Query evaluation against outdated version of entity in nested component removal requests
evstinik opened this issue · 3 comments
First of all, thanks for your library. I found it useful.
Now, to the problem:
Imagine that you have a following effect: When component A is removed, remove also component B.
At the time we remove component B all queries are evaluated against a version of entity that doesn't have B (correct), but still has A (incorrect).
I'm having a similar issue, it seems to be a race condition with removing and reindexing:
When removing an entity from the world, some onRemoved
subscribers will try to do some cleanup and run a world.removeComponent
. This will reindex the removed component, adding it back to any buckets it was just removed from. Then world.remove
will finish and the entity will no longer be in the world bucket. so each loop of a system, when I pull something out of a query I have to check if the entity is still in the world and if not I remove it from the bucket.
There should be a way to skip reindexing entities that are being removed from the root world bucket
Otherwise the library has been very helpful =D
I'm assuming these are two separate (even if similar sounding) issues -- would you be willing to provide a (failing) test case? I'd be happy to fix and improve these situations, but I need to better understand what's going on.
In the meantime, it sounds like a potential workaround for these issues would be to queue the follow-up work (like removing the second component) using queueMicrotask.
Hello! Great project, @hmans! I really enjoy the simplicity, yet still having strong type safety! 🚀
I have the same issue as reported by @abraini-Nascent.
I would guess it's the same issue as reported by @evstinik, although that one looks like it might be triggered by a world.removeComponent()
, and I haven't investigated that situation.
I'll present some diffs (rather than a PR) because I'm not sure of the accuracy of the proposed solutions. The reorder change drastically changes the behaviour of things and might break downstream code / listeners, and my gut tells me the isRemoving
"feels" a bit band-aid-ey 🩹
Here's a diff that adds a failing unit test demonstrating the bug when triggered by a world.remove(entity)
:
main...joelmalone:miniplex:add-failing-test
Here's a diff showing a solution that fixes it by reordering the some code in remove()
:
joelmalone/miniplex@add-failing-test...joelmalone:miniplex:fix-bug-by-reordering
Here a diff showing a solution that fixes the bug by refusing to re-add a removed entity during index evaluation:
joelmalone/miniplex@add-failing-test...joelmalone:miniplex:fix-bug-by-isRemoving