next »

[Ordnas]

 

[JWki]

Pretty fucking happy about having written a big part of a new reflection system for C++, again based on libclang and code generation. Already has proper enum reflection with enum constants, class reflection including (single) inheritance, field reflection, ability to create (default constructible) types by name, etc.
Content with the current state given that it was mostly a day worth of work.

[InfiniteStateMachine]

Is the code open source? I've been dealing with C++ reflection systems a LOT lately and I'm interested to see how people implemented them.

Funny I dislike how so many things are going into modern c++ but I WISH it had a decent reflection system like most other languages.

[JWki]

Is the code open source? I've been dealing with C++ reflection systems a LOT lately and I'm interested to see how people implemented them.

Funny I dislike how so many things are going into modern c++ but I WISH it had a decent reflection system like most other languages.

The code is literally no older than two days, so I'm not confident in releasing it to a public audience atm but I'm happy to talk about how it works.
The basic gist of it is I wrote some relatively straightforward code that manually registers types with a template based interface to access that information and then I used libclang to parse my headers and generate that code for me.

And yeah the lack of reflection is what bugs me most about C++.

[InfiniteStateMachine]

Cool. So if you had class X it would kind of look like this? :

class SomeClass
{
public :
int X();
int Y();

private:
int Z();
}

And then somewhere else

RegisterReflectedClass<SomeClass>();

[JWki]

Sort of yeah, I'll post some code later tonight if I remember.

[InfiniteStateMachine]

So I guess this would go in both the happy and grumpy programmer threads.

I'm having to learn regular expressions for real as I have to do language parsing in a language that doesn't have pattern matching (up until now I've always been able to get away with that).

It's going pretty good but I wish the language I was parsing wasn't C++. Thankfully I only need to do an extremely limited subset.

So anyways. Regular expressions are interesting and they can be pretty cool in some situations and do the job a lot better than me manually parsing with basic language constructs. On the other hand some problems include different languages have different implementations of regex, the severe performance issues, the global flags and just the general readability of an expression is grumpy.

I do see a lot of value in it and I'm glad I'm finally getting around to learning them. I've been putting this off for over a decade :D

[JWki]

Oh is this related to reflection too?

Anyways so here's some code

Code:

class MetaType;

template <class T>
MetaType* GetMetaType();

This is the basic interface to retrieve the meta type for a type.
Not how it is only declared, but not defined.

A definition looks like this:

Code:

template <>
reflect::MetaType* reflect::GetMetaType<Mesh>()
{
    static MetaProperty properties[] = {
        MetaProperty(GetMetaType<gfx::Buffer>(), "vertexBuffer", 0),
        MetaProperty(GetMetaType<gfx::Buffer>(), "indexBuffer", 0),
        MetaProperty(GetMetaType<unsigned int>(), "vertexOffset", 0),
        MetaProperty(GetMetaType<unsigned int>(), "vertexStride", 0),
        MetaProperty(GetMetaType<unsigned int>(), "numElements", 0),
    };
    auto ctor = []() { return static_cast<void*>(new Mesh()); };
    static auto metaType = new MetaClass("Mesh", "Mesh", sizeof(Mesh), properties, 5, nullptr, ctor);
    return metaType;
}

I actually started out with just typing that out for a few types to see what the API should look like. Obviously that's not really feasible for day to day use so I wrote a rough something using libclang that can generate those definitions. Basically it runs as a pre-built step for each project and generates a single .cpp file that contains these defintions for all the types.
Would work as well with multiple .cpp files, like one for each header or something.

Then there's a really simple runtime API and also a mechanism to register type instances with a per-binary global list so you can iterate all the types to retrieve all implementations of a given interface for example.
This also supports registration across ABI boundaries - atm I'm experimenting with a setup where the main app loads modules dynamically and modules can basically bind their type information into the main apps registry and then use that registry for itself so basically all modules have access to all the type information even when they don't have access to the actual c++ definitions.

[gimymblert]

I can finally move on with my planet generation, I got stuck on a hard problem for very long and it didn't go away easily after I solve it, it really tried to screw me by sending a last curveball in the form of a trolling with a trivial problem, but damn I'm happy I'm pas that, everything is square now, I can port all my experiment verbatim. Although there is a last challenge, I must see if I can connect properly the different tile on the face of the cube across seam.

[InfiniteStateMachine]

Oh is this related to reflection too?

Haha not directly but reflection macros are something I have to take into account :D

Interesting implementation. Is this for self study mainly or are you doing something like a game editor?

[JWki]

Oh is this related to reflection too?

Haha not directly but reflection macros are something I have to take into account :D

Interesting implementation. Is this for self study mainly or are you doing something like a game editor?

This is for a game related project editor is only one aspect of it tho.
Not sure if I'm going to go down this route however. There's a bit of bloat smell due to the tooling and stuff around it.

[gimymblert]

after years YEARS



Okay now the crossing of edge and corner 

[gimymblert]

Wrote a simple voxel landscape renderer ala comanche on blitz3D, took age to debug

[InfiniteStateMachine]

cool!

[gimymblert]

I'm getting somewhere at least, now let' see if I can get adjacency right

[oahda]

I've quite forgotten what you were going to be using this for!

[gimymblert]

no man's sky clone

[gimymblert]

Progress +1,
got the projection to work in all direction:

remain in order of complexity
1 - face adjacency (check of boundary + hardcoded selection)
2 - tile adjacency bleed over (basically assigning tile to face when bleeding over the current one see 1)
3 - circular array update (basically when moving only update tile that are in the offset direction)

The big test is if I can iterate the neighbor of a tile to spawn connection line between them, with consistent hashed seed.

Then I'll stop a bit to tackle convexhull again, which is needed for further pcg

[ferreiradaselva]

All I know is that that is a beautiful ball.

I have a question. Those quads that form the sphere, does the triangles bend or are they perfectly plane?

[gimymblert]

I place all vertex on a sphere position they follow the curvature, so I guess the QUAD bend, triangle don't, they can't. BUT I'm not sure the curvature will be any significant.