Zig ECS is a zig port of the fantasic Entt. Entt is highly templated C++ code which depending on your opinion is either a good thing or satan itself in code form. Zig doesn't have the same concept as C++ templates so the templated code was changed over to use Zig's generics and compile time metaprogramming.

Below are examples of a View and a Group, the two main ways to work with entities in the ecs along with the scaffolding code.

Declare some structs to work with:

pub const Velocity = struct { x: f32, y: f32 };
pub const Position = struct { x: f32, y: f32 };

Setup the Registry, which holds the entity data and is where we run our queries:

var reg = ecs.Registry.init(std.testing.allocator);

Create a couple entities and add some components to them

const entity = reg.create();
reg.add(entity, Position{ .x = 0, .y = 0 });
reg.add(entity, Velocity{ .x = 5, .y = 7 });
...

Create and iterate a View that matches all entities with a Velocity and Position component:

var view = reg.view(.{ Velocity, Position }, .{});
var iter = view.entityIterator();
while (iter.next()) |entity| {
    const pos = view.getConst(Position, entity); // readonly copy
    var vel = view.get(Velocity, entity); // mutable
}

The same example using an owning Group and iterating with each:

var group = reg.group(.{ Velocity, Position }, .{}, .{});
group.each(each);
fn each(e: struct { vel: *Velocity, pos: *Position }) void {
    e.pos.*.x += e.vel.x;
    e.pos.*.y += e.vel.y;
}

• View: stores no data in the ECS. Iterates on the fly with no cache to speed up iteration. Start with a View for most things and you can always upgrade to a Group/OwningGroup if you need more speed.
• Group: stores and maintains an accurate list of all the entities matching the query. This query cache speeds up iteration since it already knows exactly which entities match the query. The downside is that it requires memory and cpu cycles to keep the cache up-to-date.
• OwningGroup: for any components in an OwningGroup the actual component storage containers are constantly reordered as entities are added/removed from the Registry. This allows direct iteration with no gaps of the component data. An OwningGroup does not require much extra memory but it is more cpu intensive if there is a lot of component churn (adding/removing of the owned components).