Extending modules defined in a ruby library via C#

Hi,

I think I’ve stumbled on a bug of IronRuby, but I would like to hear
your thoughts about this before filing a report on the bug tracker as
I might be missing something or doing something wrong.

I have a ruby library in which is defined a module and this module
holds a few classes and constants. Then I wrote a library in C# where
the same module is defined to add more classes other than the ones
defined on the ruby side. Now if I load the assembly first and then i
require my ruby library everything works fine, but if I require the
ruby library first and then load the assembly, the module defined by
the ruby lib gets totally wiped off leaving there only what was
defined in the assembly.

Given that in similar scenarios we obviously can’t use Extends =
typeof(…) in the RubyModuleAttribute, shouldn’t modules previously
defined in ruby code be automatically extended instead of being
erased/redefined (which is definitely broken)? Right below here you
can find the code to reproduce and verify the reported behaviour (for
convenience I also made an attachment to this mail).

=========== foo.rb ===========

def test(a)
a.each{ |s| puts “#{s} is #{begin eval(s); rescue NameError;
‘undefined’ end}”}
end

module Foo
class Bar; end
VERSION = 1
end

/* ======== Nrk.Foo.Test.cs ======== */
namespace Nrk.Foo.Test {
[RubyModule]
public static class Foo {
[RubyClass]
public static class Hoge {
[RubyMethod(“piyo?”, RubyMethodAttributes.PublicSingleton)]
public static bool Piyo(Object self) {
return true;
}
}
}
}

=========== test_assembly_first.rb ===========

load_assembly ‘Nrk.Foo.Test’, ‘Nrk.Foo.Test’
require ‘foo.rb’

test [‘Foo’, ‘Foo::Bar’, ‘Foo::VERSION’, ‘Foo::Hoge’, ‘Foo::Hoge.piyo?’]

=========== test_ruby_first.rb ===========

require ‘foo.rb’
load_assembly ‘Nrk.Foo.Test’, ‘Nrk.Foo.Test’

test [‘Foo’, ‘Foo::Bar’, ‘Foo::VERSION’, ‘Foo::Hoge’, ‘Foo::Hoge.piyo?’]

=========== OUTPUT #1 ===========

Foo is Foo
Foo::Bar is Foo::Bar
Foo::VERSION is 1
Foo::Hoge is Foo::Hoge
Foo::Hoge.piyo? is true

=========== OUTPUT #2 ===========

Foo is Foo
Foo::Bar is undefined
Foo::VERSION is undefined
Foo::Hoge is Foo::Hoge
Foo::Hoge.piyo? is true

.NET types can never be 100% compatible with Ruby classes; the semantics
are simply too different. What would you expect to happen, for
instance, if you were to say the following:

module System::IDisposable; end
class System::Collections::ArrayList; include System::IDisposable; end;
x = System::Collections::ArrayList.new
require ‘mscorlib’

Clearly, the “real” System::Collections::ArrayList can’t implement
IDisposable because the type is immutable. And the type we created to
represent the variable x can’t be a “real” ArrayList because we didn’t
know about that class when x was allocated.

So this leaves us with two possibilities:

  1. Always overwrite the previously defined constants with the newly
    imported CLR types
  2. Fail to load the redefined types

…and we’re currently doing the first. While it seems reasonable to
fail, it raises questions about what to do if 40 types can be loaded
successfully but one can’t be. Do we raise an exception even though
there were 40 successes? Or do we fail to load any types at all?

Monkey-patching CLR types is never going to be able to work identically
to monkey-patching pure Ruby types.

That’s absolutely true for arbitrary CLR types. Library types (those
marked by RubyClass attribute) are handled differently though. They can
be seen like collections of extension methods. So it makes sense to
extend a Ruby class using library methods. So far we were focused on
extending CLR types using library types. Extending Ruby classes could be
seen as a feature that is not implemented yet. Could you file a feature
request (bug) on RubyForge?

A possible workaround for you: define a C# method that takes a Ruby
class and extends it in manually using methods on RubyModule/RubyClass
(AddMethod etc.). You would call this helper after loading the assembly.

Tomas

On Sun, Feb 1, 2009 at 07:30, Tomas M.
[email protected] wrote:

That’s absolutely true for arbitrary CLR types. Library types
(those marked by RubyClass attribute) are handled differently
though. They can be seen like collections of extension methods.
So it makes sense to extend a Ruby class using library methods.

In fact, I think the described behaviour is perfectly fine for CLR
types and I was sort of expecting this, but not for those types marked
by RubyClass.

So far we were focused on extending CLR types using library
types. Extending Ruby classes could be seen as a feature that
is not implemented yet. Could you file a feature request (bug) on
RubyForge?

OK, I will file it later. Maybe I will assign a lower priority to this
as I guess it is not really a common scenario and I don’t know how
many folks are trying to build extension libs for IronRuby. But still,
IMHO it would be nice to get this fixed before going 1.0.

A possible workaround for you: define a C# method that takes
a Ruby class and extends it in manually using methods on
RubyModule/RubyClass (AddMethod etc.). You would call this
helper after loading the assembly.

Yeah I was thinking about something along that way, I was just waiting
for confirmations before going on.

FYI I got into this particular case while porting Florian F.'s json
library (see http://json.rubyforge.org/) which provides two variants
for the core bits, namely “extension” (implemented in C) or “pure”
(ruby). Both relies on a few methods, classes and constants previously
defined in the same module in which they are loaded (JSON). When
loading the JSON::Ext::Parser, I was getting everything in JSON and
JSON::Ext erased.

require ‘json/common’

module JSON
module Ext
# require ‘json/ext/parser’
# require ‘json/ext/generator’

load_assembly 'IronRuby.Libraries.Json', 'IronRuby.Libraries.Json'

$DEBUG and warn "Using c extension for JSON."
JSON.parser = Parser
JSON.generator = Generator

end
end

I could also define the Parser and Generator classes into a different
global module and then assign them (e.g. JSON.parser =
IronJson::Ext::Parser), but well there is no fun doing this :slight_smile:


Daniele A.
http://www.clorophilla.net/blog/