June 2006 - Boost-users - lists.preview.boost.org

Boost threads built from source causes app freeze, using prebuilt works fine
by Zachary Mark 28 Jun '06

28 Jun '06

Hello, We are using Boost threading as a part of our applicaiton. On multiple Linux platforms, when we compile Boost's thread library from source and use the compiled shared libraries (we are aware of the static linking issues with the thread library) with our application, we get a deadlock at different places in the application depending on which platform we are using. When using a pre-built binary (for example, from Debian's apt repository) for the thread library, the application does not deadlock. There is otherwise no difference between the application builds. We would like for us to be able to build Boost libs from source on every target platform, so I would like to figure out exactly what our build script is doing wrong with the thread library. For reference, here is the relevant portion of the Boost build bash script (this is a part of a larger script): # run configure cd libs/config || fail-configure chmod +x configure || fail "error: could not change permissions on ${PACKAGE} configure script" ./configure || fail-configure cp user.hpp ../../boost/config/ cd ../.. # build and install libraries BOOST_LIBS="date_time filesystem program_options regex thread" for LIB in ${BOOST_LIBS}; do echo "Building library ${LIB}..." bjam -sTOOLS=gcc -sBUILD="release" --with-${LIB} \ --layout="system" stage 2>&1 || \ fail "error: could not build boost library ${LIB}" bjam -sTOOLS=gcc -sBUILD="release" --with-${LIB} --layout="system" \ --prefix=`pwd`/../../${INSTALL_DIR} install 2>&1 || \ fail "error: could not install boost library ${LIB}" done Is there something that this script is missing or doing wrong that might cause some of these deadlock issues? Thanks in advance for your help. -- Zach

1 0

Re: [Boost-users] Interesting question, may be potential feature
by Ovanes Markarian 28 Jun '06

28 Jun '06

Nat, thanks for the quick reply! Unfortunately, people ideally would like to work with this data types without any functions and my job is to give them type safety ;) The scenario should be: struct PossibleValues { accessor<0> value_name_0(my_union); accessor<1> value_name_1(my_union); accessor<N> value_name_n(my_union); private: discriminated_union<mpl_vector_over_pairs> my_union; }; usage: void ValueHandler(PossibleValues& ps) { if(ps.value_name_0) ps.value_name_0->netsted_value = new_value; .... } Accessor makes havy usage of mpl::if_, enable_if, etc. so that -> operator works will regular types, pointer types, types where the -> operator is defined etc... Accessor analyses the type associated with the discrminator and overloads a correct set of functions/operators for that type. For example, if a stored type is pointer -> returns the pointer itself, if the stored type is normal object the address of it is returned, but if the type contains the -> operator then the object is returned by value (in case of shared_ptr, because you can not apply the -> operator twice ->->). Thanks, Ovanes

1 0

Re: [Boost-users] Interesting question, may be potential feature
by Nat Goodspeed 28 Jun '06

28 Jun '06

> -----Original Message----- > From: boost-users-bounces(a)lists.boost.org [mailto:boost-users- > bounces(a)lists.boost.org] On Behalf Of Ovanes Markarian > Sent: Wednesday, June 28, 2006 12:39 PM > To: boost-users(a)lists.boost.org > Subject: [Boost-users] Interesting question, may be potential feature > > But as some of you might have seen I have to pass to all accessors the > reference or pointer to the > accessed discrminated union type. So every instance requires on 32bit OS > at least 4 bytes. The > banal thing is that a group of accessors accesses the same type. The > question/idea here would be > to use or to make these objects share the reference, so that all related > accessor references > require 4 bytes. What do you think of this? Is it possible at all? [Nat] Um, make accessors be methods of a small object that contains the reference? But in that case, why wouldn't the convenience methods be attached to the referenced union object itself?

1 0

Interesting question, may be potential feature
by Ovanes Markarian 28 Jun '06

28 Jun '06

Dear Users and Developers, I am currently working on a kind of discriminated union type, e.g. XDR specific union type implementation (please refer to http://www.faqs.org/rfcs/rfc1832.html section) which maps enumeration values to types. In this case one wants to provide users as much comfort as possible for accesing this union. I have written a template class, which is defined by mpl::vector pairs of discriminant and related type. This class contains a boost::variant definition from this sequence. The problem is, that accessing the values etc. of this variant is a little bit clumsy: discriminated_union_inst.get<some const int for discrminant>(); discriminated_union_inst.set<some const int for discrminant>(new value); So I defined some accessor template. If initialized the accessor is responsible to access only one particular discriminant related type for example if discriminated_union is defined by sequence of: mpl_derived_pair<0, int>, mpl_derived_pair<1, std::string> ... mpl_derived_pair<N, MyClass> I define an accessor object: accessor<0> accessor_for_0(union_instance); later on this accessor allows the access if valid data is contained in the union. if union contains second type (mpl_derived_pair<1, string>) accessor_for_0->some_value //throws exception or if(accessor_for_0) // is true: means unions contains the correct type //do some access if true Now I must say: It was a pleasure to implement this using many boost libraries! But as some of you might have seen I have to pass to all accessors the reference or pointer to the accessed discrminated union type. So every instance requires on 32bit OS at least 4 bytes. The banal thing is that a group of accessors accesses the same type. The question/idea here would be to use or to make these objects share the reference, so that all related accessor references require 4 bytes. What do you think of this? Is it possible at all? With Kind Regards, Ovanes Markarian

1 0

shared ptr + const_cast woes
by Evan Drumwright 28 Jun '06

28 Jun '06

Hi all, I am having trouble with pointer ownership as a result of the const_cast operator. In particular, for the code below, when a const_cast is used to create a const int pointer, the use count for x does not increase. The result of this is that when x.pop_back() is called, y is no longer valid (though the use count for y remains 1). The documentation for shared_ptr does not help here. Any suggestions? Thanks, Evan #include <vector> #include <iostream> #include <boost/shared_ptr.hpp> int main() { std::vector<boost::shared_ptr<int> > x; x.push_back(boost::shared_ptr<int>(new int)); *x.front() = 4; std::cout << "use count: " << x.front().use_count() << std::endl; boost::shared_ptr<const int> y(const_cast<const int*>(x.front().get())); std::cout << "use count: " << x.front().use_count() << std::endl; std::cout << "use count: " << y.use_count() << std::endl; boost::shared_ptr<int> z(const_cast<int*>(y.get())); std::cout << "use count: " << z.use_count() << std::endl; std::cout << "use count: " << y.use_count() << std::endl; x.pop_back(); std::cout << *y << std::endl; } Output: use count: 1 use count: 1 use count: 1 use count: 1 use count: 1 0

3 2

[BGL] Two Grpah Matching Algorithms
by Hossein Haeri 28 Jun '06

28 Jun '06

Dear all, (I don't know whether I'm supposed to post BGL stuff to a more specialised list. Am I?) I was wondering if anybody here has implemented the maximum weight graph matching algorithm for bipartite graphs. Is that already a part of BGL? And, what about the stable marriage algorithm? To stop people from lynching me, this is NOT an assignement; We [1] are running out of time for a paper submission, and, we still need a few more evaluations. That's why I'm in a hurry... So, what I need for the moment is an implementation of the above two algorithms. Any help in this respct would be highly appriciated. TIA, --Hossein [1] Semantic Web Research Laboratory, Computer Engineering Department, Sharif University of Technology (SUT), Tehran, IRAN. http://sw.ce.sharif.edu ___________________________________________________________ Try the all-new Yahoo! Mail. "The New Version is radically easier to use" The Wall Street Journal http://uk.docs.yahoo.com/nowyoucan.html

2 3

[preprocessor] CAT and '-' token
by Johan Nilsson 28 Jun '06

28 Jun '06

Hi, I guess this is more of a regular preprocessor question than a Boost.Preprocessor specific one, but: Shouldn't it be possible to concat tokens starting with (or solely consisting of) a '-'? The example at the end of this posting yields the errors: error: pasting "-" and "debug" does not give a valid preprocessing token error: pasting "foo" and "-" does not give a valid preprocessing token --- #include <boost/preprocessor/cat.hpp> #include <boost/preprocessor/stringize.hpp> #include <iostream> #define MODULE foo #define VARIANT debug #define NAME BOOST_PP_CAT(MODULE, BOOST_PP_CAT(-, VARIANT)) void foo() { std::cout << "The name is: " << BOOST_PP_STRINGIZE(NAME) << '\n'; } --- Any workaround? I'm using Boost from cvs at RC_1_34_0 and g++ 4.0.2 under OpenSuse 10. / Johan

2 2

Comments for Filesystem Library Proposal for TR2 (N1975)
by Daniel Krügler 28 Jun '06

28 Jun '06

Hello! One week ago I wrote a list of comments to the above mentioned proposal provided by Beman Dawes to his email address <mangled_address>bdawes at acm dot org</mangled_address> given in the public paper http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1975.html Since I got no answer, I assume that the mail got lost due to filter barriers or something else. Is someone aware whether the above mentioned address should be still valid? I apologize for the noise in this group, but currently I'm not able to use either comp.std.c++ or comp.lang.c++.moderated due to news group rearrangements in our company to address this issue there and have hopes that Beman Dawes might read this group ;-) Thanks for your patience, Daniel Krügler

2 2

shared_ptr in a boost::any
by Thomas Matelich 28 Jun '06

28 Jun '06

I use boost::any for sending data to my thread's command queue, and I need to send a shared_ptr in. I'm looking for any one to tell me this is fine or bad. Based on my understanding of boost::any I think it will be fine, but I'm slightly concerned I'll screw up the reference count somehow. Can anyone confirm or allay my fears? Thanks, Tom

2 1

Re: [Boost-users] [date_time] extract date from string using a list of patterns
by james.jones＠firstinvestors.com 27 Jun '06

27 Jun '06

From: Jason Dolan <jason(a)pcc.com> > Thats why I was looking into it this way, I was hoping not to have to > create my own parser. I'm actually kinda of surprised that there isn't > a open source natural language date string parser out there already. Maybe take a look at the source from the GNU "at" command? It would be in C, but might yield some useful ideas. HTH, - James Jones Administrative Data Mgmt. Webmaster 375 Raritan Center Pkwy, Suite A Data Architect Edison, NJ 08837

1 0