On Friday, 19 June 2015 at 15:55:48 UTC, Andrei Alexandrescu
wrote:
On 6/19/15 8:27 AM, Andrei Alexandrescu wrote:
On 6/19/15 8:01 AM, "Ulrich =?UTF-8?B?S8O8dHRsZXIi?=
<[email protected]>" wrote:
If opAssign is allowed, a major point of functional data
structures is
lost. Client code is so much better if rebinding is off the
table.
I have the same instinct but not enough rationale. What would
be an
example in favor of the argument?
Hard to come up with a convincing example here. Any large
function that creates a data structure
auto lst = SList!int(1, 2, 3);
features some non-trivial logic
if (lst.length % 2)
{
lst = lst.tail();
}
and produces whatever result
writeln(lst);
is much simpler to reason about if the variables are all const,
aka not assignable. The above is obviously weak. The only
convincing argument I know of is to use a language that enforces
immutability and experience the lift of a mental burden.
Erlang uses single assignment, a variable can only be bound once.
The obvious counter argument seems to be performance.
FWIW Scala's immutable containers are all assignable. -- Andrei
Not knowing scala at all, to me this looks insane:
http://www.scala-lang.org/api/2.11.5/index.html#scala.collection.immutable.Vector
There is a little too much.
