Serializing Ruby code

Hi all.

A bit of philosophy.

We have Marshal[1] to serializing Ruby's objects (and also YAML and so on).
AFAIK, Rails community even uses objects serialization for program's state
saving.
But, existing marshalling/serialization can serialize only data, not plain
code; this fact is restricting for their applications. For example,
Smalltalk-style "image" (entire program state, which can be loaded and
continued) is impossible.

But! We have a method for translating code-to-data - ParseTree[2], and we
even have method to translate back data-to-code (though a bit roundabout) -
ruby2ruby[3].

Had somebody thought about gathering all this to create full solution for
entire running program "saving" and "loading"? (Or, may be, somebody already
done this and I'm missing?)

It can be really cool. And it seems completely possible.

V.

1. http://www.ruby-doc.org/core/classes/Marshal.html
2. http://rubyforge.org/projects/parsetree
3. http://seattlerb.rubyforge.org/ruby2ruby/

If I understand your question correctly, what you're looking for is
serializable continuations.

I've checked this a while back, and the current stable Ruby branch
(1.8.x) does not have those, though it supports continuations
themselves. There are no plans to add serialization for continuations
in this branch, afaik. Perhaps it would be added in the future Ruby
implementations (at least in some of those ;-)) but here's where the
implementation developers chime in...

-Chris

···

On 12/3/06, Victor Zverok Shepelev <vshepelev@imho.com.ua> wrote:

Hi all.

A bit of philosophy.

We have Marshal[1] to serializing Ruby's objects (and also YAML and so on).
AFAIK, Rails community even uses objects serialization for program's state
saving.
But, existing marshalling/serialization can serialize only data, not plain
code; this fact is restricting for their applications. For example,
Smalltalk-style "image" (entire program state, which can be loaded and
continued) is impossible.

But! We have a method for translating code-to-data - ParseTree[2], and we
even have method to translate back data-to-code (though a bit roundabout) -
ruby2ruby[3].

Had somebody thought about gathering all this to create full solution for
entire running program "saving" and "loading"? (Or, may be, somebody already
done this and I'm missing?)

It can be really cool. And it seems completely possible.

V.

1. module Marshal - RDoc Documentation
2. http://rubyforge.org/projects/parsetree
3. http://seattlerb.rubyforge.org/ruby2ruby/

You'd need more than that to get a full image to work (like recording the ObjectSpace, the symbol table, globals, ...)

Also, ParseTree doesn't capture the closure so you can't copy the full state of the interpreter:

$ cat closure.rb
def make_proc
   x = 1
   proc do |y| y + x end
end

p = make_proc

puts p.call(2)

require 'rubygems'
require 'ruby2ruby'

ruby = p.to_ruby

puts ruby

q = eval ruby

puts q.call(2)

$ ruby closure.rb
3
proc { |y|
   (y + x)
}
(eval):2: undefined local variable or method `x' for main:Object (NameError)
         from closure.rb:19:in `call'
         from closure.rb:19

···

On Dec 2, 2006, at 19:06 , Victor Zverok Shepelev wrote:

A bit of philosophy.

We have Marshal[1] to serializing Ruby's objects (and also YAML and so on).
AFAIK, Rails community even uses objects serialization for program's state
saving.
But, existing marshalling/serialization can serialize only data, not plain
code; this fact is restricting for their applications. For example,
Smalltalk-style "image" (entire program state, which can be loaded and
continued) is impossible.

But! We have a method for translating code-to-data - ParseTree[2], and we
even have method to translate back data-to-code (though a bit roundabout) -
ruby2ruby[3].

Had somebody thought about gathering all this to create full solution for
entire running program "saving" and "loading"? (Or, may be, somebody already
done this and I'm missing?)

--
Eric Hodel - drbrain@segment7.net - http://blog.segment7.net

I LIT YOUR GEM ON FIRE!