iteratively decay with half life

It seems incorrect to always just decay one step, even if the time passed is much larger than the half life interval. In particular when decaying and merging external data, it is probably more important to do this correctly.

Author: joostjager

lightning/src/routing/scoring.rs

@@ -493,14 +493,18 @@ impl ChannelLiquidities {
 				liquidity.decayed_offset(liquidity.max_liquidity_offset_msat, duration_since_epoch, decay_params);
 			liquidity.last_updated = duration_since_epoch;
 
-			// TODO: Call decay multiple times.
-			let elapsed_time =
-				duration_since_epoch.saturating_sub(liquidity.offset_history_last_updated);
-			if elapsed_time > decay_params.historical_no_updates_half_life {
-				let half_life = decay_params.historical_no_updates_half_life.as_secs_f64();
-				if half_life != 0.0 {
+			// Iteratively decay the liquidity history until we've caught up to the current time.
+			let half_life = decay_params.historical_no_updates_half_life.as_secs_f64();
+			if half_life != 0.0 {
+				loop {
+					let elapsed_time =
+						duration_since_epoch.saturating_sub(liquidity.offset_history_last_updated);
+					if elapsed_time <= decay_params.historical_no_updates_half_life {
+						break;
+					}
+
 					liquidity.liquidity_history.decay_buckets(elapsed_time.as_secs_f64() / half_life);
-					liquidity.offset_history_last_updated = duration_since_epoch;
+					liquidity.offset_history_last_updated += decay_params.historical_no_updates_half_life;
 				}
 			}
 			liquidity.min_liquidity_offset_msat != 0 || liquidity.max_liquidity_offset_msat != 0 ||