January 2009 - Boost-users - lists.boost.org

shared_ptr custom deleter implementation
by Zachary Turner 17 Jan '09

17 Jan '09

I'm trying to write a class lazy<T>, the skeleton of which is as follows: template<class T> class lazy { public: template<class F> lazy(F fetcher) //Store fetcher somehow { } T get() { if (!cache) cache = fetch(); //Access fetcher from constructor in a type-safe way return cache.get(); } private: optional<T> cache; }; My problem is how to store / access the fetch function in a type safe way. The reason I titled this message how I did is because I see that shared_ptr does almost exactly the same thing with its deleter. But I'm still kind of new to the whole metaprogramming thing and having a little difficulty really getting my head around how it works. I guess there is some general "pattern" that applies in both of these cases as well as any other case where you want to store some value, the type of which is deduced by the constructor, and then later accessing that value in a type safe way. This message could probably have been equally well fitting in a general C++ forum, but since this is a problem that seems to arise really frequently in boost and other code using heavy templates, I guess I'm in pretty good company here as well. The following seems close to a solution: template<class T> class lazy_impl_base { virtual T fetch() = 0; }; template<class T, class F> class lazy_impl : public lazy_impl_base<T> { // etc }; and then storing a lazy_impl_base<T>* in lazy<T>. But there are strange compiler warnings, and I fear I will encounter subtle errors when I attempt to use it with functors, binders, and other non trivial "functions". Can anyone offer some guidance here? Maybe a quick high level overview of how shared_ptr and other boost libraries solve this problem, or even an implementation of lazy that works :P

2 1

Optionally enforce const correctness in argument dependent lookup
by er 17 Jan '09

17 Jan '09

Hi All, The example below illustrates my intent, but obviously it's not the right approach... so I'm hoping someone has a suggestion. template<typename X> struct enforce_constness{ typedef add_reference<const X> type; }; template<template<typename> class T = enforce_constness> struct A{ template<typename X> void operator()(typename T<X>::type x)const{ // Under default I'd like const X& }; } X x; A<> a; a(x); no match for call to (A<T>)(X&)

2 2

problem with boost::is_same
by Zachary Turner 17 Jan '09

17 Jan '09

I have a question about the following code: typedef void(intArrayParamFunc)(int[]); typedef void(intPtrParamFunc)(int*); BOOST_STATIC_ASSERT((boost::is_same<intArrayParamFunc,intPtrParamFunc>::value)); BOOST_STATIC_ASSERT((boost::is_same<int*,int[]>::value)); cout << typeid(intArrayParamFunc).name() << endl; cout << typeid(intPtrParamFunc).name() << endl; cout << typeid(int*).name() << endl; cout << typeid(int[]).name() << endl; The first assert compiles but the second fails, which means that boost::is_same<> thinks that int* and int[] are different, but void(int[]) and void(int*) are the same. Why is this? The typeids print out the following: void __cdecl(int * const) void __cdecl(int *) int * int [0] I'm using Visual Studio 2008. I notice that typeid tells me that intArrayParamFunc is void(int* const). So why does boost::is_same<> compare them equal? Surely void(int* const) should not be the same type as void(int*). Unless this is just some obscure language subtlety I'm not aware of. So is this a bug in boost::is_same<>, a bug in my compiler, or no bugs at all and everything is working perfectly?

2 1

[date_time] Error compiling greg_month with Intel C++ 11.0
by Remko Tronçon 17 Jan '09

17 Jan '09

Hi, I'm having a problem compiling greg_month.cpp with Intel C++ 11.0. I used bcp to create a local version of boost 1.37.0. When I run icc -I. -c libs/date_time/src/gregorian/greg_month.cpp I get: (0) internal error: backend signals (and nothing more). Is this a known problem? It works fine with other compilers, and the rest of boost compiles fine for me with ICC as well. thanks, Remko

1 0

Serialization on Mac Followup
by Brandon Aubie 17 Jan '09

17 Jan '09

I have gotten everything working correctly now. The first time when I installed boost, I downloaded the source, did the regular configure, make, install into /usr/local. I then tried using MacPorts to install Boost 1.37 and this time, using the MacPorts installation, everything is working fine!

1 0

[boost][iostreams] Crash when using bzip2 codec
by Marc Viala 17 Jan '09

17 Jan '09

Hi, During our migration from Boost 1.34.1 to Boost 1.37.0 we are facing a bug in bzip2 codec (boost::iostreams::bzip2_compressor & boost::iostreams::bzip2_compressor). To demonstrate it, you will find hereafter a code snippet: #include <fstream> #include <boost/iostreams/filtering_stream.hpp> #include <boost/iostreams/combine.hpp> #include <boost/iostreams/filter/bzip2.hpp> #include <boost/iostreams/filter/gzip.hpp> #include <boost/iostreams/filter/zlib.hpp> void main() { using namespace boost::iostreams ; // Run smoothly { filtering_stream<bidirectional> chain ; chain.push(combine(gzip_decompressor(), gzip_compressor())) ; chain.push(std::fstream("test.gz", std::ios::binary|std::ios::out)) ; chain << "test" ; } // Run smoothly { filtering_stream<bidirectional> chain ; chain.push(combine(zlib_decompressor(), zlib_compressor())) ; chain.push(std::fstream("test.zip", std::ios::binary|std::ios::out)) ; chain << "test" ; } // Crash in bzip2_decompressor_impl::close { filtering_stream<bidirectional> chain ; chain.push(combine(bzip2_decompressor(), bzip2_compressor())) ; chain.push(std::fstream("test.bz2", std::ios::binary|std::ios::out)) ; chain << "test" ; } } The crash in bzip2_decompressor_impl::close is due to an access violation @ line 359 in boost/iostreams/filter/bzip2.hpp. For the other codec, zlib & gzip, everything runs smoothly. Set-up: Boost 1.37.0 Windows XP SP3 Visual C++ 7.1 Bzip2 Library 1.0.5 With Boost 1.34.1 the bzip2 codec seems to run smoothly. Is there any work around for this codec w/ Boost 1.37.0? Thanks in advance. Best regards, Marc Viala

2 1

Boost::Serialize on OS X 10.5.6 with g++ 4.3.1 -- Fails to initialize
by Brandon Aubie 17 Jan '09

17 Jan '09

Hello all, I have an application that works fine in Linux that I'm trying to port to OS X. Since it mostly relies on Boost, I was hoping this would be easy, but Boost::Serialization is causing me grief. I have tried this with both binary_oarchive and text_oarchive to the same results (I am using Boost 1.37). Everything compiles fine, but I get an "Abort trap" program crash at this line: ofstream ofs("output"); boost::archive::text_oarchive oa(ofs); // <--- HERE I know it's that line since I put a cout << "Above" << endl; above it and a cout << "Below" << endl; below it and the only thing that gets printed in "Above". I'm using g++ 4.3.1 from MacPorts and I compiled Boost 1.37 with that compiler. Like I said, everything compiles fine, but the program crashes at this point. It works fine in Linux. Any ideas? Thanks

1 0

[Random] Bernoulli generalization.
by Rodolfo Federico Gamarra 17 Jan '09

17 Jan '09

Hi!! I was working with boost::bernoulli_distribution, and needed a generalization which instead of receiving a single probability (and return true/false) would receive N probabilities (adding less than 1) and would return a whole number in the interval [0 N]. It seems that's usually called "categorial distribution" and that could be easily implemented, generalizing boost::bernoulli_distribution. For now, it seems that the workaround, so to say, is to use a uniform [0 1] (boost::uniform_01) and then externally to the distro compare its results with the N probabilities. Thanks a lot to you all. -- Rodolfo Federico Gamarra

1 0

[pool] swap function
by Simon Brown 17 Jan '09

17 Jan '09

Hi, I'm using boost::pool for unordered malloc/free and find myself wanting two additional features: * A nothrow swap function (c.f. scoped_ptr) to swap two pools with equal chunk size. This is a bit of a show-stopper for me at the moment, as I don't want to add the extra level of indirection from allocating the pool object itself on the heap. Has anyone already implemented this for boost::pool, or know if it is on the roadmap for a future release? * A function to reset the pool, but without actually deallocating the memory blocks. This would be the same functionality as purge_memory, but without deallocation. Perhaps "purge_chunks" is a reasonable name for this feature? This isn't a show-stopper, but I want to reset everything and allocate a similar numbe of objects, and having good cache coherency for the allocations would be a benefit. Again, before I delve into the internals, has anyone already had a go at this? Many thanks, Simon

1 0

Accumulators: Adding features at runtime
by dhruva 16 Jan '09

16 Jan '09

Hello, Accumulator features can be marked as droppable so that at runtime, one could drop some calculations from being performed. Is there some way to add features at runtime. I need to do something like this in a generic method. The caller will provide me the calculations they would want to perform in a bitwise ORed format. Based on that, I need to create an accumulator_set object in my constructor. The accumulator_set object will be a member of my class. A crude way I can think of is to add all features that I intend to support and drop them based on user input. Is there a better way of doing it, is this not an optimal way to do it? Looking forward for your feedback. // Some pseudo code to show my intentions class calcstat { public: void SetCalc(int type) { // I want to add features here based on user input } void compute(void) { // If I cannot add based on user input, I will drop based on // user input. I will then not be able to add if the user input // changes. For now, I am not interested in its reuse! } private: accumulator_set<double> m_acc; }; -dhruva -- Contents reflect my personal views only!

1 0