Comments on: Objects Are Not Abstract Data Types

By: IL

IL — Fri, 15 Jan 2010 22:28:17 +0000

Sounds like erlang behavior but no dynamic dispatching.

By: Yo

Yo — Fri, 15 Jan 2010 02:12:35 +0000

Scala solves all of this via traits.

By: Rick Minerich's Development Wonderland : F# Discoveries This Week 12/21/2009

Rick Minerich's Development Wonderland : F# Discoveries This Week 12/21/2009 — Mon, 21 Dec 2009 17:11:28 +0000

[...] Stuart Sierra on why Objects are not Abstract Data Types [...]

By: William Cook

William Cook — Fri, 18 Dec 2009 19:20:23 +0000

I wrote the article “On Understanding Data Abstraction, Revisited” mentioned above. I also responded to Liskov when she presented her talk at OOPSLA a few days before my presentation. In my article I completely refute Liskov’s negative comments about object-oriented programing. I identify a notion of pure OO in which only *interfaces* are used as types. In this approach, there is no conflation of “concrete representation of a type with the interface used to access it”. The problem comes from using classes as types. The overall point of my article is:
* CLU introduced the idea of user-defined abstract data types (ADT)
* Objects are NOT abstract data types (they are fundamentally different)
* Java supports a blend of ADT and object-oriented programming
(but the ADT support is weak)
* Abstract data types and objects both have advantages and disadvantages
Adding ADT forms to Java will improve some things, but make other things worse. Be careful!
* Haskell type classes are a form of abstract data type (but without much information hiding)
Once you get the pure form of both concepts clear, then the various hybrids are easy to recognize. Unfortunately, most people only know Java, which is a mixed up hash of both forms, so using Java as your foundation leads to confusion. I’m sorry my article is not quite as accessible as I had hoped it would be. Since there seems to be so much interest in the topic, I’m thinking about writing something that is easier to digest.

By: banq

banq — Wed, 16 Dec 2009 08:04:48 +0000

just like Mixin in AOP

By: Stuart

Stuart — Tue, 15 Dec 2009 18:06:19 +0000

James Swift: Yes, I’ve read that. Rich Hickey mentioned it in connection with his work on datatypes/protocols in Clojure.

By: Stuart

Stuart — Tue, 15 Dec 2009 18:02:15 +0000

Noel Grandin wrote: “So this proposal is basically a restricted Java class and a Java interface?”

Not really restricted, because you can do all the same things you can do with Java classes/interfaces. Just a different approach, and more flexible because you can modify datatypes/protocols dynamically.

By: Stuart

Stuart — Tue, 15 Dec 2009 17:58:50 +0000

polypus wrote: “any idea what the ETA is on this?”

Don’t hold your breath. I’m guessing it will be 2-3 months from now.

By: polypus

polypus — Tue, 15 Dec 2009 17:55:33 +0000

“Not yet. The implementation of datatypes and protocols is still being developed. I’m sure that once it is finished there will be a raft of articles describing how to use it.”

any idea what the ETA is on this?

By: Noel Grandin

Noel Grandin — Tue, 15 Dec 2009 07:48:12 +0000

So this proposal is basically a restricted Java class and a Java interface?

Which is not a surprising choice – James Gosling has said that he would have liked to make Java’s original type system interface based, but they ran out of time to do the necessary research.