forked from Alepha/Alepha
Document the aggregate analysis algorithm.
(I still don't actually deal with the weird case of bases also being members... but that's fairly rare.)
This commit is contained in:
@ -23,8 +23,43 @@ namespace Alepha::Hydrogen::Reflection
|
||||
|
||||
using Meta::overload;
|
||||
|
||||
// Basic methodology here: The number of members in an aggregate is equal to the number of initializer parameters
|
||||
// it takes less the number of empty base classes it has.
|
||||
/*!
|
||||
* Basic methodology here.
|
||||
*
|
||||
* The number of members in an aggregate is equal to the number of initializer parameters it takes less
|
||||
* the number of empty base classes it has. In simple terms, this would be `init_terms< T > - empty_bases< T >`,
|
||||
* However, it's not that simple. To do that the easy way, one might need get `std::tuple< InitTerms... >` and then compute
|
||||
* which terms were bases. Now that gets a bit complicated, as in C++ one can't just directly get a tuple of initializer
|
||||
* arguments (portably) to scoop out the arguments and analyze them one-by-one. One can, however, constrain the arguments
|
||||
* one-by-one in templates. Those constraints cannot directly leak out the types they conclude, as that requires side
|
||||
* effects. (Yes, template-friend-injection can be used here, but these mechanisms are extremely delicate. They're
|
||||
* not really portable. Further, the way that constraints get instantiated for matching is prone to complications.)
|
||||
*
|
||||
* Instead, a side-stepping approach is required. It's trivial to ask: "Can this object be constructed from these
|
||||
* N adaptive types, where the first one is constrained to be a base class of your object's type?" If yes, then
|
||||
* this proves a (likely) empty base. One can just recursively iterate through more an more constrained adaptive
|
||||
* types until the first non-base type is reached. At this point, there are no more than that many base classes.
|
||||
*
|
||||
* There may actually be fewer base classes, however. Consider:
|
||||
*
|
||||
* ```
|
||||
* struct SneakyBase {};
|
||||
*
|
||||
* struct Complicated : SneakyBase
|
||||
* {
|
||||
* SneakyBase member;
|
||||
* };
|
||||
* ```
|
||||
* In that case, a constrained adaptable argument would see two base types. Here is where a bit of C++ trivia and
|
||||
* knowledge comes into play. C++ forbids repetition of a base class's type. Therefore the sequence of base classes
|
||||
* cannot have repeats. The solution is to perform a nested exploration of instantiations of `checker` types which
|
||||
* has each descent disable casting to `std::is_base_of_v` types which already have been expanded. Thus whatever
|
||||
* count this expands to, it will be the correct empty-bases count. Then that count can be subtracted from the
|
||||
* initializer list count.
|
||||
*
|
||||
* Note that this will not work with types that have non-empty bases, but those types cannot be decomposed,
|
||||
* anyhow. Such types cannot have C++17 reflection performed on them.
|
||||
*/
|
||||
|
||||
// The basic adaptable argument. Because it pretends to be anything, it can be used as a parameter in invoking
|
||||
// any initialization method.
|
||||
|
Reference in New Issue
Block a user