In Vizkit::ConnectionManager: Using port as key leads to duplicates

[vbargsten]

The hash @on_data_listeners in Vizkit::ConnectionManager uses port as key. However, if the port is e.g. of type SubPortProxy, and if the user connects again to the same port (determined from type, task, and name), the port is added multiple times (and we get multiple listeners for the same data). Is this the intended behavior? When should we consider port to be equal?

I found that Orocos::PortBase has a definition

        # True if +self+ and +other+ represent the same port
        def ==(other)
            return false if !other.kind_of?(PortBase)
            other.task == self.task && other.name == self.name
        end

Would adding a similar function as e.g. SubPortProxy.eql? (hash uses key.eql?) be reasonable?

[doudou]

Would adding a similar function as e.g. SubPortProxy.eql? (hash uses key.eql?) be reasonable?

This would definitely not make sense. Objects that are different should look different from the API perspective. Changing the notion of equality globally for a specific use-case would be a really bad choice.

Disclaimer: this is not my code, so the following comes from what I remember about discussions with @D-Alex.

The sub-port proxying is there so that fields from port and full ports looks all the same from the listening side - which allows Vizkit to display e.g. an image that is a field within a bigger structure. Nonetheless, a given port proxy will only create a single connection and be polled only once, regardless of the number of listeners (full or partial) that are on it, i.e. it's not more costly to have it the way it currently is. orocos.rb is doing the "reference counting", so that Vizkit can stop the listeners when it needs to, and have the connection severed only when all listeners stopped listening.

Some evidence supporting this: code-wise, SubPortProxy is actually a delegate on top of the "full" port. SubPortProxy#on_raw_data calls PortProxy#on_raw_data. This ones creates only one connection.