[Where is type inherence ?] Re: Proposal: Simplified syntax for dealing with parameterized types. (original) (raw)

Neal Gafter neal at gafter.com
Mon Mar 23 14:51:21 PDT 2009

On Mon, Mar 23, 2009 at 2:12 PM, <rssh at gradsoft.com.ua> wrote:

BTW, are type inference proposal was submitted ? May be somebody have plans to  do this ? Or exists some consensus that this is not in project coin ?

Some links: http://gafter.blogspot.com/2007/07/constructor-type-inference.html // Neal, what you think now ?

Joe Darcy and Jeremy Manson are working on a detailed proposal. I still think it's a good idea.

Exists implementation http://weblogs.java.net/blog/forax/archive/2006/12/callmesanta.html

That's a different approach to shorten some of the same use cases. For those who aren't aware, Rémi was looking at using "final" instead of a type on the variable declaration. See the discussion and comments on my blog post for opinions about the two alternatives. Rémi's approach is much simpler (which is a good thing), and has the advantage that it is usable when the right-hand-side is something other than just a constructor invocation. On the other hand, the constructor-type-inference approach is useful in a few contexts where the object creation appears somewhere other than directly on the right-hand-side of an assignment. If I were to write up the approach Rémi implemented, the disadvantages section would read

"This is a departure from Java's previous philosophy that all named entities should be given an explicit type. Theoretically, it is possible to intentionally write difficult-to-understand code by using this construct where the types of the intermediate results are not clear from context. In addition, this creates language support for an idiom that departs from the advice in Effective Java (2nd edition) Item 52 (use interface types for variable declarations): local code might unintentionally depend on features of the implementation class rather than the abstract type. On the other hand, a few years of experience with this change in a Java-like language has yielded virtually uniform positive feedback, suggesting that these are not problems in practice."

More information about the coin-dev mailing list