From: Robert Klemme <shortcutter@...>
Date: 2006-08-13T07:25:09+09:00
Subject: Re: PATCH to make internal Hash class retain order...

Thies C. Arntzen wrote:
> Brian McCallister schrieb:
> 
>> On Aug 12, 2006, at 9:55 AM, Thies C. Arntzen wrote:
>>
>>> Hi there -
>>>
>>> beeing new to ruby (thru rails) and coming from a strong php
>>> background
>> <snip />
>>
>>> * remember the order in which elements have been added to a Hash (at
>>> the cost of 2 extra pointers in every hash-bucket + 2 extra pointers
>>> per hash-table)
>>> * use this order for iterating (st_foreach will use this order)
>>> * implement a call Hash.order! allows the user to change the internal
>>> order;-)
>> Nice :-) The better place for discussion hacking on the ruby
>> interpreter itself is probably ruby-core.
> 
> i wasn't aware of that list;-) subscribed...
> 
>> How difficult would it be to do an OrderedHash data structure as a C
>> extension instead of in Ruby?
> 
> i really believe that it needs to be a 1st class citizen, - the reason
> it's done in C is because userland imposes too much overhead (i
> believe) and converting forward<->back to the normal Hash would be
> slow, memory intensive and ordering information would get lost on the
> way...
> 
> also the amount of code added to the Hash class is ~20 lines of C-code
> - this code adds zero overhead. the bigger chunk of the code was added
> to the st.c which is the underlying library for handling hash-tables -
> and duplicating > 600 lines of code to add ~60 lines didn't seem very
> cool to me;-)

Adding to David's comments: you're changing semantics of one of the core 
classes which is in itself a dangerous thing.  Plus, you increase the 
memory footprint of a hash which is IMHO a very bad idea for such a 
widely used data structure (and some of these hashes grow huge).  So, if 
you want to patch ruby core, make it at least a subclass of Hash so 
people can choose.

Then of course there's a whole bunch of other things that one should 
consider.  Is the name properly chosen?  Do we have to provide a means 
for the user to *define* the standard order on creation?  This is e.g. 
the way Java's TreeSet works - and IMHO it's superior vs. having to call 
order! manually. etc.

Kind regards

	robert