So here again with some more opinion/discussion related topic about what features should/are normally be related into a game engine's core system from the implementation side of view. With a huge time I invested into writing and even more into research for my very own game engine project, there were as many differences as there are to coding guidelines. Especially open source engines (Urho3D, Lumberyard) and those that offer there source code (Unreal, CryEngine, Phyre) change during the centuries depending on current technical, coding and hardware standards. They all consist of third party code and libraries that are set apart from the core system bur rely on it.

Now I'm on a point where I want restructure/cleanup my project(s), removing obsolete and integrate prototyping code as same as change the growing code base to have a better clean managed file structure. The question that has had to emerge for long or short is what parts of the code base should be a core feature in order to keep the core system clean and flexible. Again reading many cross references it pointed some similarity out but also huge differences (like Unreals hashing/encryption code beeing in core) so I made a list of important stuff any game needs (I left out platform specifics and audio/renderer here because they are platform dependent and should stay in an own module/system)

• Allocator (memory management)
• Math (Vector, Matrix, Quaternion ...)
• Threading (Threading, Locks)
• Container (apart from the STL, Queue, Buffers, certain types of Array ...)
• String (management, encoding, utils)
• Input (seen a game without input?)
• IO (reading/writing files and manage filesystem structure)
• Type Handling

And a list that is optional but may/may not be inside the core code

• Logging (because it is a development tool)
• Profiler (see logging)
• Serialization
• Plugins
• Events

In my opinion, logging and profiler code is a heavyly used feature in development but is stripped out mostly when deploying a release version, not all games rely on events rather than using other interaction (like polling, fixed function loops) and also the same for plugins and serializing/deserializing data; so what do you think should also be a must have in core system or should go from the list into core system as well as a must have? What else should go outside as an own module/system (into an own subfolder inside the engine project's code base)?

Courious to read your opinions

Forgot the memory management/allocator stuff so added it to the list I'm on the hop to force strict memory management at all levels of the engine to at least throw ya an assertion failure arround the head when leaving anything on the heap on scope exit.
 

59 minutes ago, Hodgman said:

Cross platform GPU wrapper (preferably low level in the core module - with models/scenes/materials as a higher level module)

I think you mean the graphics API (DX, GL, VK, whatever) here?

1 hour ago, Hodgman said:

3D collision and rigid body dynamics

Highly depends on the game, Puzzle or 2D platformer might not need any of these so I see them in an optional external instead of core where I'm not sure about input system here (what is a reason to write this post). But I think you were right and it depends a level upward than placing it in core directly.

Networking should stay external as not every game needs it but as I have seen for example Unreal hosts its HTTP stuff inside the core module. I personally would avoid this and instead keep networking low level on the one hand (to have possibility to write also server code) but also more high level as a game client.

Same thoughts for the crypto module. It consist of a lot of static functions to utilize AES, ECDSA, XTEA as same as different hashing functions SHA, RIPEMD and MD5 (as short hash) where the core system contains some text related 32/64 bit hashing functions used for example to identify names.

I agree on your arguments about strings and IO in general. Assets should be converted to packages of engine preprocessed data and savegames should be serialized/ deserialized in order to keep a level of error proofness but these systems may need to interact themselfs with the disk/ network/ cloud storage so I keeped this in my entire list Nearly anything inside my code is already using streams/ buffered reading and static shared memory (aka memory mapped files) as needed and passing a stream interface is preferreable by any data processing function.

Under threading I have currently threads and locks as utility code wrapping the underlaying OS APIs where my JobPool and event system is in an own module but I see that this might go as well into core too. What do you think about event driven engines vs. polling/ update driven ones?

Profiler and logging were currently own modules as I thought of having them in build while development but strip them out when shipping. Profiler may be a point of diskussion but I see logging at least for support purposes as a core feature now so maybe move it there. Dont know about profiler yet

Reflection is currently realized as an external module named MetaType that is intendet to provide some meta information and C# like function invocation (for runtime class manipulation) as same as some serialization frontends. As I have seen this in core for various projects, I will potentially move it into core level, for at least support scripting and maybe editor UI interaction from C# level.

Hardware/OS specific items should still have an interface layer in your core such that it is consistent no matter how you implement the backend.  In regards to that, you really should add input device handling (keyboard, mouse & joystick) and window management items.  Depending on goals, a lot of folks roll window management into the rendering engine but I tend to think it should be separated for quite a few reasons.  Just my $0.02.