From: Clemens Hintze <c.hintze@...>
Date: 2002-03-28T01:49:34+09:00
Subject: Re: Hash.new(Hash.new) doesn't use Hash.new as default value

In <a7sjhc$k1c$1@sunsite.dk> "Jonas Delfs" <jonas@NOSPAMdelfs.dk> writes:

> Hi -

Hi too,

> In the documentation for the 'new' method in the Hash class in Programming
> Ruby it says that 'anObject' will be used as default value, but consider the
> following example:

> var = Hash.new(Hash.new)

The documentation is true. It says that *anObject*, i.e. the reference
to an (better *one* here) object is used as default value!

So a new Hash will be created. That *one* instance will be passed to
Hash::new as default value to be used, if a certain key does not exist.

> var['index1'] = 'value1'
> var['index2']['index22'] = 'value22'

Because 'index2' does not exist, the instance serving as default value
will be returned. *That* instance will get sent the []= message and
therefore insert a association between key 'index22' and 'value22'.

The default Hash instance contains now { 'index22' => 'value22' }

> var['index3']['index32'] = 'value32'

And here happens the same again. So after this line the Hash instance
used as default value contain now: 

  { 'index22' => 'value22', 'index32' => 'value32' }

> var['index4'] = 'value4'
> p var['donotexist']

> Why does "p var['donotexist']" output {"index32" => "value32", "index22" =>
> "value22"} - and _not_ just {} (a new Hash)?

You have passed a default *value* to Hash::new above. That value
happens to be the reference to a new created Hash instance. The
reference of this one Hash instance is returned every time you try to
access a key that does not exist.

Look this:

  puts( var['index3'].id == var['donotexist'].id )

you see? Every time the *same* Hash instance is returned to you if you
try to access a key that does not exist.

In Ruby 1.7.x it is possible to do this:

  var = Hash.new { Hash.new }

Now every time a key was not found, the block will be executed
returning a brand new Hash instance.

> If it's no bug, I think it's a bug in documentation. - and not a very logic
> behavior.

OTOH, I think it is a very logic and usefull behavior. Especially as
we have now the block supporting Hash::new method :-)


HTH,
\cle
-- 
Clemens Hintze  mailto: c.hintze@gmx.net