Mirror of forum.nim-lang.org

34 :: OOP in Nimrod looks strange, value semantics

• exhu (orginal) [2012-06-13T18:29:09+02:00] view original

  OOP looks nasty because multimethods get lost when there are many of them. Also I don't understand why TStream uses procvar fields rather than multimethods.

  Could anyone explain the idea?

  Also as far as I understand TTable is expensive to assign, e.g.

  var t = toTable([("abc", "val1"), ("bbb", "val2")]) var t2 = t # is expensive

  So which is the way to declare a reference to it?

---

• Araq (orginal) [2012-06-13T20:11:16+02:00] view original

  > Multimethods get lost when there are many of them? Isn't that the same as saying "dynamic binding gets confusing"? Well yes, non local control flow is confusing... That's hardly news.

  Streams use proc vars because methods weren't ready back then and I don't feel like changing working code when there are much more important things left to do. Also dead code elimination doesn't work yet with multi methods, but it works with proc vars.

  The convention is to have newT that returns a ref T to get reference semantics and initT that returns T to get value semantics. Currently you need to explicitly derefence when passing a ref T to a routine that expects T which is ugly. It's planned to make this an implicit conversion for convenience.

  In general value semantics can be used to avoid additional indirections and thus expensive memory accesses (when you know what you're doing) at the expense of some copying. I also chose value semantics for consistency with the rest of the language. Copying is also easier to reason about than aliasing.

---

• exhu (orginal) [2012-06-13T20:35:53+02:00] view original

  Ok, the first item seems confusing to me because I've got used to java/c++ style of declaring methods, so probably it's a matter of a habbit.

  Thanks for explaining the rest, i'll keep learning by experimenting further because the language is very interesting and many ideas are close to what I want (I'm looking for a good alternative to C++/Java).

---

• Araq (orginal) [2012-06-13T21:03:18+02:00] view original

  Well if it's only syntax that concerns you, try to implement the class macro I mentioned in some other thread. It'll be a welcome addition to the stdlib. ;-)

---

• leledumbo (orginal) [2012-06-14T06:05:32+02:00] view original

  Ok, the first item seems confusing to me because I've got used to java/c++ style of declaring methods, so probably it's a matter of a habbit.

  Actually Nimrod isn't the first language to implement methods this way. Oberon, Component Pascal, Object Pascal (via class helper), Fortran (since it supports OOP, 2003 version if I'm not mistaken), Lua, JavaScript and all other prototype based OOP languages does it similarly (well... more or less).

---