On Mon, 2006-02-20 at 16:28 +0900, SASADA Koichi wrote:

> Please teach your YARV's revision.

Before it was the 0.3.3 tarball
Updated to the 0.4.0 tarball now (congratulations on 0.4.0 btw!)

kig@bauhaus:~$ uname -a
Linux bauhaus 2.6.15 #1 PREEMPT Thu Jan 12 20:04:01 EET 2006 i686
GNU/Linux

kig@bauhaus:~$ ruby-yarv -v
ruby 2.0.0 (Base: Ruby 1.9.0 2006-02-14) [i686-linux]
YARVCore 0.4.0 Rev: 466 (2006-02-22) [opts: ]


> > 1. found when compiling rubysdl, mkmf.rb contains code like:
> > 
> > for d,*f in [[1,2,3]]
> > ...
> > 
> > kig@bauhaus:~$ ruby1.8 -e 'for d,*f in [[1,2,3]]; p d,f; end'
> > 1
> > [2, 3]
> > 
> > kig@bauhaus:~$ ruby1.9 -e 'for d,*f in [[1,2,3]]; p d,f; end'
> > [1, 2, 3]
> > nil
> > 
> Maybe I fixed it.

broken in another way

kig@bauhaus:~$ ruby-yarv -e 'for d,*f in [[1,2,3]]; p d,f; end'
1
nil


> > 
> > 2. had some incompletely arg'd blocks in my code and ran into:
> > 
> > kig@bauhaus:~$ ruby1.8 -e '[[1,2,3]].map{|a,b| p [a,b]}'
> > [1, 2]
> > 
> > kig@bauhaus:~$ ruby1.9 -e '[[1,2,3]].map{|a,b| p [a,b]}'
> > [2, 3]
> 
> ditto

this works with 466

>  
> > 3. define_method has bug with splat args
> > 
> > kig@bauhaus:~$ ruby1.8 -e 'class Object; define_method(:foo){|*args| 
> > p args}; end; 0.foo(1,2,3)'
> > [1, 2, 3]
> > 
> > kig@bauhaus:~$ ruby1.9 -e 'class Object; define_method(:foo){|*args| 
> > p args}; end; 0.foo(1,2,3)'
> > -e:1:in `<main>': wrong number of arguments (3 for 0) (ArgumentError)
> 
> ditto.

broken in another way

kig@bauhaus:~$ ruby-yarv -e 'class Object; define_method(:foo){|*args| p
args}; end; 0.foo(1,2,3)'
nil

actually,

kig@bauhaus:~$ ruby-yarv -e 'p lambda{|*a| a}[1]'
nil

> 
> > 4. method_missing and blocks
> > 
> > kig@bauhaus:~$ ruby1.8 -e 'def method_missing(m,&b) b end; p f{}'
> > #<Proc:0x00000000@-e:1>
> > 
> > kig@bauhaus:~$ ruby1.9 -e 'def method_missing(m,&b) b end; p f{}'
> > nil
> > 
> ditto

this works with 466

> > x. YARV GC runs take longer than Ruby1.8 GC runs (not really a bug) 
> >   (ruby1.8 GC run: 45fps -> 17fps, YARV GC run: 60fps -> 10fps)

> It's a valuable information.

> BTW, do you have profiler?  If you have it, please check bottleneck of
> GC.

I probably have some profiler, but don't know how to use it :I

rev 466 doesn't seem to have the problem anymore; now the GC run frames
are 25fps.

Stats with the test scene:
Athlon XP 1733MHz
ruby1.8:
 normal fps: ~45
 GC run fps: ~17
 GC run time: ~0.037s
 allocs per frame:
  Arrays: ~250
  Floats: ~390
 GC run every 10 frames, freeing 2370 arrays and 3650 floats

ruby-yarv:
 normal fps: ~60
 GC run fps: ~25
 GC run time: ~0.023s
 allocs per frame:
  Arrays: ~140
  Floats: ~1400 (are floats primitives in YARV or boxed?)
 GC run every 30 frames, freeing 4600 arrays and 40380 floats


> Thanks,
> -- 
> // SASADA Koichi at atdot dot net
> 
> 

Likewise,
-Ilmari


-- 
ML: yarv-dev-en quickml.atdot.net
使い方: http://www.atdot.net/~ko1/quickml