From: Eivind Eklund <eeklund@...>
Date: 2007-12-12T16:29:52+09:00
Subject: Re: Dynamically generate a symbol name

On Dec 11, 2007 4:45 PM, Gary Wright <gwtmp01@mac.com> wrote:
> On Dec 11, 2007, at 7:00 AM, Sam Smoot wrote:
> >
> > :"object#{n}" is the syntax. Or just "object#{n}".to_sym.
> >
> > I would avoid doing this though. Symbols aren't meant to represent
> > infinite variations. They're stored in a Hash internally, so the more
> > symbols you add, the slower they get.
>
> Hash table performance shouldn't be affected by size, that is the
> whole point of using a hash table.

The memory footprint will increase.  Unless you're already past the
limit where each cache line on average only contains one accessed
symbol, this memory footprint increase will decrease cache locality
which again will decrease performance.  In the general case: Depending
on hash table implementation, you could also have chaining costs as
your hash table grows to be full (which gives conflicts that need
either chaining, a tree, or a rebuild of the table to handle).

"Hash table performance is not affected by size" is a simplification
that hold when you have a pre-90s microprocessor or a microcontroller
and the entire hash table is in memory to start with and you have no
two entries that hash to the same slot.  This is a restricted case.
Assuming some speed loss for increased size is a good assumption.

As for the point of a hash table, I tend to think of it as "Fast
access to data where the ordering property don't matter" (or lookup is
convenient on something that isn't a number).

Eivind.