diff options
| author | (quasar) nebula <qznebula@protonmail.com> | 2025-11-19 22:34:17 -0400 |
|---|---|---|
| committer | (quasar) nebula <qznebula@protonmail.com> | 2025-11-25 07:06:48 -0400 |
| commit | aefb53631314295e1574847683bbfa5fec98d536 (patch) | |
| tree | 3080a14161af7ce9d2beea7da540700fd8dcfc9f /src/data/patches.js | |
| parent | 0b963b6e6b4f90c80e7369a804407f32e69c7ede (diff) | |
data, infra: better CacheableObject prototype inheritence
Basically, properties (as written in source code) are only registered once EVER, instead of "also" on subclasses, which was previously the case - we were literally copying over the descriptors and then re-evaluating as though they were new. That's the goal, but to make it happen, we had to redo how default property values are initialized. Things no longer self- prepare by making a "storage" for all their cached values with null, or defaults, filled in - instead they create a new object which inherits those defaults from their prototype. This object respectively has the prototype of the superclass' prototype's defaults, and so on. A nice effect is that validating defaults against update.validate specs now happens during prototype finalization, instead of every time any instance is instantiated. We also had to do the same kind of inheritence shenanigans with the full propertyDescriptors specs themselves. The real-world performance boost for this is probably more or less zero RIGHT NOW, because Thing subclassing is used crazy sparingly (and there's still work to make it really nice). But if [[Prototype]] chains don't really take memory, then possibly this cleans up a lot of extremely similar objects where every new Thing property multiplies memory usage across all instances of that thing, regardless how few actually set a value. Maybe! Who knows what V8 really optimizes, but maybe!
Diffstat (limited to 'src/data/patches.js')
0 files changed, 0 insertions, 0 deletions