Bug #13113: Performance issue, const_set Module.new is too slow in Ruby 2.4

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 2.4.0p0 (2016-12-24 revision 57164) [x86_64-linux]
• Backport: 2.2: DONTNEED, 2.3: DONTNEED, 2.4: DONE

[pocke (Masataka Kuwabara)]

In Ruby 2.4, the following code is too slow.

5000000.times { Module.new.const_set(:X, Module.new) }

In Ruby 2.4, it took 115.68 seconds to run this code.
However, in Ruby 2.3.3, it took 7.36 seconds.

And if const_set value is not Module.new, it is not slow. For example

5000000.times { Module.new.const_set(:X, 2) }

Ruby versions

ruby 2.3.3p222 (2016-11-21 revision 56859) [x86_64-linux]
ruby 2.4.0p0 (2016-12-24 revision 57164) [x86_64-linux]

See also. https://github.com/rspec/rspec-core/issues/2365

[nobu (Nobuyoshi Nakada)]

• Backport changed from 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN to 2.2: DONTNEED, 2.3: DONTNEED, 2.4: REQUIRED

Since r53376, anonymous module is resolved immediately by assignment to a constant, including the outer module.

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset r57283.

---

benchmarks for [Bug #13113] [ci skip]

[nobu (Nobuyoshi Nakada)]

benchmark results:
Execution time (sec)

name	2.3.3	r57281	r57284
loop_whileloop2	0.146	0.135	0.106
vm2_module_ann_const_set*	9.072	107.850	6.889
vm2_module_const_set*	3.954	13.374	3.468

Speedup ratio: compare with the result of `2.3.3' (greater is better)

name	r57281	r57284
loop_whileloop2	1.075	1.373
vm2_module_ann_const_set*	0.084	1.317
vm2_module_const_set*	0.296	1.140

[ko1 (Koichi Sasada)]

中田さん：

(1) ann_ って何？
(2) Module.new.const_set(:X, Module.new) だと、
　　Module.new が支配的だと思うんだけど、何が計りたいの？

On 2017/01/08 10:46, nobu wrote:

nobu 2017-01-08 10:46:10 +0900 (Sun, 08 Jan 2017)

New Revision: 57283

https://svn.ruby-lang.org/cgi-bin/viewvc.cgi?view=revision&revision=57283

Log:
benchmarks for [Bug #13113] [ci skip]

Added files:
trunk/benchmark/bm_vm2_module_ann_const_set.rb
trunk/benchmark/bm_vm2_module_const_set.rb
Index: benchmark/bm_vm2_module_ann_const_set.rb

--- benchmark/bm_vm2_module_ann_const_set.rb (revision 0)
+++ benchmark/bm_vm2_module_ann_const_set.rb (revision 57283)
@@ -0,0 +1,5 @@ https://github.com/ruby/ruby/blob/trunk/benchmark/bm_vm2_module_ann_const_set.rb#L1
+i = 0
+while i<6_000_000 # benchmark loop 2

• i += 1
• Module.new.const_set(:X, Module.new)
  +end

Property changes on: benchmark/bm_vm2_module_ann_const_set.rb

---

Added: svn:eol-style

• LF

Index: benchmark/bm_vm2_module_const_set.rb

--- benchmark/bm_vm2_module_const_set.rb (revision 0)
+++ benchmark/bm_vm2_module_const_set.rb (revision 57283)
@@ -0,0 +1,8 @@ https://github.com/ruby/ruby/blob/trunk/benchmark/bm_vm2_module_const_set.rb#L1
+i = 0
+module M
+end
+$VERBOSE = nil
+while i<6_000_000 # benchmark loop 2

• i += 1
• M.const_set(:X, Module.new)
  +end

Property changes on: benchmark/bm_vm2_module_const_set.rb

---

Added: svn:eol-style

• LF

--
ML: [e-mail]
Info: http://www.atdot.net/~ko1/quickml/

--
// SASADA Koichi at atdot dot net

[nobu (Nobuyoshi Nakada)]

On 2017/01/19 10:40, SASADA Koichi wrote:

中田さん：

(1) ann_ って何？

ANoNymous

(2) Module.new.const_set(:X, Module.new) だと、
　　Module.new が支配的だと思うんだけど、何が計りたいの？

無名モジュールの名前解決がそれ以上に支配的だった、というのが問題なので
Module.newだけが支配的になればそれで充分。

--
--- 僕の前にBugはない。
--- 僕の後ろにBugはできる。
中田 伸悦

[ko1 (Koichi Sasada)]

On 2017/01/19 10:47, Nobuyoshi Nakada wrote:

(1) ann_ って何？

ANoNymous

わかんねーです。
せめてコメントとか。

(2) Module.new.const_set(:X, Module.new) だと、
　　Module.new が支配的だと思うんだけど、何が計りたいの？

無名モジュールの名前解決がそれ以上に支配的だった、というのが問題なので
Module.newだけが支配的になればそれで充分。

うーん、ベンチマークじゃなくて、テストに入れても良さそうな...。

--
// SASADA Koichi at atdot dot net

[naruse (Yui NARUSE)]

• Backport changed from 2.2: DONTNEED, 2.3: DONTNEED, 2.4: REQUIRED to 2.2: DONTNEED, 2.3: DONTNEED, 2.4: DONE

ruby_2_4 r57850 merged revision(s) 57283,57284.