There isn't such thing as the only true and single way of doing OOP.
Also without OOP, good luck doing GUI programming in C++. Even Motif and GTK+ are OOP based, in C, because immediate UIs suck for anything beyond simple game UIs.
More specifically, C++-style OO couples (1) in-memory layout of types, (2) behaviour of objects, (3) encapsulation boundaries, and (4) the hierarchical "is-a" relationship. In Cforall, (1) and (3) are handled by the usual C mechanisms of struct types and translation-unit boundaries [we go to great effort to maintain C-style separate compilation, though we will likely build a better namespace/module system at some point], (2) by traits, and (4) by a yet-to-be-implemented system of RTTI for specific "inherited" types [likely when the undergrad RA who started our exceptions implementation comes back as a Master's student this fall].
As you mention, many use-cases (including widgets in GUIs, but also abstract syntax trees in compilers) require some sort of heterogeneous collection type to work properly, and for this OO is the right model. We just don't want to force the entirety of that model on every use case, so we can just take traits when we want to write a polymorphic function over related types or an ADT like "Set" which requires its keys be sortable, or just take RTTI for exception handling, when the set of exception types we're handling is known statically, but we need to determine if the thrown exception inherits from one of the caught types.
[1] https://golang.org/doc/faq#Is_Go_an_object-oriented_language
It doesn't matter what libraries one uses, they don't remove the OOP model out of JavaScript.
Additionally all those react ideas already existed in Smalltalk as well.