From: "matz (Yukihiro Matsumoto)" <matz@...>
Date: 2012-10-17T23:22:29+09:00
Subject: [ruby-core:48051] [ruby-trunk - Bug #7166] Speed up Hash#dup by patching Hash#initialize_copy


Issue #7166 has been updated by matz (Yukihiro Matsumoto).


Compatibility is important. Call rehash in initialize_copy and see how performance changes.

Matz.

----------------------------------------
Bug #7166: Speed up Hash#dup by patching Hash#initialize_copy
https://bugs.ruby-lang.org/issues/7166#change-30968

Author: tenderlovemaking (Aaron Patterson)
Status: Assigned
Priority: Normal
Assignee: tenderlovemaking (Aaron Patterson)
Category: 
Target version: 
ruby -v: ruby 2.0.0dev (2012-10-15 trunk 37193) [x86_64-darwin12.2.0]


Hash#dup can be sped up.  Hash#initialize_copy will iterate over each pair in the hash, inserting in to the new hash.  I think we can speed up hash duping by using st_copy and copying the underlying hash table.

Here is the benchmark I've been using:

  https://gist.github.com/3893852

When you pass a hash to Hash.[], it just uses st_copy to copy the hash.  If you run the benchmark, you'll see a fairly large difference between the performance of using Hash#dup and copying the hash via Hash.[].

I've attached a patch that changes Hash#initialize_copy to use st_copy.  Here is a plot of the performance difference:

  http://i.imgur.com/ai9Am.png

The blue line is the old Hash#dup, the green line is Hash#dup after my patch is applied, and the red line is copying via Hash.[].

I'm not 100% confident in this patch, so I hope someone can review more closely before applying (or rejecting!).  Thanks.


-- 
http://bugs.ruby-lang.org/