July 2018 - Boost - lists.preview.boost.org

[iostreams] libstdc++ vs MSVC exception handling behavior in seekg, seekp
by James E. King III 17 Jul '18

17 Jul '18

I found a disabled test in Boost.Iostreams called stream_state_test. It was disabled because it was thought it only tests the standard library implementation. Partially true, but it also tests some parts of iostreams so I have been working to re-enable it along with the generic CI upgrades. It looks like libstdc++ and the standard library that comes with MSVC (is it still Dinkumware? It used to be...) handle exceptions on stream seekg and seekp differently. I haven't consulted the standards documents yet, but I found this reference for exception handling: http://www.cplusplus.com/reference/istream/istream/seekg/ which states, " Any exception thrown by an internal operation is caught and handled by the function, setting badbit <http://www.cplusplus.com/ios_base::badbit>. " Sitting in the debugger where the test throws an exception in error_device::seek, there is no exception handling within iostreams or in msvcp140d code paths for a call to seekg, while the libstdc++ implementation is able to pass all of the tests. This leads to a couple questions: 1. Are the standards clear in this area? 2. Is someone violating them? A cursory review (such as this) leads me to believe that the MSVC C++ runtime implementation is not properly handling exceptions here. The sources for seekg in libstdc++ wrap the internals in a try catch in order to restore state. The sources for seekg in the MSVCPRT do not. - Jim

2 1

Safe Float design question
by Damian Vicino 17 Jul '18

17 Jul '18

Hi, I looking for some design advice. The context: - I'm revisiting SafeFloat after some time and, while doing so, I'm rethinking all the decisions from the past. Idea is to sent for review in the following months. - The goal of the library is being a drop-in replacement for float, that adds checks to floating point operations and reports when a check failed. Without SafeFloat, someone could write something like this: #include <iostream> #include <limits> #include <cfenv> using namespace std; int main(){ float a = 1.0f; float b = numeric_limits<float>::max(); feclearexcept(FE_ALL_EXCEPT); float c = a/b; if(fetestexcept(FE_UNDERFLOW)) { cout << "underflow result\n"; } } What I expect when using safe_float is to declare upfront "what checks" I care about, "what operations" to check, and "what to do when a check fails". Some intention examples: - I would like to check there was no "overflow_to_infinite" in "addition operations", otherwise I "throw exception".. - I would like to check there was no "division by zero" in "division", otherwise I will "log to cerr and ignore it". - I would like to check there was no "inexact" "addition", otherwise I return an boost::unexpected. I have at least 5 things to check (there is 5 flags in c++11::fenv). I have at least 4 places to check (+/-/*//) and I want to keep what to do about it customizable. In addition, sometimes I want to check multiple things "overflow and underflow", etc... So, the question is how the user can pass all that information to the type and it doesn't look as horrible nonsense. My original option was: int main(){ using CHECK = compose_policy<check_addition_overflow, check_division_by_zero, check_division_underflow>::type; using REPORT = report_throw_on_failure; using sf = safe_float<float, CHECK, REPORT>; try{ sf a = 1.0_sf; sf b = numeric_limits<sf>::max(); auto c = a/b; } catch (safe_float_exception e){ cout << e.message(); //this outputs there was a underflow } } Please comment in what you think about this way to use. Is there a better way to specify the policies to apply that I should try? Best regards, Damian I expect when using safe float to write some code like this: int main(){ safe_float<float> }

2 3

[system] Clang CI builds broken because of constexpr
by Andrey Semashev 17 Jul '18

17 Jul '18

Hi, I'm seeing these errors in Travis CI builds with Clang 3.8, 3.9, 4.0 and 5.0: ./boost/system/error_code.hpp:232:41: error: constexpr constructor never produces a constant expression [-Winvalid-constexpr] BOOST_SYSTEM_CONSTEXPR explicit std_category( boost::system::error_category const * pc ): pc_( pc ) ^ ./boost/system/error_code.hpp:232:41: note: non-constexpr constructor 'error_category' cannot be used in a constant expression /usr/bin/../lib/gcc/x86_64-linux-gnu/4.8/../../../../include/c++/4.8/system_error:69:5: note: declared here error_category() noexcept; ^ I'm assuming this is because the compiler uses libstdc++ from gcc 4.8, in which error_category() is not marked constexpr. Can this be fixed somehow?

1 0

[graph] disjoint_sets
by James E. King III 17 Jul '18

17 Jul '18

Hello folks in the Boost.Graph team, The CMT currently maintains Boost.DisjointSets which contains a class that is only used by Boost.Graph and Boost.GraphParallel. The headers are published in "boost/pending" right now. Given the CMT has a considerable workload already I would like to discuss moving disjoint_sets into Boost.Graph, as long as it satisfies the requirements for the two repositories currently consuming it. This would be for 1.69.0. Any thoughts, objections, considerations on this (open to anyone)? Thanks, Jim

2 3

Heads up about future standards annexing new identifiers
by Marshall Clow 16 Jul '18

16 Jul '18

As facilities get added to the standard library, they need names. When this happens, it restricts what other developers can and cannot do in their programs. For example, one cannot do this and expect most anything in the standard library to work: #define allocate(T) . new T [ But you knew about that ]. What you may not have known about was that the standards committee has voted contract enforcement/checking facilities into the draft standard (which will probably be C++20). This includes attributes named: 'expects' and 'ensures'. (all lower case) I have checked all of boost, and with the possible exception of Boost.Contract, we do not have any macros with those names. But you might want to keep your eyes open. -- Marshall P.S. These are not keywords - so it's perfectly fine to have member functions with these names. Variables, too. Probably even free functions - though I haven't tried that. But macros are right out ;-) P.P.S. For people who are interested, the paper is http://wg21.link/P0542

2 1

Searching for a intrusive::avltree_algorithms/bstree_algorithms dev
by Thimothée Virgoulay 16 Jul '18

16 Jul '18

Hello I'm working with the boost library (intrusive: : avltree_algorithms/bstree_algorithms ) and I found some bugs (not sure, maybe it's me who haven't understood the algorithm logic ). I have made hotfixes , add functions and it's working. But I don't know if this change were fine or needed. So I 'm searching for someone who can exchange with me about this library. Thimothée Virgoulay, second's year master degree in Bioinformatics at Montpellier University, France.

4 3

[rational] 1.68.0 merge permission
by James E. King III 16 Jul '18

16 Jul '18

The addition of pow() for rational was identified as a problem in beta testing. It has been reverted for now, and an issue added to the github repository to revisit it after 1.68.0. There are two commits I would like to merge to master: 1. The revert of pow which caused problems 2. A CI-only change to .travis.yml that fixes codecov.io and coverity scan build jobs in light of recent security upgrades in travis CI. No production code changes, zero affect on the release build workflow. This will allow the build status to remain green for the repository as it is merged to master. Thanks, Jim

2 1

[CMT] 1.68.0 merge permission
by James E. King III 16 Jul '18

16 Jul '18

The CI builds in CMT repositories are failing since Travis CI updated their backend last week. The fix is simple and isolated in the .travis.yml file in each repository which is not part of the boost release build cycle. All repositories develop branch have been updated with the fix. I'd like to merge the fix in the CMT repositories into master to eliminate the remaining failed build indicators. Given the change is only in .travis.yml I hope that will be acceptable. Without the fix any cron job on master in these repositories will fail in codecov.io and covscan. Thanks, Jim

2 1

Invalid arguments
by randy 16 Jul '18

16 Jul '18

I’ve the following function but i get invalid arguments error , can please someone tell me where’s this invalid param or argument : int Asio_Send_Pck(int iSendt,int Lsocket,char * Lbuff,int Len) { session* iTempSession = reinterpret_cast<session*>(Lsocket); boost::asio::async_write(iTempSession->socket() , boost::asio::buffer(Lbuff, Len), boost::bind(&session::handle_write, iTempSession, boost::asio::placeholders::error, iTempSession->data, iSendt )); return 0; };

2 1

Boost.Signals deprecation and removal
by James E. King, III 15 Jul '18

15 Jul '18

This first implementation of signals has been deprecated since the 1.54 release. (When) are we going to remove it? It is under management of the CMT, and it has been deprecated for over 10 major releases. I would like to suggest we remove it in the Boost 1.68.0 release. - Jim

9 9

Reporting accurate code coverage
by James E. King III 13 Jul '18

13 Jul '18

General Code Coverage Tip (for everyone): Measure your coverage using branches, not lines. Branch coverage tells you how well error cases are tested and therefore a far better measure of quality than line-based. I have seen over 90% line coverage but the branch coverage is more like 60% to 70% on average over the boost CMT-managed repositories. Here's the shell script I use on travis to generate the data and feed it to codecov.io: https://github.com/jeking3/boost-ci/blob/master/ci/travis/codecov.sh The trick is to feed --rc lcov_branch_coverage=1 to the correct commands to generate and preserve the branch information. Thanks, - Jim

3 3

[boost-ci] code coverage build fix
by James E. King III 13 Jul '18

13 Jul '18

For folks using boost-ci integration, here is a fix to your .travis.yml file to enable code coverage builds to work again. I received confirmation from Travis CI that a back-end update on their end corrected some security issues which necessitates this change. https://github.com/boostorg/rational/pull/22/files In the script section of the coverage job, replace the commands: cd lcov && sudo make install with: export PATH=/tmp/lcov/bin:$PATH Thanks, Jim

1 0

[boost-ci] codecov.io integration
by James E. King III 13 Jul '18

13 Jul '18

Folks using boost-ci integration may see the code coverage builds failing. Part of the .travis.yml script includes logic to update lcov to a newer version for branch coverage to work. These build jobs are breaking due to enforcement of the sudo restriction. I am working on a way to fix this and will send out an update when I have a solution. - Jim

1 0

[Range] 1.68 master/beta MSVC compilation error with /fpermissive-
by Vinnie Falco 13 Jul '18

13 Jul '18

When compiling using /fpermissive- (enable strict conformance) I get the following compilation error: boost\boost\range\concepts.hpp(257): error C3646: 'n': unknown override specifier boost\boost\range\concepts.hpp(264): note: see reference to class template instantiation 'boost::range_detail::RandomAccessIteratorConcept<Iterator>' being compiled boost\libs\beast\include\boost\beast\core\string.hpp(87): note: see reference to class template instantiation 'boost::basic_string_view<char,std::char_traits<char>>' being compiled boost\boost\range\concepts.hpp(257): error C4430: missing type specifier - int assumed. Note: C++ does not support default-int boost\boost\range\concepts.hpp(257): error C3646: 'n': unknown override specifier boost\boost\range\concepts.hpp(264): note: see reference to class template instantiation 'boost::range_detail::RandomAccessIteratorConcept<Iterator>' being compiled I'm using the tip of master, digest 000bd8507f8a17ada2c1fef19158e16cdeabd631 Thanks

6 14

[Boost.Serialization] Help with PR fixing a memory leak in the current version
by Alexander Grund 12 Jul '18

12 Jul '18

Hi, TLDR: I would like https://github.com/boostorg/serialization/pull/105 to be merged for the next Boost release to fix a memory leak and need someone to review this or help getting it merge-ready. Details: Some time ago I made an analysis of a crash introduced in Boost 1.65.0 in the serialization part. The details are quite complicated as the cause is static initialization/destruction order in shared libraries which may even depend on compilers (could not confirm) or standard libraries (dito, observed in all tested versions). It boils down to a violation of the expected destruction order in the case of shared libraries. I observed how 2 global instances of the same type but from different shared libs are destroyed together although the 2nd should not yet be which causes a use-after-free from another global instance. The Singletons involved did have methods to detect, whether they are active or destroyed but they were kinda optimized away in Boost 1.65 leading to a crash which makes the whole library unusable. Not understanding the root cause of the crash lead to changing the singleton to use dynamic allocation, but not freeing it which leads to the memory leak that is currently in Boost. The current state of the develop-branch changed this back to the crash situation we had before making it unsuitable for the next release. Another pair of eyes would be great to check the PR and get this finally fixed. Thanks, Alex

6 33

[rational, spirit] 1.68.0 beta 2 bug report
by James E. King III 12 Jul '18

12 Jul '18

A bug reported by someone using a version of 1.68.0 (not specified): https://github.com/qtumproject/cpp-eth-qtum/issues/31#issuecomment-404257256 Seems to have something to do with the new pow template that was added to rational. This came from an old pull request that was simplified, however it seems pow is now a template defined in the namespace boost, and seems to have caused some issues. I'd guess this is only an issue inside boost namespace (so all boost headers) and not external namespaces. Not sure at the moment if the fix should be in rational or spirit. We could instead back out the change to rational for 1.68.0 where pow() was added into namespace boost. https://github.com/boostorg/rational/pull/21 - Jim

2 1

boost/context/all.hpp missing in 1.68.0beta1
by Roger Leigh 11 Jul '18

11 Jul '18

Is this intentional, or an unintentional regression? I was using this header in the CMake component checks to verify the presence of Boost.Context, and this is now breaking. I can update it to use a different header which has been present since 1.61, but this seems like a compatibility break, hence the question. Thanks, Roger

2 1

Fwd: [Range] 1.68 master/beta MSVC compilation error with /fpermissive-
by degski 11 Jul '18

11 Jul '18

It did go off-list. ---------- Forwarded message ---------- From: degski <degski(a)gmail.com> Date: 11 July 2018 at 19:16 Subject: Re: [boost] [Range] 1.68 master/beta MSVC compilation error with /fpermissive- To: Mateusz Loskot <mateusz(a)loskot.net> On 11 July 2018 at 18:56, Mateusz Loskot <mateusz(a)loskot.net> wrote: > (was your response intentionally direct and off-list?) > No, not intentionally, but it depends on how you send the email (because I pushed reply), like on this one (it would have gone off-list again). I had to add the mailing list back in. This doesn't happen with most posts. I'm not sure what you mean, but I meant the driver that Vinnie mentioned > I'm not totally sure what Vinnie mentions. But I'm talking about this one for stable <https://releases.llvm.org/6.0.1/LLVM-6.0.1-win64.exe> and this one for (almost) trunk <http://prereleases.llvm.org/win-snapshots/LLVM-7.0.0-r336178-win64.exe>. I have the impression Vinnie is talking about Clang/C2 <https://blogs.msdn.microsoft.com/vcblog/2016/04/06/clangc2-we-need-your-adv…>, but like I said it's not clear to me, that compiler, however, is discontinued. degski --- *"If something cannot go on forever, it will stop" - Herbert Stein*

1 0

[Histogram] Status of the review schedule
by Mateusz Loskot 11 Jul '18

11 Jul '18

Hi, Following some discussion in January this year, Hans Dembinski's Histogram library was scheduled for review in March. AFAICT, no review happened. What is the current status of the proposal? Best regards, -- Mateusz Loskot, http://mateusz.loskot.net

2 4

[geometry][units]
by Pranam Lashkari 11 Jul '18

11 Jul '18

I am trying to use boost.geometry with the integration of boost.units. What I want to do is store a coordinate point (e.g: Cartesian or spherical) which should be unit safe.(e.g a cartesian point with length unit meter should give error with cross product is done with cartesian point with length unit millimeter). For this I am using boost::geometry::model::point. Problems I am facing: 1. If I use something like boost::geometry::model::point<UnitClass, 3 ,boost::geometry::cs::cartesian> then all the operations like cross product and dot product will fail as units can not be multiplied and if we multiply then dimensions and units will change. 2. suppose I want to use spherical 3d coordinate then boost::geometry::model::point is cable accommodate only one type of class for all the underlying dimension(eg: distance, longitude and latitude). But in real case distance should be in length and longitude and latitude should be in angle unit. Is there any way to integrate boost.units with boost.geometry for the purpose I mentioned above. Thank you, Pranam Lashkari

1 0

Boost.Graph depth_first_visit algorithm
by Brook Milligan 09 Jul '18

09 Jul '18

The documentation on the Boost.Graph depth_first_visit algorithm indicates that the first argument is a non-const reference to an Incidence Graph. See https://www.boost.org/doc/libs/1_67_0/libs/graph/doc/depth_first_visit.html. However, in contrast the code in boost/graph/depth_first_search.hpp only implements the function for a _const_ reference to an Incidence Graph. I can see that changing the graph structure mid-traversal is a problem, but I have use cases in which graph properties (not graph structure) must be changed along the traversal. Thus, it seems that a valid use case is for this algorithm to work as documented, i.e., on a non-const graph. The attached patch implements overloads for the non-const versions of depth_first_{visit|search}, including overloads for the dfs_visitor member functions. Is there some reason not to extend the implemented interface in this way to better agree with the documentation? Cheers, Brook

1 0

Clang 600 and BOOST_SYSTEM_CONSTEXPR causes error_code.hpp compile failure.
by Paul A. Bristow 09 Jul '18

09 Jul '18

Having failed to build Boost libraries using Clang 6.0.0 using b2/bjam on Windows, I am now trying instead to use Codeblocks on Windows 10 to build static libraries. It comes to grief with these messages (the berbose gory details can be posted if anyone can help). In file included from I:\modular-boost\boost/system/system_error.hpp:14: I:\modular-boost\boost/system/error_code.hpp:409:88: error: variable does not have a constant initializer template<class T> BOOST_SYSTEM_REQUIRE_CONST_INIT system_error_category cat_holder<T>::system_category_instance; ^~~~~~~~~~~~~~~~~~~~~~~~ I:\modular-boost\boost/system/error_code.hpp:409:19: note: required by 'require_constant_initialization' attribute here template<class T> BOOST_SYSTEM_REQUIRE_CONST_INIT system_error_category cat_holder<T>::system_category_instance; ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ I:\modular-boost\boost/system/error_code.hpp:392:44: note: expanded from macro 'BOOST_SYSTEM_REQUIRE_CONST_INIT' # define BOOST_SYSTEM_REQUIRE_CONST_INIT [[clang::require_constant_initialization]] ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This seems to be the ultimate cause I:\modular-boost\boost/system/error_code.hpp:226:41: note: non-constexpr constructor 'error_category' cannot be used in a constant expression BOOST_SYSTEM_CONSTEXPR explicit std_category( boost::system::error_category const * pc ): pc_( pc ) Is this because BOOST_SYSTEM_CONSTEXPR should not be defined? I note this in error_code.hpp #if defined(__GNUC__) && (__GNUC__ == 7 || __GNUC__ == 8) && __cplusplus >= 201700L // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83835 # undef BOOST_SYSTEM_HAS_CONSTEXPR #endif should this undef list also include clang, and if so which versions 600? Or is there something about Boost.config and Codeblocks on Windows? Adding unconditional # undef BOOST_SYSTEM_HAS_CONSTEXPR seems to work. Suggestions welcome. Paul --- Paul A. Bristow Prizet Farmhouse Kendal UK LA8 8AB +44 (0) 1539 561830

7 13

toolset=clang-win uses clang-linux-win?
by Peter Dimov 09 Jul '18

09 Jul '18

When I try `b2 toolset=clang-win` on Appveyor, I get `clang-linux-win` instead, that is, `-win` is interpreted as version for some reason, not as platform: https://ci.appveyor.com/project/pdimov/system/build/job/sjpdq5yrir2ewuhh Is this normal, or is it a bug in the clang toolset? Is there a magic command line incantation that would let me select clang-win?

3 7

[Boost Build] clang-win, <compatibility> property?
by degski 08 Jul '18

08 Jul '18

error: Please set <compatibility> property for visual studio version! How and to what value (VS2017 - 15.7.4, Compiler Version 19.14.26431 ) do I set the <compatibility> property using a project-config.jam file? The documentation <https://boostorg.github.io/build/manual/develop/index.html#bbv2.reference.t…> does not list clang at all, which probably means it works like gcc, there is obviously no <compatibility> property there. Help would be appreciated, then I can post some build results of 1.68-rc2. degski -- *"If something cannot go on forever, it will stop" - Herbert Stein*

1 0

python::numpy not built on windows binary installers
by Nicholas Devenish 07 Jul '18

07 Jul '18

Hi All, I just downloaded the windows boost binaries from https://dl.bintray.com/boostorg/release/1.67.0/binaries/, linked from the boost downloads page. They are missing the boost_numpy library. This (silently) doesn't build if the python installation that boost finds is missing numpy. This is listed as "Third Party" so apologies if this is the wrong place to report it - but it's not clear from the page who makes them. I also couldn't find any mention of this, so don't know if it's a deliberate exclusion. Nick

2 3