- Boost-users - lists.boost.org

How do you handle state-independent events efficiently in MSM?
by Peter Gissel 19 Sep '11

19 Sep '11

I have a state machine that has a large number of event types but only a few different states. Some of the events can be received regardless of the state and the same action is taken regardless of which state the machine is in, so the event is really unrelated to the state. I there an efficient way of handling the events that are state independent rather than defining a transition from every possible state for each state independent event? A number of the events do transition the state machine and depend on the current state, so a transition is necessary. But if no state change is required for the event and the current state is not important, I'm not sure how to define event handling without defining a transition from all possible states. This winds up creating a much larger number of transitions than I really need, and a much longer compile time than I would like. I am looking for some way to handle these states that would still give me good run time speed, put not pay the penalty that defining a large number of transitions would impose. Thanks! Peter Gissel

2 6

threads: shared lock question
by Mark Higgins 19 Sep '11

19 Sep '11

I'm new to threads and boost, so I'm sure this is a real newbie question and hopefully quite easy to answer! I'm trying to set up a pool of threads that do a calculation, and I want to wait for those calculations to finish. Normally I'd do thread_group.join_all() to wait for the threads to exit - but in this case I want my threads hanging around so that I can redo that calculation over and over again when I ask it to do. What's the best way to signal back to the main thread that all calculations are complete? Should I use a barrier?

5 4

What concepts/interface must the element type of multi_index_container<> support?
by Chris Cleeland 19 Sep '11

19 Sep '11

I'm trying to use boost::multi_index_container on a type that I've defined. I want the container to have two indices: 1. insertion order 2. hashed_unique Further, I want the hash to use an sha1 hash value, which I have represented as a struct. So, I have something like the following: struct sha1_hash_value { ... }; class MyRecord { public: ... sha1_hash_value hash() const { return _sha1_hash; } ... private: sha1_hash_value _sha1_hash; }; using namespace boost::multi_index; typedef boost::multi_index_container< MyRecord, indexed_by< sequenced<>, hashed_unique<const_mem_fun<MyRecord,sha1_hash_value,&MyRecord::hash> > > > RecordSet; When I compile this, I get lots of complaints regarding the conversions to/from the 'sha1_hash_value' type. I understand that I need to define appropriate member functions or specializations for this type. What I would like to know is what those member functions/specializations are and where that is documented so that I don't have to decipher compiler error gibberish iteratively and just cross my fingers that the errors are really about missing functionality and not about misuse. FWIW, if I change sha1_hash_value to a 'long', the code compiles, so it doesn't appear as if I'm completely misusing multi_index_container<>, though I'm certainly still open to that possibility as well. Thanks for any help. The brick wall I'm banging my head against is really starting to hurt.

2 2

boost::bind replacements
by Littlefield, Tyler 18 Sep '11

18 Sep '11

Hello all: I really like boost::bind because I use an event system in my game engine, and it allows me to bind callbacks in classes, rather than having everything as global functions, like they were before. My problem is this. Boost::bind incurs a lot of overhead, both during compilation and in binary size. Is there any way to decrease this? If not, I'm curious if someone could explain how boost::bind works so that I can create my own? I understand function objects can hold function pointers and boost::bind returns a function object, but I'm really confused as to how placeholders work. -- Take care, Ty Web: http://tds-solutions.net The Aspen project: a light-weight barebones mud engine http://code.google.com/p/aspenmud Sent from my toaster.

4 4

[fusion] [range] range::erase vs. fusion::erase
by Nathan Ridge 17 Sep '11

17 Sep '11

Hi, I have an std::vector<S> and I use boost::erase() from Boost.Range on it. Now I changed S to be a Fusion sequence type, and as a result, boost::fusion::erase() is being found by ADL and called instead of boost::range::erase(). Which of course gives all kinds of errors because std::vector is not a Fusion sequence. Is there a way to avoid this, besides qualifying every call to boost::erase() as boost::range::erase()? Basically I want to disable the finding of boost::fusion:: algorithms by ADL. Thanks, Nate

2 2

Question on std::binary_function template arguments
by John M. Dlugosz 17 Sep '11

17 Sep '11

You may recall my earlier question asking if there was an easy way to "bind" a member function to a weak_ptr. Looking at the lower-level STL features, I came up with this idea, to prepare an adaptor in the same manner as mem_fun and mem_fun_ref. Generalizing the idea, it takes a member function and produces a functor that takes the object as a first argument, for various ways of presenting the object. I extended it with mem_fun_weak, where the resulting functor is called with a weak_ptr as the first argument. Below is one form (the one I need). To work along with the other STL primitives, there are a total of 4 forms (const/non-const, one arg or no args). That doesn't have arbitrary signatures like the more general "bind", which is something I'd like to discuss further. For now, my question concerns the first template argument to std::binary_function. Looking at the declarations for the standard mem_fun1_t and mem_fun1_ref_t, I don't know if this should be: 1) the full type passed as the first argument to the resulting functor, 2) such, but without the final &, 3) just the object's type, 4..n) something else? Can someone clarify that for me? Here is my code (using #1 above): template <typename Result, typename Type, typename Arg> class mem_fun1_weak_t : public std::binary_function<const std::tr1::weak_ptr<Type>&, Arg, Result> { Result (Type::*p )( Arg ); public: explicit mem_fun1_weak_t( Result (Type::*Pm )( Arg ) ) : p(Pm) {} Result operator() (const std::tr1::weak_ptr<Type>& left, Arg right) const { std::tr1::shared_ptr<Type> sp (left.lock()); if (sp) { Type* object= sp.get(); // shared_ptr doesn't overload operator->*. return (object->*p)(right); } else return Result(); } }; template<typename Result, typename Type, typename Arg> mem_fun1_weak_t<Result, Type, Arg> mem_fun_weak (Result ( Type::*Pm )( Arg ) ) { return mem_fun1_weak_t<Result, Type, Arg> (Pm); } Thanks, —John

3 4

complex matrix operations - eigen decomposition, singular value decomposition, complex inverse, cholesky decomposition etc
by sairamesh raghuraman 17 Sep '11

17 Sep '11

Hello, I am looking for a C++ library that has routines for a variety of complex matrix decompositions (eigen, cholesky, singular value etc) and even complex matrix inverse, principal matrix logarthm of a complex matrix etc. And the matrices which I am talking are pretty huge..can go upto 1024 x 1024 x 32. I have already found a library that can do all of the above..used it..it works fine..but a disadvantage is that it cannot handle multidimensional arrays. Now the question is, does Boost C++ library have functions for the above? Looking forward to a reply and thanks in advance Ramesh

2 1

[Boost.Log] Compiling
by Leo Cacciari 17 Sep '11

17 Sep '11

Hi, I'm working on an application that'll use the boost libraries, and I thought using boost.Log for logging (yes, I'm aware the API will change when boost.Log will be included in the boost distribution). My problems start when I try to compile the library. I downloaded the source and copied the relevant files in a 1.42.0 source tree (1.42.0 is the default version of boost on my Ubuntu 10.10 system) then run bjam --with-log This works, and it also builds a bunch of other libraries, the ones Boost.Log depends upon i.e., as per Boost.Log documentation, Boost.Filesystem, Boost.System, Boost.DateTime, Boost.Thread and Boost.Regex. However, these libraries are already installed on my system, using my distribution packages. Installing the newly compiled version of these libraries alongside the pre-existing ones will confuse my linker, and replacing the latter with the former will confuse my package manager... To me the alternatives seem to be the following: - install a full boost distribution replacing the one furnished by Ubuntu (plus: I could use 1.47, minus: I prefer using software in the Ubuntu repositories, if possible) - manually copying the compiled libboost_log* from the stage directory and header files from the boost/log directory into the target directory, skipping the bjam installation phase (pro: quickly done, minus: I'm not sure that the install process does not perform other tasks, beyond copying the files and running ldconfig, nor I'm sure that there is no compatibility issues) Obviously, I could miss other approach, especially one that simply told bjam to use the already installed libraries instead of compiling new versions. Could anyone offer suggestions? -- Leo Cacciari Aliae nationes servitutem pati possunt populi romani est propria libertas

2 1

"apply" member function over collection
by John M. Dlugosz 17 Sep '11

17 Sep '11

Consider this: // new way, so far so good boost::range::for_each (mylist, std::tr1::bind(&C::foo, _1)); // original: // for(Collection::iterator fi = mylist.begin(); fi != mylist.end(); ++fi) { // Collection::reference item = *fi; // item->foo(); // } I've got a zillion of these one-shot "for" loops, and I was able to change this sample to use a range::for_each. But, it's still too ackward and will be unfamiliar to other users, so it has to be "sleek or bust"! The item type is actually a smart pointer. Is there a more succinct way of writing it, without needing the extra placeholder to be mentioned? The for_each works find as an "apply" except that the member function needs to be adapted; it won't do that by itself. I found that boost::range::for_each (Filters, std::tr1::mem_fn(&VseeMediaFilter::Uninitialize)); compiles too, even though I'll be passing a smart pointer to the resulting adapted function. But, as mentioned on another thread, mem_fn is deprecated? How might I write this more tersely, or provide an 'apply' utility of my own that does what I want? Thanks, —John

1 0

Re: [Boost-users] boost 1.47: gcc file dependency generation trips over BOOST_PARAMETER_MAX_ARITY
by Ivan Le Lann 17 Sep '11

17 Sep '11

----- "Ytsen de Boer" <yrdeboer(a)gmail.com> a écrit : > Thank you for your answer. > I believe we are not on the same page. > Obviously, when I define the max arity to a value of 7 or more, the > preprocessor will finish properly. > > But the point I am trying to make is more subtle, namely, that the > generation of the project dependencies has not a thing to do with the > boost max arity. I think it does actually. How could gcc track inclusions without preprocessing ? #ifdef USE_LIB_A # include "a.h" #else # include "b.h" #endif It seems very dangerous to me to use "-MMD" with different preprocessor definitions than at real compile time. But I'm not an expert. Anyway : > > I don't know what the boost max arity means or what its for, but it > seems to me that it is a run time variable. Have you checked that ? My guess is that it is a compile time variable controlling the number of template arguments for a particular class or function. Pure compile time. Regards, Ivan

3 2