mirror of
https://github.com/bitcoin/bitcoin.git
synced 2026-09-13 06:04:42 +02:00
b9d573e4a9
policy_estimator_io deliberately reuses a CBlockPolicyEstimator because constructing one for every fuzz input severely reduces throughput. However, Read() returns true for an incompatible old fee estimates file without replacing the estimator state. The target then calls Write() with state loaded by a previous input, making coverage depend on corpus order. Return false for incompatible files so the target skips Write() when no state was loaded. This keeps the estimator reuse optimization instead of resetting the expensive object before every fuzz input. For the in-tree production caller, incompatible files remain non-fatal and the estimator still starts from its default state. Read() now reports failure, so startup emits one additional non-fatal warning. Node startup and estimator state are unchanged, as is RPC behavior. Co-authored-by: maflcko <6399679+maflcko@users.noreply.github.com>