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

[type_erasure] BOOST_TYPE_ERASURE_MEMBER with template?
by Nat Goodspeed 15 Aug '15

15 Aug '15

The code with which I'm working has two similar template classes: they have a common subset of 8 methods that could be used on an instance of either. For performance reasons, these methods are not virtual. As far as the compiler knows, the two template classes are unrelated. Naturally we can write a template function to accept an instance of either template class, as long as it uses only methods from the common subset. I also want to support runtime polymorphism, permitting a non-template function to accept an instance of either template class. Obviously that will incur some performance hit, but the coder can decide whether or not his/her function is performance-critical. If so, s/he can write that function to accept one or the other specific template class -- or write a template function. We want to incur the performance hit only when the coder deems that an acceptable tradeoff. The classic solution would be to derive both template classes from a common base class and make the common methods virtual -- but then that penalizes every consumer. I haven't yet used Boost.TypeErasure, but it seems to me potentially applicable to this problem. I see the example[0] illustrating use of BOOST_TYPE_ERASURE_MEMBER(). I was going to define requirements for each of the 8 common methods ... but then I became confused. Our common methods use a template parameter from the parent template class as a method parameter type. Can I invoke BOOST_TYPE_ERASURE_MEMBER in a generic way that would support all valid template parameters? Two of the 8 common methods are template methods, accepting a parameter whose type is a template parameter for the method rather than the containing class. I don't know whether that introduces further complexity. Thank you for any help you can provide! [0] http://www.boost.org/doc/libs/1_59_0/doc/html/boost_typeerasure/basic.html

2 1

Making the Windows binaries 'official'
by Tom Kent 15 Aug '15

15 Aug '15

<history> After the boostpro.com organization dissolved, there was never an official place to go to get pre-built binaries for the releases. Some random components were made available on the sourceforge download page and I had been hosting them on my own website (http://boost.teeks99.com) since I started building them around the release of 1.37. A couple years ago, before the release of 1.54, I asked the community if they would like my builds to be hosted on the sourceforge site, and the answer was positive: http://lists.boost.org/Archives/boost/2013/04/202460.php Since the 1.54 release, I've generally been uploading the windows binaries based on the RCs and enabling the downloads just a few minutes after the release notice goes out. </history> I would now like to propose that we make these libraries an official part of each Boost release. I don't think this entails much, just the following: 1) Have the release team coordinate with me to ensure that the binaries are ready before announcing the release. 2) Include a link to them in the release announcement. 3) Link to them from the website. It would also be good to get others who can run the scripts to perform the build, if I am someday unable to do this. The biggest hurdle here is having a (virtual) machine with all six versions of visual studio on it. Is this something the community would be in favor of? I know that many of the people on these lists don't use windows and if they do are generally capable of building the windows binaries themselves, however there is a large group out there that prefers to just get the installer for them (like they do for other projects like Qt, wxWidgets, etc). Thoughts? Tom Kent

1 0

[Graph] Descriptor invalidation on remove_vertex() with adjacency_list VL=vecS
by Chris Dellin 14 Aug '15

14 Aug '15

Hello, The documentation for the adjacency_list [0] is clear that if the vertex list container is vecS, then calling remove_vertex() will invalidate any vertex/edge descriptors into the graph. However, I was wondering if this is actually the case if the removed vertex was the one most recently added, that is, remove_vertex(vertex(num_vertices(g)-1)). In this case, I would think that no shifting of vertex data or adjustment of edge source/target indices would be necessary, and existing descriptors are guaranteed to be still valid. Is this safe to do with the current implementation? If not, why not? Thanks, - Chris Dellin [0] http://www.boost.org/doc/libs/1_59_0/libs/graph/doc/adjacency_list.html

2 1

[shared_ptr] boost::shared_ptr deleter question
by Tamas.Ruszkai＠leica-geosystems.com 14 Aug '15

14 Aug '15

Hi All, I wonder when the shared_ptr deleter is destroyed. I have a shared_ptr with a custom deleter. I would expect when the last shared pointer to the object is gone then not only the deleter method will be called, but the deleter function object is also released. As I see the deleter is taken by value and stored by value in the reference counter, so I do not understand what is going on. How is the deleter mean to be freed? Thanks, Tamas struct MyClassDeleter { ~MyClassDeleter() { // Break point is set here } void operator()(MyClass* item) { // Break point is set here } } { auto sharedPtr = boost::shared_ptr<MyClass>(new MyClass, MyClassDeleter) } // shared_ptr leaves the scope here, the custom deleter is called (parenthesis operator) // the deleter instance is however not destroyed, the breakpoint in the destructor does not hit

2 2

Log: Overwrite format of sink
by Florian Lindner 14 Aug '15

14 Aug '15

Hello, I have a console sink: boost::log::add_console_log(std::cout, keywords::format = fmtStream); with fmtStream being an expressions stream. After setting up the sinks I load a configuration file: std::ifstream file("log.settings"); init_from_stream(file); Is there a way to overwrite the format of this sink from within in the configuration file. Or, alternatively disable this sink. What I want to achieve is that a user have the option to modify the format without having two console outputs... Thanks, Florian

1 0

Query regarding interest in my submitting plf::colony to Boost
by Soul Studios 14 Aug '15

14 Aug '15

Hi all, I've had some expressions of support outside of the Boost community for submission of the plf::colony container and it's associated container plf::stack, mainly from the review judges for my cppcon submission on the same topic (topic didn't make it through to the finals but got a positive response). I'd like to guage reactions here to see whether it is worth spending the time necessary to port it to Boost, with all the associated reviews etc. plf::colony is a container for unordered data, where erasure and addition do not cause iterator/pointer invalidation, with superior performance characteristics to all std:: containers when either (a) Pointers and iterators which point to container elements must stay valid regardless of container additions and deletions and/or (b) Additions to or deletions from the container will be occurring in performance-critical code. Erasure performance in particular can be in advance of 6000x of vectors, while still being faster than lists, maps, deques etc for add and iteration. Full details and benchmarks across four compilers are here: http://www.plflib.org/colony.htm#benchmarks It has been thoroughly tested, including integrating it into the game engine I constructed last year as well as the usual unit tests. It supports both C++0x and C++11, and has been tested on the following compilers so far: Clang 3.4, GCC 4.6, 4.7, 4.8, 4.9 and 5.1 (both 32-bit and 64-bit), MS Visual C++ 2010 & 2013 (2015 supported but untested as yet). Internally a colony is a combination of a linked-list of increasingly-larger memory blocks combined with boolean erasure fields, combined with a custom memory location recycle stack. The most common comment I get is "that sounds kinda like a deque", so to save some time here are the differences (from the FAQ): "A double-ended queue requires a very different internal framework. It may or may not use a linked-list of memory blocks, dependent on the compiler, and has no comparable performance characteristics. Deque element erasure is very slow, similar to a vector, and it does not free unused memory blocks to the OS once they are empty, unlike a colony. In addition a deque invalidates references to subsequent container elements when erasing elements, which a colony does not. A colony never invalidates pointers/iterators to elements unless the element being referred to is erased." The rest of the faq is here and I would appreciate it if people read it before casting judgement: http://www.plflib.org/colony.htm I am aware that it will be a lot of work to port colony to Boost, so I will only undertake this if there is genuine interest. In addition, plf::stack - which colony must use internally - is a std::stack equivalent container but with much better performance characteristics, which can be used outside of plf::colony. Asides from the performance characteristics, colonies have the following positive and negative properties: Positive: Lack of pointer/iterator invalidation makes interrelating between collections of data easier, without the cache misses associated with vectors of pointers to dynamically-allocated objects. Negative: Due to the additional overhead, performance can be worse for scalar types. Thanks in advance- Matt B

5 13

[units] Conversions from quantities
by Michael Powell 14 Aug '15

14 Aug '15

Hello, Just wanted to clarify something. These should convert from given quantities to the desired units? double unit_conversion::from_length_to_inches(quantity<length> const & L) { const quantity<inch_base_unit::unit_type> result_(L); return result_.value(); } double unit_conversion::from_mass_to_pounds(quantity<mass> const & M) { const quantity<ounce_base_unit::unit_type> result_(M); return result_.value(); } >From an earlier post, I think maybe so, but I was interested in expert opinions, as well. https://groups.google.com/forum/#!topic/boost-list/YDJoMcu1Xx4 Thank you... Best regards, Michael Powell

4 4

Boost 1.59.0 has been released
by Marshall Clow 13 Aug '15

13 Aug '15

Release 1.59.0 of the Boost C++ Libraries is now available. These open-source libraries work well with the C++ Standard Library, and are usable across a broad spectrum of applications. The Boost license encourages both commercial and non-commercial use. This release contains two new libraries and numerous enhancements and bug fixes for existing libraries. New Libraries * Convert: An extendible and configurable type-conversion framework, from Vladimir Batov. * Coroutine2: from Oliver Kowalke For details, including download links, see http://www.boost.org/users/news/version_1.59.0 You can also download directly from SourceForge: http://sourceforge.net/projects/boost/files/boost/1.59.0/ To install this release on your system, see http://www.boost.org/doc/libs/release/more/getting_started/index.html Thanks, --The Boost release team Vladimir Prus Rene Rivera Marshall Clow Eric Niebler Daniel James Beman Dawes

3 2

[Interprocess][Container] nested containers and scoped allocators
by Malko 13 Aug '15

13 Aug '15

Hi, First of all my setup: I am compiling on redhat 6 with clang 3.4 in C++03, using boost_1_58. In a project of mine I am using a map<int, vector<int>> (types to be stored in shared memory and I am using the proper allocators as per interprocess documentation). If mymap is an instance of such a type, then mymap[0] is an invalid syntax because of issues that, to the best of my understanding, are related to the fact that the map tries to use the default constructor (with the default allocator) for the nested type as operator[] also accounts for the case in which the entry is missing and must be constructed. Using at() instead of operator[] makes the whole thing compile. On construction, I give the usual instance of allocator<void, segment_manager> to mymap. According to what I read in Boost.Container docs: http://www.boost.org/doc/libs/1_58_0/doc/html/container/Cpp11_conformance.h… It would seem to me that a scoped allocator is a mechanism by which the allocator I gave to mymap is passed down to the nested vector. In which case I would expect operator[] to try to use such allocator instead of the default one. But then again, I always do mymap.insert(std::pair<const int, vector<int,...>>(0, vector<...>(my_void_alloc))) to add an element to the map, since I need to specify the allocator (and the examples in interprocess use this syntax). Given all of this, I would be grateful for any insight into the following: 1. If the scoped allocator is meant to be passed from the map to the vector, why can I not use operator[] and I necessarily need to use insert() to create a new vector in the map? Have I misunderstood what a scoped allocator is? Thanks in advance. -- View this message in context: http://boost.2283326.n4.nabble.com/Interprocess-Container-nested-containers… Sent from the Boost - Users mailing list archive at Nabble.com.

1 1

1.59.0 rc3 is now available
by Marshall Clow 13 Aug '15

13 Aug '15

New Release candidate files for 1.59.0 are available at: http://boost.cowic.de/rc/ RC3 should fix the problem with VC14 and MSPDB140.DLL. 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. This helps ensure the candidates build OK before we push them out to SourceForge. The files (and associated md5s) are: MD5 (boost_1_59_0_rc3.7z) = 0a2e512844f3e30a6240f8139ee983f3 MD5 (boost_1_59_0_rc3.tar.bz2) = 6aa9a5c6a4ca1016edd0ed1178e3cb87 MD5 (boost_1_59_0_rc3.tar.gz) = 51528a0e3b33d9e10aaa311d9eb451e3 MD5 (boost_1_59_0_rc3.zip) = 08d29a2d85db3ebc8c6fdfa3a1f2b83c Thanks! -- The release managers P.S. Release notes are here: http://www.boost.org/users/history/version_1_59_0.html

5 4