From: William James <w_a_x_man@...>
Date: 2006-07-26T14:05:12+09:00
Subject: Re: Hash order bug?

Hal Fulton wrote:
> William James wrote:
> > No.  What language has ordered hashes?
> >
> > When you say
> > a = {:gr => false, :und => false, :det => true, :sta => false, :inv
> >   => false}
> > you are saying that you want to access the values in this fashion:
> > a[:und]
> > So the order in which they are stored is irrelevant.
> > If you want to access them this way
> > a[1]
> > then you set them up like this
> > [ false, false, true, false, false ]
> >
> > If you want to get a sequence of values in a certain order:
> > a.values_at( :und, :det, :sta )
>
> Your point is well taken. David Alan Black also argues eloquently
> against the concept of an ordered hash.
>
> But to me there is psychological weight in seeing a literal of this form:
>
>     {a=>b, c=>d, e=>f}
>
> There is in fact a syntactic order (which turns out to be meaningless
> internally).

In awk, such hash literals don't exisit; so when using that language,
one is less tempted to imagine that hashes are ordered.  Perhaps
we should think of the above as 'syntactic sugar' for

  a = Hash.new; a[a]=b; a[c]=d; a[e]=f

>
> > If you want to have your cake and eat it too, you could
> > use an association list:

Lisp also uses these.

> >
> >
> >>>as = [[:foo,22], [:bar,33], [:baz,44]]
> >
> > => [[:foo, 22], [:bar, 33], [:baz, 44]]
>
> [snip]
>
> That works fine, but I dislike the ugly syntax.

Maybe this is better.

  a = [ :foo,22,  :bar,33,  :baz,44 ].to_assoc

  a.assoc_set :bar, 99
  a.assoc_set :yes, -1
  a.assoc_del :baz

The prerequisite:

class Array
  def to_assoc
    f=nil  ;  partition{f=!f}.transpose
  end
  def assoc_set k, v
    (pair = assoc(k))  ?  self[ index(pair) ] = [k,v]  :  push( [k,v] )
  end
  def assoc_del k
    (pair = assoc( k ))  &&  slice!( index( pair ) )
  end
end

> 
> 
> Hal