On Sat, Jun 7, 2008 at 10:53 AM, Joel FALCOU <joel.falcou@u-psud.fr> wrote:
the meta_class specializations for the templates like pair, vector, tuple, etc. are currently hand-coded, and using some common boilerplate to handle the template param names, but I'm working on some registering macros similar to Joel's. see for example: http://svn.boost.org/svn/boost/sandbox/mirror/boost/mirror/meta_classes/std_...
Seems mirror has a far broader usage than my humnle identify<> ! I like the possibility of reflecting any parts of a type adn I also understand the gap between this task and mine.
My macro in fact really simple, I "cheat" by using MPL_AUX_LAMBDA_ARITY to detect the number of tempalte parameters and generate the proper specialisation. When it fails, I use a small preprocessor macro using sequence to generate the proper display.
Other difference is my use of std::string as a return type. Seems you have a complex system to compute string length and use char*. Not sure which one is the best but I'm willing to hear the rationale behind this choice.
Well there is no bulletproof ;) rationale behind this. I've made some performance tests with both char* buffer and with ::std::string and char* buffer did better. When formatting the type name of a complicated template there is a lot of string copying and realocation. Calculating the type-name string lenght was not a big issue. The problem with the current implementation is that there are some memory leaks at the program exit but they are not cumulative, but still, fixing this is on my TODO list ;)
-- Joel FALCOU Research Engineer @ Institut d'Electronique Fondamentale Université PARIS SUD XI France
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
-- ________________ ::matus_chochlik