- Boost - lists.preview.boost.org

[Boost-announce] [review] Metaparse Review Result
by Christophe Henry 13 Jul '15

13 Jul '15

Hi all, First of all, the final result. Metaparse is formally unconditionally accepted into Boost. We had 11 reviews, all "Yes" votes. 2 "No" reviews based on formality were later reversed to a "Yes". The review was lively, with a general agreement that the library is very useful, the design of high quality and the documentation great. Here are some noticeable quotes which helped me take my decision: - "thank you Abel for submitting such an amazing library. The quality of the documentation and code is wonderful and it has been a pleasure to review Metaparse." (Michael Caisse) - (Design) "Of high quality. Based on the MPL paradigm, but I don't view this as a problem." (Peter Dimov) - "Part of the role of boost is to push compilers to the limit (and sometimes beyond). This library certainly has the capabilities to do so." (Roland Bock) Worth noticing is the review of Gordon Woodhull. I myself wrote a new front-end for MSM using Metaparse, and other use cases were very welcome and Gordon managed to produce one for his compile-time graph library by the end of the review, which I found impressive (for both the reviewer and the library). Thank you for investing time in this review. Another important factor for me for acceptance is how well the author is going to maintain his library. We have too many abandoned libraries. Abel went out of his way to oblige remarks and demands made by reviewers. Long before the reviews he also had obliged my many critics and demands. He also showed himself open for experiments. All this while keeping a very calm and professional tone. This is for me a very good sign of a library which will be well-maintained. Finally, there were some repeated important comments which I see as adressed to the rest of Boost that the library is targeted towards library writers. >From Peter Dimov: > - Are you knowledgeable about the problem domain? "I'm not sure if anybody is. :-)" It all depends on us, library and potential library writers to become knowledgeable about this domain and make this library move from potentially useful to very useful. We need to use it to write more cool libraries. I notice a recurring theme where Boost would be dead or on the way to doom etc etc. Some very popular Boost libraries are now part of the Standard, and so what? We need to make new ones. Metaparse is one of them. Let's build more, using it! Yes Reviews(11): Andrzej Krzemienski Bjorn Reese Gordon Woodhull Lee Clagett Louis Dionne Michael Caisse Paul A. Bristow Peter Dimov Roland Bock Edward Diener Niall Douglas I want to thank all the reviewers who invested their time in this review and those who provided some usefuls comments. I also want to thank Abel for presenting this great library. Christophe Henry Metaparse Review Manager

3 2

Re: [boost] [Boost-users] Boost.DLL formal review is ending soon - submit your reviews
by Klaim - Joël Lamotte 12 Jul '15

12 Jul '15

On 12 July 2015 at 21:24, Edward Diener <eldiener(a)tropicsoft.com> wrote: > The name of the library seems Windows specific, since DLL is the term that > Windows uses for shared libraries. Would not Boost DLM ( for dynamic > loadable modules ) or Boost Loadable Library be clearer ? If I remember correctly, there was a long discussion about the naming of the library before the majority seemed ok with DLL because it's explicitely only about loading libraries, not "modules"(which is misleading) nor "plugins" (which would require a lot more features). The name, if my memory is correct, was not particularly thought to be about the Windows context. Other libraries also already took the other names like Boost.SharedObject so it was not an option.

1 0

[msm][eUML2][metaparse] Compile time
by Christophe Henry 12 Jul '15

12 Jul '15

Michael Caisse wrote: >> >> Boost.Msm v3 also implements a new compile-time strings based front-end >> called eUML2 to demonstrate the power of the library and the conciseness >> and expressiveness it allows. >> Please have a look at the documentation: >> >> https://htmlpreview.github.io/?https://raw.githubusercontent.com/boostorg/m… >> >> > >Christophe, this is wonderful work and a great example of the usefulness >of Metaparse. Can you comment on the compile time compared to eUML? Hi Michael, sorry, I had completely forgotten this post and saw it while preparing the review results (which will be done soon). Please allow me to answer this question, with my apologies. I added in the msm3 branch a CompareFrontEnds.cpp example. It is a simple fsm with 12 transitions. I tried with gcc-4.9 and clang-3.5, using -O2 on a i7-5960X processor, suse linux 13.2. Here are the results. eUML clang 4s g++ 5s eUML2 clang 9s g++ 14s Clang is clearly much faster than g++, with about 0.5s added compile-time per transition. With g++ it does not look so good. I already noticed that compile times of most production code at work with g++ is much higher than with clang. Regards, Christophe

2 1

[Better Enums] Maps between enums and other types
by Anton Bachin 12 Jul '15

12 Jul '15

Hello, In past exchanges on this list, several people expressed a desire for bidirectional maps on enums. I've made a simple implementation [1] that relies on constexpr functions containing a switch. The switch forces case exhaustiveness checking. This method therefore requires C++14 to be truly useful. Usage is: ENUM(Channel, int, Red, Green, Blue) constexpr const char* describe(Channel channel) { switch(channel) { case Channel::Red: return "the red channel"; case Channel::Green: return "the green channel"; case Channel::Blue: return "the blue channel"; } } constexpr auto descriptions = better_enums::make_map(describe); int main() { std::cout << descriptions.from_enum(Channel::Red) << std::endl; std::cout << descriptions.to_enum("the green channel") << std::endl; } This prints the red channel Green In principle, make_map should enumerate the function by passing every possible value of Channel, and use the results to build up some lookup data structure. I am not doing that right now because I haven't found a compile-time data structure with fast enough generation to be practical. I suppose with reflective information, there are other ways to check for exhaustiveness besides using a switch. The implementation is currently inefficient (it uses a linear scan to do the reverse mapping). There is also no support for multimaps or partial maps. I could probably support the latter easily by having the function return an option type. I am not sure multimaps are a good idea at all. The only point of them would be to do a reverse mapping from multiple values to the same enum. Interested to hear any opinions. Anton [1] http://aantron.github.io/better-enums/tutorial/Maps.html

1 0

Another Problem (was RE: [type_traits] Another big update in develop)
by Fletcher, John P 11 Jul '15

11 Jul '15

A recent change on develop is breaking something in boost function: _In file included from ../../../boost/function/function_base.hpp:22:0, from ../../../boost/function/detail/prologue.hpp:17, from ../../../boost/function.hpp:24, from ../../../boost/thread/future.hpp:52, from ../../../libs/thread/src/pthread/thread.cpp:19: ../../../boost/type_traits/has_trivial_copy.hpp:47:47: error: template argument 1 is invalid This was not there yesterday. John Fletcher

3 6

Boost.Hana
by Glen Fernandes 11 Jul '15

11 Jul '15

Dear Ron, Louis, and Boost developers, The review period for Hana has concluded. 18 reviews were submitted: - 17 votes for acceptance. (1 of these was conditional upon Hana wholly supporting a second C++ implementation however). - 1 vote for rejection; for lack of compiler support. I submit that Hana be accepted into Boost. I want to congratulate Louis: the Boost community wants Hana to become an official Boost library. Thank you again, Louis, for your work on Hana and for contributing it to Boost. I propose we require, prior to Hana's inclusion in a Boost release, only that: 1. Hana's unit tests to be integrated into Boost's regression tests. If this requires maintaining both CMakeLists.txt and .b2 files, it is still worth it. I am willing to assist with this effort (and the invitation extends to anyone in the community to also contribute to the task). For release managers, potential Hana contributors, or just the general developer community in Boost, to see Hana's test cases in http://boost.org/development/tests/master/developer/summary.html will be useful. I trust, but do not mandate before Hana is officially accepted into Boost, that: 1. All GitHub issues pertaining to implementation and documentation that were opened by Louis in response to review feedback (or by others, such as Vincente, that were approved by Louis) will be addressed eventually (as long as they remain relevant). I personally would ask for, but do not require for Hana's acceptance, that: 1. That Hana's non-reference documentation live outside of the source code. The presentation of Hana's documentation is great; it looks modern, is easy to browse. I am not a fan of the prose living inside hana.hpp though (API reference documentation is the only exception). 2. Louis to investigate further the possibility of a Hana core library if such can be achieved with zero overhead (instead of focusing that effort onto a separate library that provides only a subset of Hana's functionality). On the note of Hana's current support in existing C++ implementations: - While Hana is wholly only supported by clang at present, gcc is not far behind. Given the trend of language conformance, it is reasonable to expect Hana will be entirely supported in a future gcc release. - Hana targets the C++ standard (i.e. This isn't a case of targeting any compiler-vendor specific extensions) and any highly-contentious C++ language features within (e.g. This isn't something like a pre-C++11 library that uses 'export' that only builds with an EDG compiler). - The community now has expressed sufficient interest in both kinds of libraries being part of Boost: Highly portable libraries that target implementation-specific features, and maintain compatibility with even legacy compilers; standard conforming, modern, libraries that are free of the maintenance debt of ancient compiler support. - We need more libraries taking advantage of C++14 language features to drive vendor conformance. A standard may only be as potent as the C++ implementations that support support it, but when one (such as clang does) does, others will be more driven to follow suit when real (and highly desired) projects take advantage of it. My thanks to everyone that participated in Hana's review with a review (Edouard, John Fletcher, John Bytheway, Niall, Krzysztof, Manuel, Roland, Christophe, David Stone, David Sankel, Lorenzo, John, Paul, Zach, Kohei, Charley, Vincente, Abel) as well as discussion and bug reports. Best, Glen

7 9

[BestPracticeHandbook] Run this code online
by Krzysztof Jusiak 11 Jul '15

11 Jul '15

Hi, I've recently added a new functionality to proposed boost.di library documentation. After `try it online` the natural next step was to add `run this code` functionality like in http://en.cppreference.com/w/ or http://www.rust-lang.org/ programming language. Powered by Wandbox and CodeMirror approach, let you play with the particular example and switch between normal quick book programming listing mode and the editor/compiler mode. You can check it here: http://krzysztof-jusiak.github.io/di/cpp14/boost/libs/di/doc/html/di/exampl… Just click on `Run this code` button and you will be switched to the editor mode, where you would be able to edit the code and compile it online using the newest version of the library. Implementation details: https://github.com/krzysztof-jusiak/di/blob/cpp14/doc/js/run_this_code.js It's basically the same solution as `try it online`, just implemented in js instead. Additionally, it also takes the newest version of the library from github. Minimal html: <link rel="stylesheet" href="CodeMirror/lib/codemirror.css"> <link rel="stylesheet" href="CodeMirror/theme/zenburn.css"> <script src="CodeMirror/lib/codemirror.js"></script> <script src="CodeMirror/mode/clike/clike.js"></script> <script src="CodeMirror/addon/selection/active-line.js"></script> <script src="run_this_code.js"></script> <button class="TryItBtn" id="run_it_btn" onclick="show(' https://raw.githubusercontent.com/krzysztof-jusiak/di/cpp14/examples/hello_…)">Run this code!<\/button> <textarea style="display: none" id="code"></textarea> <textarea style="display: none" id="output"></textarea> <div id="code_listing"> //quick book programming listing here or anything else, which should be hidden when editor mode is enabled </div> Let me know, what do you think. Cheers, Kris

3 2

Using clang on Windows with mingw-64
by Edward Diener 11 Jul '15

11 Jul '15

This is just a heads up about testing with clang on Windows. The latest clang, when built from the latest source, is now capable on Windows of using the mingw-64/gcc RTL. Furthermore it is no longer necessary to have the mingw/mingw-64 implementation hardcoded or linked from c:\mingw. With the latest clang the executable will look for a mingw/mingw-64 implementation on the PATH, meaning that the if the implementation's 'bin' directory is on the PATH clang will find it and use its RTL. The ability of the latest clang to use mingw-64 means that clang can use the latest gcc-5.1 RTL on Windows. The mingw-64 implementation offers more up to date and better implementations of gcc than mingw. There is one "gotcha" I have found with this setup on Windows when using clang on Windows with gcc-4.9 on up. The gcc intrin.h implementation of mingw-64 for gcc-4.9 on up has a bug where if the gcc version is less than 4.9 a compiler error will occur. I have reported this to mingw-64 on their developer's mailing list but it has been ignored. Unfortunately clang still sets itself to gcc version 4.2.1. But a workaround solves the problem and that workaround is that when using clang you define '__MINGW_FORCE_SYS_INTRINS', which causes the bug in the mingw-64 intrin.h header file to not trigger. Previous versions of clang on Windows ( 3.4, 3.5, 3.6 ) of course still have the limitation of only working with mingw and needing the mingw implementation hardcoded at c:\mingw. But at least with the latest clang built from source this has changed as explained above. The Boost build clang toolset, which on Windows is really clang-linux, should continue to work with this latest change of clang.

2 2

[1.59] Boost 1.59 schedule?
by Joaquin M López Muñoz 09 Jul '15

09 Jul '15

Hi, The official release schedule at http://www.boost.org/development/index.html seems to indicate we're now really close to the first beta of Boost 1.59, yet I don't see much activity in the mailing list and, for instance, the release notes are almost empty now: https://github.com/boostorg/website/blob/master/feed/history/boost_1_59_0.q… Is this schedule still valid? Thank you, Joaquín M López Muñoz Telefónica

6 7

Problem cross-compiling boost.context for Raspberry Pi (2)
by Biddiscombe, John A. 09 Jul '15

09 Jul '15

Hello list, I am able to compile the full set of boost libraries on the Raspberry Pi, and I get all the libraries successfully built, including context and coroutine which I need. However, when I cross compile on an Ubuntu virtual machine setup for the job, all the libraries except context/coroutine are built. The problem seems to be that when cross compiling, the file "unsupported.cpp" is being pulled in (here is a verbose output showing flag I'm using) "/home/biddisco/pi/tools/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64/bin/arm-linux-gnueabihf-g++" -ftemplate-depth-128 -O3 -finline-functions -Wno-inline -Wall -pthread -fPIC -D__arm__ -std=c++11 -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard -DBOOST_ALL_NO_LIB=1 -DBOOST_CHRONO_DYN_LINK=1 -DBOOST_CONTEXT_DYN_LINK=1 -DBOOST_CONTEXT_SOURCE -DBOOST_SYSTEM_DYN_LINK=1 -DBOOST_SYSTEM_NO_DEPRECATED -DBOOST_THREAD_BUILD_DLL=1 -DBOOST_THREAD_POSIX -DBOOST_THREAD_USE_DLL=1 -DNDEBUG -I"." -c -o "bin.v2/libs/context/build/gcc-arm/release/threading-multi/unsupported.o" "libs/context/src/unsupported.cpp" which emanates from the jamfile boost_1_58_0/libs/context/build/Jamfile.v2 where the following alias asm_context_sources : unsupported.cpp ; is in a section marked #COMBINED Now, when I build on the pi itself, this file is not being added to the build, so it seems to me that during the bootstrap phase on the pi, different flags are being used which trigger the correct behaviour of bjam my user-config.jam looks like this using gcc : : /home/biddisco/pi/tools/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64/bin/arm-linux-gnueabihf-g++ : <compile-flags>-D__arm__ -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard ; (NB. I added -D__arm__ to see if it helps, but it does not make a difference without it). I have no knowledge of how bjam / boost.build do their things, so I do not know what to try next. It seems that I need to bootstrap the build using the flags for the pi, but built for the host so that it can generate the correct makefiles and this looks like where the problem is arising. the build process seems good apart from this wrinkle The bootstrap command I'm using is ./bootstrap.sh --with-toolset=gcc and the build appears sensible when I use ./b2 cxxflags="-D__arm__ -std=c++11 -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard" --layout=versioned threading=multi link=shared variant=release -j8 Performing configuration checks - 32-bit : yes - arm : yes .... etc etc Can anyone see what I'm doing wrong? (Note : With another project which depends on libpthread, I had to copy libpthread from the pi to the host and force the linker to use the pthread lib from the pi and not the host one, does bjam do something internally which might require a trick like this? - note2 - the binaries generated via this process run without error) Many thanks JB -- John Biddiscombe, email:biddisco @.at.@ cscs.ch http://www.cscs.ch/ CSCS, Swiss National Supercomputing Centre | Tel: +41 (91) 610.82.07 Via Trevano 131, 6900 Lugano, Switzerland | Fax: +41 (91) 610.82.82

4 17