Hi all!
Happy to announce the release of Ark.jl 0.5. It was a slow burning effort this time, where we tried to make the package more robust in several aspects. It took time and effort, but I can say we made some good progress 
We haven’t introduced any big feature in this release, only small ones, a lot of performance improvements and some breaking changes to improve the API. Performance-wise maybe the most important improvement was due to some internal redesign which allowed us to remove some inference failures causing unnecessary allocations. I was somewhat surprised not to have catched this previously since we have a lot of benchmarks in CI, apparently applying the package to some big simulations helped!
So here are the major changes:
Breaking Changes
The relation API has been greatly improved. Instead of a keyword argument relations, which required to repeat the component information, now, relations are passed directly as pairs without the need of extra information, e.g. before one would had need to write new_entity!(world, (ChildOf(),); relations=(ChildOf => parent,)), while now one needs only new_entity!(world, (ChildOf() => parent,)).
Besides that, instead of requiring components to be defined as relations at struct definition time through a Relationship abstract type, now relation components can have any abstract type and are defined as relations when the World is initialized. This should help extensibility, since any struct can be used as relations, also ones which are naturally so but come from other packages.
Performance Improvements
While moving BeforeIT.jl (a big macroeconomics simulation model) to use Ark.jl, we found some inference issues with the API, but, thanks to some (more) @generated magic we were able to fix those little problematic aspects causing dynamic dispatch. Also, thanks to this, we also noticed some allocations in queries using relation components, and so, we fixed also that for the cause.
Apart from these issues which came from battle testing the package, we also found that many batch operations could have been made faster, achieving 3x faster results in some cases. And also some more caching could have helped in single entity operations, getting a small 10/20% improvement, still significant I would say since that part of the package is already very optimized (I looped through it trying to improve transition handling tens of times).
Features
We added support for sorting (and partitioning) at the level of tables, this way for each table one can sort the entities in it with a single sort_entities! call, which follows the sort! interface.
In general, I think that Ark.jl is getting very robust, the bugs we found were minor things (and so I didn’t described them here), and, also, its API is now in a state with which I don’t have any big complain with.
In any case, happy to receive any suggestions or discuss anything about it!
For a more extensive list of changes, see the CHANGELOG.
I think some of the generated functions could be spared, by trusting a bit more Julia to optimize the code by itself. Though, some others (the majority I think) are instead really fundamental for performance.