Re: [pclasses-devel] Rewrite of Factory and PluginManager

[Prochnow, C. <pro...@se...>]

Am Freitag 27 Mai 2005 16:06 schrieb stephan beal:
> i'm going through the code now, and i've got a couple of concerns:
>
> #1:
> 	return static_cast<Factory&>(
> 		FactoryBase::instance(typeid(InterfaceT).name(),
> 			typeid(ContextT).name(), &createHook));
>
> typeid::name() OFFICIALLY provides undefined behaviour. Under gcc i have
> seen cases where typeid(T*).name() will return 2 different values,
> depending on whether T comes from a DLL or was statically linked.
>
> i spent months fighting with typeid::name() in s11n before finally
> dropping it altogether. i promise: it cannot be trusted at all.

that's not fully true. it's right that gcc emits type_info objects in the 
compilation unit where they are used, therefore you cannot simply compare two 
type_info objects, cause gcc does a address comparsion. but when using 
type_info::name().. the type_info object returns the mangled c++ ABI name of 
the type which _must_ be the same in all compilation units. 

see this: http://gnu.open-mirror.com/software/gcc/faq.html#dso

> Also, why do you need the static cast there?

the static_cast is needed to up-cast the FactoryBase to the Factory.

> #2:
>
> template <class TypeName>
> void registerType(const std::string& key, TypeName)
> {
> 		registerFactory(key, &typeCreateHook<TypeName>);
> }
>
> Why are we forced to pass a TypeName object? That is not at all suitable
> for large types. If i'm not mistaken, it also means that we cannot
> call registerType if TypeName is an incomplete type, where we can
> (i think) if the user does not pass a TypeName object.

i think we can make the second argument a const reference type.


Greetings,

Christian