I have an application where I am performing an operation on a series of elements and the exact nature of the operation depends on the type of the element being operated upon. For reasons of encapsulation, it's not appropriate for the element to implement the operation; this means it can't be a virtual method on the element type and so 'standard' polymorphism doesn't work. I posed a previous question related to this and was informed that this was known as the Visitor Pattern.
I had previously always implemented this using an if/elseif
dispatcher method based on the type of the object, then calling an appropriate implementation. Recently, however, I noticed that the same thing could be accomplishing using the dynamic
keyword, like this:
private void ReconcileTips()
{
foreach (var step in _definition.Steps)
{
ReconcileTips((dynamic)step);
}
}
private void ReconcileTips(IBulkDispenseDefinition bulkDispense)
{
bulkDispense.TipType = ReconcileTip(bulkDispense.TipType);
}
private void ReconcileTips(ImportScreenDefinition importScreen)
{
foreach (var usage in importScreen.ReagentUsages)
usage.TipType = ReconcileTip(usage.TipType);
}
private void ReconcileTips(BuildScreenDefinition buildScreen)
{
foreach (var function in buildScreen.Functions)
function.TipType = ReconcileTip(function.TipType);
}
A similar pattern could be used for other operations parallel to the class structure, like creating view models for each element of _definition.Steps
. The thought is that the compiler basically transforms this into the same if/elseif
logic I wrote before, saving me the effort. So, a few questions:
Are there any gotchas with the dynamic dispatching that I haven't considered? I believe this is equivalent to performing a series of if (x is TypeA) Do((TypeA)x) else...
, but I could be wrong.
Is this actually cleaner and easier to understand than a long if/elseif
method?
Are there any gotchas with the dynamic dispatching that I haven't considered? I believe this is equivalent to performing a series of if (x is TypeA) Do((TypeA)x) else..., but I could be wrong.
The main gotcha would be if a type implements more than one interface in your visitor pattern - the compiler will probably pick the one you want, but it may not be the same choice you'd make if you use if (x is TypeA)
/else if (x is TypeB)
logic, as you'd control the order the checks occur.
Is this actually cleaner and easier to understand than a long if/elseif method?
I personally think so. This provides a very clean, fairly decently performing dispatch determined by the runtime type, and "just works." Hard to beat simple, short, clean code. Just make sure to (potentially) handle the case where you get a runtime error from the wrong type being passed in.
Yes, I was considering this approach, but decided against it for a more traditional approach. I derive each visitor from an interface that has Visit methods for each type I want to implement the operation for.
If you have a number of different operations you want to implement as visitors, eg. Save and Load operations, and you might want to add more in the future; with the dynamic approach you don't get compilation errors if you forget to implement your operation for one of the types you need to handle. You'll only find out when the program crashes and burns at runtime.
I want to be sure at compile time that any operation has been implemented for all of the possible types.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With