GECS 5.0.0-rc4 - Query Performance Optimization

[csprance]

[5.0.0-rc4] - 2025-10-14 - Query Performance Optimization

🚀 Performance Improvements

Massive Query Cache Key Optimization

85% faster cache key generation leading to dramatic query performance improvements:

• Cache key generation: 283ms → 43ms (85% faster)
• Query caching: 500ms → 4.7ms (99% faster)
• Query with all: 13ms → 0.55ms (96% faster)
• Query with any: 27ms → 5.6ms (79% faster)
• Complex queries: Significantly improved scaling

Technical Details:

• Replaced expensive str(comp) fallbacks with direct get_instance_id() calls
• Eliminated conditional checks in cache key generation hot path
• Implemented polynomial rolling hash with XOR for collision resistance
• Used different prime multipliers (31, 37, 41) for component type separation

Impact:

• Query system now scales linearly instead of exponentially
• ECS performance optimized for large-scale applications (10,000+ entities)
• No performance regressions in any core ECS operations
• Cache effectiveness dramatically improved

See performance test results in reports/perf/ for detailed metrics.

Full Changelog: v5.0.0-rc3...v5.0.0-rc4

---

This discussion was created from the release GECS 5.0.0-rc4 - Query Performance Optimization.

[nageslan]

I have a question.
Since I am also trying to implement ECS myself, I would like to offer the following suggestions. I have attempted to analyze the code by observing it to see if there are any unreasonable aspects, but my brain capacity is not sufficient, so I have tried to directly ask questions. Please forgive my being unreasonable.
Is there a mechanism like (entities) that creates an index when the component changes and the entity changes, so that the system can directly traverse (omitting the step of filtering before each traversal) this dynamically conforming and rule-compliant collection every time?
Also, when querying, use the mask (indexer, there are many optimization methods for this mask-based filtering approach) method to perform batch filtering. When obtaining components, you can optionally directly use the digital index (I found that the efficiency difference between digital indexing and string indexing is too large).
Also, for example, when obtaining the value of a component, store the data of the entity in a continuous array. For instance, query the index of the component in the ready stage of the system, and directly obtain the data of the component through the index in the process stage, etc.