18 KiB
Pandemonium Engine TODOs
Core
- Add a core binder class for Math. This would allow for simpler script to engine side code ports. Maybe it could also smplify scripting language code.
- Maye threading could be de-modernized. So the implementation could use the old os level constructs. (like pthreads). An implementation that uses std::threads could be kept around though. (without it being compiled). -> Just add back and update the old os api implementations (pthreads etc), keep the core classes, and make every single class that uses std threads overridable similarly to what can be done with thread now (mutexes etc).
- Maybe use NULLs again instead of nullptrs? TODO check if there is an actual difference, other than requiring cpp11.
- Add error macro like methods to global scope for scripting
- Add containers that are a drop in replacement for stl containers.
- Add compiler traits header.
WFC module
- Need to rework the readme. Also link to the sample repo. (https://github.com/Relintai/wfc_module_samples)
- TilingWaveFormCollapse should not generate the images themeslves, rather it should just use internal ids, and return those to you. It could store variants. -> a derived class chould be mamde that adds image generation on top, but in a friendlier way.
- Array2D and 3D's getters and setters that need coordinates use a reversed order compared to everything in the engine. This is super dangerous, and should be changed. (Currently: get(y, x). Should be get(x, y)).
- The classes need smaller fixes and touchups.
- There are probably a few lingering bugs, as some examples give bad results.
- Need to make sure wave size, out size, and size related things are consistent, and intuitive, as right now they are neither.
- The bindings will likely need more improvements.
- Would be nice, if the algorithm could be parametrized from gdscript as well.
- Maybe adding Array2D, and Array3D like classes to the engine would be a good idea.
- Replace std::vectors to engine vectors.
Physics
- Add a multi thread wrapper around the physics server (and nav server).
- Add a method to the physics server that lets the caller thread wait for the physics update, and wait for each thread to finish.
- These 2 changes would enable heavy simplifications to most Props, Voxelman, and Terraman generation code, as they could just directly use the physics server in multi threaded jobs without crashing. The entire update mode thing could be removed.
- Add dummy physics servers, so they can be turned off when not needed. It should return rids that seem valid, so app code doesn't freak out anywhere.
Navmeshes
- Add dummy nav server, so it can be turned off.
- The current nav server uses worker threads (ThreadWorkPool), maybe it should use ThreadPool.
- The current nav server might not work if threading is disabled. TODO check.
- Maybe ThreadWorkPool could be removed.
Terraman, Voxelman, Props
- MaterialCache class (classes) should be either core or just an another module. (I think the build system has module dependencies implemented.)
Terraman, Voxelman
-
The rid storage API is using dicts currently, and it's pretty awful to mess with. A new mesh storage class should be added instead. Since more than one module needs it either should be core or a new module. The api could just return a Ref<> to the storage class. And the class could be inherited from when needed. Should probably contain some generic api which allows you to store most things without really messing around too much. Like an api for mesh rids, navmesh rids, and collider rids, and generic rids.
-
Currently the library only supports images for the top, all 4 sides, and the bottom. It should have image support for all 6 sides at least. I think surface definitions should have a count variable, which sets the stored image count. It could work similarly to lod materials, so you can just use indexes, and it will return what it can. I might also be a good idea to make the image getter scriptable (Or provide one which is).
Editor
- Add support for enabling / disabling editor plugins, and also add a gui for this. (Like how the gdscript addons can be enabled/disabled, but for built in c++ stuff.) It will need support for either exclusion, or dependencies. (Dependencies would be better.)
- Modularize the editor.
Engine
- (Maybe) add an SDL backend.
- Reorganize the project. Except for core, and a few smaller things like main everything should be a module that can be disabled.
- Add more hooks where needed, to let stuff register itself. Like visual server backends, so adding them is simple, as I think this is the best way forward, so you can just create your custom rendering backend, when needed. Like if you want lightmaps, add a module that has the tooling + visualserver that has it implemented.
- Port the DisplayServer from Godot4, as I think having display and windowing in a DisplayServer singleton instead of in OS is better architecturally.
- Port the capability to have more than one windows. Multi windowing however should be disabled by default, also I don't want to port the full multi windowing stuff for the editor, as it's a lot leaner the way it is. However multi windowing would be cool for creating tools. Would also make it easier to not have a window at all.
- The DisplayServer at first could just use stuff in OS, so it doesn't have to be ported all at once.
- Should backport the tests and the test suite from Godot4 as well.
- Backport optimizations to the hashmap (or maybe just add it as another option), and also replace usage where applicable (
8b7c7f5a75
), and (45af29da80
) - Make operator='s return void (
7da392bcc5
). - Switch back to the old c++ version (pre c++11) for the codebase. I'm already targeting low end hardware, I think it will likely come in handy later.
- Implement something similar to this, so scripting language code can be simplified. (
5288ff538d
)(Create Variant built-in functions.) - Move DynamicFont to a module, and make tools depend on that.
- Maybe backport the TextServer from Godot4?
- Make modules from everything under the thirdparty folder.
- Collect license information from modules and merge it back and handle it like the current thirdparty.md file.
- Clean up the third party licenses.
Scene folder
- Rework it into modules.
GUI
- Rename the EditorZoomWidget to something like ZoomWidget, and make it available for the games as well. Also I think I did smaller improvements to it in the world_generator addon. Also port those.
Graphics
- Refactor platform code so it's not automatically creating a window, instead the VS should ask for it's creation, passing along the requested graphics api.
- Add a dummy VisualServer implementation, and a way to switch between this and gles2. Similar to how godot4 can have a dummy renderer.
Rendering
- Move CPUParticles code into the VisualServer.
- Add support for generating vertex colors from lights in the VisualServer. (Also do this for 2d)
- Add LOD support to the visualserver. (Or maybe the meshes) - Add core LOD generation support (probably VisualServer). (This and vertex color baking could drastically reduce the amount of code in Terraman, Voxelman, and terraman 2d, props) (
d2302f53d6
) - Maybe add a texture (mergedTexture) that makes it easier to handle texture merges.
- Add gles3 support, but just expand gles2 for it. (Like it could use better multimesh support, better particles etc), but it shouldn't be that different from gles2.
- Probably add support for other renderers, like vulkan, but unlike godot4 this should be realtively simple, with the visualserver design. (In the same veins as gles2/3)
- (And/or other highly portable graphics apis, that is if they exist.)
- Add a raw shader type that can be written in the native graphics api language (support per backend shader implementation for it). It will likely need some easily parsable header for shader properties. This might simplify the creation of other backends.
- Support for adding graphics backends from modules. This will require platform specific creation funcs.
- Could add an api to query gpus, and optionally to mess with them. Like you could use an off gpu to to do something, maybe like calculations, etc. (Needs more research, not even sure if this is possible or not, or how hard it would be.)
- Rename VisualServer to RenderingServer like on Godot 4. I think it's a better long term name for what it does.
- Port the RenderingDevice API from Godot 4 alongside a RenderingServer implementation that uses it once it's stable. However keep the GLES backends around the same way as they are now, so RenderingServer would be Pluggable as well: RenderingServer <- RenderingSreverGLES2, RenderingServer <- RenderingSreverGLES3 (eventually), RenderingServer <- RenderingSreverRenderingDevice (This would then just use an available RenderingDevice).
- Port the newer godot rendering backends once they are stable enough. (With at least some of their features.)
- Somehow make VisualServer's implementation more modular.
- Add a scriptable VisualServer implementation. Similar idea as the scriptable networking api. Alongside with a similar interface for GLES. Would help with experimentation.
material_maker module
- Polygon editor needs fixes, and likely an overhaul.
- Curve editor needs smaller fixes, and likely an overhaul.
- Port the rest of the nodes from material maker.
Databases
- Table builder should have a method that can automatically create tables from any Object using it's properties. Like tb->add_object_properties("obj type"). You could the still mess with the query, add fields, and set keys. It's purpose should be simple automation when you don't need handoptimized tables.
- Query builder should have more helper methods to automate simpler queries as much as possible. Like: sfw("col", "table", "a < 3") -> SELECT col FROM table WHERE a < 3;, or sf()... etc
- Query builder should also have an update and create, (and also get) method which generates a query using properties to automatically create sql to save objects. They could also take a pk argument to tell which entry is the key. save method would also be nice that decides whether to update or create an entry.
- No full ORM functionality though, it should be really close to SQL queries, but with lots of helper methods, for the more trivial stuff.
- Go through and double check that all methods for the QueryBuilder are properly set up, and is using escapes as they should.
- Fix postgres and mysql backends, so they can be built. (They should be off be default.)
Web
- Move simple web server to a new module.
- Implement big (or not in-memory) file upload support.
- Move HTMLParser, BBCodeParser, and MarkDownRenderer to a new (parsers?) module.
- BlogWebPage <- BlogWepBageFile, BlogWebPageDB
- NewsWebPage <- NewsWebPageFile, NewsWebPageDB
- AdminWebPage <- UserAdminWebPage, AdminPanelWebPage
- Add epoll api support to sockets, and make the web server use it.
- Add range http header support.
- Finish all the todos in the http parser.
- Could add a simple templating language. Probably would not be more than a few hundred lines. WebPageTemplate (Resource), with an importer. Most webnodes could have assignable properties for them. Could be similar to laravel's for example. Also it could be pre compiled in an extremely simple fashion. Vector StringStatement, IFStatement, LoopStatement, CallStatement etc. Other than bool evaluation every statement type is extremely trivial. Also bool evaluation logic could be reused elsewhere (Also gdscript already have it implemented).
Users
- Clean / flesh out the api of UserManager / UserDB. Also make UserManager register itself into UserDB.
Networking
- Maybe port https://github.com/godotengine/godot/pull/50710 ? It adds some lower level stuff which should be useful. It also adds mesh networking which I'm not really sure about that it's really that useful for most games, as you probably want want authoritative servers in most cases.
- Rename the networking keywords (puppet, master etc) to something that is a lot easier to understand at a glance. Maybe something like CLINET_CAN_CALL_SERVER etc. Should call local maybe could be made a separate bool param like in Godot 4, although I think I like the current solution better.
- Somehow add helper methods that do what the RPC macros do in Entity.
Ground Clutter
- Add new ground clutter module. It should have classes that help with spawning clutter like small objects, or foliage.
- Add GroundClutterData. It should contain all data that is needed to spawn in everything.
- Add GroundClutter Spatial node. Similar to TiledWalls, just adds in meshed.
- Add GroundClutterMerger Spatial node, which is the same as GroundClutter except it merges everything together.
- GroundClutterData should have a bool property, which causes spawners to raycast the ground and spawn stuff on hit.
- Should support foliage weights mode, which adds weights that shaders can use to add fake wind. (eventually)
- For Terraman/Voxelman they have to be registered and freed when chunks unload, also they should eventually handle chunk mesh generation events, if they need to raycast ground.
- When they need to raycast ground, they should wait until a raycast hits, maybe a few params could be added to how long they should try before giving up.
- Remove Ground clutter classes from terraman and voxelman.
- Should have prop support (by spawning in Nodes). It should not be merged.
TiledWalls
- Finish implementing all collider types.
- Finish tiling types + z index support for other tiling that horizontal.
- Flavours + y_size probably has quite a few bugs.
Platforms
- Add export platforms for frt, frt_sdl, and the linux server, so their binaries can be easily added and used from export template (tpz) files.
Binaries, Containers, Releases
- Set up containers and build scripts to automatically build frt and frt_sdl binaries.
- Set up automatically building ios binaries as well.
- Try to set up godot's UWP platform again. Maybe.
- Add all binaries to the project side export and release helper scripts.
- Maybe split TPZ files, so you can install the export templates that you want.
Ports
Enable type information on release
Not sure whether to have this or not. It does increase binary size for release builds by quite a bit.
226103d166
663978e8ba
f0ca73892a
3a8f2dda01
Physics interpolation - Move out of Scenario
- Move VisualServer interpolation data out of Scenario and into VisualServerScene, so the interpolation data and enabled status is now common to all Scenarios.
- Fix physics interpolation in multithreaded mode by ensuring tick and pre-draw are called.
At first glance this might have the side effect of not being able to run multiple physics worlds. If this is true this is not needed.
The fix might be needed though. Double check.
Fix the issue causing the screen to be black after resuming when in low processor mode.
I did the low processor mode fix differently than in upstream godot, so need to check.
If needed then it's probably only the Main::force_redraw() call.
Physics Interpolation - add helper warnings
Not sure if it's needed.
Remove duplicate editor settings definitions
Has to be hand ported, however some of my modules will likely need a similar cleanup, so todo.
Display a modal dialog if threads are unavailable in the web editor
Not sure, if I end up porting it, I'll likely simplify it.
Use parameter classes instead of arguments for all physics queries
Seems like a good idea to have eventually.
Also it should be expanded on, for the apis that returns arrays and dicts. Maybe those could just write results directly to the passed parameter classes, so no allocations occur.
Remove redundant thread sync counter draw_pending
The functions that used it already use a threadsafe FIFO queue to communicate between threads and a sync to have the main thread wait for the render thread.
Fixes #35718
The bugfix is likely needed, however I did not move physics interpolation out of scenario, so this one needs more thought.
[Core] Use std type traits to check operations triviality.
possibly
Added support for building shared libraries on linuxbsd
possibly
This adds shared build support for the scons build.
The build can now be built shared by passing shared=True in the scons command. This does drastically speed up the link time for the godot bin (anecdotally for my machine from 5 seconds to less than a second).
This PR mainly is to aid in faster development as outlined in godotengine/godot-proposals#1796