April 2012 - Boost - lists.boost.org

[config] 4 new C++11 macros added
by John Maddock 14 May '12

14 May '12

A long time back I was asked for a bunch of new config macros which have finally now been added: BOOST_NO_CXX11_ALLOCATOR The standard library does not provide a C++11 version of `std::allocator` in <memory>. BOOST_NO_CXX11_ATOMIC_SP The standard library <memory> does not support atomic smart pointer operations. BOOST_NO_CXX11_HDR_FUNCTIONAL The standard library does not provide a C++11 compatible version of <functional>. BOOST_NO_CXX11_SMART_PTR The standard library header <memory> has no shared_ptr and unique_ptr. Enjoy, John.

6 11

[complex] Feedback and Potential Review Manager
by Christopher Kormanyos 13 May '12

13 May '12

Hi Matthieu, I would like to provide a detailed feedback on the proposed Boost.Complex. I am terribly sorry about the late response on this, as I was not doing well in winter when this topic was more active. First of all, great work with this, Matthieu. I believe the community desperately requires this facility. Boost.Complex appears to have two main goals: A) Provide acceleration for pure imaginary numbers. B) Allow complex to be used with user-defined types. For goal B), I have briefly tested parts of the Boost.Complex code with the cpp_dec_float multiprecision data type proposed for Boost.Multiprecision by John Maddock. My Tests include: * Test some elementary transcendental functions. * Test the overall package via computation of orthogonal polynomials.* My test file (complex.cpp) is included in this e-mail. * Visual Studio 2010, SP1. * Proposed Boost.Multiprecision and boost trunk. Remarks: 1) Various functions use macros like M_PI_2. These macros do not provide enough precision for a multiprecision type. And they will be inefficient for some fixed-point types. Perhaps you need to "bite the bullet" and use boost::math::constants. (Is M_PI_2 even part of C++, or is it only lurking around in <math.h> of GCC?) 2) Ditto for the LN_10 and LN_2 constants. Does boost::math::constants support ln_ten()? I do know that it has ln_two(). 3) I disagree with explicit complex(const T& re = T(0.), const T& im = T(0.)) I believe this constructor should be non-explicit. See ISO/IEC 14882:2011, Section 26.4.2. 4) I would really like to see global operators for seamless interaction of complex<T> with integer PODs. The proposed code of boost::complex forces explicit construction from integer "all the way" to boost::complex<T> just to, for example, multiply with 2. This is extremely costly as most multiprecision or fixed-point types have specially optimized routines for multiplication with integer PODs. Perhaps something like this would be good template<typename T> complex<T> operator*(const complex<T>& z, int n) { T re(z.real()); T im(z.imag()); re *= n; im *= n; return complex<T>(re, im); } 5) In the exp() function, polar() can not be resolved without explicit template parameter. Instead of this: template<typename T> inline complex<T> exp(const complex<T>& x) { return polar(exp(x.real()), x.imag()); } I needed this: template<typename T> inline complex<T> exp(const complex<T>& x) { return polar<T>(exp(x.real()), x.imag()); } I don't know who's right here, your code or the compiler. 6) The tanh() function and some other functions compute the exponent twice. By that I mean both exp(x) as well as exp(-x). I believe that division would be faster if exp(-x) were computed with (1 / exp(x)). Or is there an issue with branch cuts that mandates the potentially redundant calculation? 7) I hope to find time to test this code further with a tiny fixed-point class. But I might be too busy for that extra work. REVIEW PROGRESS: I may potentially be qualified for review manager. I do know the math and the code. I am also well-versed with all aspects of software review, assessment and audit situations. BUTTT!!! I'm not as good at C++ as some of you folks and I'm a real beginner at boost. Nonetheless, If nobody better steps forward, I could potentially manage the review of Boost.Complex. Best regards, Chris.

8 33

[hash] Turn on BOOST_HASH_NO_IMPLICIT_CASTS by default
by Daniel James 12 May '12

12 May '12

Hello everyone, There have been occasional issues where Boost.Hash will incorrectly hash an object that doesn't have a hash function but is convertible to bool. I added an extra templated overload of hash_value which contains a static assert. The idea is that it's a better match than the implicit cast, so it will cause a compile error in this case. This is currently only activated when a macro is defined, as it could break otherwise valid code. I was a bit worried about making any interface changes because at this stage, Boost.Hash should be very stable from release to release. The problem turned up again yesterday, so I'm starting to think it's enough of a problem that it's worth activating this be default. A different solution might be to change the implement of Boost.Hash for bools, so that the bool overload isn't required. But that would be treating the sympton, as other implicit casts could cause problems. An example of valid code that will now break is something like: struct base; std::size_t hash_value(base const&); struct derived : base { void some_method_but_no_extra_state(); }; boost::hash<derived> derived_hash; Currently that will automatically use base's hash_value, with this change it will be a compile error. Although that might be a good thing, as sometime it is an error to use the base's hash function. Of course, there will be a macro to keep the old behaviour. Any objections to this change?

3 5

Re: [boost] [Review Request] Multiprecision Arithmetic Library
by Christopher Kormanyos 11 May '12

11 May '12

Sorry, John my file was too big and admin rejected post. The candidate file is now 7-zip and should make it through. <snip> >> I experimented with FFTs. We need your inputs if we *really* >> want to make very high-precision available at this stage. <snip> >> maybe about 30 lines of code. > Nice, had no idea it could be implemented in such a small space! Ooops, my bad. It was 100 lines. I guess I was in a daze by then. <snip> > So my gut feeling is to hold off for now, and maybe do things > properly later (dynamic allocation, better std lib support etc)? I agree. We should pull the reigns on it for now. We won't get the reliability we need if we try to do something fast. I could go on indefinitely with big-numbs, but we have to get that first version potentially reviewed. I should really get back to my book project soon. But I'm definitely in on any relevant review issues. > I think the existing ones work out quite nicely now thanks, > I could use some ideas for integer and maybe rational number examples, > but that's a whole other ball game ;-) I'm not an integer guy. Maybe someone could help us out here. But what springs to my mind is: * Prime factorization * Greatest common denominator (GCD) * Primality testing for Mersenne Primes (has an exceptionally simple algo) * MD5, SHA-1, 2, etc. (but cool std::uint32_t reference apps already exist)<snip> >> I resolved the excessive guard digit issue and reduced the guard digits >> to the necessary amount (and provide rationale in a code comment). <snip> > Sure, or mail me a patch and I'll run the full tests > (or you can - just cd into libs/multiprecision/test and > do a "bjam --enable-specfun" and then wait for a long time!) > Thanks, John. OK. I attached my proposed version in /big_number/multiprecision/cpp_dec_float.hpp. The diffs include the removal of some non-needed constants, the trimming of max_digits10 and a minor improvement to the multiplication. Thanks, Chris.

4 29

[reference_wrapper] add operator()(Args...) to reference_wrapper
by Philipp Moeller 10 May '12

10 May '12

Given that Boost now has perfect forwarding emulation through Boost.Move and a better result_of, would it be possible to enable reference_wrapper to forward the function call operator? The current behavior is somewhat surprising and makes a few of the most promising applications of reference_wrapper (reference wrap polymorphic boost::function objects) impossible with code that is no prepared to work with reference_wrapper. It would also be more consistent with reference_wrapper of the stdlib. I can file a corresponding bug report and add a patch, if there is interest. Cheers, Philipp Moeller

2 5

[coroutine] request for review manager
by Oliver Kowalke 10 May '12

10 May '12

Hi, I want to ask who wants to be the review manager for boost.coroutine? The documentation is updated and available at: http://ok73.ok.funpic.de/boost/libs/coroutine/doc/html/index.html best regards, Oliver

2 4

Iterator Range and operator==
by Olaf van der Spek 08 May '12

08 May '12

Hi, What do you expect this code to do? Is b true or false? And why? Is this expected behaviour? #include <boost/range/iterator_range.hpp> #include <string> int main() { std::string s = "Olaf"; boost::iterator_range<std::string::iterator> r(s); bool a = r == s; bool b = r == "Olaf"; assert(a); assert(b); return 0; } -- Olaf

9 17

Latest version and status of lockfree
by Julien Nitard 06 May '12

06 May '12

Hi All, Where can I find the latest version of boost.lockfree ? There's nothing in the sandbox or the vault (it seems). The latest commit on Tim's public git repo was done more than 6 month ago. Now that atomic has been accepted, are there any roadblocks to have boost.lockfree released ? Many thanks, Julien

4 9

[contract] syntax redesign
by Lorenzo Caminiti 04 May '12

04 May '12

Hello all, After a long time of inactivity for Boost.Contract, I am considering how to re-design its macro syntax to make it simpler. I can eliminate a large number of extra parenthesis based on a few pp tricks I've learned in processing Boost.Local's `const bind(type)& var`. Plus, I can provide (optional) support for variadics macros so to use comma-separated pp tuplets instead of parenthesized pp sequences on preprocessors with variadics. For example, I _think_ (but I've not implemented this yet) that I can simplify the syntax like this: #include <contract.hpp> #include <vector> #include "myvector/pushable.hpp" // Wrapper class that adds contracts to std::vector. CONTRACT_CLASS( template( typename T ) class (myvector) extends( public pushable<T> ) // Subcontracting. ) { CONTRACT_CLASS_INVARIANT( empty() == (size() == 0) // More invariants here (comma separated)... ) CONTRACT_FUNCTION( public void (push_back)( (const T&) element ) precondition( size() < max_size() // More preconditions here (comma separated)... ) postcondition( auto old_size = CONTRACT_OLDOF(size()), // Old value. size() == old_size + 1 // More postconditions here (comma separated)... ) ) { vector_.push_back(element); // Implementation. } // Rest of the class here (with more contracts if needed)... public: typedef typename std::vector<T>::size_type size_type; size_type size(void) const { return vector_.size(); } size_type max_size(void) const { return vector_.max_size(); } bool empty(void) const { return vector_.empty(); } const T& back(void) const { return vector_.back(); } private: std::vector<T> vector_; }; For a side-by-side comparison with N1962 syntax: http://svn.boost.org/svn/boost/sandbox/contract/libs/contract/doc/html/inde… For a comparison of the entire std::vector (plus Boost.ConceptCheck concepts): http://svn.boost.org/svn/boost/sandbox/contract/libs/contract/doc/html/cont… For a comparison for all Boost.Contract features (plus Boost.ConceptCheck concepts and Boost.Parameter named parameters): http://svn.boost.org/svn/boost/sandbox/contract/libs/contract/doc/html/cont… (Don't pay attention to the rest of the docs that are way out of date.) A few notes: * Extra parenthesis are still needed around class and function names. * Extra parenthesis are still needed around types (like for the parameter `(T) element`) but only if the type is not a C++ keyword (no parenthesis needed aournd int, double, void, etc) or it contains a symbol (parenthesis needed if & or * are used `(int&) x`, etc). * In postconditions, the `auto old_size = CONTRACT_OLDOF(size())` only copies the vector size (and not the entire vector :)) ). Similarly, `auto result = return` can be used to access the return value for non-void functions. (The `auto ...` is parsed by the pp and it's implemented using Boost.Typeof so no C++0x is required.) * Template declaration (but not instantiation) angular parenthesis <> are replaced by round parenthesis (). All of this _should_ be implementable in pure ISO C++. What do you think? Based on the discontent that the previous "parenthesized" syntax generated among Boosters, the key question is: Would you consider this new syntax usable?? Thanks, --Lorenzo

5 45

[Countertree + Suballocator] New Version
by Francisco José Tapia 03 May '12

03 May '12

Hi this message is to announce the new version of the [countertree + suballocator] library. You can find the zip file with the code and the documentation in https://docs.google.com/file/d/0B9lWJ1Nhb_RhRXVHSS14aGxUZjJjWFdRVC0xZ2czZw/… or if you want a quick look , you can see in my web page http://fjtapia.webs.com/works/countertree/index.html *COUNTERTREE * This library is an implementation of a binary red-black counter tree. This tree have an additional counter in each leaf. This permit the access to the elements by the position, like in a vector. It is a random access container with random access iterators . With this tree we have vectors (cntree::vector_tree) with identical interface than std::vector. The vector_tree have the same speed inserting and deleting in any position (all the operations are O(log N)).It is slower than std:vector inserting and deleting at end, but much faster for to insert and delete in any other position. With ordered information, we have in the cntree namespace the classes set, multiset, map and multimap, with identical interface than the STL classes, with the plus of access to the elements by position, like in a vector. The iterators are random access , and you can subtract two iterators in a O(log N) time for to know the number of nodes between them (even with the end( ) and rend( ) iterators) For to show the utility of this, imagine a large unordered file with the information about the employees of a big company. You have indexes implemented with multimaps, with pairs [value, position], and use the upper_bound and lower_bound functions. You need to select the employees in a range of age and weight. With the boost:multimap you can know the number of elements of the selection. In the weight query we found 100 elements and in the age query 10000 elements. To travel the elements of the weight query asking for the age, is faster than the opposite. *SUBALLOCATOR * A very hard test for the allocators are the data structures which need a very high number (several millions) of elements of the same size, like the STL list, set, multiset, map or multimap. When the number of elements grow, many allocators begin to have speed problems. The solution proposed for this problem are the custom allocators. These allocators are very fast, but many of them present additional problems related with the memory consumption. And they don't work well when they must manage many types of data. When the allocators receive the request from the program, they request memory to the Operating System, and the memory consumed by the program grows up. When the program return to the allocators the request, many allocators have problems for to return the memory freed to the Operating System (The std::allocator of GCC 4.6 don't return well the memory, and the std::allocator of Visual C++, return, but it is slower than the GCC allocator) If you use only one allocator, you have a bad management or in the speed of equal size elements, or with a wide variety of elements. If you use two allocators, the memory freed by an allocator can't be used by the other, and the memory consumption is very high, and when decrease the number of elements to manage, the memory don't decrease. These problems are specially important in two environments : a) When you have a limited resources, as occurs in the mobile phones, pads ... b) When the programs must run continuously in a multitask environment The solution is the suballocator. It is a mechanism for to manage in a fast way a greater number (hundred millions) of elements with the same size. The suballocator is a layer over the allocator (any allocator with the STL interface). The suballocator receives an allocator as template parameter. When the suballocator needs memory, request memory from the allocator, and when the memory is not used, is returned to the allocator. The suballocator present the same interface than the STL allocator. Form the view point of the data structures, the suballocator is other allocator The suballocator always return the first element free. This improve the cache performance and the speed of the data structures. This simple memory schema permit to the allocators return memory to the Operating System and decrease the memory used by the program, as demonstrate the programs in the benchmarks, and you can see in the benchmark point. With the suballocator a) We have a very *fast allocation* (3 times faster than GCC 4.6 std::allocator, 3.5 times faster than Visual Studio 10 std::allocator ) b) Return memory to the allocator, for to be used by others types of data. Many allocators, return this memory to the Operating System, and *decrease the memory used by the program * c) You *can use with any allocator* if according with the STL definition. The suballocator provides speed and memory management to any allocator. d) In the insertion in a std::set<uint64_t> , the time spent by the allocator is around 1%, but the suballocator save around 35% of time . The suballocator always provide the first position free, This provide us a very compact data are, which *improve the cache performance* All the information showed here is explained and detailed in the point Suballocator. All the information is extracted from the programs in the benchmarks folder. *COUNTERTREE + SUBALLOCATOR* The join of the two ideas provide us data structures with a suballocator built-in. They are, in the namespace cntree, the vector_tree_pool, set_pool, multiset_pool,map_pool and multimap_pool, with identical interface than the STL classes but better performance. Tray it. It fast, useful and easy to understand and use,. They are the STL classes with a few additional functions. I had checked this code with GCC 4.6 (32 and 64) , CLANG/LLVM 3.0 and Visual Studio 1010 (32 and 64). If you find any error or problem in the code , the documentation or in any other part ( as friendly do Mr Rene Rivera in the first version). Please mail me ( fjtapia(a)gmail.com) or if you mail Boost, please insert [countertree] in the head of the message. Some days I have not time to read all messages arrived to my mail, but if I see that I detect and respond first. If you have some idea, or have any comment, please mail me, always two legs run more than only one in the pursuit of the goal, to be useful to the community of programmers in the same way than our predecessors had been useful to us. Sincerely yours Francisco Tapia fjtapia(a)gmail.com

7 19

[release] scope_exit, local_function, utility/identity_type, and functional/overloaded_function
by lcaminiti 02 May '12

02 May '12

Hello release managers, I've been working in the trunk on a few libraries which are almost ready to be released: a) When is the release date? b) What do I have to do to merge into release? (This will be my 1st release merge :) ). 1) ScopeExit (improved) I'm waiting on MVSC 9 and 10 tests to cycle to make sure they all pass. Other compilers OK. 2) Utility/IdentityType (new) Should be ready to be merged into release. 3) Functional/OverloadedFunction (new) Should be ready to be merged into release. 4) LocalFunction (new) I'm making final fixes for some tests on some of compilers. Hopefully, I'll be done in 2 weeks max. For all of those, I've followed the checklist so I've run the code inspection tool, updated all docs, librariers.htm, and maintainers.txt. Thanks a lot. --Lorenzo -- View this message in context: http://boost.2283326.n4.nabble.com/boost-release-scope-exit-local-function-… Sent from the Boost - Dev mailing list archive at Nabble.com.

4 9

[uuids] On boost::uuids::uuid operator == performance.
by Michael Kochetkov 02 May '12

02 May '12

Hello. I would like to discuss a performance issue with boost::uuids::uuid operator ==. Let us consider the following program: --------------------------------------------------- #include <boost/uuid/uuid.hpp> #include <boost/uuid/uuid_generators.hpp> int main() { boost::uuids::uuid id1((boost::uuids::random_generator()())), id2((boost::uuids::random_generator()())); return id1 == id2; // we will consider this line } --------------------------------------------------- That is compiled with maximum optimizations the following way with Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 16.00.40219.01 for 80x86: cl -EHsc -Ox -GL -Ob2 -Oi -Fa -I..\3dparty\boost_1_48_0 m051.cpp boost uses the following comparison operator: inline bool operator==(uuid const& lhs, uuid const& rhs) /* throw() */ { return std::equal(lhs.begin(), lhs.end(), rhs.begin()); } which end up with the following code: lea edx, DWORD PTR _id2$[esp+92] lea ecx, DWORD PTR _id1$[esp+108] lea eax, DWORD PTR _id1$[esp+92] call ?_Equal@std@@YA_NPBE00@Z ; std::_Equal The whole story looks the following way (the code above plus std::_Equal call expansion): 0117C6AB lea edx,[esp+24h] 0117C6AF lea ecx,[esp+48h] 0117C6B3 lea eax,[esp+38h] 0117C6B7 call 01171000 01171000 push esi 01171001 mov esi,eax 01171003 sub ecx,esi 01171005 push edi 01171006 cmp ecx,4 01171009 jb 01171024 0117100B jmp 01171010 0117100D lea ecx,[ecx] 01171010 mov eax,dword ptr [esi] 01171012 cmp eax,dword ptr [edx] 01171014 jne 01171028 01171016 sub ecx,4 01171019 add edx,4 0117101C add esi,4 0117101F cmp ecx,4 01171022 jae 01171010 01171024 test ecx,ecx 01171026 je 01171073 01171028 movzx eax,byte ptr [esi] 0117102B movzx edi,byte ptr [edx] 0117102E sub eax,edi 01171030 jne 01171063 01171032 cmp ecx,1 01171035 jbe 01171073 01171037 movzx eax,byte ptr [esi+1] 0117103B movzx edi,byte ptr [edx+1] 0117103F sub eax,edi 01171041 jne 01171063 01171043 cmp ecx,2 01171046 jbe 01171073 01171048 movzx eax,byte ptr [esi+2] 0117104C movzx edi,byte ptr [edx+2] 01171050 sub eax,edi 01171052 jne 01171063 01171054 cmp ecx,3 01171057 jbe 01171073 01171059 movzx eax,byte ptr [esi+3] 0117105D movzx ecx,byte ptr [edx+3] 01171061 sub eax,ecx 01171063 sar eax,1Fh 01171066 or eax,1 01171069 xor edx,edx 0117106B test eax,eax 0117106D pop edi 0117106E sete al 01171071 pop esi 01171072 ret 01171073 xor eax,eax 01171075 xor edx,edx 01171077 test eax,eax 01171079 pop edi 0117107A sete al 0117107D pop esi 0117107E ret I do understand that the boost::uuids::uuid implementation is conceptually correct and inoculate the best programming practices. But the following equivalent is too much faster (you may probably want to ensure a proper alignment for some processor architectures or even get even faster 64-bit version): inline bool comp(const boost::uuids::uuid& l, const boost::uuids::uuid& r) { return *reinterpret_cast<const uint32_t*>(l.data) == *reinterpret_cast<const uint32_t*>(r.data) && *reinterpret_cast<const uint32_t*>(l.data+4) == *reinterpret_cast<const uint32_t*>(r.data+4) && *reinterpret_cast<const uint32_t*>(l.data+8) == *reinterpret_cast<const uint32_t*>(r.data+8) && *reinterpret_cast<const uint32_t*>(l.data+12) == *reinterpret_cast<const uint32_t*>(r.data+12); } the assembler output is: mov ecx, DWORD PTR _id1$[esp+92] cmp ecx, DWORD PTR _id2$[esp+92] jne SHORT $LN37@main mov edx, DWORD PTR _id1$[esp+96] cmp edx, DWORD PTR _id2$[esp+96] jne SHORT $LN37@main mov eax, DWORD PTR _id1$[esp+100] cmp eax, DWORD PTR _id2$[esp+100] jne SHORT $LN37@main mov ecx, DWORD PTR _id1$[esp+104] cmp ecx, DWORD PTR _id2$[esp+104] jne SHORT $LN37@main mov eax, 1 jmp SHORT $LN40@main $LN37@main: xor eax, eax $LN40@main: movzx eax, al that the question comes: is boost supposed to be used in commercial applications or it is just a testing area for some concepts? You have probably never thought of the fact that Debug (unoptimized configurations) are sometimes just unusable with such approach on real (big) data. If you are interested you may want to compile the program above this way and see the resulting code: cl -EHsc -Zi -Fa -I..\3dparty\boost_1_48_0 m051.cpp Thank you in advance for your considerations. -- Michael Kochetkov

7 27

[uuid] Unused variable warnings
by Andrey Semashev 02 May '12

02 May '12

Hi, With recent versions of boost I get lots of warnings like the following: /.../boost/uuid/seed_rng.hpp: In member function ‘void boost::uuids::detail::seed_rng::sha1_random_digest_()’: /.../boost/uuid/seed_rng.hpp:157:24: warning: variable ‘not_used’ set but not used [-Wunused-but-set-variable] I'm using GCC 4.6 with -Wall. It's trivial to mask this warning out by attaching __attribute__((unused)) to the variable, which I've done locally. Ok to commit?

2 2

[Smart Ptr] make_shared slower than shared_ptr(new) on VC++9 (and 10) with fix
by Ivan Erceg 02 May '12

02 May '12

Hi all, Before I switch to using boost::make_shared<> I wanted to test its purported performance advantage. I created a simple benchmark for measuing raw allocation throughput for 3 classes of different sizes with a common base class (constructors and destructors trivial). The number of allocations was set to 40,000,000 as it was roughly giving me 10 seconds running time per test. Unfortunatelly it turns out that on VC++9 (release target with default optimizations) boost::make_shared is significantly slower than simply doing boost::shared_ptr(new). Here's the benchmark output: TestBoostMakeShared 10.577s 3.78179e+006 allocs/s TestBoostSharedPtrNew 8.907s 4.49085e+006 allocs/s As you can see boost::make_shared is over 15% slower than boost::shared_ptr(new) idiom. Having available VC++10 compiler as well I then compared these results with std::shared_ptr and std::make_shared implementations that come with that compiler (but not VC++9). Here are the results: TestBoostMakeShared 9.688s 4.12882e+006 allocs/s TestBoostSharedPtrNew 8.252s 4.84731e+006 allocs/s TestStdMakeShared 5.07s 7.88955e+006 allocs/s TestStdSharedPtrNew 8.159s 4.90256e+006 allocs/s While std::shared_ptr(new) performs about the same as boost::shared_ptr(new), std::make_shared really blows away boost::make_shared and both shared_ptr(new) tests, being almost twice as fast as boost::make_shared. I then profiled the boost::make_shared test to see what's the biggest performance bottleneck when compared to boost::shared_ptr(new) profiler run. The culprit was immediately obvious: boost::make_shared test was spending above 25% of its time in "type_info::operator==(class type_info const &) const" function. This function was being called indirectly from boost::make_shared through boost::get_deleter. After digging some more through the implementation I came to the conclusion that, in this particular case, we are guaranteed to always be requesting deleter for the right class (namely T from boost::make_shared<T>). Since boost::shared_ptr doesn't have a way to retrieve the deleter without using RTTI I decided to add one and use it from an alternative boost::make_shared. So I did the following: 1. I added a virtual function to detail::sp_counted_base (detail\sp_counted_base_w32.hpp): virtual void * get_raw_deleter( ) = 0; 2. I implemented get_raw_deleter() function in sp_counted_impl_p (detail\sp_counted_impl.hpp): virtual void * get_raw_deleter( ) { return 0; } 3. I implemented get_raw_deleter() function in sp_counted_impl_pd (detail\sp_counted_impl.hpp): virtual void * get_raw_deleter( ) { return &reinterpret_cast<char&>( del ); } 4. I implemented get_raw_deleter() function in sp_counted_impl_pda (detail\sp_counted_impl.hpp): virtual void * get_raw_deleter( ) { return &reinterpret_cast<char&>( d_ ); } 5. I added the following function to detail::shared_count: void * get_raw_deleter( ) const { return pi_? pi_->get_raw_deleter( ): 0; } 6. I added the following function to shared_ptr<>: void * _internal_get_raw_deleter( ) const { return pn.get_raw_deleter( ); } 7. I made a separate copy of boost::make_shared function and replaced a single line from: boost::detail::sp_ms_deleter< T > * pd = boost::get_deleter< boost::detail::sp_ms_deleter< T > >( pt ); to: boost::detail::sp_ms_deleter< T > * pd = static_cast<boost::detail::sp_ms_deleter< T > *>(pt._internal_get_raw_deleter()); Benchmarking the results afterwards gave me the following results on VC++9: TestBoostSharedPtrNew 9.204s 4.34594e+006 allocs/s TestBoostMakeShared 10.499s 3.80989e+006 allocs/s TestBoostMakeSharedAlt 7.831s 5.1079e+006 allocs/s My changes translated into almost 35% improvement in allocation speed over the current implementation of boost::make_shared. Or to put it differently, they amount to 25+% decrease in running time as we could have supposed from the profiling results. Results on VC++10 are similar: TestBoostSharedPtrNew 8.487s 4.71309e+006 allocs/s TestBoostMakeShared 9.609s 4.16276e+006 allocs/s TestStdSharedPtrNew 8.283s 4.82917e+006 allocs/s TestStdMakeShared 5.039s 7.93808e+006 allocs/s TestBoostMakeSharedAlt 6.802s 5.88062e+006 allocs/s VC++10's std::make_shared is still much faster (almost 35% faster than boost::shared_ptr) and we will be switching to it once we switch to VC++10. But in the meantime it seems to me that boost::make_shared should be fixed to improve the performance. Again, this is only one compiler and other compilers might not have such a severe RTTI performance issue but I still think it would be well worth avoiding unnecessary calls to RTTI during performance-relevant operations such as heap allocations. The testing and changes were done on Boost 1.48.0 but I compared Smart Ptr library sources with Boost 1.49.0 and the above changes should work there equally well. Thanks, Ivan

15 40

Re: [boost] [preprocessor] Sequences vs. All Other Data Structures
by Brian Wood 02 May '12

02 May '12

Paul Mensonides: > As some sort of technological preview? It is not portable enough to be > used within Boost, though perhaps a slew of regression failures will cause > that to change (unlikely). <rant>MS, in particular, is too busy taking it > upon themselves to "ensure C++ is relevant" to prioritize implementing the > language.</rant> Also, one of the principles of Chaos is to have > absolutely zero compiler workarounds, and that would have to stay if it > was ever part of Boost. Given your desire to avoid compiler workarounds, an on line code generator may be helpful. One of the happy results of going on line is kissing compiler workarounds good-bye. I've worked on the distribution of code generated by an on line code generator -- http://webEbenezer.net/build_integration.html . You're welcome to use that as a starting point if you like. Shalom, Brian Wood Ebenezer Enterprises http://webEbenezer.net

4 4

Review request: extended complex number library
by Matthieu Schaller 02 May '12

02 May '12

Dear all, Following the comments from V. Escriba, I formally propose my complex number library for review. The library is an extension of the std::complex class addressing two issues: - The standard does not guaranty the behaviour of the complex class if instantiated with types other than float/double/long double. - Some calculation where pure imaginary numbers (i.e. multiples of sqrt(-1)) appear are unnecessarily slowed down due to the lack of support for these numbers. The code I submit contains two interleaved classes boost::complex and boost::imaginary which can be instantiated with any type T provided T overloads the usual arithmetic operators and some basic (real) mathematical functions depending on which complex function will be used. It is thus an extended version of Thorsten Ottosen's n1869 proposal (http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2005/n1869.html) Performance tests show some non-negligible speed-ups compared to std::complex for calculations where pure imaginary numbers are involved. A speed-up of 25% has been observed when solving the Schroedinger equation explicitly on a regular mesh and some comparable figures can be observed when computing the Mandelbrot set (the two examples snippets provided in the archive). Furthermore, the functions (sin(), exp(), log(),...) involving boost::imaginary numbers are more precise than their equivalent using a std::complex with the real part set to 0. The code and (doxygen) documentation is available in the repository http://code.google.com/p/cpp-imaginary-numbers/ A comprehensive zip archive can be found in the "download" and the code can also be checked-out via the SVN repository. The archive contains the class header, two examples, a comprehensive precision test, a brute-force performance test and the documentation. I'd be happy to answer any question from your side and to provide more detailed information if required. Regards, Matthieu -- Matthieu Schaller PhD student - Durham University

8 23

Array in Unordered Map
by Olaf van der Spek 01 May '12

01 May '12

Hi, What's the right way to get an array into an unordered map? This one works on VC10, but GCC appears to look for the hash_value function in namespace std. What's the right way to do this? Olaf #include <array> #include <boost/unordered_map.hpp> namespace boost { template<class T, size_t N> size_t hash_value(const std::array<T, N>& v) { return boost::hash_range(v.begin(), v.end()); } } int main() { boost::unordered_map<std::array<char, 20>, int> c; c[std::array<char, 20>()] = 0; return 0; } -- Olaf

3 6

Re: [boost] [exception] warning about non-virtual destructor - resolution?
by Hite, Christopher 30 Apr '12

30 Apr '12

This is annoying me too: boost_1_49_0/boost/checked_delete.hpp:34:5: warning: deleting object of polymorphic class type 'boost::error_info<boost::tag_original_exception_type, const std::type_info*>' which has non-virtual destructor might cause undefined behaviour [-Wdelete-non-virtual-dtor] A) easy solution /boost_1_49_0/boost/exception/detail/error_info_impl.hpp It's probably easiest to just make ~error_info_base() virtual. It doesn't cost much since there's already a virtual table. B) harder solution The other cooler C++11 alternative is declare error_info final. I tried it on gcc-4.7.0 and it works wonderfully. Since the compiler knows you can not derive a class from that type it knows a that deconstructor is complete. struct base{ virtual void foo(); }; struct derived final : base { }; //struct illegal : derived {}; // error: cannot derive from 'final' base 'derived' in derived type 'illegal' void kill(derived* d){ delete d; // no complaint } I vote for A, because you'd anyway have to detect support for final. Chris

12 48

Re: [boost] [type_traits] has_equal_to< std::vector<T> > always true_
by Frédéric Bron 30 Apr '12

30 Apr '12

> We could use <boost/detail/container_fwd.hpp> here (rather like Boost.Hash > does, which has a similar issue) and get the best of both worlds? My > concern about making it a separate header, is that it runs the risk of ADL > errors if the #includes are inconsistent across translation units. I assume > that only forward declarations are needed for the containers? This looks a very good idea. It is a pity that the standard does not provide such header as it does for iostreams (iosfwd). However, I see that <boost/detail/container_fwd.hpp> has some troubles with many compilers, in particular with g++. // GNU libstdc++ 3 // Disable forwarding for all recent versions, as the library has a // versioned namespace mode, and I don't know how to detect it. This is a major drawback but maybe we can do something for it. I see also that with the new standard, this header is now not exhaustive; we have to include all of these: * <array> - <deque> * <forward_list> - <list> - <vector> - <map> - <set> * <unordered_map> * <unordered_set> - <queue> - <stack> - <string> - <bitset> *: missing headers in <boost/detail/container_fwd.hpp> Frédéric

2 1

Exception visibility and non-GCC compilers
by Mathias Gaunard 30 Apr '12

30 Apr '12

Exceptions types need to be exported so that shared libraries work correctly. Currently, the only code I see that does this is in Boost.Exception : #if defined(__GNUC__) # if (__GNUC__ == 4 && __GNUC_MINOR__ >= 1) || (__GNUC__ > 4) # pragma GCC visibility push (default) # endif #endif ... some exception type being declared or defined #if defined(__GNUC__) # if (__GNUC__ == 4 && __GNUC_MINOR__ >= 1) || (__GNUC__ > 4) # pragma GCC visibility pop # endif #endif But shouldn't we have macros to make this work with other compilers too? I'm not sure all unix-like compilers pretend to be GCC 4.1+.

2 3

[pool] Definition of a static_pool ?
by DUPUIS Etienne 30 Apr '12

30 Apr '12

Greetings, The Boost.Pool class describes a pool of buffer, pool whose size grows as needed. This is very convenient but there usage cases where the pool should not perform any dynamic memory allocation, i.e. the maximum number of simultaneously required buffers is known at initialization. I have hence defined a simple class 'static_pool' (see below), which (i) allocate all requested buffers at initialization; (ii) fails (returns null) if more buffers are being asked for at runtime. Does such a class has its place within the Boost Pool library ? Regards, Étienne #ifndef BOOST_STATIC_POOL_HPP #define BOOST_STATIC_POOL_HPP #include <boost/pool/pool.hpp> namespace boost { template<typename UserAllocator> class static_pool : protected pool<UserAllocator> { public : // Overridden functions explicit static_pool(const size_type nrequested_size) : pool<UserAllocator>(nrequested_size, nrequested_size) { free(malloc()); } void * malloc() { return store().empty() ? NULL : store().malloc(); } public : // Inherited functions void free(void * const chunk) { return pool<UserAllocator>::free(chunk); } size_type get_requested_size() const { return pool<UserAllocator>::get_requestedSize(); } bool is_from(void * const chunk) const { return pool<UserAllocator>::is_from(chunk); } }; } #endif Ce message et toutes les pièces jointes sont confidentiels et établis à l'intention exclusive de ses destinataires. Toute modification, édition, utilisation ou diffusion non autorisée est interdite. Si vous avez reçu ce message par erreur, merci de nous en avertir immédiatement. ATEME décline toute responsabilité au titre de ce message s'il a été altéré, déformé, falsifié ou encore édité ou diffusé sans autorisation. This message and any attachments are confidential and intended solely for the addressees. Any unauthorized modification, edition, use or dissemination is prohibited. If you have received this message by mistake, please notify us immediately. ATEME decline all responsibility for this message if it has been altered, deformed, falsified or even edited or disseminated without authorization. Note: To protect against computer viruses, e-mail programs may prevent sending or receiving certain types of file attachments.

7 18

[filesystem] [system] Suddenly, dynamic linking by default
by Lars Viklund 30 Apr '12

30 Apr '12

Hello list. I have a habit of tracking trunk to find out if there's any strangeness or mistakes done on the platforms I target. A while ago I encountered some strange linking errors on some of my projects, which in the end turned out to be because for some reason, Boost.Filesystem and Boost.System now attempt to be dynamically linked by default, both symbol-decoration-wise and auto-link filename-wise. I happened to have some older shared libraries from an older trunk version lying around, so they were suddenly and unexpectedly picked up instead of my fresh static libraries. Even without my library hygiene deficiencies, the decision (which I have yet to see a rationale or announcement about) will cause a lot of problems. Why is it that System/Filesystem suddenly default to dynamic linking? Are there problems about linking statically? Why are there then static libraries built at all? Is this just some personal preference of the author that goes contrary to the general approach of Boost? Such choices have _massive_ impacts on consumers of the code, they should not be made lightly and if there is a really concrete need to do so, why is it not documented and thoroughly explained? Linking dynamically requires you to deploy the Boost DLLs to the end user, which prevents any plugin-like deployment, as they tend to require a single self-contained shared library with no dependencies, which implies linking to a static language runtime of choice. Dynamic linking of Boost means that you need to share a CRT between the project and Boost, which implies that the only CRT choice is the dynamic one, which prevents many deployment scenarios, as you have to deploy _that_ as well to the end user, either as a redistributable install or as a private deployment. This also has an impact on code size, as there is no dead-code elimination possible when the code resides in a DLL, you have to ship _all_ of the library even if you use a few convenience functions. For one of my projects (an extension to a popular music player), this would inflate its size many times for each and every download, as the software is deployed as a single ZIP archive through automatic update services. Additionally, the autolink macros to force symbol decoration and linking for a static link are undocumented in the Boost.Config table [1], which explicitly states that all libraries that can be statically linked will be statically linked. There is no mention of libraries that arbitrarily decide to go dynamic, just that some libraries are only available in dynamic form. For future reference for anyone finding this in archives, the undocumented, and in this case required, macros to force a static link and symbol decoration are: BOOST_<WHATEVER>_STATIC_LINK or BOOST_ALL_STATIC_LINK [1] http://www.boost.org/doc/libs/1_49_0/libs/config/doc/html/index.html#boost_… -- Lars Viklund | zao(a)acc.umu.se

5 6

[iostream] tests broken due to dependency on removed filesystem internals
by Jürgen Hunold 30 Apr '12

30 Apr '12

Hi, almost all Boost.Iostreams test fail with: bin.v2/libs/iostreams/test/array_test.test/gcc-4.6/debug/array_test.o In file included from ../array_test.cpp:14:0: ../detail/temp_file.hpp:16:46: fatal error: boost/filesystem/v3/operations.hpp: No such file or directory compilation terminated. The v3/ was removed recently as v2 was finally dropped. Patch to fix this attached. This gets a lot of the tests running. Okay to commit? There are some remaining failures. Yours, Jürgen -- * Dipl.-Math. Jürgen Hunold ! * voice: ++49 4257 300 ! Fährstraße 1 * fax : ++49 4257 300 ! 31609 Balge/Sebbenhausen * jhunold(a)gmx.eu !

3 4

[multiprecision] Looking for a review manager
by John Maddock 29 Apr '12

29 Apr '12

Just a quick line to remind/hassle you all that I'm still looking for a review manager for the multiprecision library: http://svn.boost.org/svn/boost/sandbox/big_number/libs/multiprecision/doc/h… All volunteers/victims gratefully accepted ;-) Many thanks in advance, John.

1 0

Re: [boost] Slow Bugs Tracker
by Olaf van der Spek 29 Apr '12

29 Apr '12

Hi, A few months ago there seemed to be plans to improve Trac. What's the status of those plans? Trac is still slow as hell. http://lists.boost.org/Archives/boost/2011/12/188639.php http://lists.boost.org/Archives/boost/2011/12/188937.php -- Olaf

7 44