April 2013 - Boost - lists.boost.org

[operators] The future of Boost.Operators
by Daniel Frey 26 Apr '13

26 Apr '13

Hello, for a long time there hasn't been any significant update of Boost.Operators and I had an easy job being its maintainer for the past 10+ years. I guess that will change soon. :) I would like to discuss the future of Boost.Operators, as several issues and options have piled up in the past. Here are some points to take into account: 1) Support for rvalue references to generate fewer temporaries. (where available) 2) Support for noexcept. (where available) 3) Support for constexpr if this is applicable at all. I haven't found the time to properly research it yet. 4) Allow dedicated commutative and non-commutative versions for several operators, leading to fewer temporaries for commutative versions in several cases. 5) Remove work-arounds for ancient compilers, general cleanup. I worked on a clean rewrite (called df.operators) which supports 1), 2) and 4) (and obviously 5)) and published it a few weeks ago on GitHub, see <http://github.org/d-frey/operators/>. It is meant to show one option, but it is not a suitable replacement for Boost.Operators in its current form. If it turns out to be useful as a test-bed for the next version of Boost.Operators, I will be glad to improve it into this direction but the future of Boost.Operators does not depend on it. That said, I'd like to invite anyone interested to have a look. If applicable to the future of Boost.Operators, please provide feedback on this mailing list. If feedback only concerns df.operators but not Boost, please don't spam the list and write to my private email, thanks! Some issues which need to be discussed/solved for Boost.Operators' future: 1.1) Which implementation should be used for maximum performance/compatibility? 1.1.1) Pass-By-Value as suggested by Dave Abrahams' article. (<http://cpp-next.com/archive/2009/08/want-speed-pass-by-value/>) 1.1.2) Overloads for Pass-By-Const-LValue-Reference and Pass-By-RValue-Reference. (as in df.operators) See below for why I decided to create 1.1.2) as an alternative to 1.1.1). 4.1) Decide for a default. 4.1.1) Commutative by default (as the current version of Boost.Operators assumes), non-commutative if explicitly requested by the user. 4.1.2) Non-commutative per default, commutative versions need to be explicitly requested by the user. (as in df.operators) 5.1) The main issue is probably the base-class-chaining. 5.1.1) Keep base-class-chaining, makes VC++ happy. (apply EBO and keep the object size small) 5.1.2) Drop base-class-chaining, stop punishing the rest of the world for VC++'s inability to apply the EBO for multiple inheritance. (I might be a bit biased here…) 5.1.3) Any other option I fail to see? (If a MS-employee would like to provide some information on if/when the EBO-problem will be fixed (here or in private email), I'd be grateful. If it is already fixed, I apologize, but the last information I got about VS2012 suggests EBO is not applied for multiple inheritance.) I think this should be enough to get some discussion going… Best regards, Daniel Why 1.1.2) as an alternative to 1.1.1)? Consider the following test program: #include <iostream> #include <utility> struct A { A() { std::cout << "A::A()" << std::endl; } A( const A& ) { std::cout << "A::A(const A&)" << std::endl; } A( A&& ) { std::cout << "A::A(A&&)" << std::endl; } ~A() { std::cout << "A::~A()" << std::endl; } A& operator+=( const A& ) { std::cout << "+=" << std::endl; return *this; } }; // #define BY_VALUE #ifdef BY_VALUE A operator+( A lhs, const A& rhs ) { lhs += rhs; return lhs; } #else A operator+( const A& lhs, const A& rhs ) { A nrv( lhs ); nrv += rhs; return nrv; } A&& operator+( A&& lhs, const A& rhs ) { lhs += rhs; return std::move( lhs ); } #endif int main() { A a, b, c, d; A r = a + b + c + d; } The above code, compiled with G++ or Clang prints: A::A() A::A() A::A() A::A() A::A(const A&) += += += A::A(A&&) A::~A() A::~A() A::~A() A::~A() A::~A() A::~A() If you define BY_VALUE, it prints: A::A() A::A() A::A() A::A() A::A(const A&) += A::A(A&&) += A::A(A&&) += A::A(A&&) A::~A() A::~A() A::~A() A::~A() A::~A() A::~A() A::~A() A::~A()

7 30

Re: [boost] [gsoc-2013] Boost.Thread/ThreadPool project
by Niall Douglas 26 Apr '13

26 Apr '13

> > I'm not ruling out the merit of a Boost.Thread threadpool. For me personally > > though, the hurdle is very significantly raised: you need to strongly prove > > to me why your proposed thread pool is much superior to a Boost.ASIO based > > threadpool. Otherwise I will zero score the GSoC proposal, because in the > > end we need other proposed projects more. > > Superior in what way? Performance-wise, interface-wise? What advanced > features would you want from a thread pool to make it superior to the > presented implementation? That's for the applicant to decide. Any proposal needs to be very significantly better than something based on existing features to pass muster not just with Boost, but with Google themselves. There is always an assumption against duplication of existing functionality just for the sake of it. Niall

1 0

[type_erasure] error in type-erased iterators
by Adam Wulkiewicz 26 Apr '13

26 Apr '13

Hi, iterator< ::boost::no_traversal_tag, ...> defines constructible<T()> requirement. This however musn't be meet in the case of incrementable and single-pass iterators. It should probably be moved to the iterator< ::boost::forward_traversal_tag, ...> Patch attached. Regards, Adam Wulkiewicz

1 0

Fwd: [gsoc 2013] boost.multiprecision
by Sagar Massand 26 Apr '13

26 Apr '13

Hi, I wanted to submit a proposal but I am not sure right now if I am going in the right direction.It would be great if you could guide me here. I was thinking of implementing karatsuba multiplication in the initial stages, complete with a testing of the implementation, and then move on to Schonhage-Strassen and other fast-fourier transform algorithms in the last one and half months seeing that our primary requirement is for numbers uptil 1000 digits. I was also thinking of the implementation of some other division algorithms like Goldschmidt, Newton-Raphson etc. A complete link to my proposal can be found below: https://docs.google.com/file/d/0B5iapFjtoO7GTVR2eGhJd2dXZHM/edit It would be great if you could give suggestions to improve this.Thanks. Regards, Sagar Massand

1 0

[gsoc-2013] Boost.Thread/ThreadPool project
by Dan Lincan 26 Apr '13

26 Apr '13

Hello, My name is Dan Lincan. I'm a 4th year undergratuate student in Computer Science at the "Politehnica" University of Bucharest and I'm interested in the project "Boost.Thread/ThreadPool". I have studied the documents provided as resouces and I would really appreciate if you helped me better understand this project. Are there any tasks that are meant to get solved for the proposal? How do I start? I am aware that it will not be an easy task at all, especially because I don't have much experience with boost, but I'm willing to learn and work as much as it is needed to get it done. Thank you for your time, Dan Lincan

3 15

[Runtime Parameters Reference]Command Line Argument
by Hemanth Choudary M V (RBEI/ETA1) 26 Apr '13

26 Apr '13

Hi, I would like to run selected test cases from multiple tests suits. Pl let me know the command line argument to be passed. Mit freundlichen Grüßen / Best Regards, Hemanth Choudary MV RBEI/ETA1

1 0

Re: [boost] Live read-only GIT mirrors of Boost trunk SVN
by Niall Douglas 25 Apr '13

25 Apr '13

> >The live GIT mirrors of Boost trunk SVN have become a bit stale in recent > >months [...] > > I tried making a mercurial mirror of the Boost trunk SVN and found > that it took so long to do the conversion process, that I could never > get it to complete on my home computer with cable modem internet > access, but I could get it to complete on a shell account with faster > trunk access. I simply allocated an OpenVZ container on my cloud with 64Mb of RAM and 128Mb of swap. > I also found the resulting repository was *huge* because I had *every* > revision of boost ever recorded in the svn repository. Github reckons the entire mirror consumes 700.40Mb. No idea if that's good or bad. > I'm curious if you had different experiences when making a git > repository. To be honest, the process was very straightforward. I configured it as per subgit's instructions. I set it running. It OOMed a few times, but hardly unsurprising with 64Mb of RAM for the VM. I simply restarted when it did. And after a day or two it completed (note that the cloud hardware is a 2Ghz Core 2 without SSD). Better hardware would be far faster. > I am more familiar with mercurial than git, and I wanted to make a > local mercurial repository so that I could work on my changes locally > and checkpoint them using the version control system instead of manual > backups. AFAIK, there's no way to record local changes with > versioning using subversion, there's only committing to the remote > server. Ah, getting subversion commits to match git commits that's subgit's problem :). Subgit gives you three sync options, "safe", "somewhat safe" and "not safe". "safe" locks both subversion and git repos during the sync, however long that takes, and during which no pushes nor pulls nor checkouts may happen. So long as you can live with that (I can), the live two way mirroring works well, but it could be frustrating for users of busy repos. Niall

2 1

Re: [boost] [operators] The future of Boost.Operators
by Andrew Ho 25 Apr '13

25 Apr '13

> Maybe we're talking past each other. I have 3 overload set variants in > mind: > > (A) > T operator?(T, T); > > (B) > T operator?(T const &, T const &); > ... > T operator?(T&&, T&&); > > (C) > T operator?(T const &, T const &); > ... > T&& operator?(T&&, T&&); > > AFAICT, neither (A) nor (B) have any safety issues. Oh, but now I see the > problem with (C): T&&'s are implicitly convertible to T&'s...that's what I > was missing before. Sorry for being dense. > > So if you do > > T& a = b + c; > This succeeds in VS2012 with all (no compile or runtime problems, oddly VS2012 extends the lifetime of ref returned by C to be valid), and fails with all in gcc 4.7.2 (compiler error) :P > invalid initialization of non-const reference of type 'T&' from type 'T&&' Yep, this is what happens in gcc for me. > > I swear I just tried a T& y = static_cast<T&&>(x) on gcc and finding it > compiled fine. This gave a compiler error for me. > Now you can ask the same question with T const& instead... Compile succeeds, with only two operands the called overload returns a T, not T&& so there's no problems. If you had: const T& r = a + b + c; (A) and (B) are perfectly safe, but (C) will have runtime issues. Same goes for: T&& a = b + c; T&& r = a + b + c; First line doesn't have any problems with (A), (B), or (C) because of which overload is used. The second line (A) and (B) have no problems but (C) will have runtime issues. Of course, (C) will have a few extra temporaries created (which may or may not be optimized away). The question originally posed is do we: 1. Ignore this problem with (C) to get better performance and just document the problem? 2. Implement (B) to take advantage of move semantics with no internal problems, albeit with a few temporaries being created (if they aren't optimized away by the compiler)? 3. Implement a user-accessible macro switch to let the user decide globally As Jeff said, (3) is ugly especially when linking different codes together, but does provide the most options to the user. (2) isn't necessarily the most efficient but is completely safe. It may also end up being just as efficient as (1) if compilers support the optimization to remove unnecessary temporaries. Daniel's argument is that the problems which occur with (C) are typically a result of user error, and (C) in general guarantees the least amount of temporaries being used. Are there any cases where the problems described above with (C) would be a result of intended behavior with no acceptable work-around? Hopefully we're on the same page now.

1 0

Google Summer of Code Proposal Feedback
by Niall Douglas 25 Apr '13

25 Apr '13

> aware of GSOC until recently. I had strongly considered writing a proposal > for Boos.ASIO, but I fear my lack of experience with Linux, POSIX, and > concurrency in general might prove to be too great of an obstacle. If the > subject matter can be learned relatively quickly, I would love to discuss > what such a project would require with the mentor, Niall Douglas, or a > community member who has is familiar with the project. I'm hardly "the" mentor in Boost.ASIO. I'm probably just the only one to have registered as a potential mentor so far. It's still early days. A lot more potential mentors will register in time, then depending on project scoring and people's availability and of course Google's preferences, we'll see what turns out. > My original proposal probably isn't unique, and much to my chagrin as I > read the hints for successful proposals it seems a popular topic. I am of > course suggesting a project dealing with Trees. Rather than suggest > building a generic tree, I am suggesting the start of a tree library, or > the addition of several types of trees to existing libraries, as I am not > sure which is more appropriate. An awful lot of students expose their ignorance of what Boost, or indeed the C++ standard library, provides during GSoC. It really doesn't help them to be scored well which is why Boost strongly recommends that students discuss their ideas here first before submission. 95% of the time their idea is already implemented. For example, there are multiple tree implementations in the C++ STL and Boost. Look into Boost.Intrusive for example. > Because trees are such a fundamental data type in computer science, I have > found it difficult to understand why there are no explicit tree classes in > the c++ standard. They generally don't have "tree" in their name ;) > tested in an afternoon, while implementing a Red Black Tree would take > considerably longer, yet the C++ standard has linked lists and not Red > Black Trees. Red-black trees are called std::map<>, std::set<> et al. in C++ since 1998. Of course std::map<> et al. doesn't *have* to be implemented as red-black, but most implementations do since the SGI STL defined std::map<> using stl_tree<>. > The library I am proposing should eventually be able to satisfy almost > anyone requiring the use of a tree. It should have many different types of > trees, Binary trees, AVL Trees, Red Black Trees, Treeps, B-Trees, > Skip-Lists, etc. It should support common methods of traversal, inorder, > preorder, postorder, and user defined traversal methods. Eventually, the > library should support an n-ary trees and their operations. Trees in the > library should also be performant, fail gracefully, be well documented, > unit and integration tested, as exception safe as possible, and, someday, Look into Boost.Intrusive. They're the most appropriate category for what you propose. Niall

1 0

Discussion Policy
by Rob Stewart 25 Apr '13

25 Apr '13

If you're relatively new to this list, you may not have taken the time to read Boost's discussion policies[1]. Even if you've been on the list for a while, reviewing the policies may be educational. Thanks for your efforts to make discussions easier and more consistent for all. ___ Rob (Sent from my portable computation engine) [1] http://www.boost.org/community/policy.html

1 0