Re: [Boost-users] [multi_index] custom template class for handle multi_index map

Hello Daniel ________________________________________ De: boost-users-bounces@lists.boost.org [boost-users-bounces@lists.boost.org] En nombre de damny@web.de [damny@web.de] Enviado el: sábado, 19 de abril de 2008 17:42 Para: boost-users@lists.boost.org Asunto: Re: [Boost-users] [multi_index] custom template class for handle multi_index map

hi joaquin,

a) here's a ready to use code snippet (see below). iisolating the problem by writing the snippet to the reuse of objects or using local temporary objects; i thought multi_index maps store copies of data objects, that where inserted. am i wrong?

As you correctly point out, multi_index_containers store copies of the data objects you pass them, that is right. Regarding the particular piece of code where you're getting the segfault: objClass01 = cClass01("elm01_01", 11); objClass00 = cClass00("elm00_01", 1); objClass01.setmapelm(objClass00); objClass00 = cClass00("elm00_02", 2); objClass01.setmapelm(objClass00); objClass01 = cClass01("elm01_02", 12); objClass00 = cClass00("elm00_03", 4); objClass01.setmapelm(objClass00); objClass00 = cClass00("elm00_04", 3); objClass01.setmapelm(objClass00); ptrClass02->setmapelm(objClass01); printf("START\n"); printf( "dat = %d\n", ptrClass02->getmapelm("elm01_01")->getdat() ); // seg_vault Please note that you've invoked ptrClass02->setmapelm only once to insert a copy of objClass01 when this object was storing "elm01_02" as its _nme member: no wonder that ptrClass02->getmapelm("elm01_01") returns NULL (which is what's happening) as you haven't stored any object with _nme=="elm01_01" inside it!! The segfault comes from trying to dereference NULL (you invoke ->getdat() on it). Now, if you insert two objects into ptrClass02 as follows: objClass01 = cClass01("elm01_01", 11); objClass00 = cClass00("elm00_01", 1); objClass01.setmapelm(objClass00); objClass00 = cClass00("elm00_02", 2); objClass01.setmapelm(objClass00); ptrClass02->setmapelm(objClass01); // insert objClass01 objClass01 = cClass01("elm01_02", 12); objClass00 = cClass00("elm00_03", 4); objClass01.setmapelm(objClass00); objClass00 = cClass00("elm00_04", 3); objClass01.setmapelm(objClass00); ptrClass02->setmapelm(objClass01); // inserted object different than above you won't get any segfault: ptrClass02's internal multi_index_container will have two objects with _nme=="elm01_01" and _nme=="elm01_02", respectively. Not sure whether this is the configuration you had in mind, though. There is one more issue that has caught my eye: From the way you've defined cClass01: class cClass01 : public ... { private: STRING _nme; UINT32 _dat; public: cClass01() {} cClass01(STRING nme, UINT32 dat) { this->_nme=nme; this->_dat=dat; this->_scNme = &(this->_mpMap.get<0>()); this->_scDat = &(this->_mpMap.get<1>()); } you're relying on _scNme and _scDat members to point to the tCanMap object cClass01 derives from: but you failed to guarantee this invariant when copying or assigning cClass01's: cClass01 x(...); cClass01 y(...); // y->_scNme points to y's tCanMap base object x=y; // wrong!! x->_scNme points to y's tCanMap base object Default copy ctor and assignment operator will just copy the _scNme and _scDat pointers along, which is not what you want. You have the same problem with cClass02. I've attached a modified version of your code with correct copy ctors and assignment operators for these classes.

b) why it isn't possible to write (with "?" is a class attribute or template parameter; see snippet):

template const ELM *tCanMap::getmapelm(STRING stNme) const { IT1 itr = this->_mpMap.get<?>().find(stNme); if ( itr != this->_mpMap.get<?>().end() ) { return &(*itr); } }

You have to insert a "template" keyword for the syntax to be correct, for instance: template const ELM *tCanMap::getmapelm(STRING stNme) const { try { IT1 itr = this->_mpMap.template get<0>().find(stNme); if ( itr != this->_mpMap.template get<0>().end() ) { return &(*itr); } } catch(...) {} return NULL; } Note the "_mpMap.template get<0>()" bit. Why "template" is needed in so-called dependent contexts is explained at http://www.comeaucomputing.com/techtalk/templates/#templateprefix The attached version also includes this fix for you to experiment with. BTW, having getmapelm() as a member of tCanMap as you tried in variants VERSION01 and VERSION02 is a superior solution as you can dispense with _scNme and _scDat, which pose problems of their own as commented above. Good luck with your project, Joaquín M López Muñoz Telefónica, Investigación y Desarrollo