[Ironruby-core] Protocols and implicit conversions

John Lam (IRONRUBY) jflam at microsoft.com
Tue Apr 22 09:55:56 EDT 2008


I wanted to call out a fairly simple pattern that shows up in our libraries that folks largely haven't caught onto in the patches that they've been submitting:

Wayne recently submitted a patch for exist? and exists? which fixes bug #19626:

[RubyMethod("exist?", RubyMethodAttributes.PublicSingleton)]
[RubyMethod("exists?", RubyMethodAttributes.PublicSingleton)]
public static bool Exists(object self/*!*/, MutableString/*!*/ path) {
    return File.Exists(path.ToString()) || Directory.Exists(path.ToString());
}

While this works fine, if you pass it legal arguments, it blows up when the invariants are broken (eg path == null). You can see it blow up when you run the specs for this method from the command line via:

rake spec file exist dox

To make this patch complete, you'll need to add an overload that accepts a nullable object parameter, and add a [NotNull] attribute to the previous overload:

[RubyMethod("exist?", RubyMethodAttributes.PublicSingleton)]
[RubyMethod("exists?", RubyMethodAttributes.PublicSingleton)]
public static bool Exists(object self/*!*/, [NotNull]MutableString/*!*/ path) {
    return File.Exists(path.ToString()) || Directory.Exists(path.ToString());
}

[RubyMethod("exist?", RubyMethodAttributes.PublicSingleton)]
[RubyMethod("exists?", RubyMethodAttributes.PublicSingleton)]
public static bool Exists(CodeContext/*!*/ context, object self/*!*/, object path) {
    return Exists(self, Protocols.CastToString(context, path));
}

Notice how the null case is taken care of by Protocols.CastToString(). It also will handle the case where the user passes a string that can 'act like' a string by implementing to_str. This is an *extremely* common case in the Ruby libraries.

The [NotNull] attribute is used by the binder to ensure that a null is never passed as a parameter to the first overload, but instead directs the caller to the second overload. This is the case which correctly handles null (and is also reflected by the fact that we are declaring via comments that path is *nullable* eg missing the spec# bang).

Let me know if this isn't clear or needs some further clarification.

Thanks!
-John



More information about the Ironruby-core mailing list