[ruby-core:26495] Re: HashWithIndifferentAccess to core

On Tue, Nov 3, 2009 at 6:12 AM, Urabe Shyouhei <[e-mail]> wro=
te:
> Hello,
>
> I got a pull request for a HashWithIndifferentAccess implementation to me=
rge
> into core. =A0I think it's almost OK to make it so (except few minor bugs=
 that can
> easily be handled). =A0How about it? =A0I'm also attaching a patch file f=
or people who
> are not used to git (mainly nobu).

> +++ b/test/ruby/test_hash.rb

> +class TestStrHash < Test::Unit::TestCase
> + =A0def setup
> + =A0 =A0@strings =3D { 'a' =3D> 1, 'b' =3D> 2 }.strhash
> + =A0 =A0@symbols =3D { :a =A0=3D> 1, :b =A0=3D> 2 }.strhash
> + =A0 =A0@mixed =A0 =3D { :a =A0=3D> 1, 'b' =3D> 2 }.strhash
> + =A0 =A0@fixnums =3D { =A00 =A0=3D> 1, =A01 =A0=3D> 2 }.strhash
> + =A0end

> +
> + =A0def test_keys
> + =A0 =A0assert_equal ["a", "b"], @strings.keys
> + =A0 =A0assert_equal [:a, :b], @symbols.keys
> + =A0 =A0assert_equal [:a, "b"], @mixed.keys
> + =A0 =A0assert_equal [0, 1], @fixnums.keys
> + =A0end

So judging by the tests, and without going through the implementation
in detail. It appears that this proposal goes to lengths to preserve
the class of a key in the hash.

I'm not sure that this is a real requirement.  I contrast consider the
current HashWithIndifferentAccess(HWIA) in Rails/ActiveSupport, here's
a snippet from a rails console irb session:

=3D> {}
>> h[:a] =3D 1
=3D> 1
>> h['b'] =3D 2
=3D> 2
>> h.keys
=3D> ["a", "b"]

HWIA converts symbol arguments to keys in the access methods, and
stores the keys internally as strings.

I think picking a single internal representation for the keys is
simpler, and probably a bit faster.  The main decision is whether to
pick strings or symbols for that internal representation.

ActiveSupport picked strings, primarily from what I understand over a
concern that symbols once interned can't be GCed and the key values in
the Rails use cases come from arbitrary user inputs (e.g. query
parameters in URIs). So converting such strings to symbols could be
the basis of a DOS attack.  As a coincidencs I just posted about this
in reply to a thread on the ruby-lang group.

I think that if the key class is preserved there are other
ramifications to ponder, for example what should h.keys be after

h =3D  { 'a' =3D> 1, 'b' =3D> 2 }.strhash
h[:a] =3D 3

I think that any of  ['a', 'b'] or [:a, 'b'] or ['b', :a]  all of
which might seem to be plausible answers, would surprise someone, and
that the POLS would be better preserved by conversion to a consistent
class of key.

If the current proposal were accepted, I'm not sure that it would be
useful to Rails as a replacement implementation of HWIA, but I'm
hoping that Yehuda or one of his cohorts would comment on that.


--=20
Rick DeNatale

Blog: http://talklikeaduck.denhaven2.com/
Twitter: http://twitter.com/RickDeNatale
WWR: http://www.workingwithrails.com/person/9021-rick-denatale
LinkedIn: http://www.linkedin.com/in/rickdenatale