In school I learned about inheritance, not composition. The usual: base "shape" class, then subclasses triangle, square, circle, etc. Then you can have a collection of "shape"s that know what they are. What I've read online indicates it's the same for most other people (who learned in a formal context, anyway). This way of doing things necessitates downcasting; at some point you're converting shape into a triangle or whatever, to access some specific method it has.
I'm not saying this is the right way of doing things (I don't think so), but many people expect this to be possible to do easily -- and it isn't. For them this is normal and not "something very exotic". I guess this is the source of the misunderstanding and the complaints.