Report inconsistencies between derivative cache linker and compute() …

…early

src/main/java/com/machinezoo/foxcache/CachedData.java

@@ -9,33 +9,37 @@
 
 @CodeIssue("Downgrade softref to weakref after one second. Anything that needs longer caching should be persisted.")
 class CachedData {
-	static final ConcurrentMap<LazyCache<?>, Object> lazy = new ConcurrentHashMap<>();
-	/*
-	 * Soft-valued cache may cause extremely inefficient GC behavior:
-	 * https://bugs.openjdk.java.net/browse/JDK-6912889
-	 * 
-	 * It is however very simple and it will use all RAM that is allocated to Java process,
-	 * which is usually some fraction of physical RAM.
-	 * This cache can be tuned indirectly with -Xmx and -XX:SoftRefLRUPolicyMSPerMB.
-	 * 
-	 * Cached value is wrapped in Optional, because Guava cache does not tolerate null values.
-	 */
-	static final LoadingCache<ComputeCache<?>, Optional<?>> compute = CacheBuilder.newBuilder()
-		.softValues()
-		.build(CacheLoader.from(k -> Optional.ofNullable(k.compute())));
-	/*
-	 * Like in ComputeCaches, just specialized for DerivativeCache.
-	 */
-	static final LoadingCache<DerivativeCache<?>, ReactiveLazy<CacheDerivative<Object>>> derivative = CacheBuilder.newBuilder()
-		.softValues()
-		.build(CacheLoader.from(k -> materialize(k)));
-	private static <T> ReactiveLazy<CacheDerivative<Object>> materialize(DerivativeCache<T> cache) {
-		return new ReactiveLazy<>(() -> CacheDerivative.capture(() -> {
-			/*
-			 * Touch the cache just in case there are extra dependencies there.
-			 */
-			cache.touch();
-			return cache.compute();
-		}));
-	}
+    static final ConcurrentMap<LazyCache<?>, Object> lazy = new ConcurrentHashMap<>();
+    /*
+     * Soft-valued cache may cause extremely inefficient GC behavior:
+     * https://bugs.openjdk.java.net/browse/JDK-6912889
+     * 
+     * It is however very simple and it will use all RAM that is allocated to Java process,
+     * which is usually some fraction of physical RAM.
+     * This cache can be tuned indirectly with -Xmx and -XX:SoftRefLRUPolicyMSPerMB.
+     * 
+     * Cached value is wrapped in Optional, because Guava cache does not tolerate null values.
+     */
+    static final LoadingCache<ComputeCache<?>, Optional<?>> compute = CacheBuilder.newBuilder()
+        .softValues()
+        .build(CacheLoader.from(k -> Optional.ofNullable(k.compute())));
+    /*
+     * Like in ComputeCaches, just specialized for DerivativeCache.
+     */
+    static final LoadingCache<DerivativeCache<?>, ReactiveLazy<CacheDerivative<Object>>> derivative = CacheBuilder.newBuilder()
+        .softValues()
+        .build(CacheLoader.from(k -> materialize(k)));
+    private static <T> ReactiveLazy<CacheDerivative<Object>> materialize(DerivativeCache<T> cache) {
+        return new ReactiveLazy<>(() -> CacheDerivative.capture(() -> {
+            /*
+             * Touch the cache just in case there are extra dependencies there.
+             */
+            cache.touch();
+            /*
+             * Report inconsistencies between linker and compute() early.
+             */
+            CacheInput.get().freeze();
+            return cache.compute();
+        }));
+    }
 }