andlon / dynamecs Goto Github PK

Currently we use type_name() for identifying systems. However, this has several issues. For one, it's not necessarily unique for two different types T1, T2. However, a potentially bigger problem is that the same type could technically produce different names when type_name() is called from different locations in a program. Certainly there are no guarantees here.

It appears a better way might be to make a tag available in the Component trait, which can be used to produce storage tags. I'm thinking something like domain.ComponentName. Then we can provide a derive(Component) macro which allows you to specify domain and derives ComponentName from the type name.

Cool project!

There is missing README.md and the crates.io / docs.rs doesn't have the link to this repository.

This is necessary so that Systems that use entities as references consistently work with the same entity no matter if the universe was initialized by the scenario initializer or restored from a checkpoint.

We also need to store scenario config, maybe app settings etc. in the serialized checkpoint data.

Should also warn or even outright refuse to continue if there are mismatches.

Best idea so far: Refuse to run, but add CLI flag to override this behavior and run anyway (but log the mismatch with a warning).

Currently inserting several components for the same entity is very verbose:

universe.register_insert_component(obj, Name::from("box"));
universe.register_insert_component(obj, initial_state);
universe.register_insert_component(
    obj,
    MassDistribution::for_mesh_with_uniform_density(&mesh, density),
);
universe.register_insert_component(
    obj,
    ElasticSolidMaterial::for_mesh_with_uniform_parameters(&mesh, NeoHookeanMaterial, lame),
);
universe.register_insert_component(obj, FiniteElementDomain { mesh });

We could provide [register_]insert_components functions that allow you to simplify the above to

universe.register_insert_components(obj, (
    Name::from("box"),
    initial_state,
    MassDistribution::for_mesh_with_uniform_density(&mesh, density),
    ElasticSolidMaterial::for_mesh_with_uniform_parameters(&mesh, NeoHookeanMaterial, lame),
    FiniteElementDomain { mesh }
));

The snap documentation suggests to use a buffered reader when decompressing. However, the FrameEncoder docs suggest that buffering is not necessary for writing. So we should insert an intermediate BufReader when restoring, but not when writing.

Need to install a custom panic handler that logs panics (and possibly flushes log writers if necessary)