From: Gene Tani <gene.tani@...>
Date: 2005-10-31T23:12:06+09:00
Subject: Re: determining whether an object is an immediate?

################ benchmark, round 2
OBJECTS = [1, 3.14159, [], :sym, "astring", true, {}, false, 1..3, nil,
/x/]
ITERATIONS = 10000          ## change
bm(2) do |x|
        x.report("VClass ") do
                ITERATIONS.times do
                        OBJECTS.each do |obj|
                                begin
                                        class<<obj; end
                                rescue TypeError
                                        # nothing
        end;    end;    end;    end
        x.report("===    ") do
                ITERATIONS.times do
                        OBJECTS.each do |obj|
                                case obj
                                        when Fixnum, Symbol, TrueClass,
FalseClass, NilClass
                                                # nothing
                                        else
                                                # nothing
        end;    end;     end;   end
        x.report("objSpac") do
                ITERATIONS.times do
                        OBJECTS.each do |obj|
                            begin
                                if ObjectSpace.include?(obj)
                                   # nothin
                                end
                            rescue TypeError
                                # noting
        end;    end;    end;end
end
######################
        user     system      total        real
VClass   0.937000   0.000000   0.937000 (  0.937000)
===      0.219000   0.000000   0.219000 (  0.219000)
objSpac  4.734000   0.000000   4.734000 (  4.734000)


Eric Mahurin wrote:
> On 10/31/05, Brian Mitchell <binary42@gmail.com> wrote:
>
> > The clone test does not work. Reason? There are plenty of
> > non-immediate objects that won't clone like Float and Bignum
> > instances.
>
>
> Didn't notice that. I wonder why they didn't make this work. If it were up
> to me, I would make all immutable objects (including immediate objects)
> return self for clone.
>
> One test you could use is the singleton class:
> >
> > # Should raise a TypeError
> > class << :immediate_object; end
>
>
> I thought of that one previously too. The problem with this one is that for
> non-immediates, it will create singleton/meta classes for them when I don't
> need them. Maybe not that big a deal.
>
> Of course immediate object could also be found missing from
> > ObjectSpace, but I would imagine that test to be very slow.
>
>
> Probably the best test for now, but slow.
>
> May I ask why you need to know the immediacy of an object?
> >
>
> Performance. I don't really "need" to know - since this is an optimization.
> I'm doing some code generation (parser generator). When I'm given an
> immediate object instead of storing it in a variable and having the code use
> the variable, I want immediates to appear directly in the code (because
> there is no object creation penalty and I know it is immutable). My
> generated code doesn't reassign these variables (but it may modify the
> objects) so this works.
>
> Looking at the C code (ruby.h: SPECIAL_CONST_P), another option would be
> this:
>
> id = obj.__id__
> (id&3).nonzero? || (id&~4).zero?
>
> As long as ruby continues to encode immediates as it does now (can add
> more), this should work.