Notes on implementing concise calls to constructors with type parameters

Alex Buckley Alex.Buckley at Sun.COM
Fri May 15 10:05:37 PDT 2009

Ulf Zibis wrote:
>>> It should have stopped them more cold in experiencing a consequently 
>>> occurring CCE at runtime.  ;-)
>> Which they knew might happen because of the unchecked warning at the 
>> assignment, giving them a chance to identify potential heap pollution 
>> and upgrade certain uses of raw types at their own pace.
> With this same argument, following code should similar output a warning 
> instead of a compile time error:
> int i;
> ...
> if (i == 7) {  // error: variable not initialized
>   ...
> }

No, it shouldn't. An unchecked warning indicates a hole is being punched 
in the generic type system. Assigning a List to a List<..> is not 
statically type correct (List is not a subtype of List<..>) and can lead 
to a loss of dynamic type safety. But migration compatibility desires 
the assignment to be legal, so an error is impractical. A warning is the 
only thing left.

(For the same reason, you ought to get a warning every time you assign 
into an array, but there's a level of realism which prevents that.)

Whereas, the assignment type system - where the "type" of i is ASSIGNED 
or UNASSIGNED - is much more conservative. No holes may be punched 
through it. Either code is statically checkable (==> dynamic type 
safety, all variables definitely assigned at runtime), or it's not. 
Error, or not.


More information about the coin-dev mailing list