I'm building my own pet engine and after struggling with how to handle resource ownership for a while, I came across this question and specifically Josh and Sean's responses (as well as Sean's blog on shared_ptr), which really made a whole lot of sense to me. So I went ahead and implemented my own similar system, and so far it's working great except for one minor snag - what do I do when I need to access a resource of a derived type when the handle only refers to the base type?

I'm building my own pet engine and after struggling with how to handle resource ownership for a while, I came across this question and specifically Josh and Sean's responses (as well as Sean's blog on shared_ptr), which really made a whole lot of sense to me. So I went ahead and implemented my own similar system, and so far it's working great except for one minor snag - what do I do when I need to access a resource of a derived type when the handle only refers to the base type?

• When I dereference the handle, use dynamic_cast to cast the pointer to a RenderTarget* as necessary.

  • RTTI is "Bad" TM, which is to say that it very often gets disabled (along with C++ exceptions, etc.) in game applications due to unnecessary overhead, particularly on consoles. Plus resorting to dynamic_cast feels like I'm "doing it wrong," architecturally speaking.

• I know it's a RenderTarget; it was when I created it, and if it weren't, I wouldn't be passing it to SetRenderTarget in the first place! reinterpret_cast!

  • Do I REALLY know? If I'm wrong, then I'm casting something else to something it's not, and I can't even check if it's nullptr if I'm wrong. Also, again, I'd like to avoid casts if possible. But I don't entirely disagree with the logic here...

• Implement a lightweight homegrown alternative to RTTI, and use that instead.

  • I've seen this done in one or two engines, but it sounds like some fairly heavy-duty wheel-reinventing...

• Don't derive RenderTarget from Texture; instead, add a flag that says whether or a Texture is a RenderTarget or not. If it's not, then SetRenderTarget will simply fail.

  • Potentially lots of excess baggage added to the Texture class. (What's another pointer to an ID3D11RenderTargetView among friends?) Doesn't sound like it scales very well.

• Derive RenderTarget from Texture, but add a virtual method to Texture called GetRenderTarget() that normally returns nullptr but is overridden in RenderTarget to return this. Similar to the flag method above, but without the excess baggage.

  • Feels like a poor man's dynamic_cast. Again, doesn't scale well (add a new method for each derived class that needs it?). But if my RenderTarget scenario is maybe the only case I actually need something like this, I'm beginning to feel like maybe this solution isn't so bad...

• When I dereference the handle, use dynamic_cast to cast the pointer to a RenderTarget* as necessary.

  • RTTI is "Bad" TM, which is to say that it very often gets disabled (along with C++ exceptions, etc.) in game applications due to unnecessary overhead, particularly on consoles. Plus resorting to dynamic_cast feels like I'm "doing it wrong," architecturally speaking.

• I know it's a RenderTarget; it was when I created it, and if it weren't, I wouldn't be passing it to SetRenderTarget in the first place! reinterpret_cast!

  • Do I REALLY know? If I'm wrong, then I'm casting something else to something it's not, and I can't even check if it's nullptr if I'm wrong. Also, again, I'd like to avoid casts if possible. But I don't entirely disagree with the logic here...

• Implement a lightweight homegrown alternative to RTTI, and use that instead.

  • I've seen this done in one or two engines, but it sounds like some fairly heavy-duty wheel-reinventing...

• Don't derive RenderTarget from Texture; instead, add a flag that says whether a Texture is a RenderTarget or not. If it's not, then SetRenderTarget will simply fail.

  • Potentially lots of excess baggage added to the Texture class. (What's another pointer to an ID3D11RenderTargetView among friends?) Doesn't sound like it scales very well.

• Derive RenderTarget from Texture, but add a virtual method to Texture called GetRenderTarget() that normally returns nullptr but is overridden in RenderTarget to return this. Similar to the flag method above, but without the excess baggage.

  • Feels like a poor man's dynamic_cast. Again, doesn't scale well (add a new method for each derived class that needs it?). But if my RenderTarget scenario is maybe the only case I actually need something like this, I'm beginning to feel like maybe this solution isn't so bad...