DumiFlex
fdf3d03235
Pure refactor — wildcard_handler.py imports from new shared module instead of defining inline. Combine + fixed_values cycles will reuse the same helper for their seed-lock support. Identical SHA-256 derivation, same call sites. All 210 engine tests pass after the lift.
31 lines
1.2 KiB
Python
31 lines
1.2 KiB
Python
"""Shared seed-derivation helper for module handlers.
|
|
|
|
Every kind that needs reproducible randomness (wildcard pick, combine
|
|
{a|b|c} resolution, fixed_values value alternations) routes through
|
|
this same function so the same `(seed, key)` pair always yields the
|
|
same RNG state. Lifting it out of wildcard_handler.py prevents drift
|
|
when more kinds gain seed-lock support.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import hashlib
|
|
import random
|
|
|
|
|
|
def derive_module_rng(seed: int, key: str) -> random.Random:
|
|
"""Per-module RNG derived from (seed, key).
|
|
|
|
`key` is a stable per-module string — wildcard uses `binding`,
|
|
combine uses `output_var`, fixed_values uses module id. Different
|
|
keys under the same seed yield independent streams so two
|
|
co-resident modules binding different vars roll independently.
|
|
|
|
Used for BOTH locked and unlocked picks — applying the same
|
|
derivation either way is what lets lock capture the visible
|
|
roll: locking with `locked_seed = chain_seed` reproduces the
|
|
unlocked pick exactly because both paths derive from
|
|
`sha256(chain_seed:key)`.
|
|
"""
|
|
digest = hashlib.sha256(f"{int(seed)}:{key}".encode()).hexdigest()
|
|
return random.Random(int(digest[:16], 16))
|