/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
 * vim: set ts=8 sts=4 et sw=4 tw=99:
 * This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */

#ifndef ds_OrderedHashTable_h
#define ds_OrderedHashTable_h

/*
 * Define two collection templates, js::OrderedHashMap and js::OrderedHashSet.
 * They are like js::HashMap and js::HashSet except that:
 *
 *   - Iterating over an Ordered hash table visits the entries in the order in
 *     which they were inserted. This means that unlike a HashMap, the behavior
 *     of an OrderedHashMap is deterministic (as long as the HashPolicy methods
 *     are effect-free and consistent); the hashing is a pure performance
 *     optimization.
 *
 *   - Range objects over Ordered tables remain valid even when entries are
 *     added or removed or the table is resized. (However in the case of
 *     removing entries, note the warning on class Range below.)
 *
 *   - The API is a little different, so it's not a drop-in replacement.
 *     In particular, the hash policy is a little different.
 *     Also, the Ordered templates lack the Ptr and AddPtr types.
 *
 * Hash policies
 *
 * See the comment about "Hash policy" in HashTable.h for general features that
 * hash policy classes must provide. Hash policies for OrderedHashMaps and Sets
 * must additionally provide a distinguished "empty" key value and the
 * following static member functions:
 *     bool isEmpty(const Key&);
 *     void makeEmpty(Key*);
 */

#include "mozilla/Move.h"

using mozilla::Forward;
using mozilla::Move;

namespace js {

namespace detail {

/*
 * detail::OrderedHashTable is the underlying data structure used to implement both
 * OrderedHashMap and OrderedHashSet. Programs should use one of those two
 * templates rather than OrderedHashTable.
 */
template <class T, class Ops, class AllocPolicy>
class OrderedHashTable
{
  public:
    typedef typename Ops::KeyType Key;
    typedef typename Ops::Lookup Lookup;

    struct Data
    {
        T element;
        Data* chain;

        Data(const T& e, Data* c) : element(e), chain(c) {}
        Data(T&& e, Data* c) : element(Move(e)), chain(c) {}
    };

    class Range;
    friend class Range;

  private:
    Data** hashTable;           // hash table (has hashBuckets() elements)
    Data* data;                 // data vector, an array of Data objects
                                // data[0:dataLength] are constructed
    uint32_t dataLength;        // number of constructed elements in data
    uint32_t dataCapacity;      // size of data, in elements
    uint32_t liveCount;         // dataLength less empty (removed) entries
    uint32_t hashShift;         // multiplicative hash shift
    Range* ranges;              // list of all live Ranges on this table
    AllocPolicy alloc;

  public:
    explicit OrderedHashTable(AllocPolicy& ap)
        : hashTable(nullptr), data(nullptr), dataLength(0), ranges(nullptr), alloc(ap) {}

    bool init() {
        MOZ_ASSERT(!hashTable, "init must be called at most once");

        uint32_t buckets = initialBuckets();
        Data** tableAlloc = alloc.template pod_malloc<Data*>(buckets);
        if (!tableAlloc)
            return false;
        for (uint32_t i = 0; i < buckets; i++)
            tableAlloc[i] = nullptr;

        uint32_t capacity = uint32_t(buckets * fillFactor());
        Data* dataAlloc = alloc.template pod_malloc<Data>(capacity);
        if (!dataAlloc) {
            alloc.free_(tableAlloc);
            return false;
        }

        // clear() requires that members are assigned only after all allocation
        // has succeeded, and that this->ranges is left untouched.
        hashTable = tableAlloc;
        data = dataAlloc;
        dataLength = 0;
        dataCapacity = capacity;
        liveCount = 0;
        hashShift = HashNumberSizeBits - initialBucketsLog2();
        MOZ_ASSERT(hashBuckets() == buckets);
        return true;
    }

    ~OrderedHashTable() {
        for (Range* r = ranges; r; ) {
            Range* next = r->next;
            r->onTableDestroyed();
            r = next;
        }
        alloc.free_(hashTable);
        freeData(data, dataLength);
    }

    /* Return the number of elements in the table. */
    uint32_t count() const { return liveCount; }

    /* True if any element matches l. */
    bool has(const Lookup& l) const {
        return lookup(l) != nullptr;
    }

    /* Return a pointer to the element, if any, that matches l, or nullptr. */
    T* get(const Lookup& l) {
        Data* e = lookup(l, prepareHash(l));
        return e ? &e->element : nullptr;
    }

    /* Return a pointer to the element, if any, that matches l, or nullptr. */
    const T* get(const Lookup& l) const {
        return const_cast<OrderedHashTable*>(this)->get(l);
    }

    /*
     * If the table already contains an entry that matches |element|,
     * replace that entry with |element|. Otherwise add a new entry.
     *
     * On success, return true, whether there was already a matching element or
     * not. On allocation failure, return false. If this returns false, it
     * means the element was not added to the table.
     */
    template <typename ElementInput>
    bool put(ElementInput&& element) {
        HashNumber h = prepareHash(Ops::getKey(element));
        if (Data* e = lookup(Ops::getKey(element), h)) {
            e->element = Forward<ElementInput>(element);
            return true;
        }

        if (dataLength == dataCapacity) {
            // If the hashTable is more than 1/4 deleted data, simply rehash in
            // place to free up some space. Otherwise, grow the table.
            uint32_t newHashShift = liveCount >= dataCapacity * 0.75 ? hashShift - 1 : hashShift;
            if (!rehash(newHashShift))
                return false;
        }

        h >>= hashShift;
        liveCount++;
        Data* e = &data[dataLength++];
        new (e) Data(Forward<ElementInput>(element), hashTable[h]);
        hashTable[h] = e;
        return true;
    }

    /*
     * If the table contains an element matching l, remove it and set *foundp
     * to true. Otherwise set *foundp to false.
     *
     * Return true on success, false if we tried to shrink the table and hit an
     * allocation failure. Even if this returns false, *foundp is set correctly
     * and the matching element was removed. Shrinking is an optimization and
     * it's OK for it to fail.
     */
    bool remove(const Lookup& l, bool* foundp) {
        // Note: This could be optimized so that removing the last entry,
        // data[dataLength - 1], decrements dataLength. LIFO use cases would
        // benefit.

        // If a matching entry exists, empty it.
        Data* e = lookup(l, prepareHash(l));
        if (e == nullptr) {
            *foundp = false;
            return true;
        }

        *foundp = true;
        liveCount--;
        Ops::makeEmpty(&e->element);

        // Update active Ranges.
        uint32_t pos = e - data;
        for (Range* r = ranges; r; r = r->next)
            r->onRemove(pos);

        // If many entries have been removed, try to shrink the table.
        if (hashBuckets() > initialBuckets() && liveCount < dataLength * minDataFill()) {
            if (!rehash(hashShift + 1))
                return false;
        }
        return true;
    }

    /*
     * Remove all entries.
     *
     * Returns false on OOM, leaving the OrderedHashTable and any live Ranges
     * in the old state.
     *
     * The effect on live Ranges is the same as removing all entries; in
     * particular, those Ranges are still live and will see any entries added
     * after a successful clear().
     */
    bool clear() {
        if (dataLength != 0) {
            Data** oldHashTable = hashTable;
            Data* oldData = data;
            uint32_t oldDataLength = dataLength;

            hashTable = nullptr;
            if (!init()) {
                // init() only mutates members on success; see comment above.
                hashTable = oldHashTable;
                return false;
            }

            alloc.free_(oldHashTable);
            freeData(oldData, oldDataLength);
            for (Range* r = ranges; r; r = r->next)
                r->onClear();
        }

        MOZ_ASSERT(hashTable);
        MOZ_ASSERT(data);
        MOZ_ASSERT(dataLength == 0);
        MOZ_ASSERT(liveCount == 0);
        return true;
    }

    /*
     * Ranges are used to iterate over OrderedHashTables.
     *
     * Suppose 'Map' is some instance of OrderedHashMap, and 'map' is a Map.
     * Then you can walk all the key-value pairs like this:
     *
     *     for (Map::Range r = map.all(); !r.empty(); r.popFront()) {
     *         Map::Entry& pair = r.front();
     *         ... do something with pair ...
     *     }
     *
     * Ranges remain valid for the lifetime of the OrderedHashTable, even if
     * entries are added or removed or the table is resized. Don't do anything
     * to a Range, except destroy it, after the OrderedHashTable has been
     * destroyed. (We support destroying the two objects in either order to
     * humor the GC, bless its nondeterministic heart.)
     *
     * Warning: The behavior when the current front() entry is removed from the
     * table is subtly different from js::HashTable<>::Enum::removeFront()!
     * HashTable::Enum doesn't skip any entries when you removeFront() and then
     * popFront(). OrderedHashTable::Range does! (This is useful for using a
     * Range to implement JS Map.prototype.iterator.)
     *
     * The workaround is to call popFront() as soon as possible,
     * before there's any possibility of modifying the table:
     *
     *     for (Map::Range r = map.all(); !r.empty(); ) {
     *         Key key = r.front().key;         // this won't modify map
     *         Value val = r.front().value;     // this won't modify map
     *         r.popFront();
     *         // ...do things that might modify map...
     *     }
     */
    class Range
    {
        friend class OrderedHashTable;

        OrderedHashTable& ht;

        /* The index of front() within ht.data. */
        uint32_t i;

        /*
         * The number of nonempty entries in ht.data to the left of front().
         * This is used when the table is resized or compacted.
         */
        uint32_t count;

        /*
         * Links in the doubly-linked list of active Ranges on ht.
         *
         * prevp points to the previous Range's .next field;
         *   or to ht.ranges if this is the first Range in the list.
         * next points to the next Range;
         *   or nullptr if this is the last Range in the list.
         *
         * Invariant: *prevp == this.
         */
        Range** prevp;
        Range* next;

        /*
         * Create a Range over all the entries in ht.
         * (This is private on purpose. End users must use ht.all().)
         */
        explicit Range(OrderedHashTable& ht) : ht(ht), i(0), count(0), prevp(&ht.ranges), next(ht.ranges) {
            *prevp = this;
            if (next)
                next->prevp = &next;
            seek();
        }

      public:
        Range(const Range& other)
            : ht(other.ht), i(other.i), count(other.count), prevp(&ht.ranges), next(ht.ranges)
        {
            *prevp = this;
            if (next)
                next->prevp = &next;
        }

        ~Range() {
            *prevp = next;
            if (next)
                next->prevp = prevp;
        }

      private:
        // Prohibit copy assignment.
        Range& operator=(const Range& other) = delete;

        void seek() {
            while (i < ht.dataLength && Ops::isEmpty(Ops::getKey(ht.data[i].element)))
                i++;
        }

        /*
         * The hash table calls this when an entry is removed.
         * j is the index of the removed entry.
         */
        void onRemove(uint32_t j) {
            MOZ_ASSERT(valid());
            if (j < i)
                count--;
            if (j == i)
                seek();
        }

        /*
         * The hash table calls this when the table is resized or compacted.
         * Since |count| is the number of nonempty entries to the left of
         * front(), discarding the empty entries will not affect count, and it
         * will make i and count equal.
         */
        void onCompact() {
            MOZ_ASSERT(valid());
            i = count;
        }

        /* The hash table calls this when cleared. */
        void onClear() {
            MOZ_ASSERT(valid());
            i = count = 0;
        }

        bool valid() const {
            return next != this;
        }

        void onTableDestroyed() {
            MOZ_ASSERT(valid());
            prevp = &next;
            next = this;
        }

      public:
        bool empty() const {
            MOZ_ASSERT(valid());
            return i >= ht.dataLength;
        }

        /*
         * Return the first element in the range. This must not be called if
         * this->empty().
         *
         * Warning: Removing an entry from the table also removes it from any
         * live Ranges, and a Range can become empty that way, rendering
         * front() invalid. If in doubt, check empty() before calling front().
         */
        T& front() {
            MOZ_ASSERT(valid());
            MOZ_ASSERT(!empty());
            return ht.data[i].element;
        }

        /*
         * Remove the first element from this range.
         * This must not be called if this->empty().
         *
         * Warning: Removing an entry from the table also removes it from any
         * live Ranges, and a Range can become empty that way, rendering
         * popFront() invalid. If in doubt, check empty() before calling
         * popFront().
         */
        void popFront() {
            MOZ_ASSERT(valid());
            MOZ_ASSERT(!empty());
            MOZ_ASSERT(!Ops::isEmpty(Ops::getKey(ht.data[i].element)));
            count++;
            i++;
            seek();
        }

        /*
         * Change the key of the front entry.
         *
         * This calls Ops::hash on both the current key and the new key.
         * Ops::hash on the current key must return the same hash code as
         * when the entry was added to the table.
         */
        void rekeyFront(const Key& k) {
            MOZ_ASSERT(valid());
            Data& entry = ht.data[i];
            HashNumber oldHash = prepareHash(Ops::getKey(entry.element)) >> ht.hashShift;
            HashNumber newHash = prepareHash(k) >> ht.hashShift;
            Ops::setKey(entry.element, k);
            if (newHash != oldHash) {
                // Remove this entry from its old hash chain. (If this crashes
                // reading nullptr, it would mean we did not find this entry on
                // the hash chain where we expected it. That probably means the
                // key's hash code changed since it was inserted, breaking the
                // hash code invariant.)
                Data** ep = &ht.hashTable[oldHash];
                while (*ep != &entry)
                    ep = &(*ep)->chain;
                *ep = entry.chain;

                // Add it to the new hash chain. We could just insert it at the
                // beginning of the chain. Instead, we do a bit of work to
                // preserve the invariant that hash chains always go in reverse
                // insertion order (descending memory order). No code currently
                // depends on this invariant, so it's fine to kill it if
                // needed.
                ep = &ht.hashTable[newHash];
                while (*ep && *ep > &entry)
                    ep = &(*ep)->chain;
                entry.chain = *ep;
                *ep = &entry;
            }
        }
    };

    Range all() { return Range(*this); }

    /*
     * Change the value of the given key.
     *
     * This calls Ops::hash on both the current key and the new key.
     * Ops::hash on the current key must return the same hash code as
     * when the entry was added to the table.
     */
    void rekeyOneEntry(const Key& current, const Key& newKey, const T& element) {
        if (current == newKey)
            return;

        Data* entry = lookup(current, prepareHash(current));
        if (!entry)
            return;

        HashNumber oldHash = prepareHash(current) >> hashShift;
        HashNumber newHash = prepareHash(newKey) >> hashShift;

        entry->element = element;

        // Remove this entry from its old hash chain. (If this crashes
        // reading nullptr, it would mean we did not find this entry on
        // the hash chain where we expected it. That probably means the
        // key's hash code changed since it was inserted, breaking the
        // hash code invariant.)
        Data** ep = &hashTable[oldHash];
        while (*ep != entry)
            ep = &(*ep)->chain;
        *ep = entry->chain;

        // Add it to the new hash chain. We could just insert it at the
        // beginning of the chain. Instead, we do a bit of work to
        // preserve the invariant that hash chains always go in reverse
        // insertion order (descending memory order). No code currently
        // depends on this invariant, so it's fine to kill it if
        // needed.
        ep = &hashTable[newHash];
        while (*ep && *ep > entry)
            ep = &(*ep)->chain;
        entry->chain = *ep;
        *ep = entry;
    }

  private:
    /* Logarithm base 2 of the number of buckets in the hash table initially. */
    static uint32_t initialBucketsLog2() { return 1; }
    static uint32_t initialBuckets() { return 1 << initialBucketsLog2(); }

    /*
     * The maximum load factor (mean number of entries per bucket).
     * It is an invariant that
     *     dataCapacity == floor(hashBuckets() * fillFactor()).
     *
     * The fill factor should be between 2 and 4, and it should be chosen so that
     * the fill factor times sizeof(Data) is close to but <= a power of 2.
     * This fixed fill factor was chosen to make the size of the data
     * array, in bytes, close to a power of two when sizeof(T) is 16.
     */
    static double fillFactor() { return 8.0 / 3.0; }

    /*
     * The minimum permitted value of (liveCount / dataLength).
     * If that ratio drops below this value, we shrink the table.
     */
    static double minDataFill() { return 0.25; }

    static HashNumber prepareHash(const Lookup& l) {
        return ScrambleHashCode(Ops::hash(l));
    }

    /* The size of hashTable, in elements. Always a power of two. */
    uint32_t hashBuckets() const {
        return 1 << (HashNumberSizeBits - hashShift);
    }

    static void destroyData(Data* data, uint32_t length) {
        for (Data* p = data + length; p != data; )
            (--p)->~Data();
    }

    void freeData(Data* data, uint32_t length) {
        destroyData(data, length);
        alloc.free_(data);
    }

    Data* lookup(const Lookup& l, HashNumber h) {
        for (Data* e = hashTable[h >> hashShift]; e; e = e->chain) {
            if (Ops::match(Ops::getKey(e->element), l))
                return e;
        }
        return nullptr;
    }

    const Data* lookup(const Lookup& l) const {
        return const_cast<OrderedHashTable*>(this)->lookup(l, prepareHash(l));
    }

    /* This is called after rehashing the table. */
    void compacted() {
        // If we had any empty entries, compacting may have moved live entries
        // to the left within |data|. Notify all live Ranges of the change.
        for (Range* r = ranges; r; r = r->next)
            r->onCompact();
    }

    /* Compact the entries in |data| and rehash them. */
    void rehashInPlace() {
        for (uint32_t i = 0, N = hashBuckets(); i < N; i++)
            hashTable[i] = nullptr;
        Data* wp = data;
        Data* end = data + dataLength;
        for (Data* rp = data; rp != end; rp++) {
            if (!Ops::isEmpty(Ops::getKey(rp->element))) {
                HashNumber h = prepareHash(Ops::getKey(rp->element)) >> hashShift;
                if (rp != wp)
                    wp->element = Move(rp->element);
                wp->chain = hashTable[h];
                hashTable[h] = wp;
                wp++;
            }
        }
        MOZ_ASSERT(wp == data + liveCount);

        while (wp != end)
            (--end)->~Data();
        dataLength = liveCount;
        compacted();
    }

    /*
     * Grow, shrink, or compact both |hashTable| and |data|.
     *
     * On success, this returns true, dataLength == liveCount, and there are no
     * empty elements in data[0:dataLength]. On allocation failure, this
     * leaves everything as it was and returns false.
     */
    bool rehash(uint32_t newHashShift) {
        // If the size of the table is not changing, rehash in place to avoid
        // allocating memory.
        if (newHashShift == hashShift) {
            rehashInPlace();
            return true;
        }

        size_t newHashBuckets = 1 << (HashNumberSizeBits - newHashShift);
        Data** newHashTable = alloc.template pod_malloc<Data*>(newHashBuckets);
        if (!newHashTable)
            return false;
        for (uint32_t i = 0; i < newHashBuckets; i++)
            newHashTable[i] = nullptr;

        uint32_t newCapacity = uint32_t(newHashBuckets * fillFactor());
        Data* newData = alloc.template pod_malloc<Data>(newCapacity);
        if (!newData) {
            alloc.free_(newHashTable);
            return false;
        }

        Data* wp = newData;
        Data* end = data + dataLength;
        for (Data* p = data; p != end; p++) {
            if (!Ops::isEmpty(Ops::getKey(p->element))) {
                HashNumber h = prepareHash(Ops::getKey(p->element)) >> newHashShift;
                new (wp) Data(Move(p->element), newHashTable[h]);
                newHashTable[h] = wp;
                wp++;
            }
        }
        MOZ_ASSERT(wp == newData + liveCount);

        alloc.free_(hashTable);
        freeData(data, dataLength);

        hashTable = newHashTable;
        data = newData;
        dataLength = liveCount;
        dataCapacity = newCapacity;
        hashShift = newHashShift;
        MOZ_ASSERT(hashBuckets() == newHashBuckets);

        compacted();
        return true;
    }

    // Not copyable.
    OrderedHashTable& operator=(const OrderedHashTable&) = delete;
    OrderedHashTable(const OrderedHashTable&) = delete;
};

}  // namespace detail

template <class Key, class Value, class OrderedHashPolicy, class AllocPolicy>
class OrderedHashMap
{
  public:
    class Entry
    {
        template <class, class, class> friend class detail::OrderedHashTable;
        void operator=(const Entry& rhs) {
            const_cast<Key&>(key) = rhs.key;
            value = rhs.value;
        }

        void operator=(Entry&& rhs) {
            MOZ_ASSERT(this != &rhs, "self-move assignment is prohibited");
            const_cast<Key&>(key) = Move(rhs.key);
            value = Move(rhs.value);
        }

      public:
        Entry() : key(), value() {}
        template <typename V>
        Entry(const Key& k, V&& v) : key(k), value(Forward<V>(v)) {}
        Entry(Entry&& rhs) : key(Move(rhs.key)), value(Move(rhs.value)) {}

        const Key key;
        Value value;
    };

  private:
    struct MapOps : OrderedHashPolicy
    {
        typedef Key KeyType;
        static void makeEmpty(Entry* e) {
            OrderedHashPolicy::makeEmpty(const_cast<Key*>(&e->key));

            // Clear the value. Destroying it is another possibility, but that
            // would complicate class Entry considerably.
            e->value = Value();
        }
        static const Key& getKey(const Entry& e) { return e.key; }
        static void setKey(Entry& e, const Key& k) { const_cast<Key&>(e.key) = k; }
    };

    typedef detail::OrderedHashTable<Entry, MapOps, AllocPolicy> Impl;
    Impl impl;

  public:
    typedef typename Impl::Range Range;

    explicit OrderedHashMap(AllocPolicy ap = AllocPolicy()) : impl(ap) {}
    bool init()                                     { return impl.init(); }
    uint32_t count() const                          { return impl.count(); }
    bool has(const Key& key) const                  { return impl.has(key); }
    Range all()                                     { return impl.all(); }
    const Entry* get(const Key& key) const          { return impl.get(key); }
    Entry* get(const Key& key)                      { return impl.get(key); }
    template <typename V>
    bool put(const Key& key, V&& value)             { return impl.put(Entry(key, Forward<V>(value))); }
    bool remove(const Key& key, bool* foundp)       { return impl.remove(key, foundp); }
    bool clear()                                    { return impl.clear(); }

    void rekeyOneEntry(const Key& current, const Key& newKey) {
        const Entry* e = get(current);
        if (!e)
            return;
        return impl.rekeyOneEntry(current, newKey, Entry(newKey, e->value));
    }
};

template <class T, class OrderedHashPolicy, class AllocPolicy>
class OrderedHashSet
{
  private:
    struct SetOps : OrderedHashPolicy
    {
        typedef const T KeyType;
        static const T& getKey(const T& v) { return v; }
        static void setKey(const T& e, const T& v) { const_cast<T&>(e) = v; }
    };

    typedef detail::OrderedHashTable<T, SetOps, AllocPolicy> Impl;
    Impl impl;

  public:
    typedef typename Impl::Range Range;

    explicit OrderedHashSet(AllocPolicy ap = AllocPolicy()) : impl(ap) {}
    bool init()                                     { return impl.init(); }
    uint32_t count() const                          { return impl.count(); }
    bool has(const T& value) const                  { return impl.has(value); }
    Range all()                                     { return impl.all(); }
    bool put(const T& value)                        { return impl.put(value); }
    bool remove(const T& value, bool* foundp)       { return impl.remove(value, foundp); }
    bool clear()                                    { return impl.clear(); }

    void rekeyOneEntry(const T& current, const T& newKey) {
        return impl.rekeyOneEntry(current, newKey, newKey);
    }
};

}  // namespace js

#endif /* ds_OrderedHashTable_h */