Files
mruby-mruby/include
Yukihiro "Matz" Matsumoto 2371b52ab4 khash: optimize structure size by 50% with single data pointer
BREAKING CHANGE: khash field access macros now require type name parameter

Replace individual pointer fields (keys, vals, ed_flags) with single data
pointer and address calculation functions. This reduces khash structure
size from 32 to 16 bytes (50% reduction) while maintaining performance
through pointer caching in hot paths.

Structure changes:
- Single void *data field replaces keys/vals/ed_flags pointers
- Address calculation functions compute array locations on demand
- Hot path functions cache calculated pointers for performance

API changes (BREAKING):
- kh_key(h, x)      -> kh_key(typename, h, x)
- kh_val(h, x)      -> kh_val(typename, h, x)
- kh_exist(h, x)    -> kh_exist(typename, h, x)
- kh_value(h, x)    -> kh_value(typename, h, x)
- KHASH_FOREACH()   -> KHASH_FOREACH(typename, ...)

Migration required:
- mruby-metaprog: 4 call sites updated (familiar macro names, just add type parameter)
- mruby-array-ext: no changes needed (uses function-style API)
- External users: add type name as first parameter to field access macros

Benefits:
- 50% memory reduction per hash table (32 -> 16 bytes)
- 464 bytes total memory savings in mrbtest execution
- Better cache locality with smaller structures
- Optimized hot path performance with pointer caching
- Consistent with mruby memory-first design priority

Co-authored-by: Claude <noreply@anthropic.com>
2025-08-14 10:53:02 +09:00
..