ThreadLocal

ThreadLocal不是用来解决对象共享访问问题的，而主要是提供了保持对象的方法和避免参数传递的方便的对象访问方式。归纳了两点：
1。每个线程中都有一个自己的ThreadLocalMap类对象，可以将线程自己的对象保持到其中，各管各的，线程可以正确的访问到自己的对象。
2。将一个共用的ThreadLocal静态实例作为key，将不同对象的引用保存到不同线程的ThreadLocalMap中，然后在线程执行的各处通过这个静态ThreadLocal实例的get()方法取得自己线程保存的那个对象，避免了将这个对象作为参数传递的麻烦。
当然如果要把本来线程共享的对象通过ThreadLocal.set()放到线程中也可以，可以实现避免参数传递的访问方式，但是要注意get()到的是那同一个共享对象，并发访问问题要靠其他手段来解决。但一般来说线程共享的对象通过设置为某类的静态变量就可以实现方便的访问了，似乎没必要放到线程中。

ThreadLocal的应用场合，我觉得最适合的是按线程多实例（每个线程对应一个实例）的对象的访问，并且这个对象很多地方都要用到。
1.42
/*
* @(#)ThreadLocal.java 1.42 06/06/23
*
* Copyright 2006 Sun Microsystems, Inc. All rights reserved.
* SUN PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.
*/

package java.lang;
import java.lang.ref.*;
import java.util.concurrent.atomic.AtomicInteger;

/**
* This class provides thread-local variables.  These variables differ from
* their normal counterparts in that each thread that accesses one (via its
* <tt>get</tt> or <tt>set</tt> method) has its own, independently initialized
* copy of the variable.  <tt>ThreadLocal</tt> instances are typically private
* static fields in classes that wish to associate state with a thread (e.g.,
* a user ID or Transaction ID).
*
* <p>For example, the class below generates unique identifiers local to each
* thread.
* A thread's id is
* assigned the first time it invokes <tt>UniqueThreadIdGenerator.getCurrentThreadId()</tt> and remains unchanged on subsequent calls.
* <pre>
* import java.util.concurrent.atomic.AtomicInteger;
*
* public class UniqueThreadIdGenerator {
*
*     private static final AtomicInteger uniqueId = new AtomicInteger(0);
*
*     private static final ThreadLocal &lt; Integer > uniqueNum =
*         new ThreadLocal &lt; Integer > () {
*             &#64;Override protected Integer initialValue() {
*                 return uniqueId.getAndIncrement();
*         }
*     };
*
*     public static int getCurrentThreadId() {
*         return uniqueId.get();
*     }
* } // UniqueThreadIdGenerator
* </pre>
* <p>Each thread holds an implicit reference to its copy of a thread-local
* variable as long as the thread is alive and the <tt>ThreadLocal</tt>
* instance is accessible; after a thread goes away, all of its copies of
* thread-local instances are subject to garbage collection (unless other
* references to these copies exist).
*
* @author  Josh Bloch and Doug Lea
* @version 1.42, 06/23/06
* @since   1.2
*/
public class ThreadLocal<T> {
    /**
     * ThreadLocals rely on per-thread linear-probe hash maps attached
     * to each thread (Thread.threadLocals and
     * inheritableThreadLocals).  The ThreadLocal objects act as keys,
     * searched via threadLocalHashCode.  This is a custom hash code
     * (useful only within ThreadLocalMaps) that eliminates collisions
     * in the common case where consecutively constructed ThreadLocals
     * are used by the same threads, while remaining well-behaved in
     * less common cases.
     */
    private final int threadLocalHashCode = nextHashCode();

    /**
     * The next hash code to be given out. Updated atomically. Starts at
     * zero.
     */
    private static AtomicInteger nextHashCode =
new AtomicInteger();

    /**
     * The difference between successively generated hash codes - turns
     * implicit sequential thread-local IDs into near-optimally spread
     * multiplicative hash values for power-of-two-sized tables.
     */
    private static final int HASH_INCREMENT = 0x61c88647;

    /**
     * Returns the next hash code.
     */
    private static int nextHashCode() {
return nextHashCode.getAndAdd(HASH_INCREMENT);
    }

    /**
     * Returns the current thread's "initial value" for this
     * thread-local variable.  This method will be invoked the first
     * time a thread accesses the variable with the {@link #get}
     * method, unless the thread previously invoked the {@link #set}
     * method, in which case the <tt>initialValue</tt> method will not
     * be invoked for the thread.  Normally, this method is invoked at
     * most once per thread, but it may be invoked again in case of
     * subsequent invocations of {@link #remove} followed by {@link #get}.
     *
     * <p>This implementation simply returns <tt>null</tt>; if the
     * programmer desires thread-local variables to have an initial
     * value other than <tt>null</tt>, <tt>ThreadLocal</tt> must be
     * subclassed, and this method overridden.  Typically, an
     * anonymous inner class will be used.
     *
     * @return the initial value for this thread-local
     */
    protected T initialValue() {
        return null;
    }

    /**
     * Creates a thread local variable.
     */
    public ThreadLocal() {
    }

    /**
     * Returns the value in the current thread's copy of this
     * thread-local variable.  If the variable has no value for the
     * current thread, it is first initialized to the value returned
     * by an invocation of the {@link #initialValue} method.
     *
     * @return the current thread's value of this thread-local
     */
    public T get() {
        Thread t = Thread.currentThread();
        ThreadLocalMap map = getMap(t);
        if (map != null) {
            ThreadLocalMap.Entry e = map.getEntry(this);
            if (e != null)
                return (T)e.value;
        }
        return setInitialValue();
    }

    /**
     * Variant of set() to establish initialValue. Used instead
     * of set() in case user has overridden the set() method.
     *
     * @return the initial value
     */
    private T setInitialValue() {
        T value = initialValue();
        Thread t = Thread.currentThread();
        ThreadLocalMap map = getMap(t);
        if (map != null)
            map.set(this, value);
        else
            createMap(t, value);
        return value;
    }

    /**
     * Sets the current thread's copy of this thread-local variable
     * to the specified value.  Most subclasses will have no need to
     * override this method, relying solely on the {@link #initialValue}
     * method to set the values of thread-locals.
     *
     * @param value the value to be stored in the current thread's copy of
     *        this thread-local.
     */
    public void set(T value) {
        Thread t = Thread.currentThread();
        ThreadLocalMap map = getMap(t);
        if (map != null)
            map.set(this, value);
        else
            createMap(t, value);
    }

    /**
     * Removes the current thread's value for this thread-local
     * variable.  If this thread-local variable is subsequently
     * {@linkplain #get read} by the current thread, its value will be
     * reinitialized by invoking its {@link #initialValue} method,
     * unless its value is {@linkplain #set set} by the current thread
     * in the interim.  This may result in multiple invocations of the
     * <tt>initialValue</tt> method in the current thread.
     *
     * @since 1.5
     */
     public void remove() {
         ThreadLocalMap m = getMap(Thread.currentThread());
         if (m != null)
             m.remove(this);
     }

    /**
     * Get the map associated with a ThreadLocal. Overridden in
     * InheritableThreadLocal.
     *
     * @param  t the current thread
     * @return the map
     */
    ThreadLocalMap getMap(Thread t) {
        return t.threadLocals;
    }

    /**
     * Create the map associated with a ThreadLocal. Overridden in
     * InheritableThreadLocal.
     *
     * @param t the current thread
     * @param firstValue value for the initial entry of the map
     * @param map the map to store.
     */
    void createMap(Thread t, T firstValue) {
        t.threadLocals = new ThreadLocalMap(this, firstValue);
    }

    /**
     * Factory method to create map of inherited thread locals.
     * Designed to be called only from Thread constructor.
     *
     * @param  parentMap the map associated with parent thread
     * @return a map containing the parent's inheritable bindings
     */
    static ThreadLocalMap createInheritedMap(ThreadLocalMap parentMap) {
        return new ThreadLocalMap(parentMap);
    }

    /**
     * Method childValue is visibly defined in subclass
     * InheritableThreadLocal, but is internally defined here for the
     * sake of providing createInheritedMap factory method without
     * needing to subclass the map class in InheritableThreadLocal.
     * This technique is preferable to the alternative of embedding
     * instanceof tests in methods.
     */
    T childValue(T parentValue) {
        throw new UnsupportedOperationException();
    }

    /**
     * ThreadLocalMap is a customized hash map suitable only for
     * maintaining thread local values. No operations are exported
     * outside of the ThreadLocal class. The class is package private to
     * allow declaration of fields in class Thread.  To help deal with
     * very large and long-lived usages, the hash table entries use
     * WeakReferences for keys. However, since reference queues are not
     * used, stale entries are guaranteed to be removed only when
     * the table starts running out of space.
     */
    static class ThreadLocalMap {

        /**
         * The entries in this hash map extend WeakReference, using
         * its main ref field as the key (which is always a
         * ThreadLocal object).  Note that null keys (i.e. entry.get()
         * == null) mean that the key is no longer referenced, so the
         * entry can be expunged from table.  Such entries are referred to
         * as "stale entries" in the code that follows.
         */
        static class Entry extends WeakReference<ThreadLocal> {
            /** The value associated with this ThreadLocal. */
            Object value;

            Entry(ThreadLocal k, Object v) {
                super(k);
                value = v;
            }
        }

        /**
         * The initial capacity -- MUST be a power of two.
         */
        private static final int INITIAL_CAPACITY = 16;

        /**
         * The table, resized as necessary.
         * table.length MUST always be a power of two.
         */
        private Entry[] table;

        /**
         * The number of entries in the table.
         */
        private int size = 0;

        /**
         * The next size value at which to resize.
         */
        private int threshold; // Default to 0

        /**
         * Set the resize threshold to maintain at worst a 2/3 load factor.
         */
        private void setThreshold(int len) {
            threshold = len * 2 / 3;
        }

        /**
         * Increment i modulo len.
         */
        private static int nextIndex(int i, int len) {
            return ((i + 1 < len) ? i + 1 : 0);
        }

        /**
         * Decrement i modulo len.
         */
        private static int prevIndex(int i, int len) {
            return ((i - 1 >= 0) ? i - 1 : len - 1);
        }

        /**
         * Construct a new map initially containing (firstKey, firstValue).
         * ThreadLocalMaps are constructed lazily, so we only create
         * one when we have at least one entry to put in it.
         */
        ThreadLocalMap(ThreadLocal firstKey, Object firstValue) {
            table = new Entry[INITIAL_CAPACITY];
            int i = firstKey.threadLocalHashCode & (INITIAL_CAPACITY - 1);
            table[i] = new Entry(firstKey, firstValue);
            size = 1;
            setThreshold(INITIAL_CAPACITY);
        }

        /**
         * Construct a new map including all Inheritable ThreadLocals
         * from given parent map. Called only by createInheritedMap.
         *
         * @param parentMap the map associated with parent thread.
         */
        private ThreadLocalMap(ThreadLocalMap parentMap) {
            Entry[] parentTable = parentMap.table;
            int len = parentTable.length;
            setThreshold(len);
            table = new Entry[len];

            for (int j = 0; j < len; j++) {
                Entry e = parentTable[j];
                if (e != null) {
                    ThreadLocal key = e.get();
                    if (key != null) {
                        Object value = key.childValue(e.value);
                        Entry c = new Entry(key, value);
                        int h = key.threadLocalHashCode & (len - 1);
                        while (table[h] != null)
                            h = nextIndex(h, len);
                        table[h] = c;
                        size++;
                    }
                }
            }
        }

        /**
         * Get the entry associated with key.  This method
         * itself handles only the fast path: a direct hit of existing
         * key. It otherwise relays to getEntryAfterMiss.  This is
         * designed to maximize performance for direct hits, in part
         * by making this method readily inlinable.
         *
         * @param  key the thread local object
         * @return the entry associated with key, or null if no such
         */
        private Entry getEntry(ThreadLocal key) {
            int i = key.threadLocalHashCode & (table.length - 1);
            Entry e = table[i];
            if (e != null && e.get() == key)
                return e;
            else
                return getEntryAfterMiss(key, i, e);
        }

        /**
         * Version of getEntry method for use when key is not found in
         * its direct hash slot.
         *
         * @param  key the thread local object
         * @param  i the table index for key's hash code
         * @param  e the entry at table[i]
         * @return the entry associated with key, or null if no such
         */
        private Entry getEntryAfterMiss(ThreadLocal key, int i, Entry e) {
            Entry[] tab = table;
            int len = tab.length;

            while (e != null) {
                ThreadLocal k = e.get();
                if (k == key)
                    return e;
                if (k == null)
                    expungeStaleEntry(i);
                else
                    i = nextIndex(i, len);
                e = tab[i];
            }
            return null;
        }

        /**
         * Set the value associated with key.
         *
         * @param key the thread local object
         * @param value the value to be set
         */
        private void set(ThreadLocal key, Object value) {

            // We don't use a fast path as with get() because it is at
            // least as common to use set() to create new entries as
            // it is to replace existing ones, in which case, a fast
            // path would fail more often than not.

            Entry[] tab = table;
            int len = tab.length;
            int i = key.threadLocalHashCode & (len-1);

            for (Entry e = tab[i];
e != null;
e = tab[i = nextIndex(i, len)]) {
                ThreadLocal k = e.get();

                if (k == key) {
                    e.value = value;
                    return;
                }

                if (k == null) {
                    replaceStaleEntry(key, value, i);
                    return;
                }
            }

            tab[i] = new Entry(key, value);
            int sz = ++size;
            if (!cleanSomeSlots(i, sz) && sz >= threshold)
                rehash();
        }

        /**
         * Remove the entry for key.
         */
        private void remove(ThreadLocal key) {
            Entry[] tab = table;
            int len = tab.length;
            int i = key.threadLocalHashCode & (len-1);
            for (Entry e = tab[i];
e != null;
e = tab[i = nextIndex(i, len)]) {
                if (e.get() == key) {
                    e.clear();
                    expungeStaleEntry(i);
                    return;
                }
            }
        }

        /**
         * Replace a stale entry encountered during a set operation
         * with an entry for the specified key.  The value passed in
         * the value parameter is stored in the entry, whether or not
         * an entry already exists for the specified key.
         *
         * As a side effect, this method expunges all stale entries in the
         * "run" containing the stale entry.  (A run is a sequence of entries
         * between two null slots.)
         *
         * @param  key the key
         * @param  value the value to be associated with key
         * @param  staleSlot index of the first stale entry encountered while
         *         searching for key.
         */
        private void replaceStaleEntry(ThreadLocal key, Object value,
                                       int staleSlot) {
            Entry[] tab = table;
            int len = tab.length;
            Entry e;

            // Back up to check for prior stale entry in current run.
            // We clean out whole runs at a time to avoid continual
            // incremental rehashing due to garbage collector freeing
            // up refs in bunches (i.e., whenever the collector runs).
            int slotToExpunge = staleSlot;
            for (int i = prevIndex(staleSlot, len);
(e = tab[i]) != null;
                 i = prevIndex(i, len))
                if (e.get() == null)
                    slotToExpunge = i;

            // Find either the key or trailing null slot of run, whichever
            // occurs first
            for (int i = nextIndex(staleSlot, len);
(e = tab[i]) != null;
                 i = nextIndex(i, len)) {
                ThreadLocal k = e.get();

                // If we find key, then we need to swap it
                // with the stale entry to maintain hash table order.
                // The newly stale slot, or any other stale slot
                // encountered above it, can then be sent to expungeStaleEntry
                // to remove or rehash all of the other entries in run.
                if (k == key) {
                    e.value = value;

                    tab[i] = tab[staleSlot];
                    tab[staleSlot] = e;

                    // Start expunge at preceding stale entry if it exists
                    if (slotToExpunge == staleSlot)
                        slotToExpunge = i;
                    cleanSomeSlots(expungeStaleEntry(slotToExpunge), len);
                    return;
                }

                // If we didn't find stale entry on backward scan, the
                // first stale entry seen while scanning for key is the
                // first still present in the run.
                if (k == null && slotToExpunge == staleSlot)
                    slotToExpunge = i;
            }

            // If key not found, put new entry in stale slot
            tab[staleSlot].value = null;  
            tab[staleSlot] = new Entry(key, value);

            // If there are any other stale entries in run, expunge them
            if (slotToExpunge != staleSlot)
                cleanSomeSlots(expungeStaleEntry(slotToExpunge), len);
        }

        /**
         * Expunge a stale entry by rehashing any possibly colliding entries
         * lying between staleSlot and the next null slot.  This also expunges
         * any other stale entries encountered before the trailing null.  See
         * Knuth, Section 6.4
         *
         * @param staleSlot index of slot known to have null key
         * @return the index of the next null slot after staleSlot
         * (all between staleSlot and this slot will have been checked
         * for expunging).
         */
        private int expungeStaleEntry(int staleSlot) {
            Entry[] tab = table;
            int len = tab.length;

            // expunge entry at staleSlot
            tab[staleSlot].value = null;  
            tab[staleSlot] = null;
            size--;

            // Rehash until we encounter null
            Entry e;
            int i;
            for (i = nextIndex(staleSlot, len);
(e = tab[i]) != null;
                 i = nextIndex(i, len)) {
                ThreadLocal k = e.get();
                if (k == null) {
                    e.value = null;
                    tab[i] = null;
                    size--;
                } else {
                    int h = k.threadLocalHashCode & (len - 1);
                    if (h != i) {
                        tab[i] = null;

                        // Unlike Knuth 6.4 Algorithm R, we must scan until
                        // null because multiple entries could have been stale.
                        while (tab[h] != null)
                            h = nextIndex(h, len);
                        tab[h] = e;
                    }
                }
            }
            return i;
        }

        /**
         * Heuristically scan some cells looking for stale entries.
         * This is invoked when either a new element is added, or
         * another stale one has been expunged. It performs a
         * logarithmic number of scans, as a balance between no
         * scanning (fast but retains garbage) and a number of scans
         * proportional to number of elements, that would find all
         * garbage but would cause some insertions to take O(n) time.
         *
         * @param i a position known NOT to hold a stale entry. The
         * scan starts at the element after i.
         *
         * @param n scan control: <tt>log2(n)</tt> cells are scanned,
         * unless a stale entry is found, in which case
         * <tt>log2(table.length)-1</tt> additional cells are scanned.
         * When called from insertions, this parameter is the number
         * of elements, but when from replaceStaleEntry, it is the
         * table length. (Note: all this could be changed to be either
         * more or less aggressive by weighting n instead of just
         * using straight log n. But this version is simple, fast, and
         * seems to work well.)
         *
         * @return true if any stale entries have been removed.
         */
        private boolean cleanSomeSlots(int i, int n) {
            boolean removed = false;
            Entry[] tab = table;
            int len = tab.length;
            do {
                i = nextIndex(i, len);
                Entry e = tab[i];
                if (e != null && e.get() == null) {
                    n = len;
                    removed = true;
                    i = expungeStaleEntry(i);
                }
            } while ( (n >>>= 1) != 0);
            return removed;
        }

        /**
         * Re-pack and/or re-size the table. First scan the entire
         * table removing stale entries. If this doesn't sufficiently
         * shrink the size of the table, double the table size.
         */
        private void rehash() {
            expungeStaleEntries();

            // Use lower threshold for doubling to avoid hysteresis
            if (size >= threshold - threshold / 4)
                resize();
        }

        /**
         * Double the capacity of the table.
         */
        private void resize() {
            Entry[] oldTab = table;
            int oldLen = oldTab.length;
            int newLen = oldLen * 2;
            Entry[] newTab = new Entry[newLen];
            int count = 0;

            for (int j = 0; j < oldLen; ++j) {
                Entry e = oldTab[j];
                if (e != null) {
                    ThreadLocal k = e.get();
                    if (k == null) {
                        e.value = null; // Help the GC
                    } else {
                        int h = k.threadLocalHashCode & (newLen - 1);
                        while (newTab[h] != null)
                            h = nextIndex(h, newLen);
                        newTab[h] = e;
                        count++;
                    }
                }
            }

            setThreshold(newLen);
            size = count;
            table = newTab;
        }

        /**
         * Expunge all stale entries in the table.
         */
        private void expungeStaleEntries() {
            Entry[] tab = table;
            int len = tab.length;
            for (int j = 0; j < len; j++) {
                Entry e = tab[j];
                if (e != null && e.get() == null)
                    expungeStaleEntry(j);
            }
        }
    }
}