Re: The fastest way to read files

Ryan Davis <[e-mail]> wrote:
> On Nov 11, 2011, at 15:20 , Robert Klemme wrote:
> > On Fri, Nov 11, 2011 at 10:07 PM, Ryan Davis <[e-mail]> wrote:
> >> On Nov 11, 2011, at 04:50 , Robert Klemme wrote:
> >> 
> >>> # ensure every word is only once in memory
> >>> words = Hash.new {|h,k| k.freeze; h[k] = k}
> >> 
> >> AFAIK, Ruby hashes have (almost) always frozen their keys.
> > 
> > I know.  That's the reason why I do the freeze in the block.
> 
> I'm confused. If you know that the key is going to be frozen anyways,
> why freeze it?

The implicit freeze from Hash#[]= duplicates the string and freezes
the duplicate, not the same object given by the user.

Explicitly freezing the key before Hash#[]= prevents MRI[1] from
duplicating the string.

------------------------ freeze_example.rb --------------------
h = {}
frozen = "foo".freeze
h[frozen] = true

# explicitly freezing the key means the String object is stored as-is
p [ :frozen_original_key, frozen.object_id ]
p [ :frozen_key_after_aset, h.keys[0].object_id ]

h = {}
not_frozen = "foo"
h[not_frozen] = true

# Not freezing means the key stored in the hash key is a different
# object than the one provided by the user.
p [ :not_frozen_original_key, not_frozen.object_id ]
p [ :not_frozen_key_after_aset, h.keys[0].object_id ]

------------------------ Output --------------------------------
[:frozen_original_key, 70096844693360]
[:frozen_key_after_aset, 70096844693360]
[:not_frozen_original_key, 70096844693120]
[:not_frozen_key_after_aset, 70096844694120]



[1] - Verified by reading rb_hash_aset() in hash.c which eventually
      calls rb_str_new_frozen() in string.c (ruby/trunk):

rb_str_new_frozen(VALUE orig)
{
    VALUE klass, str;

    if (OBJ_FROZEN(orig)) return orig;

    ...