December 2019 - Boost - lists.boost.org

GSoc 2020
by Aniswar SK 19 Dec '19

19 Dec '19

Hello everyone, I am Aniswar Krishnan, a 2nd year Computer Science student at Indian Institute of Technology, Madras. I am have been coding in c++ and python for the past 15 months. I am interested in Mathematics and Algorithms and have done courses on same along with Data Structures, Linear Algebra. I got interested int the projects like Boost.Real and Boost GIL and would like to contribute more to the community. Please guide me and share the relevant resources needed. Will be grateful. Regards, Aniswar Krishnan II Year Computer Science IIT Madras Chennai, Tamil Nadu

3 2

Re: [boost] Proposed legacy macro renaming policy
by Glen Fernandes 18 Dec '19

18 Dec '19

I've fixed the macros in: * boostorg/multi_array * boostorg/iterator * boostorg/parameter * boostorg/polygon Glen

2 1

What to do about Boost libraries which use user-defined macros which are not BOOST_ prefixed
by Niall Douglas 18 Dec '19

18 Dec '19

Acting on guidance given to my recent supply to this list of unprefixed macros, namespaces etc injected into the global namespace by Boost, I have logged a list of all issues with each library, and linked to all those from a single issue logged at https://github.com/boostorg/boost/issues/352. A problem has arisen which requires input from boost-dev. Some Boost libraries define as part of their public API a non-BOOST_-prefixed macro which changes the definition of their header files e.g. PHOENIX_LIMIT (just to be clear, Phoenix is not the only library doing this) If we change such macros to be like BOOST_PHOENIX_LIMIT instead, all existing code using Boost which sets PHOENIX_LIMIT would potentially *silently* break. What we could decide to do instead is: #if defined(PHOENIX_LIMIT) && !defined(BOOST_PHOENIX_LIMIT) #error In Boost 1.xx PHOENIX_LIMIT was renamed to BOOST_PHOENIX_LIMIT. Please upgrade your code to reflect the new name. #endif Or one could just leave PHOENIX_LIMIT named as is. What would Boost developers prefer we do? Niall

8 19

Re: [boost] [review] [STLInterfaces] STLInterfaces review starts today
by Hadriel Kaplan 18 Dec '19

18 Dec '19

Howdy, For this review I've really only tried out the iterator_interface part. I'm sorry I can't provide feedback on the container_interface nor view_interface, as I just don't have the free cycles. > - Your name Hadriel Kaplan > - Your knowledge of the problem domain I've been using boost's iterator_facade and adaptor for a long time; I've written my own custom iterators as well, of both traditional and proxy-types. The usual stuff. > - Whether you believe the library should be accepted into Boost (be clear about this) Definitely. It could use word-smithing and minor cleanup stuff, but presumably that happens later. (?) > - What is your evaluation of the library's > * Design? Good overall. > * Implementation? Good. I've got some nitpicks, but honestly none are a big deal. minor nitpicks: 1) I didn't use the proxy_iterator_interface one, only looked at the code. The proxy_arrow_result storing a local T value that it only returns a pointer to, means it's not optimal in some cases. Obviously for something like std::vector<bool> it won't matter, but some value-types are not so lightweight. On the other hand, it looks like we can specialize this for our type (or not even use proxy_arrow_result to begin with)? 2) Can the stuff in fwd.hpp be split out? I could be wrong, but I don't think most of it is needed/used by iterator_interface, and only used by the other two, correct? Reducing what gets included would be nice, for people who don't need the container/view stuff. The old boost iterator_facade/adaptor headers included so much that it drove some people to stop using them; for example see the comments here: https://github.com/facebook/folly/blob/master/folly/detail/Iterators.h 3) This lib hasn't been tested for C++20, and obviously can't be yet, but it makes assumptions that it can safely replace some of its internals with C++20's new ones. Things like concepts and ranges. I'm not sure that's advisable, until all the major compilers support them and can be tested. It would be a shame if people using this lib can't upgrade to C++20 because it breaks this lib, until they upgrade boost as well. (not a big deal, just more noting it) 4) The files use a macro to control doxygen output. Is this still required for boost libs? Doxygen has had the cond/endcond commands for a long, long time. It's kinda ugly to see these macros, and kinda silly that everyone's compiler will be evaluating them all the time. (I know that's super-nitpicky) 5) I assume all the _has_include(<stl2/ranges.hpp>) checks will be removed, since that's a thirdparty header? 6) This lib requires C++14, but I don't see a macro checking that anywhere and providing a simple error to the user. 7) I can't decide if the BOOST_STL_INTERFACES_STATIC_ASSERT_ITERATOR_TRAITS() thing is a good idea or a waste of time. I tried it, it passed. But of course it did, because it's just checking that I got the API that I told iterator_interface to create, no? This just seems like needless repetition to me. I mean if I screwed it up to begin with, I'll screw it up in this macro too, no? But maybe I'm missing something obvious. (could be - it's getting late) > * Documentation? It's OK. I enjoyed some of the humor. :) It's missing some info for things that are customization points. For example base_reference() is shown in an example, but not really documented anywhere as far as I can tell. The tables aren't as useful/explanatory as one would hope. I mean I knew what they meant, but I'm not sure someone newer to this area would. (but then again, that's what blogs/articles are for I guess) I think it might also help if there were some explanation for what happens under-the-hood, in terms of what public methods get auto-generated from what, for which iterator types. For example, it's surprising that for a bidirectional iterator the user defines operator==(), but not for random-access iterator. (that particular case happens to be mentioned in a note, but you get my point) There are also typos and such, but I presume that gets checked later. > * Tests? Good. I copied them over and ran them as well. (but not the v2 tests, nor the ones for container_interface or view_interface) I think it would be better if there were some tests for iterator_interface with non-copyable value-types. Testing it only for `int`s is cheating. :) Also, random question: for the container_interface, did you try using it with the boost::fixed_string (or static_string, or whatever it's called)? I mean that _is_ a contiguous container, and it's in boost review too, so might it be a good test candidate? :) > * Usefulness? Very useful. iterator_facade/adaptor were showing their age. It's time for something new and shiny. I was happy that it reduced the number of functions we needed to implement in our current usages from 5 to 3. > - Did you attempt to use the library? Yup. I replaced three existing boost::iterator_facade usages in my company's code-base with this one, to try it out. Ran our unit tests with and without valgrind as well. But to be honest these weren't very complicated iterators. I did _not_ try replacing boost::iterator_adaptor, but I hope to try that eventually. > * Which compiler(s)? (old ones, so I could build my employer's codebase and run all their unit tests...) 1) gcc 7.3.1 with c++17 2) clang 5.0.0 with C++17, using gcc's libstdc++ Both using boost 1.55 (sorry!), with the stl_interfaces headers copied over manually, without using the CMake files in the github repo. Neither compiler has the early Concepts support. > * What was the experience? Any problems? Not really. The hardest part was figuring out the new model of what the derived had to define, compared to iterator_facade. (i.e., I was so used to the old model that it took me a bit to grok this one) > - How much effort did you put into your evaluation of the review? Only a few hours, maybe 4-5 total. Sorry, I wish I had more free time. :( > - Do you like the name container_interface? sequence_container_interface > is more precise, but seems a bit long. I'm not sure the length matters much for this use-case, as it's a one-time thing (generally), and not used by users of the actual container; and can anyway be aliased. (I mean, the "stl_interfaces" sub-namespace is annoying too, but it's not a big deal in my opinion) > - Would you use something like container_interface, but for associative > containers? If so, should it assume a node-based interface, a la std::map > and std::set? This assumption would preclude alternative associative > container-like types, such as flat_map. Yes I think so, depending on what it did... but for *unordered* ones (i.e., hash-maps/sets), rather than rb-trees. I often find myself encapsulating them, and then having to write a lot of boilerplate public member functions. And several functions can be boiled down for hash-maps. For example count() and contains() can just use find(). I don't know why an unordered_container_interface wouldn't be do-able, including heterogenous lookups, for various hash-map/set flavors. (except, of course, that boost::unordered_map/set's API is different than std's for heterogeneous lookup; which is unfortunate) -hadriel

1 0

[review] [STLInterfaces] STLInterfaces iterator_interface mini mini review
by Arthur Gruzauskas 18 Dec '19

18 Dec '19

On Wednesday, 11 December 2019 11:21:23 AM AEDT Barrett Adair via Boost wrote: > Dear Boost, > > The formal review of Zach Laine's STLInterfaces library begins now, and > will run through December 19. Please participate in this review if you can. > To submit a review, please reply to this email with the following > information: > - Your name Arthur Gruzauskas > - Your knowledge of the problem domain I've written a number of iterator interfaces over the years. Nothing fancy, but have dipped my toes in iterator implementation. > - Whether you believe the library should be accepted into Boost (be clear > about this) ACCEPT, simply because I am using it happily. but I have only used container_interface, not the rest. > > In addition, you are strongly encouraged to answer the following questions: > - What is your evaluation of the library's > * Design? simple interface, small learning curve, great value for effort. A small number of methods to implement the essential concepts. A small enough concept chunk to be easily manageable. simple to use, just copied the code in the readme.md and adapted it. > * Implementation? worked for me. I didn't examine the code much. I like its layout and it looks clean to me. > * Documentation? The readme.md with that simple example was enough for me to adapt the intro sample iterator to my code. I note the example has been criticised by Gavin Lambert a few posts ago. But I love it as it made everything simple and clear. Reading the main doc page filled in a few conceptual gaps, but that readme.md gave me the overview that made the rest easy. > * Tests? So far brief tests have all worked well. I did not exhaustively test the iterators. They also worked fine in a couple STL algorithms. Again, I have only dealt with iterator_interface.hpp > * Usefulness? My experimental class would have stalled for some time if this easy iterator interface had not been there. Encouraged, I'm now looking at using it for more complicated structures. > - Did you attempt to use the library? If so: > * Which compiler(s)? clang-8 for c++17 on debian linux. > * What was the experience? Any problems? very straightforward. a few small glitches, as I just grabbed iterator_interface.hpp and fwd.hpp and made small changes for it to work inside my project. https://github.com/tzlaine/stl_interfaces/issues describes two minor issues, which could be well related to me just grabbing those 2 files only, outside the boost structure that would normally sustain it. > - How much effort did you put into your evaluation of the review? Several hours reading, adapting and getting it working for me. > > STLInterfaces is a C++14 library targeting ISO standardization. The > following templates are provided, all C++20-friendly: > > 1. iterator_interface - a modern version of the iterator_facade and > iterator_adaptor parts of Boost.Iterator > 2. view_interface - a pre-C++20 implementation of C++20's eponymous feature > 3. container_interface - a tool to eliminate boilerplate when writing new > containers > > We would appreciate answers to these library-specific questions: > - Do you like the name container_interface? sequence_container_interface > is more precise, but seems a bit long. It doesn't clash with anything in my mental namespace, so I'm happy with the name container_interface. BTW, I have not used container_interface, I just saw it now and had a look. I can see this simplifying some upcoming containers, and will certainly be giving it a go. view_interface is not something I have a need for currently. > - Would you use something like container_interface, but for associative > containers? > If so, should it assume a node-based interface, a la std::map > and std::set? This assumption would preclude alternative associative > container-like types, such as flat_map. > > Code: https://github.com/tzlaine/stl_interfaces > Docs: https://tzlaine.github.io/stl_interfaces/doc/html/index.html > > Thanks, > Barrett > My one small concern is in the required method: constexpr auto operator-(repeated_chars_iterator other) const noexcept instead of: constexpr auto operator-(const repeated_chars_iterator& other) const noexcept I may soon have a need for an iterator that holds a lot of state, and copying that iterator for each subtraction makes me wonder about a performance penalty in inner loops. It is not a dealbreaker for this review, and I didn't examine Zach's tradeoffs for doing it this way. I see many of my use cases where this approach will be fine. This is my first review, incomplete as it is, mainly motivated because no one else has done one yet, and because I found iterator_interface useful and want to encourage Zach for this lovely simplifying interface. Thanks very much, Zach. Arthur

3 4

Unmerged change in boostcpp.jam
by Peter Dimov 17 Dec '19

17 Dec '19

boostcpp.jam on develop contains the following patch, which is not merged to master: diff --git a/boostcpp.jam b/boostcpp.jam index efc41c1ed..299401778 100644 --- a/boostcpp.jam +++ b/boostcpp.jam @@ -607,7 +607,7 @@ rule address-model ( ) return <conditional>@boostcpp.deduce-address-model ; } -local deducable-architectures = arm mips1 power sparc x86 combined ; +local deducable-architectures = arm mips1 power riscv s390x sparc x86 combined ; feature.feature deduced-architecture : $(deducable-architectures) : propagated optional composite hidden ; for a in $(deducable-architectures) { @@ -618,11 +618,13 @@ rule deduce-architecture ( properties * ) { local result ; local filtered = [ toolset-properties $(properties) ] ; - local names = arm mips1 power sparc x86 combined ; + local names = arm mips1 power riscv s390x sparc x86 combined ; local idx = [ configure.find-builds "default architecture" : $(filtered) : /boost/architecture//arm : /boost/architecture//mips1 : /boost/architecture//power + : /boost/architecture//riscv + : /boost/architecture//s390x : /boost/architecture//sparc : /boost/architecture//x86 : /boost/architecture//combined ] ; This seems to add support for "riscv" and "s390x" as architectures. Should this be merged to master?

1 0

Test Matrix not Updating
by Michael Caisse 16 Dec '19

16 Dec '19

We are aware that the test matrix has not been updating. The problem is caused by a server that we do not control. Things should be resolved in the next couple days. Thank you for your patience and sorry about the disruption. michael -- Michael Caisse Ciere Consulting ciere.com

1 0

Enabling warnings for library sources and tests but not for dependencies
by Alexander Grund 15 Dec '19

15 Dec '19

Hi, I want to build at least the tests but better also the library sources itself with warnings enabled and warnings-as-errors to avoid regressions. >From BoostAssert I found the following snipped added to the top of test/Jamfile.v2: project : requirements <warnings>pedantic <warnings-as-errors>on ; Further down I have a dependency on BoostFilesystem like: run test_stdio.cpp : : : <library>/boost/filesystem//boost_filesystem ; But now the build of BoostFilesystem fails: gcc.compile.c++ bin.v2/libs/filesystem/build/gcc-5.4.0/release/cxxstd-11-iso/threading-multi/visibility-hidden/directory.o libs/filesystem/src/directory.cpp:217:48: error: unused parameter 'buffer' [-Werror=unused-parameter] Is there any way to restrict the warnings to my project only? Or at least the warnings-as-errors option? Is there a way to set warnings-as-errors via b2 command only for this library? This would avoid putting it in the Jamfile which could affect end-users, but having it on CI is actually enough. Alex

2 3

bug-fix release 1.72.1?
by Oliver Kowalke 15 Dec '19

15 Dec '19

Hello, unfortunately I accepted a faulty pull request and the commit that reverted it was not included into release 1.72 (for instance spawn() of Boost.Asio will fail to compile). Oliver

3 4

Assuming maintainership of boostorg/io
by Glen Fernandes 15 Dec '19

15 Dec '19

I'd like to take on maintainership of IO, as suggested by Peter. So far, I've already: * Implemented the Fundamentals TS v2 ostream_joiner as boost::io::ostream_joiner * Converted the documentation to Asciidoc * Updated the test cases to use Lightweight Test and build faster * Updated the CI configurations I also plan to: * Finish the boost::io::quoted implementation started by Beman but currently in detail. i.e. Update and improve the implementation. Add better documentation and more tests, and make it non-detail (public). * Migrate my boost::ostream_string facility from Boost.Utility (which is more general purpose) to Boost.IO (because, well, IO). * Make some minor improvements to the implementation of boost::io::ios_state. Glen

2 1