Stories
Slash Boxes
Comments
NOTE: use Perl; is on undef hiatus. You can read content, but you can't post it. More info will be forthcoming forthcomingly.

All the Perl that's Practical to Extract and Report

The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
 Full
 Abbreviated
 Hidden
More | Login | Reply
Loading... please wait.
  • Not completely into Moose yet, but isn't a role the equivalent of a Java interface?

    In that case, the behaviour you describe has been implemented there as a intended feature: a regular class *implements* the interface.

    Additionally, a role seems to also have the function of an abstract class as well. In Java, overriding concrete methods is, again, a feature you want.

    If you don't want to override a method simply do not include a overloading method in the implementing class.

    It looks like a (sensible) design dec

    • Roles are far more than interfaces. I strongly recommend reading up on them to understand what they do. The protect (usually) against method collision and they provide a default implementation. If you want to use one like an interface, it's simple:

      package MyRole::DoesSomething;

      use Moose::Role;
      requires qw(
          save
          search
      );

      And now your classes (or some other role composed into said classes) must provide save and search methods.

      However, if you have this:

      package MyRole::DoesSomething;

      use Moose::Role;
      requires qw(
          save
          search
      );

      sub is_deletable {
          my ($self,$id) = "_;
          return if $self->search( children => $id );
          return 1;
      }

      And the class using the role also has an 'is_deletable' method, than this potentially very important behavior is silently discarded. You need to rewrite it as something like this:

      package MyRole::DoesSomething;

      use Moose::Role;
      requires qw(
          is_deletable
          save
          search
      );

      around is_deletable => sub {
          my ($is_deletable,$self,$id) = "_;
          return if not $self->$is_deletable;

          return if $self->search( children => $id );
          return 1;
      };

      And that solves your problem. However, silently discarding behavior is not something I expect from roles. So yes, this is defined as a "feature", but it's one that has repeatedly led to many hours of debugging for me.

      If there was simply a normal "method collision" when the roles are composed, I could easily resolve this in my role composition declarations and maintainers of my code would see it when they edit the file.

      • So it's indeed also like an Abstract class in Java. The solution for the problem in the Java world is the @Override annotation before the overriding method.

        It overrides silently, but it's clear in your code and your IDE will warn you if you don't add the @Override annotation.

        Something like that would indeed save debugging time.

        C.

        • So it's indeed also like an Abstract class in Java.

          A role is like an abstract class in Java in the same way lightning is like a lightning bug.

        • No, it's not like an abstract class. I can compose several roles into my class in Perl, but I can't multiply inherit from several abstract classes in Java. Further, if I have two roles which provide the same method name, this fails at composition time but roles (particularly with Moose) offer you a rich variety of ways of resolving conflicts. You don't get that benefit with abstract classes.

          You also mention the silent overriding and the "clear in your code" and "your IDE will warn you". Two problems wit

          • I see your point about multiple inheritance.

            However, annotations are a language feature and not something IDE-centric.

            • Ah, my apologies about misunderstanding the annotations. It's been too long since I've programmed Java (er, or something like that :)

      • And the class using the role also has an 'is_deletable' method, than this potentially very important behavior is silently discarded.

        For a definition of "silently" which means "Code I explicitly wrote in my class itself", perhaps.

        What if you extracted all behavior into roles and composed them into classes? I think that would produce the desired results.

        • Still if you redefine a sub in standard procedural code - just as explicitely - you'll get a warning if you have warnings on.
          • Still if you redefine a sub in standard procedural code - just as [explicitly]...

            If you override a method in OO code (the paradigm under discussion) explicitly, you get no warning or error.

            Changing this default behavior is like resolving method dispatch at compile time. That's fine for certain languages, none of which have "Perl" in their names.

            • Overriding methods in subclasses is a standard technique - there is no warning, because it is needed and there is no other way. In the case of roles there is a different way (as the recently anounced solution shows). So why not have this useful warning?
              • Class-local overriding of composed role methods is a standard technique too. It's been in the Perl 6 role specification for years. I have code which relies on it -- and that code predates Moose.

                I fail to understand the value of adding a default warning for intended, specified, useful behavior -- especially behavior you can't accidentally back into.

                • We get warnings for undefined variables because that might be a problem. It's trivial to suppress if you really meant it.

                  If your class silently overrides a role's method, that might be a problem. By issuing a warning, I can be alerted to a potential issue and it's trivial to suppress the warning if I really meant it. I've lost too many hours debugging this silent behavior to think I'll be convinced otherwise.

                  • If your class silently overrides a role's method....

                    How do you silently write a method? Every time I've written a method, I've done so deliberately.

                    Should there be a warning if you accidentally inherit from the wrong class? How about a warning if you mistype the name of your class? What if the name of the file and its directory path doesn't match the name of the package or class contained? I've made all of those mistakes, and they've all been difficult to debug -- but I don't know any usable heuristic

                    • How do you silently write a method? Every time I've written a method, I've done so deliberately.

                      He didn't say silently write a method he said silently override a method. Big difference. On a big project if there are several classes using a Role and that Role in the future adds a method the author now has to check every class that uses that Role to make sure there are no name clashes. If he ads a method that clashes with other Roles being used at the same time he'll get an error. But if that clash is with an class method he gets no such help.

                      Now imagine the author of the Role has no control over classe

                • chromatic,

                  FWIW, I agree with you, however a warning cannot hurt and can possibly make it easier to spot problems during heavy role refactoring. And if we find it is more annoying then expected and/or not as helpful as expected, we can remove it as easily as we put it in. The decision made on #moose was that it was a simple enough change and an innocuous enough thing that it was at least worth giving a try. Roles are still a new "thing" and so I still feel like we are exploring some unknown territory here

                  • ... a warning cannot hurt...

                    That's completely untrue. This particular warning may not hurt (I don't believe that, but I acknowledge the possibility), but warnings can hurt. I should write a longer post about this. For now, consider take Vista's UAC for examples of supposedly helpful warnings that do, in fact, hurt.

        • You know, I've considered that, but only to work around this issue. I don't want to extract behavior into roles unless I intend and need for that behavior to be shared.