You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Lightyear has a bunch of component-specific systems that are the same for every component: interpolation, prediction, replication, events systems, etc.
Pros
code is simpler?
access to type information
Cons
many different systems: performance issues
potentially cannot run in parallel
This PR in bevy: bevyengine/bevy#12936 shows that this can have a significant costs.
There was system per event type, and the PR updates it to only one system that runs on all events.
Maybe we can do something simialr?
Maybe we could use bevyengine/bevy#13123 to build a system that queries all the relevant types into a single system while still retaining some parallelism
Create a system that queries any of A,B,C (using dynamic query builder)
Use the dynamic query to get a Ptr/component_id, and we can then use the component_registry
The text was updated successfully, but these errors were encountered:
cBournhonesque
changed the title
Maybe simplify component-specific systems with https://github.com/bevyengine/bevy/pull/13123
Maybe parallelize component-specific systems with https://github.com/bevyengine/bevy/pull/13123
Apr 30, 2024
cBournhonesque
changed the title
Maybe parallelize component-specific systems with https://github.com/bevyengine/bevy/pull/13123
Avoid having one system per component
May 27, 2024
Lightyear has a bunch of component-specific systems that are the same for every component: interpolation, prediction, replication, events systems, etc.
Pros
Cons
This PR in bevy: bevyengine/bevy#12936 shows that this can have a significant costs.
There was system per event type, and the PR updates it to only one system that runs on all events.
Maybe we can do something simialr?
Maybe we could use bevyengine/bevy#13123 to build a system that queries all the relevant types into a single system while still retaining some parallelism
The text was updated successfully, but these errors were encountered: