[erlang-questions] closure on function argument

[Richard O'Keefe]

On May 2, 2010, at 8:56 AM, Robert Virding wrote:

> The variables in the arguments of a fun, Y in your example, are always
> "fresh" variables and never imported from the funs environment.

For me personally, this is perhaps THE most confusing thing
about Erlang.

I have *no* problem with the same rule in ML or Haskell because they
apply it *consistently*: every pattern everywhere introduces fresh
variables, it is never ever possible in those languages for a
variable occurrence in a pattern to be an applied occurrence; all
pattern variables are binding occurrences.  This applies to OCaml and
Clean as well.

But Erlang does NOT apply this rule consistently.
Variables in a pattern of a 'case' are *not* fresh, they *may* be
applied occurrences.  And even in a fun, *some* variable occurrences
may be applied ones:  fun (Y,Y) -> ... end.

It's especially confusing because variables in fun bodies and guards
are imported as usual, so they are clearly in scope.

It would be *so* nice if ALL variables in Erlang followed exactly
the same rules.

Here's a little example.  Suppose someone has
	case E of P -> B end
but doesn't want any variable bindings in B to be exported.
The obvious thing to do is to rewrite it as
	(fun (P) -> B end)(E)
but because Erlang is inconsistent here, that does not work.
(You can patch around it with fun, case, and a new variable,
but that has problems of its own.)

I have never been able to see any merit in this rule; its effect
on me is to make it harder to write funs because I always have to
watch out for cases where the expected thing won't happen and it
has to be patched around.