[Checkers] Arrays and checkers

[Matt Papi]

The problem seems to occur because type validator uses fromTypeTree() 
for array types but getAnnotatedType() on the element of the erased 
underlying type. fromTypeTree() doesn't add implicit annotations (and it 
shouldn't) and getAnnotatedType() does (and it should); the interned 
checker adds an implicit @Interned for the type "Class", so the "use" 
type is an array of Class but the element type is an array of @Interned 
class.

Thinking about the fix raises a design question: we need a way to get 
the annotated type from a type tree *with* implicit annotations. My 
proposal: add fromTypeTree(boolean) that calls out to fromTypeTree() and 
if its parameter is true, calls annotateImplicit(), and use that from 
the type validator. [I still think having getAnnotatedType() accept type 
trees is a bad idea, since if checker (!= framework) authors are calling 
getAnnotatedType() on type trees, they're probably doing something wrong.]

Also, a side issue: I didn't realize that the type validator issued 
javac errors. If there's a type floating around that is invalid 
(according to the type validator), is the cause ever the input program 
and not a problematic checker? If invalid types are only possible via 
checker bugs, the validator should probably throw some kind of error or 
exception instead of issuing a compile error.

- Matt

Michael Ernst wrote:
> Mahmood-
> 
> I added another failing test case for arrays.
> 
> There is another failure that I can't seem to reproduce as a small test
> case.  To reproduce it, see
>   ~mernst/wisdom/build/build-freepastry
> and follow the directions.
> 
> Also, this error message is not comprehensible:
> 
> /DS/home-0/mernst/tmp/try-freepastry/pastry/src/rice/p2p/util/XMLObjectOutputStream.java:656: invalid type, type need to be a supertype of the required type.
> found   : java.lang.Class
> required: @checkers.quals.Interned java.lang.Class
>     return (Class[]) v.toArray(new Class[0]);
>                  ^
> 
> There's an English error ("need" should be "needs"), but more seriously
> there are 4 uses of "type" in the message, and it's not clear what any of
> them refer to.
> 
>                     -Mike