[erlang-questions] A modest proposal to eliminate tuples

Fred Hebert <>
Wed Aug 27 22:39:28 CEST 2014


On 08/27, Chris Pacejo wrote:
> Tuples are clearly unnecessary and a wart in Erlang's design.
> Consider all the operations tuples provide:
> 
> Construction: X = {A,B,C}
> Destruction: {A,B,C} = X
> Pattern matching: case X of {foo,B,C} -> ... end
> 
> Note now that these operations are provided equally by lists:
> 
> Construction: X = [A,B,C]
> Destruction: [A,B,C] = X
> Pattern matching: case X of [foo,B,C] -> ... end
> 

This proposal lacks foresight. Lists can, for example, be implemented
using anonymous functions:

  -module(mod).
  -compile(export_all).

  cons(H, T) -> fun(F) -> F(H, T) end.
  hd(L) -> L(fun(H,_) -> H end).
  tl(L) -> L(fun(_,T) -> T end).

Which runs as:

  1> L = mod:cons(3, mod:cons(2, mod:cons(1, []))). %% equiv to [3|[2|[1|[]]]]
  #Fun<mod.0.125116012>
  2> mod:hd(L).
  3
  3> mod:hd(mod:tl(L)).
  2

Clearly other operations on lists are only sugar, and should be reduced.

We can also use church numerals for our purposes:

  zero() -> fun(_F) -> fun(X) -> X end end.
  succ(N) -> fun(F) -> fun(Z) -> F((N(F))(Z)) end end.
  one() -> succ(zero()).
  two() -> succ(one()).
  %% This one is a cheat to help our feeble minds
  to_int(F) -> (F(fun(X) -> X+1 end))(0).

Which runs as:

  4> mod:to_int(mod:two()).
  2
  5> mod:to_int(mod:one()).
  1
  6> mod:to_int(mod:succ(mod:succ(mod:two()))).
  4

Now lambda calculus and other similar works tell us how we can implement
entire systems that way. All we need are processes. But wait, systems
like Pi Calculus can be made somewhat equivalent to Erlang, and that one
is also turing equivalent, isn't it?

Without digressing, I'm sure this mailing list will be able to
appreciate how having only one data type -- functions -- makes for a
good base. You now only need to optimize one data type and *the entire
system* should be faster. It's tempting isn't it?

Hrm.

Regards,
Fred.



More information about the erlang-questions mailing list