Hi --
Hi --
that will change in newer version of ruby, eg ruby1.9
irb(main):015:0> RUBY_VERSION
=> "1.9.0"
irb(main):016:0> a=Hash.new
=> {}
irb(main):017:0> a[1]=2
=> 2
irb(main):018:0> a[2]=2
=> 2
irb(main):019:0> a[3]=4
=> 4
irb(main):020:0> a.select { |key,value| value > 2 }
=> {3=>4}
irb(main):021:0> a.reject { |key,value| value <= 2 }
=> {3=>4}
kind regards -botp
Thanks.
So this is just some sort of artefact/legacy code that never got changed?
You can actually make a case that select and reject are not exactly
symmetrical operations. Imagine a line of people:
Joe John Joan David Jim Jenny Jeff Matz
If I tell everyone in the line whose name does not begin with J to
step backwards (reject), the original line is smaller but it's still
the same line.
If I tell everyone whose name *does* begin with J to step forwards
(select), I've got a new line of J people.
I'm having a hard time making the connection between this analogy and the
methods.
In the first case one could say that we end up with two lines, the
'original' one and the line of rejects.
But first of all, x.reject leaves x alone so the original 'line' is
unchanged.
And why can't we see your select example exactly the same way, except with
the resultant line of 'selects' just closer to you.
"Same line" and "new line" don't necessarily map to "same object" and
"new object" in Ruby (since the post-reject hash is a different hash).
But it suggests that there's a difference, arguably, between select
and reject, in terms of the formal disturbance of the object, which in
turn makes it easier to understand why select would return objects in
a different container, while reject would leave the container in the
same form but just contain fewer things.
Except that both select and reject leave the original container the same.
The analogy might be more apt for reject! and select! (if the latter method
existed).
It's specifically not about the exact objects, though. As I said,
"same line" in this model doesn't map to "same object". Rather, the
analogy has to do with types of structure or container.
Here's another, perhaps better analogy. You've got a shelf of books,
and you've got an empty box. You select some books from the shelf, and
put them in the box. Now you've got two structures: a shelf, and a
box.
If you were describing this operation, you'd say: "I selected some
books and put them in a box." The box of books is the result of the
operation, and is structurally different from the shelf. The point is
not that this is the only way it can be done, but that it *can* be
done this way. It's reasonable to select from a collection into
something that is not only a different collection, but a different
container type.
Now, if you do it reject-wise, you take the books you don't want and
throw them out the window. The box never enters into it. You're not
selecting, so you don't care about observing or measuring the books
that you're removing from the shelf.
And the shelf is still a shelf. There's no new type of container
involved, just the same type of container with different contents. The
reject operation does not imply the transfer of objects to a different
type of container. Also, the fact that Ruby can do this in terms of
duplicate objects, rather than in place, isn't relevant. The change or
conservation of the container type is what's important.
That's all I'm saying
And thereafter, to make sense of the new 1.9 behavior, all you have to
do is decide that instead of a box, you're putting the selected books
on another shelf.
Again, this has nothing to do with the identity of objects vs. copies.
It's about the proliferation (or not) of structures.
The real change, it seems to me is that pre 1.9 the Ruby enumerator methods
had/have a preference for returning arrays rather than an instance of the
same class as the receiver, whereas 1.9 seems to be shifting to a preference
for returning an instance of the same class as the receiver where that makes
sense.
Yes, that's clearly the change. I wasn't talking about the change,
though; I was suggesting a model for understanding why the pre-change
way might not have been as irrational as has sometimes been suggested.
It's all moot as of 1.9.
David