[cousin_it]

This is similar to the feedback Dave Moon gave to PG's previous language, Arc, more than a decade ago. http://www.archub.org/arcsug.txt

Representing code as linked lists of conses and symbols does not lead to the fastest compilation speed. More generally, why should the language specification dictate the internal representation to be used by the compiler? That's just crazy! When S-expressions were invented in the 1950s the idea of separating interface from implementation was not yet understood. The representation used by macros (and by anyone else who wants to bypass the surface syntax) should be defined as just an interface, and the implementation underlying it should be up to the compiler. The interface includes constructing expressions, extracting parts of expressions, and testing expressions against patterns. The challenge is to keep the interface as simple as the interface of S-expressions; I think that is doable, for example you could have backquote that looks exactly as in Common Lisp, but returns an <expression> rather than a <cons>. Once the interface is separated from the implementation, the interface and implementation both become extensible, which solves the problem of adding annotations.

This paragraph contributed a lot to my understanding of what "separating interface from implementation" means. Basically your comment is spot on. Instead of an executable spec, there should be a spec that defines as much as users need, and leaves undefined as much as implementors need.

[Viliam1234]

In Clojure, some data structures are implemented as "sequable", which means that they are not implemented as sequences, but they can be converted to sequences if needed. Functions for head and tail are also defined for these structures, for example by internally converting them to sequences first. That means that any function that works with sequences can work with these structures, too.

This seems to me like the proper way to have "separation of interface from implementation" and "everything is a list" at the same time. Yeah, everything is an (interface) list, but not necessarily an (implementation) list.

[nikisweeting]

Python and Rust both have similar "duck"y features where you can Impl iteration or add the necessary __special__ methods to get iteration features.

[mncharity]

An object type can implement interface types, as you say.

Object type, and identity, can also be decoupled from implementation type. Dynamically, statically, or with advice. V8 javascript numbers and arrays can have several different underlying implementations, which get swapped depending on how the thing is used (and with extensions, you can even introspect and dispatch on them).

Typing can also be fine grained. So things can narrowly describe what they provide, and algorithms what they require. "I require a readable sequence, with an element type that has a partial order, and with a cursor supporting depth-one backtracking".

Object identity can also be decoupled from type. Ruby's 'refinements' permits lexically scoped alternative dispatch, so an object can locally look like something else. That, and just a few other bits, might have made the Python 2 to 3 transition vastly less painful - 'from past import python2'.

We are regrettably far from having a single language which combines the many valuable things we've had experience with.