michael@0: /*
michael@0: ******************************************************************************
michael@0: *   Copyright (C) 1997-2011, International Business Machines
michael@0: *   Corporation and others.  All Rights Reserved.
michael@0: ******************************************************************************
michael@0: *   Date        Name        Description
michael@0: *   03/22/00    aliu        Adapted from original C++ ICU Hashtable.
michael@0: *   07/06/01    aliu        Modified to support int32_t keys on
michael@0: *                           platforms with sizeof(void*) < 32.
michael@0: ******************************************************************************
michael@0: */
michael@0: 
michael@0: #include "uhash.h"
michael@0: #include "unicode/ustring.h"
michael@0: #include "cstring.h"
michael@0: #include "cmemory.h"
michael@0: #include "uassert.h"
michael@0: #include "ustr_imp.h"
michael@0: 
michael@0: /* This hashtable is implemented as a double hash.  All elements are
michael@0:  * stored in a single array with no secondary storage for collision
michael@0:  * resolution (no linked list, etc.).  When there is a hash collision
michael@0:  * (when two unequal keys have the same hashcode) we resolve this by
michael@0:  * using a secondary hash.  The secondary hash is an increment
michael@0:  * computed as a hash function (a different one) of the primary
michael@0:  * hashcode.  This increment is added to the initial hash value to
michael@0:  * obtain further slots assigned to the same hash code.  For this to
michael@0:  * work, the length of the array and the increment must be relatively
michael@0:  * prime.  The easiest way to achieve this is to have the length of
michael@0:  * the array be prime, and the increment be any value from
michael@0:  * 1..length-1.
michael@0:  *
michael@0:  * Hashcodes are 32-bit integers.  We make sure all hashcodes are
michael@0:  * non-negative by masking off the top bit.  This has two effects: (1)
michael@0:  * modulo arithmetic is simplified.  If we allowed negative hashcodes,
michael@0:  * then when we computed hashcode % length, we could get a negative
michael@0:  * result, which we would then have to adjust back into range.  It's
michael@0:  * simpler to just make hashcodes non-negative. (2) It makes it easy
michael@0:  * to check for empty vs. occupied slots in the table.  We just mark
michael@0:  * empty or deleted slots with a negative hashcode.
michael@0:  *
michael@0:  * The central function is _uhash_find().  This function looks for a
michael@0:  * slot matching the given key and hashcode.  If one is found, it
michael@0:  * returns a pointer to that slot.  If the table is full, and no match
michael@0:  * is found, it returns NULL -- in theory.  This would make the code
michael@0:  * more complicated, since all callers of _uhash_find() would then
michael@0:  * have to check for a NULL result.  To keep this from happening, we
michael@0:  * don't allow the table to fill.  When there is only one
michael@0:  * empty/deleted slot left, uhash_put() will refuse to increase the
michael@0:  * count, and fail.  This simplifies the code.  In practice, one will
michael@0:  * seldom encounter this using default UHashtables.  However, if a
michael@0:  * hashtable is set to a U_FIXED resize policy, or if memory is
michael@0:  * exhausted, then the table may fill.
michael@0:  *
michael@0:  * High and low water ratios control rehashing.  They establish levels
michael@0:  * of fullness (from 0 to 1) outside of which the data array is
michael@0:  * reallocated and repopulated.  Setting the low water ratio to zero
michael@0:  * means the table will never shrink.  Setting the high water ratio to
michael@0:  * one means the table will never grow.  The ratios should be
michael@0:  * coordinated with the ratio between successive elements of the
michael@0:  * PRIMES table, so that when the primeIndex is incremented or
michael@0:  * decremented during rehashing, it brings the ratio of count / length
michael@0:  * back into the desired range (between low and high water ratios).
michael@0:  */
michael@0: 
michael@0: /********************************************************************
michael@0:  * PRIVATE Constants, Macros
michael@0:  ********************************************************************/
michael@0: 
michael@0: /* This is a list of non-consecutive primes chosen such that
michael@0:  * PRIMES[i+1] ~ 2*PRIMES[i].  (Currently, the ratio ranges from 1.81
michael@0:  * to 2.18; the inverse ratio ranges from 0.459 to 0.552.)  If this
michael@0:  * ratio is changed, the low and high water ratios should also be
michael@0:  * adjusted to suit.
michael@0:  *
michael@0:  * These prime numbers were also chosen so that they are the largest
michael@0:  * prime number while being less than a power of two.
michael@0:  */
michael@0: static const int32_t PRIMES[] = {
michael@0:     13, 31, 61, 127, 251, 509, 1021, 2039, 4093, 8191, 16381, 32749,
michael@0:     65521, 131071, 262139, 524287, 1048573, 2097143, 4194301, 8388593,
michael@0:     16777213, 33554393, 67108859, 134217689, 268435399, 536870909,
michael@0:     1073741789, 2147483647 /*, 4294967291 */
michael@0: };
michael@0: 
michael@0: #define PRIMES_LENGTH (sizeof(PRIMES) / sizeof(PRIMES[0]))
michael@0: #define DEFAULT_PRIME_INDEX 3
michael@0: 
michael@0: /* These ratios are tuned to the PRIMES array such that a resize
michael@0:  * places the table back into the zone of non-resizing.  That is,
michael@0:  * after a call to _uhash_rehash(), a subsequent call to
michael@0:  * _uhash_rehash() should do nothing (should not churn).  This is only
michael@0:  * a potential problem with U_GROW_AND_SHRINK.
michael@0:  */
michael@0: static const float RESIZE_POLICY_RATIO_TABLE[6] = {
michael@0:     /* low, high water ratio */
michael@0:     0.0F, 0.5F, /* U_GROW: Grow on demand, do not shrink */
michael@0:     0.1F, 0.5F, /* U_GROW_AND_SHRINK: Grow and shrink on demand */
michael@0:     0.0F, 1.0F  /* U_FIXED: Never change size */
michael@0: };
michael@0: 
michael@0: /*
michael@0:   Invariants for hashcode values:
michael@0: 
michael@0:   * DELETED < 0
michael@0:   * EMPTY < 0
michael@0:   * Real hashes >= 0
michael@0: 
michael@0:   Hashcodes may not start out this way, but internally they are
michael@0:   adjusted so that they are always positive.  We assume 32-bit
michael@0:   hashcodes; adjust these constants for other hashcode sizes.
michael@0: */
michael@0: #define HASH_DELETED    ((int32_t) 0x80000000)
michael@0: #define HASH_EMPTY      ((int32_t) HASH_DELETED + 1)
michael@0: 
michael@0: #define IS_EMPTY_OR_DELETED(x) ((x) < 0)
michael@0: 
michael@0: /* This macro expects a UHashTok.pointer as its keypointer and
michael@0:    valuepointer parameters */
michael@0: #define HASH_DELETE_KEY_VALUE(hash, keypointer, valuepointer) \
michael@0:             if (hash->keyDeleter != NULL && keypointer != NULL) { \
michael@0:                 (*hash->keyDeleter)(keypointer); \
michael@0:             } \
michael@0:             if (hash->valueDeleter != NULL && valuepointer != NULL) { \
michael@0:                 (*hash->valueDeleter)(valuepointer); \
michael@0:             }
michael@0: 
michael@0: /*
michael@0:  * Constants for hinting whether a key or value is an integer
michael@0:  * or a pointer.  If a hint bit is zero, then the associated
michael@0:  * token is assumed to be an integer.
michael@0:  */
michael@0: #define HINT_KEY_POINTER   (1)
michael@0: #define HINT_VALUE_POINTER (2)
michael@0: 
michael@0: /********************************************************************
michael@0:  * PRIVATE Implementation
michael@0:  ********************************************************************/
michael@0: 
michael@0: static UHashTok
michael@0: _uhash_setElement(UHashtable *hash, UHashElement* e,
michael@0:                   int32_t hashcode,
michael@0:                   UHashTok key, UHashTok value, int8_t hint) {
michael@0: 
michael@0:     UHashTok oldValue = e->value;
michael@0:     if (hash->keyDeleter != NULL && e->key.pointer != NULL &&
michael@0:         e->key.pointer != key.pointer) { /* Avoid double deletion */
michael@0:         (*hash->keyDeleter)(e->key.pointer);
michael@0:     }
michael@0:     if (hash->valueDeleter != NULL) {
michael@0:         if (oldValue.pointer != NULL &&
michael@0:             oldValue.pointer != value.pointer) { /* Avoid double deletion */
michael@0:             (*hash->valueDeleter)(oldValue.pointer);
michael@0:         }
michael@0:         oldValue.pointer = NULL;
michael@0:     }
michael@0:     /* Compilers should copy the UHashTok union correctly, but even if
michael@0:      * they do, memory heap tools (e.g. BoundsChecker) can get
michael@0:      * confused when a pointer is cloaked in a union and then copied.
michael@0:      * TO ALLEVIATE THIS, we use hints (based on what API the user is
michael@0:      * calling) to copy pointers when we know the user thinks
michael@0:      * something is a pointer. */
michael@0:     if (hint & HINT_KEY_POINTER) {
michael@0:         e->key.pointer = key.pointer;
michael@0:     } else {
michael@0:         e->key = key;
michael@0:     }
michael@0:     if (hint & HINT_VALUE_POINTER) {
michael@0:         e->value.pointer = value.pointer;
michael@0:     } else {
michael@0:         e->value = value;
michael@0:     }
michael@0:     e->hashcode = hashcode;
michael@0:     return oldValue;
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Assumes that the given element is not empty or deleted.
michael@0:  */
michael@0: static UHashTok
michael@0: _uhash_internalRemoveElement(UHashtable *hash, UHashElement* e) {
michael@0:     UHashTok empty;
michael@0:     U_ASSERT(!IS_EMPTY_OR_DELETED(e->hashcode));
michael@0:     --hash->count;
michael@0:     empty.pointer = NULL; empty.integer = 0;
michael@0:     return _uhash_setElement(hash, e, HASH_DELETED, empty, empty, 0);
michael@0: }
michael@0: 
michael@0: static void
michael@0: _uhash_internalSetResizePolicy(UHashtable *hash, enum UHashResizePolicy policy) {
michael@0:     U_ASSERT(hash != NULL);
michael@0:     U_ASSERT(((int32_t)policy) >= 0);
michael@0:     U_ASSERT(((int32_t)policy) < 3);
michael@0:     hash->lowWaterRatio  = RESIZE_POLICY_RATIO_TABLE[policy * 2];
michael@0:     hash->highWaterRatio = RESIZE_POLICY_RATIO_TABLE[policy * 2 + 1];
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Allocate internal data array of a size determined by the given
michael@0:  * prime index.  If the index is out of range it is pinned into range.
michael@0:  * If the allocation fails the status is set to
michael@0:  * U_MEMORY_ALLOCATION_ERROR and all array storage is freed.  In
michael@0:  * either case the previous array pointer is overwritten.
michael@0:  *
michael@0:  * Caller must ensure primeIndex is in range 0..PRIME_LENGTH-1.
michael@0:  */
michael@0: static void
michael@0: _uhash_allocate(UHashtable *hash,
michael@0:                 int32_t primeIndex,
michael@0:                 UErrorCode *status) {
michael@0: 
michael@0:     UHashElement *p, *limit;
michael@0:     UHashTok emptytok;
michael@0: 
michael@0:     if (U_FAILURE(*status)) return;
michael@0: 
michael@0:     U_ASSERT(primeIndex >= 0 && primeIndex < PRIMES_LENGTH);
michael@0: 
michael@0:     hash->primeIndex = primeIndex;
michael@0:     hash->length = PRIMES[primeIndex];
michael@0: 
michael@0:     p = hash->elements = (UHashElement*)
michael@0:         uprv_malloc(sizeof(UHashElement) * hash->length);
michael@0: 
michael@0:     if (hash->elements == NULL) {
michael@0:         *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     emptytok.pointer = NULL; /* Only one of these two is needed */
michael@0:     emptytok.integer = 0;    /* but we don't know which one. */
michael@0:     
michael@0:     limit = p + hash->length;
michael@0:     while (p < limit) {
michael@0:         p->key = emptytok;
michael@0:         p->value = emptytok;
michael@0:         p->hashcode = HASH_EMPTY;
michael@0:         ++p;
michael@0:     }
michael@0: 
michael@0:     hash->count = 0;
michael@0:     hash->lowWaterMark = (int32_t)(hash->length * hash->lowWaterRatio);
michael@0:     hash->highWaterMark = (int32_t)(hash->length * hash->highWaterRatio);
michael@0: }
michael@0: 
michael@0: static UHashtable*
michael@0: _uhash_init(UHashtable *result,
michael@0:               UHashFunction *keyHash, 
michael@0:               UKeyComparator *keyComp,
michael@0:               UValueComparator *valueComp,
michael@0:               int32_t primeIndex,
michael@0:               UErrorCode *status)
michael@0: {
michael@0:     if (U_FAILURE(*status)) return NULL;
michael@0:     U_ASSERT(keyHash != NULL);
michael@0:     U_ASSERT(keyComp != NULL);
michael@0: 
michael@0:     result->keyHasher       = keyHash;
michael@0:     result->keyComparator   = keyComp;
michael@0:     result->valueComparator = valueComp;
michael@0:     result->keyDeleter      = NULL;
michael@0:     result->valueDeleter    = NULL;
michael@0:     result->allocated       = FALSE;
michael@0:     _uhash_internalSetResizePolicy(result, U_GROW);
michael@0: 
michael@0:     _uhash_allocate(result, primeIndex, status);
michael@0: 
michael@0:     if (U_FAILURE(*status)) {
michael@0:         return NULL;
michael@0:     }
michael@0: 
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: static UHashtable*
michael@0: _uhash_create(UHashFunction *keyHash, 
michael@0:               UKeyComparator *keyComp,
michael@0:               UValueComparator *valueComp,
michael@0:               int32_t primeIndex,
michael@0:               UErrorCode *status) {
michael@0:     UHashtable *result;
michael@0: 
michael@0:     if (U_FAILURE(*status)) return NULL;
michael@0: 
michael@0:     result = (UHashtable*) uprv_malloc(sizeof(UHashtable));
michael@0:     if (result == NULL) {
michael@0:         *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:         return NULL;
michael@0:     }
michael@0: 
michael@0:     _uhash_init(result, keyHash, keyComp, valueComp, primeIndex, status);
michael@0:     result->allocated       = TRUE;
michael@0: 
michael@0:     if (U_FAILURE(*status)) {
michael@0:         uprv_free(result);
michael@0:         return NULL;
michael@0:     }
michael@0: 
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Look for a key in the table, or if no such key exists, the first
michael@0:  * empty slot matching the given hashcode.  Keys are compared using
michael@0:  * the keyComparator function.
michael@0:  *
michael@0:  * First find the start position, which is the hashcode modulo
michael@0:  * the length.  Test it to see if it is:
michael@0:  *
michael@0:  * a. identical:  First check the hash values for a quick check,
michael@0:  *    then compare keys for equality using keyComparator.
michael@0:  * b. deleted
michael@0:  * c. empty
michael@0:  *
michael@0:  * Stop if it is identical or empty, otherwise continue by adding a
michael@0:  * "jump" value (moduloing by the length again to keep it within
michael@0:  * range) and retesting.  For efficiency, there need enough empty
michael@0:  * values so that the searchs stop within a reasonable amount of time.
michael@0:  * This can be changed by changing the high/low water marks.
michael@0:  *
michael@0:  * In theory, this function can return NULL, if it is full (no empty
michael@0:  * or deleted slots) and if no matching key is found.  In practice, we
michael@0:  * prevent this elsewhere (in uhash_put) by making sure the last slot
michael@0:  * in the table is never filled.
michael@0:  *
michael@0:  * The size of the table should be prime for this algorithm to work;
michael@0:  * otherwise we are not guaranteed that the jump value (the secondary
michael@0:  * hash) is relatively prime to the table length.
michael@0:  */
michael@0: static UHashElement*
michael@0: _uhash_find(const UHashtable *hash, UHashTok key,
michael@0:             int32_t hashcode) {
michael@0: 
michael@0:     int32_t firstDeleted = -1;  /* assume invalid index */
michael@0:     int32_t theIndex, startIndex;
michael@0:     int32_t jump = 0; /* lazy evaluate */
michael@0:     int32_t tableHash;
michael@0:     UHashElement *elements = hash->elements;
michael@0: 
michael@0:     hashcode &= 0x7FFFFFFF; /* must be positive */
michael@0:     startIndex = theIndex = (hashcode ^ 0x4000000) % hash->length;
michael@0: 
michael@0:     do {
michael@0:         tableHash = elements[theIndex].hashcode;
michael@0:         if (tableHash == hashcode) {          /* quick check */
michael@0:             if ((*hash->keyComparator)(key, elements[theIndex].key)) {
michael@0:                 return &(elements[theIndex]);
michael@0:             }
michael@0:         } else if (!IS_EMPTY_OR_DELETED(tableHash)) {
michael@0:             /* We have hit a slot which contains a key-value pair,
michael@0:              * but for which the hash code does not match.  Keep
michael@0:              * looking.
michael@0:              */
michael@0:         } else if (tableHash == HASH_EMPTY) { /* empty, end o' the line */
michael@0:             break;
michael@0:         } else if (firstDeleted < 0) { /* remember first deleted */
michael@0:             firstDeleted = theIndex;
michael@0:         }
michael@0:         if (jump == 0) { /* lazy compute jump */
michael@0:             /* The jump value must be relatively prime to the table
michael@0:              * length.  As long as the length is prime, then any value
michael@0:              * 1..length-1 will be relatively prime to it.
michael@0:              */
michael@0:             jump = (hashcode % (hash->length - 1)) + 1;
michael@0:         }
michael@0:         theIndex = (theIndex + jump) % hash->length;
michael@0:     } while (theIndex != startIndex);
michael@0: 
michael@0:     if (firstDeleted >= 0) {
michael@0:         theIndex = firstDeleted; /* reset if had deleted slot */
michael@0:     } else if (tableHash != HASH_EMPTY) {
michael@0:         /* We get to this point if the hashtable is full (no empty or
michael@0:          * deleted slots), and we've failed to find a match.  THIS
michael@0:          * WILL NEVER HAPPEN as long as uhash_put() makes sure that
michael@0:          * count is always < length.
michael@0:          */
michael@0:         U_ASSERT(FALSE);
michael@0:         return NULL; /* Never happens if uhash_put() behaves */
michael@0:     }
michael@0:     return &(elements[theIndex]);
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Attempt to grow or shrink the data arrays in order to make the
michael@0:  * count fit between the high and low water marks.  hash_put() and
michael@0:  * hash_remove() call this method when the count exceeds the high or
michael@0:  * low water marks.  This method may do nothing, if memory allocation
michael@0:  * fails, or if the count is already in range, or if the length is
michael@0:  * already at the low or high limit.  In any case, upon return the
michael@0:  * arrays will be valid.
michael@0:  */
michael@0: static void
michael@0: _uhash_rehash(UHashtable *hash, UErrorCode *status) {
michael@0: 
michael@0:     UHashElement *old = hash->elements;
michael@0:     int32_t oldLength = hash->length;
michael@0:     int32_t newPrimeIndex = hash->primeIndex;
michael@0:     int32_t i;
michael@0: 
michael@0:     if (hash->count > hash->highWaterMark) {
michael@0:         if (++newPrimeIndex >= PRIMES_LENGTH) {
michael@0:             return;
michael@0:         }
michael@0:     } else if (hash->count < hash->lowWaterMark) {
michael@0:         if (--newPrimeIndex < 0) {
michael@0:             return;
michael@0:         }
michael@0:     } else {
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     _uhash_allocate(hash, newPrimeIndex, status);
michael@0: 
michael@0:     if (U_FAILURE(*status)) {
michael@0:         hash->elements = old;
michael@0:         hash->length = oldLength;       
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     for (i = oldLength - 1; i >= 0; --i) {
michael@0:         if (!IS_EMPTY_OR_DELETED(old[i].hashcode)) {
michael@0:             UHashElement *e = _uhash_find(hash, old[i].key, old[i].hashcode);
michael@0:             U_ASSERT(e != NULL);
michael@0:             U_ASSERT(e->hashcode == HASH_EMPTY);
michael@0:             e->key = old[i].key;
michael@0:             e->value = old[i].value;
michael@0:             e->hashcode = old[i].hashcode;
michael@0:             ++hash->count;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     uprv_free(old);
michael@0: }
michael@0: 
michael@0: static UHashTok
michael@0: _uhash_remove(UHashtable *hash,
michael@0:               UHashTok key) {
michael@0:     /* First find the position of the key in the table.  If the object
michael@0:      * has not been removed already, remove it.  If the user wanted
michael@0:      * keys deleted, then delete it also.  We have to put a special
michael@0:      * hashcode in that position that means that something has been
michael@0:      * deleted, since when we do a find, we have to continue PAST any
michael@0:      * deleted values.
michael@0:      */
michael@0:     UHashTok result;
michael@0:     UHashElement* e = _uhash_find(hash, key, hash->keyHasher(key));
michael@0:     U_ASSERT(e != NULL);
michael@0:     result.pointer = NULL;
michael@0:     result.integer = 0;
michael@0:     if (!IS_EMPTY_OR_DELETED(e->hashcode)) {
michael@0:         result = _uhash_internalRemoveElement(hash, e);
michael@0:         if (hash->count < hash->lowWaterMark) {
michael@0:             UErrorCode status = U_ZERO_ERROR;
michael@0:             _uhash_rehash(hash, &status);
michael@0:         }
michael@0:     }
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: static UHashTok
michael@0: _uhash_put(UHashtable *hash,
michael@0:            UHashTok key,
michael@0:            UHashTok value,
michael@0:            int8_t hint,
michael@0:            UErrorCode *status) {
michael@0: 
michael@0:     /* Put finds the position in the table for the new value.  If the
michael@0:      * key is already in the table, it is deleted, if there is a
michael@0:      * non-NULL keyDeleter.  Then the key, the hash and the value are
michael@0:      * all put at the position in their respective arrays.
michael@0:      */
michael@0:     int32_t hashcode;
michael@0:     UHashElement* e;
michael@0:     UHashTok emptytok;
michael@0: 
michael@0:     if (U_FAILURE(*status)) {
michael@0:         goto err;
michael@0:     }
michael@0:     U_ASSERT(hash != NULL);
michael@0:     /* Cannot always check pointer here or iSeries sees NULL every time. */
michael@0:     if ((hint & HINT_VALUE_POINTER) && value.pointer == NULL) {
michael@0:         /* Disallow storage of NULL values, since NULL is returned by
michael@0:          * get() to indicate an absent key.  Storing NULL == removing.
michael@0:          */
michael@0:         return _uhash_remove(hash, key);
michael@0:     }
michael@0:     if (hash->count > hash->highWaterMark) {
michael@0:         _uhash_rehash(hash, status);
michael@0:         if (U_FAILURE(*status)) {
michael@0:             goto err;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     hashcode = (*hash->keyHasher)(key);
michael@0:     e = _uhash_find(hash, key, hashcode);
michael@0:     U_ASSERT(e != NULL);
michael@0: 
michael@0:     if (IS_EMPTY_OR_DELETED(e->hashcode)) {
michael@0:         /* Important: We must never actually fill the table up.  If we
michael@0:          * do so, then _uhash_find() will return NULL, and we'll have
michael@0:          * to check for NULL after every call to _uhash_find().  To
michael@0:          * avoid this we make sure there is always at least one empty
michael@0:          * or deleted slot in the table.  This only is a problem if we
michael@0:          * are out of memory and rehash isn't working.
michael@0:          */
michael@0:         ++hash->count;
michael@0:         if (hash->count == hash->length) {
michael@0:             /* Don't allow count to reach length */
michael@0:             --hash->count;
michael@0:             *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:             goto err;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* We must in all cases handle storage properly.  If there was an
michael@0:      * old key, then it must be deleted (if the deleter != NULL).
michael@0:      * Make hashcodes stored in table positive.
michael@0:      */
michael@0:     return _uhash_setElement(hash, e, hashcode & 0x7FFFFFFF, key, value, hint);
michael@0: 
michael@0:  err:
michael@0:     /* If the deleters are non-NULL, this method adopts its key and/or
michael@0:      * value arguments, and we must be sure to delete the key and/or
michael@0:      * value in all cases, even upon failure.
michael@0:      */
michael@0:     HASH_DELETE_KEY_VALUE(hash, key.pointer, value.pointer);
michael@0:     emptytok.pointer = NULL; emptytok.integer = 0;
michael@0:     return emptytok;
michael@0: }
michael@0: 
michael@0: 
michael@0: /********************************************************************
michael@0:  * PUBLIC API
michael@0:  ********************************************************************/
michael@0: 
michael@0: U_CAPI UHashtable* U_EXPORT2
michael@0: uhash_open(UHashFunction *keyHash, 
michael@0:            UKeyComparator *keyComp,
michael@0:            UValueComparator *valueComp,
michael@0:            UErrorCode *status) {
michael@0: 
michael@0:     return _uhash_create(keyHash, keyComp, valueComp, DEFAULT_PRIME_INDEX, status);
michael@0: }
michael@0: 
michael@0: U_CAPI UHashtable* U_EXPORT2
michael@0: uhash_openSize(UHashFunction *keyHash, 
michael@0:                UKeyComparator *keyComp,
michael@0:                UValueComparator *valueComp,
michael@0:                int32_t size,
michael@0:                UErrorCode *status) {
michael@0: 
michael@0:     /* Find the smallest index i for which PRIMES[i] >= size. */
michael@0:     int32_t i = 0;
michael@0:     while (i<(PRIMES_LENGTH-1) && PRIMES[i]<size) {
michael@0:         ++i;
michael@0:     }
michael@0: 
michael@0:     return _uhash_create(keyHash, keyComp, valueComp, i, status);
michael@0: }
michael@0: 
michael@0: U_CAPI UHashtable* U_EXPORT2
michael@0: uhash_init(UHashtable *fillinResult,
michael@0:            UHashFunction *keyHash, 
michael@0:            UKeyComparator *keyComp,
michael@0:            UValueComparator *valueComp,
michael@0:            UErrorCode *status) {
michael@0: 
michael@0:     return _uhash_init(fillinResult, keyHash, keyComp, valueComp, DEFAULT_PRIME_INDEX, status);
michael@0: }
michael@0: 
michael@0: U_CAPI void U_EXPORT2
michael@0: uhash_close(UHashtable *hash) {
michael@0:     if (hash == NULL) {
michael@0:         return;
michael@0:     }
michael@0:     if (hash->elements != NULL) {
michael@0:         if (hash->keyDeleter != NULL || hash->valueDeleter != NULL) {
michael@0:             int32_t pos=-1;
michael@0:             UHashElement *e;
michael@0:             while ((e = (UHashElement*) uhash_nextElement(hash, &pos)) != NULL) {
michael@0:                 HASH_DELETE_KEY_VALUE(hash, e->key.pointer, e->value.pointer);
michael@0:             }
michael@0:         }
michael@0:         uprv_free(hash->elements);
michael@0:         hash->elements = NULL;
michael@0:     }
michael@0:     if (hash->allocated) {
michael@0:         uprv_free(hash);
michael@0:     }
michael@0: }
michael@0: 
michael@0: U_CAPI UHashFunction *U_EXPORT2
michael@0: uhash_setKeyHasher(UHashtable *hash, UHashFunction *fn) {
michael@0:     UHashFunction *result = hash->keyHasher;
michael@0:     hash->keyHasher = fn;
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: U_CAPI UKeyComparator *U_EXPORT2
michael@0: uhash_setKeyComparator(UHashtable *hash, UKeyComparator *fn) {
michael@0:     UKeyComparator *result = hash->keyComparator;
michael@0:     hash->keyComparator = fn;
michael@0:     return result;
michael@0: }
michael@0: U_CAPI UValueComparator *U_EXPORT2 
michael@0: uhash_setValueComparator(UHashtable *hash, UValueComparator *fn){
michael@0:     UValueComparator *result = hash->valueComparator;
michael@0:     hash->valueComparator = fn;
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: U_CAPI UObjectDeleter *U_EXPORT2
michael@0: uhash_setKeyDeleter(UHashtable *hash, UObjectDeleter *fn) {
michael@0:     UObjectDeleter *result = hash->keyDeleter;
michael@0:     hash->keyDeleter = fn;
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: U_CAPI UObjectDeleter *U_EXPORT2
michael@0: uhash_setValueDeleter(UHashtable *hash, UObjectDeleter *fn) {
michael@0:     UObjectDeleter *result = hash->valueDeleter;
michael@0:     hash->valueDeleter = fn;
michael@0:     return result;
michael@0: }
michael@0: 
michael@0: U_CAPI void U_EXPORT2
michael@0: uhash_setResizePolicy(UHashtable *hash, enum UHashResizePolicy policy) {
michael@0:     UErrorCode status = U_ZERO_ERROR;
michael@0:     _uhash_internalSetResizePolicy(hash, policy);
michael@0:     hash->lowWaterMark  = (int32_t)(hash->length * hash->lowWaterRatio);
michael@0:     hash->highWaterMark = (int32_t)(hash->length * hash->highWaterRatio);    
michael@0:     _uhash_rehash(hash, &status);
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_count(const UHashtable *hash) {
michael@0:     return hash->count;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_get(const UHashtable *hash,
michael@0:           const void* key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.pointer = (void*) key;
michael@0:     return _uhash_find(hash, keyholder, hash->keyHasher(keyholder))->value.pointer;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_iget(const UHashtable *hash,
michael@0:            int32_t key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.integer = key;
michael@0:     return _uhash_find(hash, keyholder, hash->keyHasher(keyholder))->value.pointer;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_geti(const UHashtable *hash,
michael@0:            const void* key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.pointer = (void*) key;
michael@0:     return _uhash_find(hash, keyholder, hash->keyHasher(keyholder))->value.integer;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_igeti(const UHashtable *hash,
michael@0:            int32_t key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.integer = key;
michael@0:     return _uhash_find(hash, keyholder, hash->keyHasher(keyholder))->value.integer;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_put(UHashtable *hash,
michael@0:           void* key,
michael@0:           void* value,
michael@0:           UErrorCode *status) {
michael@0:     UHashTok keyholder, valueholder;
michael@0:     keyholder.pointer = key;
michael@0:     valueholder.pointer = value;
michael@0:     return _uhash_put(hash, keyholder, valueholder,
michael@0:                       HINT_KEY_POINTER | HINT_VALUE_POINTER,
michael@0:                       status).pointer;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_iput(UHashtable *hash,
michael@0:            int32_t key,
michael@0:            void* value,
michael@0:            UErrorCode *status) {
michael@0:     UHashTok keyholder, valueholder;
michael@0:     keyholder.integer = key;
michael@0:     valueholder.pointer = value;
michael@0:     return _uhash_put(hash, keyholder, valueholder,
michael@0:                       HINT_VALUE_POINTER,
michael@0:                       status).pointer;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_puti(UHashtable *hash,
michael@0:            void* key,
michael@0:            int32_t value,
michael@0:            UErrorCode *status) {
michael@0:     UHashTok keyholder, valueholder;
michael@0:     keyholder.pointer = key;
michael@0:     valueholder.integer = value;
michael@0:     return _uhash_put(hash, keyholder, valueholder,
michael@0:                       HINT_KEY_POINTER,
michael@0:                       status).integer;
michael@0: }
michael@0: 
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_iputi(UHashtable *hash,
michael@0:            int32_t key,
michael@0:            int32_t value,
michael@0:            UErrorCode *status) {
michael@0:     UHashTok keyholder, valueholder;
michael@0:     keyholder.integer = key;
michael@0:     valueholder.integer = value;
michael@0:     return _uhash_put(hash, keyholder, valueholder,
michael@0:                       0, /* neither is a ptr */
michael@0:                       status).integer;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_remove(UHashtable *hash,
michael@0:              const void* key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.pointer = (void*) key;
michael@0:     return _uhash_remove(hash, keyholder).pointer;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_iremove(UHashtable *hash,
michael@0:               int32_t key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.integer = key;
michael@0:     return _uhash_remove(hash, keyholder).pointer;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_removei(UHashtable *hash,
michael@0:               const void* key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.pointer = (void*) key;
michael@0:     return _uhash_remove(hash, keyholder).integer;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_iremovei(UHashtable *hash,
michael@0:                int32_t key) {
michael@0:     UHashTok keyholder;
michael@0:     keyholder.integer = key;
michael@0:     return _uhash_remove(hash, keyholder).integer;
michael@0: }
michael@0: 
michael@0: U_CAPI void U_EXPORT2
michael@0: uhash_removeAll(UHashtable *hash) {
michael@0:     int32_t pos = -1;
michael@0:     const UHashElement *e;
michael@0:     U_ASSERT(hash != NULL);
michael@0:     if (hash->count != 0) {
michael@0:         while ((e = uhash_nextElement(hash, &pos)) != NULL) {
michael@0:             uhash_removeElement(hash, e);
michael@0:         }
michael@0:     }
michael@0:     U_ASSERT(hash->count == 0);
michael@0: }
michael@0: 
michael@0: U_CAPI const UHashElement* U_EXPORT2
michael@0: uhash_find(const UHashtable *hash, const void* key) {
michael@0:     UHashTok keyholder;
michael@0:     const UHashElement *e;
michael@0:     keyholder.pointer = (void*) key;
michael@0:     e = _uhash_find(hash, keyholder, hash->keyHasher(keyholder));
michael@0:     return IS_EMPTY_OR_DELETED(e->hashcode) ? NULL : e;
michael@0: }
michael@0: 
michael@0: U_CAPI const UHashElement* U_EXPORT2
michael@0: uhash_nextElement(const UHashtable *hash, int32_t *pos) {
michael@0:     /* Walk through the array until we find an element that is not
michael@0:      * EMPTY and not DELETED.
michael@0:      */
michael@0:     int32_t i;
michael@0:     U_ASSERT(hash != NULL);
michael@0:     for (i = *pos + 1; i < hash->length; ++i) {
michael@0:         if (!IS_EMPTY_OR_DELETED(hash->elements[i].hashcode)) {
michael@0:             *pos = i;
michael@0:             return &(hash->elements[i]);
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* No more elements */
michael@0:     return NULL;
michael@0: }
michael@0: 
michael@0: U_CAPI void* U_EXPORT2
michael@0: uhash_removeElement(UHashtable *hash, const UHashElement* e) {
michael@0:     U_ASSERT(hash != NULL);
michael@0:     U_ASSERT(e != NULL);
michael@0:     if (!IS_EMPTY_OR_DELETED(e->hashcode)) {
michael@0:         UHashElement *nce = (UHashElement *)e;
michael@0:         return _uhash_internalRemoveElement(hash, nce).pointer;
michael@0:     }
michael@0:     return NULL;
michael@0: }
michael@0: 
michael@0: /********************************************************************
michael@0:  * UHashTok convenience
michael@0:  ********************************************************************/
michael@0: 
michael@0: /**
michael@0:  * Return a UHashTok for an integer.
michael@0:  */
michael@0: /*U_CAPI UHashTok U_EXPORT2
michael@0: uhash_toki(int32_t i) {
michael@0:     UHashTok tok;
michael@0:     tok.integer = i;
michael@0:     return tok;
michael@0: }*/
michael@0: 
michael@0: /**
michael@0:  * Return a UHashTok for a pointer.
michael@0:  */
michael@0: /*U_CAPI UHashTok U_EXPORT2
michael@0: uhash_tokp(void* p) {
michael@0:     UHashTok tok;
michael@0:     tok.pointer = p;
michael@0:     return tok;
michael@0: }*/
michael@0: 
michael@0: /********************************************************************
michael@0:  * PUBLIC Key Hash Functions
michael@0:  ********************************************************************/
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_hashUChars(const UHashTok key) {
michael@0:     const UChar *s = (const UChar *)key.pointer;
michael@0:     return s == NULL ? 0 : ustr_hashUCharsN(s, u_strlen(s));
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_hashChars(const UHashTok key) {
michael@0:     const char *s = (const char *)key.pointer;
michael@0:     return s == NULL ? 0 : ustr_hashCharsN(s, uprv_strlen(s));
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_hashIChars(const UHashTok key) {
michael@0:     const char *s = (const char *)key.pointer;
michael@0:     return s == NULL ? 0 : ustr_hashICharsN(s, uprv_strlen(s));
michael@0: }
michael@0: 
michael@0: U_CAPI UBool U_EXPORT2 
michael@0: uhash_equals(const UHashtable* hash1, const UHashtable* hash2){
michael@0:     int32_t count1, count2, pos, i;
michael@0: 
michael@0:     if(hash1==hash2){
michael@0:         return TRUE;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * Make sure that we are comparing 2 valid hashes of the same type
michael@0:      * with valid comparison functions.
michael@0:      * Without valid comparison functions, a binary comparison
michael@0:      * of the hash values will yield random results on machines
michael@0:      * with 64-bit pointers and 32-bit integer hashes.
michael@0:      * A valueComparator is normally optional.
michael@0:      */
michael@0:     if (hash1==NULL || hash2==NULL ||
michael@0:         hash1->keyComparator != hash2->keyComparator ||
michael@0:         hash1->valueComparator != hash2->valueComparator ||
michael@0:         hash1->valueComparator == NULL)
michael@0:     {
michael@0:         /*
michael@0:         Normally we would return an error here about incompatible hash tables,
michael@0:         but we return FALSE instead.
michael@0:         */
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     count1 = uhash_count(hash1);
michael@0:     count2 = uhash_count(hash2);
michael@0:     if(count1!=count2){
michael@0:         return FALSE;
michael@0:     }
michael@0:     
michael@0:     pos=-1;
michael@0:     for(i=0; i<count1; i++){
michael@0:         const UHashElement* elem1 = uhash_nextElement(hash1, &pos);
michael@0:         const UHashTok key1 = elem1->key;
michael@0:         const UHashTok val1 = elem1->value;
michael@0:         /* here the keys are not compared, instead the key form hash1 is used to fetch
michael@0:          * value from hash2. If the hashes are equal then then both hashes should 
michael@0:          * contain equal values for the same key!
michael@0:          */
michael@0:         const UHashElement* elem2 = _uhash_find(hash2, key1, hash2->keyHasher(key1));
michael@0:         const UHashTok val2 = elem2->value;
michael@0:         if(hash1->valueComparator(val1, val2)==FALSE){
michael@0:             return FALSE;
michael@0:         }
michael@0:     }
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: /********************************************************************
michael@0:  * PUBLIC Comparator Functions
michael@0:  ********************************************************************/
michael@0: 
michael@0: U_CAPI UBool U_EXPORT2
michael@0: uhash_compareUChars(const UHashTok key1, const UHashTok key2) {
michael@0:     const UChar *p1 = (const UChar*) key1.pointer;
michael@0:     const UChar *p2 = (const UChar*) key2.pointer;
michael@0:     if (p1 == p2) {
michael@0:         return TRUE;
michael@0:     }
michael@0:     if (p1 == NULL || p2 == NULL) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     while (*p1 != 0 && *p1 == *p2) {
michael@0:         ++p1;
michael@0:         ++p2;
michael@0:     }
michael@0:     return (UBool)(*p1 == *p2);
michael@0: }
michael@0: 
michael@0: U_CAPI UBool U_EXPORT2
michael@0: uhash_compareChars(const UHashTok key1, const UHashTok key2) {
michael@0:     const char *p1 = (const char*) key1.pointer;
michael@0:     const char *p2 = (const char*) key2.pointer;
michael@0:     if (p1 == p2) {
michael@0:         return TRUE;
michael@0:     }
michael@0:     if (p1 == NULL || p2 == NULL) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     while (*p1 != 0 && *p1 == *p2) {
michael@0:         ++p1;
michael@0:         ++p2;
michael@0:     }
michael@0:     return (UBool)(*p1 == *p2);
michael@0: }
michael@0: 
michael@0: U_CAPI UBool U_EXPORT2
michael@0: uhash_compareIChars(const UHashTok key1, const UHashTok key2) {
michael@0:     const char *p1 = (const char*) key1.pointer;
michael@0:     const char *p2 = (const char*) key2.pointer;
michael@0:     if (p1 == p2) {
michael@0:         return TRUE;
michael@0:     }
michael@0:     if (p1 == NULL || p2 == NULL) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     while (*p1 != 0 && uprv_tolower(*p1) == uprv_tolower(*p2)) {
michael@0:         ++p1;
michael@0:         ++p2;
michael@0:     }
michael@0:     return (UBool)(*p1 == *p2);
michael@0: }
michael@0: 
michael@0: /********************************************************************
michael@0:  * PUBLIC int32_t Support Functions
michael@0:  ********************************************************************/
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: uhash_hashLong(const UHashTok key) {
michael@0:     return key.integer;
michael@0: }
michael@0: 
michael@0: U_CAPI UBool U_EXPORT2
michael@0: uhash_compareLong(const UHashTok key1, const UHashTok key2) {
michael@0:     return (UBool)(key1.integer == key2.integer);
michael@0: }