hg: jdk7/tl/langtools: 2 new changesets

[Alex Buckley]

Bunch of spec issues here.

The text claiming casting conversion is more inclusive than assignment 
conversion is not new, and is only a discussion note, but I agree it's 
false - a casting conversion does not allow 
narrowing-primitive-then-boxing whereas an assignment conversion 
sometimes does.

Rather than complicating the note with if's and but's about the 
assignment/casting relationship, I will remove the note for JLS7.

The compiler was compliant in rejecting (Byte)0 in the past, and it's 
compliant in rejecting it now. We are aware of improvements possible 
when casting primitive values, but they will not be specified for SE 7.

Regarding the "compile-time legality" clauses, they are a centralization 
and rewording of existing clauses. I was too strong with "always ... 
without error", since exceptions may occur as in 5.2. I softened the 
wording so an expression of primitive type _may_ undergo casting 
conversion to a reference type without error, by boxing conversion. 
(These clauses are an example of normative text which should not be 
tested, since it's an overview of rules elsewhere.)

Alex

On 6/7/2011 5:46 PM, Neal Gafter wrote:
> One more thing: the casting conversion from "0" to "Byte" is also 
> required to be allowed by the draft JLS by this curious sentence:
> 
> The compile-time legality of a casting conversion is as follows:
> • ...
> • An expression of a primitive type can always be [sic] undergo casting 
> conversion to a reference type without error, by boxing conversion.
> 
> Since 0 is an expression of primitive type and Byte is a reference type, 
> the former can "always" undergo a casting conversion to the latter 
> "without error".
> 
> On Tue, Jun 7, 2011 at 5:38 PM, Neal Gafter <neal at gafter.com 
> <mailto:neal at gafter.com>> wrote:
> 
>     Regarding the fix for "bug" 7041730... I was under the impression
>     that recent changes to the language and compiler were intended to
>     allow assignment conversions as casting conversions too.  The
>     conversion from the constant zero (0) to a Byte is among the
>     assignment conversions.  Draft SE7 JLS 5.2 (Assignment Conversion) says
> 
>     In addition, if the expression is a constant expression (§15.28) of
>     type ... int ... a narrowing primitive conversion followed by a
>     boxing conversion may be used if the type of the variable is ...
>     Byte and the value of the constant expression is representable in
>     the type byte...
> 
>     And 5.5 (Casting Conversion) says
> 
>     ...casting conversions are more inclusive than assignment or method
>     invocation conversions: a cast can do any permitted conversion other
>     than a string conversion or a capture conversion (§5.1.10).
> 
>     Yet this change claims to "fix" the bug 7041730, which requests that
>     the casting conversion from "0" to "Byte" be illegal; that violates
>     this JLS quote.
> 
>     Cheers,
>     Neal
> 
> 
>     On Wed, May 11, 2011 at 4:16 AM, <maurizio.cimadamore at oracle.com
>     <mailto:maurizio.cimadamore at oracle.com>> wrote:
> 
>         Changeset: 95fc7fd39be2
>         Author:    mcimadamore
>         Date:      2011-05-11 13:12 +0200
>         URL:      
>         http://hg.openjdk.java.net/jdk7/tl/langtools/rev/95fc7fd39be2
> 
>         7041730: Regression: compiler accepts invalid cast from int to Byte
>         Summary: Implementation of cast conversion rules between
>         primitive and boxed types is too liberal
>         Reviewed-by: jjg
> 
>         ! src/share/classes/com/sun/tools/javac/code/Types.java
>         ! test/tools/javac/types/BoxingConversionTest.java
>         ! test/tools/javac/types/CastTest.java
> 
> 
>