_inline on compilers where it doesn't actually force inlining. If the compiler is just going to treat it as a suggestion, I don't see the point of boost headers using it and then suppressing the warning. -- Frank

1 0

[signals2][variant] inline warning with MSVC 2022
by Gavin Lambert 21 Sep '22

21 Sep '22

This may not be a new issue, as I'm trying to update a project from Boost 1.68 to 1.80 (big jump!), but in trying to compile I'm being hit with this lengthy warning (and this particular project is building with warnings as errors, so this is problematic): warning C4714: function 'class boost::variant<class boost::shared_ptr<void>,class boost::signals2::detail::foreign_void_shared_ptr> __cdecl boost::detail::variant::visitation_impl<struct boost::mpl::int_<0>,struct boost::detail::variant::visitation_impl_step<struct boost::mpl::l_iter<struct boost::mpl::l_item<struct boost::mpl::long_<3>,class boost::weak_ptr<class boost::signals2::detail::trackable_pointee>,struct boost::mpl::l_item<struct boost::mpl::long_<2>,class boost::weak_ptr<void>,struct boost::mpl::l_item<struct boost::mpl::long_<1>,class boost::signals2::detail::foreign_void_weak_ptr,struct boost::mpl::l_end> > > >,struct boost::mpl::l_iter<struct boost::mpl::l_end> >,class boost::detail::variant::invoke_visitor<class boost::signals2::detail::lock_weak_ptr_visitor const ,0>,void const *,struct boost::variant<class boost::weak_ptr<class boost::signals2::detail::trackable_pointee>,class boost::weak_ptr<void>,class boost::signals2::detail::foreign_void_weak_ptr>::has_fallback_type_>(int,int,class boost::detail::variant::invoke_visitor<class boost::signals2::detail::lock_weak_ptr_visitor const ,0> &,void const *,struct boost::mpl::bool_<0>,struct boost::variant<class boost::weak_ptr<class boost::signals2::detail::trackable_pointee>,class boost::weak_ptr<void>,class boost::signals2::detail::foreign_void_weak_ptr>::has_fallback_type_,struct boost::mpl::int_<0> *,struct boost::detail::variant::visitation_impl_step<struct boost::mpl::l_iter<struct boost::mpl::l_item<struct boost::mpl::long_<3>,class boost::weak_ptr<class boost::signals2::detail::trackable_pointee>,struct boost::mpl::l_item<struct boost::mpl::long_<2>,class boost::weak_ptr<void>,struct boost::mpl::l_item<struct boost::mpl::long_<1>,class boost::signals2::detail::foreign_void_weak_ptr,struct boost::mpl::l_end> > > >,struct boost::mpl::l_iter<struct boost::mpl::l_end> > *)' marked as __forceinline not inlined I'm not sure whether the Real Bug™ is in signals2 or in variant, but I don't think it's in any of my code since it doesn't name any of my types above that I can see. I tried the big stick of setting all of Boost to "external" (and set the external warning level to 0), but sadly this did not suppress the warning. Wrapping it in a #pragma suppression does hide the warning, which makes me sufficiently happy. But perhaps someone might be interested to poke at it?

2 2

version 1.78.0 compile failed on centos 7 with compiler clang
by David Frank 15 Sep '22

15 Sep '22

error message: boost/spirit/home/classic/phoenix/closures.hpp:274:30: error: default initialization of an object of const type 'const index<0>' without a user-provided default constructor Thanks.

1 0

version 1.78.0 compile failed on centos 7
by David Frank 15 Sep '22

15 Sep '22

error message: boost/spirit/home/classic/phoenix/closures.hpp:274:30: error: default initialization of an object of const type 'const index<0>' without a user-provided default constructor Thanks.

1 0

version 1.78.0 compile failed on centos 7
by David Frank 15 Sep '22

15 Sep '22

error message: boost/spirit/home/classic/phoenix/closures.hpp:274:30: error: default initialization of an object of const type 'const index<0>' without a user-provided default constructor On Saturday, September 3, 2022 at 08:48:54 PM PDT, David Frank <david_frank95(a)yahoo.com> wrote: I'm currently using version 60, it failed to compile with openssl 3.0.5. So I need to upgrade to a version that works. Thanks.

1 0

Redefined macro on Solaris, OmniOS, and SmartOS
by Camden Narzt 11 Sep '22

11 Sep '22

Building code (on Solaris, OmniOS, and SmartOS, using gcc) which includes boost, produces warnings such as the following: In file included from src/boost/config/platform/solaris.hpp:17:0, from src/boost/config.hpp:57, from src/boost/cstdint.hpp:36, from src/code_that_includes_boost.cpp:43: src/boost/config/posix_features.hpp:27:0: warning: "BOOST_HAS_STDINT_H" redefined [enabled by default] # define BOOST_HAS_STDINT_H ^ Not a big deal, but figured I’d report it in case it hadn’t been reported yet.

1 0