How are Java generics different from C++ templates? Why can't I use int as a parameter?

Java generics are so different from C++ templates that I am not going to try to list the differences here. (See What are the differences between “generic” types in C++ and Java? for more details.)

In this particular case, the problem is that you cannot use primitives as generic type parameters (see JLS §4.5.1: "Type arguments may be either reference types or wildcards.").

However, due to autoboxing, you can do things like:

List<Integer> ints = new ArrayList<Integer>();
ints.add(3); // 3 is autoboxed into Integer.valueOf(3)

So that removes some of the pain. It definitely hurts runtime efficiency, though.

The reason that int doesn't work, is that you cannot use primitive types as generic parameters in Java.

As to your actual question, how C++ templates are different from Java generics, the answer is that they're really, really different. The languages essentially apply completely different approaches to implementing a similar end effect.

Java tends to focus on the definition of the generic. That is, the validity of the generic definition is checked by only considering the code in the generic. If parameters are not properly constrained, certain actions cannot be performed on them. The actual type it's eventually invoked with, is not considered.

C++ is the opposite. Only minimal verification is done on the template itself. It really only needs to be parsable to be considered valid. The actual correctness of the definition is done at the place in which the template is used.

They are very different concepts, which can be used to perform some, but not all of the same tasks. As said in the other responses, it would take a quite a bit to go over all the differences, but here's what I see as the broad strokes.

Generics allow for runtime polymorphic containers through a single instantiation of a generic container. In Java, all the (non-primitive) objects are references, and all references are the same size (and have some of the same interface), and so can be handled by the bytecode. However, a necessary implication of having only instantiation of byte code is type eraser; you can't tell which class the container was instantiated with. This wouldn't work in c++ because of a fundamentally different object model, where objects aren't always references.

Templates allow for compile time polymorphic containers through multiple instantiations (as well as template metaprogramming by providing a (currently weakly typed) language over the c++ type system.). This allows for specializations for given types, the downside being potential "code bloat" from needing more than one compiled instantiation.

Templates are more powerful than generics; the former is effectively another language embedded within c++, while to the best of my knowledge, the latter is useful only in containers