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

[pool] why no object-pool with O(1) allocation AND O(1) deallocation?!?
by Andre Krause 04 Sep '06

04 Sep '06

dear boost users, boost offers in lib pool the class object_pool. as is stated in the docs, it allows for allocation with O(1) but deallocation need O(n) time. i suppose it has to search the internal list for the position of the object ro be freed. but this could be speed up a lot by using a stack of free objects. if one would like to allocate an object, a pointer to a free memory cell is popped from the stack, if the object should be returned / freed to the pool, that objects pointer is pushed onto the stack. of course this causes memory overhead, but, for example if one would like to implement a really large pool of small objects for a particle system, then O(n) deallocation is prohibitive. so my question is why boost does not offer an alternative version of object_pool with both O(1) alloc and dealloc ? below is my version of a stack based object pool with O(1) both alloc and dealloc. i really do not like my implementation, but i did not know how to do better... template<class T> class Smallobjects_Pool { protected: class Node // a node of a double linked list { public: Node() {prev=next=NULL; idx=-1; } Node* prev; Node* next; int idx; // index to the objects - vector T obj; }; //Node* next(int cur) { return objects[cur].next;} //Node* root() { return _root; } //Node* last() { return _last; } protected: Node* _root; Node* _last; vector<Node> objects; vector<int> free_objects; public: T & operator[](int i){return objects[i].obj;} bool empty() { return _root == NULL;} Smallobjects_Pool(int initial_size) { _root=NULL; _last=NULL; objects.resize(initial_size); free_objects.reserve(initial_size); for(unsigned int a=0;a<objects.size();a++) free_objects.push_back(a); } void push_back(const T& t) { //assert(!free_objects.empty() && "out of capacity"); if(free_objects.empty()) { PR("small object pool full.."); PR(objects.size() - free_objects.size()); return; } Node* new_obj = &(objects[free_objects.back()]); new_obj->idx = free_objects.back(); free_objects.pop_back(); if(_root==NULL) { _root=new_obj; new_obj->prev=NULL; new_obj->next=NULL; } else { _last->next=new_obj; new_obj->prev=_last; new_obj->next=NULL; } _last=new_obj; new_obj->obj = t; } class iterator { //protected: public: Node* i; public: iterator(Node* _i) : i(_i) {} ~iterator() {} iterator& operator=(const iterator& other) { i = other.i; return(*this); } // The assignment and relational operators are straightforward bool operator==(const iterator& it) { return(i == it.i); } bool operator!=(const iterator& it) { return(i != it.i); } iterator& operator++() { assert(i && "increment on null pointer"); i = i->next; return(*this); } // Update my state such that I refer to the next element in the SQueue. //Iterator& operator++(int) { Iterator tmp(*this); ++(*this); return(tmp);} T& operator*() { return i->obj; } // Return a reference to the value in the node. I do this instead of returning by value so a caller can update the value in the node directly. T* operator->() { return &(i->obj); } // Return the address of the value referred to. }; iterator begin() { return(iterator(_root));} iterator end() { return(iterator( NULL));} iterator erase(iterator i) { assert(i.i!=NULL && "cannot free null pointer."); free_objects.push_back(i.i->idx); if(i.i == _last) { if(i.i->prev != NULL) _last = i.i->prev; _last->next=NULL; return NULL; } if(i.i == _root) { _root = i.i->next; _root->prev=NULL; return _root; } i.i->prev->next = i.i->next; i.i->next->prev = i.i->prev; return iterator(i.i->next); } };

2 1

Can't build Boost on MSVS 2005
by Saul Spatz 03 Sep '06

03 Sep '06

I downloaded Boost and bjam today,and I'm trying to build on MSVS 2005. I run vsvars32.bat, and then I try bjam "-s VC80_ROOT= C:\Program Files\Microsoft Visual Studio 8\VC -stools=vc-8_0 --builddir=C:\temp --with-python-version=2.4 --with-pydebug" install but none of the libraries get built. I keep seeing stuff like CALL "C:\Program Files\Microsoft Visual Studio .NET 2003\VC7\bin\VCVARS32.BA T" >nul "C:\Program Files\Microsoft Visual Studio .NET 2003\VC7\bin\cl" /Zm800 -nologo <snip> which fails because there, is of course, no MSVS .NET 2003 directory. As far as I can tell, there is no environment variable on my system that mentions VS 2003. Can anyone help? Saul Spatz -- View this message in context: http://www.nabble.com/Can%27t-build-Boost-on-MSVS-2005-tf2204913.html#a6106… Sent from the Boost - Users forum at Nabble.com.

3 2

[MPL] Late-Breaking Map/Set/Performance Fixes?
by Holcomb, William (Tom ) 03 Sep '06

03 Sep '06

Greetings: Are there any map/set (or general/performance MPL) fixes for problems in earlier Boosts (such as 1.33.1) that didn't make it into the 1.34.0 Release Candidate? I ask because I'm working on a metaprogramming prototype (~200 lines of code so far) that uses associative sequences and whose compilation behavior was greatly improved by switching from 1.33.1 to 1.34.0, but now the compiler (msvc-8) runs out of heap space if I increase the size of the input sequence from 4 to 5 (in production, my solution would have to contend with sequences of size 60+, so I'm worried). I can't rule out the possibility that I've got some HUGE recursion or something going on in my prototype, but I want to pursue this possibility as well. If anyone is aware of said late-breaking fixes, I'd be grateful if you'd let me know how to get them (please be patient as I am a total CVS (not to mention Boost/MPL/newsgroup) newbie). Thanks, Tom

2 1

Date calculations with exclusions
by kingos＠optusnet.com.au 02 Sep '06

02 Sep '06

Hi, I am hoping to use boost::date_time to do some date calculations. What I want to do is basically work out the number of days to a given date with various exclusions applied. eg. How many working days are there to christmas? using namespace boost::gregorian; int CalculateDaysToChristmas() { date_period someHoliday(date(2006, Dec, 01), date_duration(1)); date today(2006, Sep, 1); date christmas(2006, Dec, 25); int days = 0; for (day_iterator i = today; i != christmas; ++i) { if (is_weekday(*i) && !someHoliday.contains(*i)) ++days; } return days; } However, I can't find a simply way to work out is_weekday ... Secondly, is there some way to put every sunday in as a date_period? Thirdly, this doesn't seem very optimal ... is there a better way of doing this? Thanks, Andrew

3 2

Re: [Boost-users] Date calculations with exclusions
by james.jones＠firstinvestors.com 01 Sep '06

01 Sep '06

From: kingos(a)optusnet.com.au > Hi, > > I am hoping to use boost::date_time to do some date calculations. What I want to do is > basically work out the number of days to a given date with various exclusions applied. > > eg. > > How many working days are there to christmas? > > using namespace boost::gregorian; > > int CalculateDaysToChristmas() > { > date_period someHoliday(date(2006, Dec, 01), date_duration(1)); > > date today(2006, Sep, 1); > > date christmas(2006, Dec, 25); > > int days = 0; > for (day_iterator i = today; i != christmas; ++i) > { > if (is_weekday(*i) && !someHoliday.contains(*i)) > ++days; > } > > return days; > } > > However, I can't find a simply way to work out is_weekday ... > Secondly, is there some way to put every sunday in as a date_period? > Thirdly, this doesn't seem very optimal ... is there a better way of doing this? First off, is_weekday is pretty simple, something like: bool is_weekday(const gregorian::date& d) { return d.day_of_week != Saturday && d.day_of_week != Sunday; } Second, I don't understand your question about Sundays. Third, you're right, it isn't very optimal. You'd be better off to note that weekdays in a period are equal to: complete weeks in the period * 5 + extra weekdays at the end of the period - holidays in the period Complete weeks in the period is easy to compute (days_between/7); the extra weekdays could be done via iterator (but guarantee to be <= 6 times through the loop) or via lookup table if you want really fast results. So this just leaves holidays in the period. It would be pretty easy to build a class which implemented a fast (constant-time) lookup for the # of holidays given a starting and ending date. Once you've got that, voila. If you're looking for some really quick way to do this in Boost::DateTime, I don't think it's possible. Holidays aren't even localizable, since different businesses in the same locale can have different holiday schedules. It might be nice to include some sort of "date-exclusion" functionality in Boost that would implement the class I propose above, but I'm not sure this would be generally useful enough to bother with. Anyone else have an opinion on this? - James Jones Administrative Data Mgmt. Webmaster 375 Raritan Center Pkwy, Suite A Data Architect Edison, NJ 08837

1 0

[Boost.Test] how to get current test outcome from within test case
by Pierre-Jules Tremblay 01 Sep '06

01 Sep '06

This must be an FAQ but I haven't found the answer in the documentation. With Boost.Test, is there a way to know whether the current test case's success/failure status? Cheers pj

2 5

[asio] how to check connection state
by Sascha Seewald 01 Sep '06

01 Sep '06

Hi boosters! I've written a simple client-server example using boost::asio::ip::tcp::socket. I establish the connection using socket::connect() and close it using socket::close(). But how can I determine the current state of connection (Am I still connected or not?) Is it sufficient to check whether the socket is open? Is there a function socket::is_open()? Or do I have to track errors for disconnection events and set some is_connected-flag somewhere. Btw what is the purpose of the keep_alive socket option? Thanks in advance Sascha

1 0

[boost::serialization]assertion in basic_iarchive.cpp
by Sascha Friedmann 01 Sep '06

01 Sep '06

Hello, when loading my application's state from a text file I get sometimes this assertion: Assertion failed: object_id_vector[i].adress >= old_adress, file ../src/basic_iarchive.cpp It's very strange because sometimes this assertion is triggered and sometimes not. What exactly does this assertion mean? Thanks. Sascha Friedmann

2 2