Bug #15552: app_aobench.rb is broken

• Status: Closed
• Assignee: mame (Yusuke Endoh)
• Target version: -
• ruby -v: ruby 2.6.0p0 (2018-12-25 revision 66547) [x86_64-linux]
• Backport: 2.4: UNKNOWN, 2.5: UNKNOWN, 2.6: UNKNOWN

[Eregon (Benoit Daloze)]

Specifically, the image generated is incorrect.
When modifying the footer of the benchmark like this:

alias printf_orig printf
def printf *args
  $fp.printf(*args)
end

File.open("ao.ppm", "w") do |fp|
  $fp = fp
  printf("P6\n")
  printf("%d %d\n", IMAGE_WIDTH, IMAGE_HEIGHT)
  printf("255\n")
  Scene.new.render(IMAGE_WIDTH, IMAGE_HEIGHT, NSUBSAMPLES)
end

undef printf
alias printf printf_orig

Here is the expected image:

I get this image with MRI 2.6.0:

And interestingly, TruffleRuby 1.0.0-rc11 renders an image closer to the expected one:

I guess this might be both an interpreter bug, and possibly also a bug in the benchmark code.

I think every benchmark should have some validation, otherwise it's prone to measure something unexpected like this.

[Eregon (Benoit Daloze)]

• Description updated (diff)

[mame (Yusuke Endoh)]

• Status changed from Open to Assigned
• Assignee set to mame (Yusuke Endoh)

And interestingly, TruffleRuby 1.0.0-rc11 renders an image closer to the expected one:

Closer :-)

This is caused by change of default encoding. The following fixes the issue. It is fortunate that the issue was not caused by wrong optimization.

diff --git a/benchmark/app_aobench.rb b/benchmark/app_aobench.rb
index 2bd6acfaf8..13cbc1657d 100644
--- a/benchmark/app_aobench.rb
+++ b/benchmark/app_aobench.rb
@@ -1,3 +1,5 @@
+# coding: US-ASCII
+
 # AO render benchmark
 # Original program (C) Syoyo Fujita in Javascript (and other languages)
 #      https://code.google.com/p/aobench/

By the way, the program uses a random number. The effect would be negligible, but I'd like to add srand(0) after the above fix.

[mame (Yusuke Endoh)]

• Status changed from Assigned to Closed

Applied in changeset trunk|r66892.

---

benchmark/app_aobench.rb: add a magic comment

To support the change of default encoding.
It had not worked correctly since 2.0 :-)
[Bug #15552]

[Eregon (Benoit Daloze)]

@mame (Yusuke Endoh) Thanks for the quick fix!
With srand(0), I think we could add some verification, e.g., by computing a checksum of the image.