August 2019 - Boost - lists.preview.boost.org

[release] Boost 1.71.0 Release Candidate 2 available
by Michael Caisse 20 Aug '19

20 Aug '19

The release candidates for the 1.71.0 release are now available at: <https://dl.bintray.com/boostorg/release/1.71.0/source/> The SHA256 checksums are as follows: 1b3949ffddf56129141032542523920034f2e69775fb9332a52ba37c548dbf35 boost_1_71_0_rc2.7z d73a8da01e8bf8c7eda40b4c84915071a8c8a0df4a6734537ddde4a8580524ee boost_1_71_0_rc2.tar.bz2 96b34f7468f26a141f6020efb813f1a2f3dfb9797ecf76a7d7cbd843cc95f5bd boost_1_71_0_rc2.tar.gz 85a94ac71c28e59cf97a96714e4c58a18550c227ac8b0388c260d6c717e47a69 boost_1_71_0_rc2.zip As always, the release managers would appreciate it if you download the candidate of your choice and give building it a try. Please report both success and failure, and anything else that is noteworthy. -- The Boost Release Managers

5 6

Potential project: Resolving boost::chrono conflicts with std::chrono
by Howard Hinnant 20 Aug '19

20 Aug '19

Possible project for someone looking for a boost project to contribute to: boost::chrono was created soon after std::chrono was proposed and served people well for experimenting with this library prior to migrating to C++11. Fast forward 8 years: Now we have two competing chrono libraries: boost::chrono and std::chrono. And it is not rare for people to use non-chrono boost libraries, which in turn use boost::chrono, and for those same people to use std::chrono. Invariably what happens is they get horribly complicated compile-time errors which boil down to: boost::chrono does not interoperate with std::chrono. And these errors often come from deep within libraries which people are simply trying to use. The pitch: boost::chrono, and other boost::libs needs to defer to std::chrono for C++11 and later. This would make boost significantly easier to use. This is likely a multi-project effort, and I don’t even have a concrete strategy in mind. This is a problem in search of a solution, not vice-versa. And I often hear of enthusiastic people who want to jump in and lend a helping hand. I think this would be a terrific area to point such talent towards. Howard

13 27

Re: [boost] [multiprecision] Proposal for enhancing cpp_int multiplication
by kotika98 19 Aug '19

19 Aug '19

-------- Original message --------From: Ankur Dua via Boost <boost(a)lists.boost.org> Date: 17/08/2019 18:17 (GMT+08:00) To: boost(a)lists.boost.org Cc: Ankur Dua <ankurdua15(a)gmail.com> Subject: [boost] [multiprecision] Proposal for enhancing cpp_int multiplication We have benchmarked cpp_int vs mpz_int (GNU GMP backend) and we found thataround 10^6 bits, cpp_int is tremendously slow (almost 1200x times). Wehave also prepared a document containing the benchmarks [1] alongwith thedetails about the idea and the project.References,[1] Introductory document -https://docs.google.com/document/d/1cclKlbBWDVmY9zKSdn0ga2yTcCkyD9ZiUUdCZZkVyaI/edit?usp=sharing__________________________________________________________________________________ Nice idea for a project! In GMP, things are organized around thresholds for switching to the next algorithm. It is impossible to predict theoretically what that threshold will be or how it will evolve in the future with changes to computer architecture.1. However, for very large input, FFT dominates over Karatsuba and Toom. The threshold is around 4000 limbs which is roughly 250,000 bits. BTW, it is below your test limits.2. For most sizes in between 0 and 4000 limbs, Toom-8 dominates, not Karatsuba.3. Karatsuba and Toom are anyway both special cases of FFT.https://gmplib.org/manual/Multiplication-Algorithms.html#Multiplication…

1 0

I'm getting some neat results by changing graph testing over to using pre-compiled headers
by jrmarsha 18 Aug '19

18 Aug '19

Shout out to Rene Rivera for helping me get this stuff together. I'm getting some useful results out of changing to using pre-compiled headers. There isn't the speed-up I expected, but it has actually turned out useful for finding out incorrect header inclusion patterns and will certainly be making the code more reliable. This is something that we ought to consider trying more broadly in the project.

1 0

[multiprecision] Proposal for enhancing cpp_int multiplication
by Ankur Dua 18 Aug '19

18 Aug '19

Hi Boost community, We are team of two students (Senior Undergraduate) from IIT(ISM) Dhanbad — Madhur Chauhan & Ankur Dua. We want to bring your attention towards the the potential project that we wish to complete — Enhancing multplication of cpp_int by using Karatsuba as the multiplication algorithm. We have benchmarked cpp_int vs mpz_int (GNU GMP backend) and we found that around 10^6 bits, cpp_int is tremendously slow (almost 1200x times). We have also prepared a document containing the benchmarks [1] alongwith the details about the idea and the project. We are awaiting response from the community after which we will prepare a formal proposal containing the details of approach and possible implementations with documentation and test code changes. Thanks & Regards Madhur Chauhan & Ankur Dua madhur4127(a)gmail.com ankurdua15(a)gmail.com References, [1] Introductory document - https://docs.google.com/document/d/1cclKlbBWDVmY9zKSdn0ga2yTcCkyD9ZiUUdCZZk…

2 1

Interest in updated, standalone iterator_facade library?
by Zach Laine 18 Aug '19

18 Aug '19

I write a lot of iterators, and I'm getting sick of doing it by hand. I can't use Boost.Iterator's iterator_facade in library code, since it knows nothing about constexpr and noexcept. The design of Boost.Iterator's iterator_facade does not lend itself very well to being modified in a direction that is standardization-friendly (reasons why are outlined in the online docs). So, here is one take on this problem: GitHub: https://github.com/tzlaine/iterator_facade Online Docs: https://tzlaine.github.io/iterator_facade For those unfamiliar with the problem, iterators are very simple, at the highest level of abstraction. An iterator's implementation does not reflect this -- it is highly repetitive and easy to get subtly wrong. This library fixes that. The library is forward-looking, since I intend to write WG21 papers to standardize iterator_facade; there is good support for upcoming C++20 library features. In particular, the iterators created with iterator_facade model the C++20 iterator concepts, and therefore do not necessarily match the requirements of the C++17-and-earlier requirement tables. I have not found this to affect the usability of iterator-facade-based iterators with pre-C++20 code. Something to note: there was a recent-ish attempt to stanrdize iterator_facade, which is now abandoned, P0186 (http://wg21.link/P0186) This was very different from what I'm proposing for Boost. I talked to Eric Niebler about this, and he favors a return to the CRTP-based approach of my lib and the original Boost.Iterator iterator_facade library. Zach

9 20

Interest in STL interface templates for iterators, views, and containers? (Was: Interest in updated, standalone iterator_facade library?)
by Zach Laine 17 Aug '19

17 Aug '19

On Wed, Aug 14, 2019 at 10:44 AM Zach Laine <whatwasthataddress(a)gmail.com> wrote: > On Wed, Aug 14, 2019 at 4:44 AM Hans Dembinski <hans.dembinski(a)gmail.com> > wrote: > > [snip] > > >> > Since iterator_interface is a lot more like Boost.Iterator's >> iterator_adaptor than my previous iterator_facade was (and I'll look into >> that), it may be possible to make one template that does everything, but I >> will not include two separate templates. >> >> If you could merge the two that would be even better, but I am not sure >> you can. >> > > I can, and I did last night. You can see an example here (not that this > is not the master branch): > > > https://github.com/tzlaine/iterator_facade/blob/interface_alternative/examp… > > When I get MSVC fixed, I'll merge all this to master. > This is now done. The focus of the library is now a bit different. All the iterator bits are still there, but now iterator_facade is called iterator_interface, and has facilities for acting more like an iterator_facade and more like an iterator_adaptor. One of the new things is an implementation of C++20's view_interface. Give it a range or view, even one with only begin() and end(), and it gives you all the view-type API that can be supported from begin() and end(), based on the iterator types and categories. The other new thing is container_interface. It is analogous to view_interface, but instead of supporting definition of new view types, it supports the definition of new containers. It has support for all the operations in the sequence container and reversible container requirements in the standard. In the example in the docs, I was able to write a static_vector, for which container_interface provided 38(!) of the member functions. Finally, the library's name has changed to reflect all this. You can now get the code here: https://github.com/tzlaine/stl_interfaces and see the online docs here: https://tzlaine.github.io/stl_interfaces/doc/html/index.html Zach

1 0

Beast/Context build failures
by Vinnie Falco 15 Aug '19

15 Aug '19

I'm seeing new build failures related to Boost.Context, could this be the recent changes there? <https://travis-ci.org/boostorg/beast/jobs/571029999> Thanks

3 2

Boost.Predef still broken in 1.71 RC 1
by Niall Douglas 13 Aug '19

13 Aug '19

As reported earlier, Boost.Build's predefined macro testing support is still misdetecting in RC 1: Using Boost 1.71 RC 1 with VS2019: b2 libs/outcome/test Performing configuration checks - default address-model : 32-bit - default architecture : x86 - !BOOST_COMP_GNUC : no - BOOST_COMP_GNUC >= 6.0 : no - !BOOST_COMP_CLANG : no - BOOST_COMP_CLANG >= 4.0 : no - Boost.Config Feature Check: cxx14_variable_templates : yes - Boost.Config Feature Check: cxx14_constexpr : yes - !BOOST_COMP_GNUC : no - BOOST_COMP_GNUC >= 6.0 : no - !BOOST_COMP_CLANG : no - BOOST_COMP_CLANG >= 4.0 : no - Boost.Config Feature Check: cxx14_variable_templates : yes - Boost.Config Feature Check: cxx14_constexpr : yes ...found 1 target... Using Boost 1.71 RC 1 with GCC 7.3 on Linux: g++ --version g++ (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0 Copyright (C) 2017 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. ./b2 libs/outcome/test Performing configuration checks - !BOOST_COMP_GNUC : no - BOOST_COMP_GNUC >= 6.0 : no - !BOOST_COMP_CLANG : no - BOOST_COMP_CLANG >= 4.0 : no - default address-model : 64-bit - default architecture : x86 - Boost.Config Feature Check: cxx14_variable_templates : yes - Boost.Config Feature Check: cxx14_constexpr : yes - !BOOST_COMP_GNUC : no - BOOST_COMP_GNUC >= 6.0 : no - !BOOST_COMP_CLANG : no - BOOST_COMP_CLANG >= 4.0 : no - Boost.Config Feature Check: cxx14_variable_templates : yes - Boost.Config Feature Check: cxx14_constexpr : yes ...found 1 target...

2 2

Re: [boost] Potential project: Resolving boost::chrono conflicts with std::chrono
by Glen Fernandes 12 Aug '19

12 Aug '19

Howard wrote: > There is truth to this rationale, but I also wanted to point out that it is not absolute. > [snip] > Not providing a boost implementation does not preclude enhancing a std::lib. > But it can certainly help in changing existing signatures, Indeed. Of course, someone who wants to prototype some breaking change to a library like chrono or smart_ptr can do it without Boost. For example, we could have reimplemented ALL of shared_ptr in github/JohnDoe/smart_ptr to prototype the extensions added in C++17 and C++20. But here were some advantages of not doing so, and using Boost instead: - NOT reimplementing ALL of shared_ptr - Access to a bigger user base than github/JohnDoe/smart_ptr would have e.g. * Boost being white-listed as approved open source in many organizations * New versions of Boost included in many Linux distribution software repositories * etc. So it was an easy choice to go through Boost because boost::shared_ptr wasn't just an alias for std::shared_ptr, and had a quality, well tested, implementation. Still, even with Smart_Ptr, you have the option to convince the maintainer of Smart_Ptr to make boost::shared_ptr just an alias for [C++17] std::shared_ptr, and make boost::allocate_shared just defer to [C++20] std::allocate_shared. It would make users unhappy. There are already users of boost::local_shared_ptr which depends on the implementation of boost::shared_ptr, that would continue using Boost even after C++20 features are implemented everywhere. Glen

1 0