Array of generics (JLS seems too restrictive)

Question

Hi everybody,
with some friends (in a lab) we currently develop a compiler generator
(a la SableCC i.e SLR, LR, LALR).
Because we start last june, we develop using jdk 1.5 and
generics. For me, i think it save us lot of time mainly because
Map<NonTerminal,Map<LRItem,Node>> is more readable
than Map :)
And this is my question : why CREATION of array of generics is unsafe ?
I don't understand why a code like below is tagged unsafe :
HashMap<State,Action>[] maps=new HashMap<State,Action>[5];
for(int i=0;i<map.length;i++)
  maps=new HashMap<State,Action>();
For me, the fact that JLS forbids array of generics creation
is too restrictive, only cases where type parameter are lost
should be tagged as unsafe.
Example :
Object[] array=maps; // unsafe
Object o=map;            // weird but unsafe because
                                      // Object[] o=(Object[])(Object)maps; must be unsafeWhat do you think about this ?
R�mi Forax

Answer 1

The question is why :
Future<Double>[] futures=new
Future<Double>[10000];
is not allowed. It seems safe !First of all, it would not be safe. The example below would exhibit exactly the same vulnerabilities with this assignment as it would with any of the allowed ones. This is because both the declared type and the runtime type would still be the same.
It seems the current implementation is not able to detect (and warn about) the unsafety of the assignment above. That's why you're not allowed to use it: allowing it would provide a very false sense of type-safety
and why :
Object[] o=futures;
is not tagged unsafe.
It's unsafe because you can write :
Object[] o=futures;
o[0]=new Future<String>(); // for the example, let
says Future is a class
R�mi ForaxThis is probably done to avoid redundancy. As far as I can see, in all cases where this could be unsafe, you have been notified about the lack of safety already when you first assigned to futures.