/**
* Assumes that the given element is not empty or deleted.
*/
static UHashTok
_uhash_internalRemoveElement(UHashtable *hash, UHashElement* e) {
UHashTok empty;
U_ASSERT(!IS_EMPTY_OR_DELETED(e->hashcode));
--hash->count;
empty.pointer = NULL; empty.integer = 0;
return _uhash_setElement(hash, e, HASH_DELETED, empty, empty, 0);
}
U_CAPI const UHashElement* U_EXPORT2
uhash_find(const UHashtable *hash, const void* key) {
UHashTok keyholder;
const UHashElement *e;
keyholder.pointer = (void*) key;
e = _uhash_find(hash, keyholder, hash->keyHasher(keyholder));
return IS_EMPTY_OR_DELETED(e->hashcode) ? NULL : e;
}
U_CAPI void* U_EXPORT2
uhash_removeElement(UHashtable *hash, const UHashElement* e) {
U_ASSERT(hash != NULL);
U_ASSERT(e != NULL);
if (!IS_EMPTY_OR_DELETED(e->hashcode)) {
return _uhash_internalRemoveElement(hash, (UHashElement*) e).pointer;
}
return NULL;
}
/**
* Look for a key in the table, or if no such key exists, the first
* empty slot matching the given hashcode. Keys are compared using
* the keyComparator function.
*
* First find the start position, which is the hashcode modulo
* the length. Test it to see if it is:
*
* a. identical: First check the hash values for a quick check,
* then compare keys for equality using keyComparator.
* b. deleted
* c. empty
*
* Stop if it is identical or empty, otherwise continue by adding a
* "jump" value (moduloing by the length again to keep it within
* range) and retesting. For efficiency, there need enough empty
* values so that the searchs stop within a reasonable amount of time.
* This can be changed by changing the high/low water marks.
*
* In theory, this function can return NULL, if it is full (no empty
* or deleted slots) and if no matching key is found. In practice, we
* prevent this elsewhere (in uhash_put) by making sure the last slot
* in the table is never filled.
*
* The size of the table should be prime for this algorithm to work;
* otherwise we are not guaranteed that the jump value (the secondary
* hash) is relatively prime to the table length.
*/
static UHashElement*
_uhash_find(const UHashtable *hash, UHashTok key,
int32_t hashcode) {
int32_t firstDeleted = -1; /* assume invalid index */
int32_t theIndex, startIndex;
int32_t jump = 0; /* lazy evaluate */
int32_t tableHash;
UHashElement *elements = hash->elements;
hashcode &= 0x7FFFFFFF; /* must be positive */
startIndex = theIndex = (hashcode ^ 0x4000000) % hash->length;
do {
tableHash = elements[theIndex].hashcode;
if (tableHash == hashcode) { /* quick check */
if ((*hash->keyComparator)(key, elements[theIndex].key)) {
return &(elements[theIndex]);
}
} else if (!IS_EMPTY_OR_DELETED(tableHash)) {
/* We have hit a slot which contains a key-value pair,
* but for which the hash code does not match. Keep
* looking.
*/
} else if (tableHash == HASH_EMPTY) { /* empty, end o' the line */
break;
} else if (firstDeleted < 0) { /* remember first deleted */
firstDeleted = theIndex;
}
if (jump == 0) { /* lazy compute jump */
/* The jump value must be relatively prime to the table
* length. As long as the length is prime, then any value
* 1..length-1 will be relatively prime to it.
*/
jump = (hashcode % (hash->length - 1)) + 1;
}
theIndex = (theIndex + jump) % hash->length;
} while (theIndex != startIndex);
if (firstDeleted >= 0) {
theIndex = firstDeleted; /* reset if had deleted slot */
} else if (tableHash != HASH_EMPTY) {
/* We get to this point if the hashtable is full (no empty or
* deleted slots), and we've failed to find a match. THIS
* WILL NEVER HAPPEN as long as uhash_put() makes sure that
* count is always < length.
*/
U_ASSERT(FALSE);
return NULL; /* Never happens if uhash_put() behaves */
}
return &(elements[theIndex]);
}
U_CAPI const UHashElement* U_EXPORT2
uhash_nextElement(const UHashtable *hash, int32_t *pos) {
/* Walk through the array until we find an element that is not
* EMPTY and not DELETED.
*/
int32_t i;
U_ASSERT(hash != NULL);
for (i = *pos + 1; i < hash->length; ++i) {
if (!IS_EMPTY_OR_DELETED(hash->elements[i].hashcode)) {
*pos = i;
return &(hash->elements[i]);
}
}
/* No more elements */
return NULL;
}
/**
* Attempt to grow or shrink the data arrays in order to make the
* count fit between the high and low water marks. hash_put() and
* hash_remove() call this method when the count exceeds the high or
* low water marks. This method may do nothing, if memory allocation
* fails, or if the count is already in range, or if the length is
* already at the low or high limit. In any case, upon return the
* arrays will be valid.
*/
static void
_uhash_rehash(UHashtable *hash) {
UHashElement *old = hash->elements;
int32_t oldLength = hash->length;
int32_t newPrimeIndex = hash->primeIndex;
int32_t i;
UErrorCode status = U_ZERO_ERROR;
if (hash->count > hash->highWaterMark) {
if (++newPrimeIndex >= PRIMES_LENGTH) {
return;
}
} else if (hash->count < hash->lowWaterMark) {
if (--newPrimeIndex < 0) {
return;
}
} else {
return;
}
_uhash_allocate(hash, newPrimeIndex, &status);
if (U_FAILURE(status)) {
hash->elements = old;
hash->length = oldLength;
return;
}
for (i = oldLength - 1; i >= 0; --i) {
if (!IS_EMPTY_OR_DELETED(old[i].hashcode)) {
UHashElement *e = _uhash_find(hash, old[i].key, old[i].hashcode);
U_ASSERT(e != NULL);
U_ASSERT(e->hashcode == HASH_EMPTY);
e->key = old[i].key;
e->value = old[i].value;
e->hashcode = old[i].hashcode;
++hash->count;
}
}
uprv_free(old);
}
static UHashTok
_uhash_remove(UHashtable *hash,
UHashTok key) {
/* First find the position of the key in the table. If the object
* has not been removed already, remove it. If the user wanted
* keys deleted, then delete it also. We have to put a special
* hashcode in that position that means that something has been
* deleted, since when we do a find, we have to continue PAST any
* deleted values.
*/
UHashTok result;
UHashElement* e = _uhash_find(hash, key, hash->keyHasher(key));
U_ASSERT(e != NULL);
result.pointer = NULL; result.integer = 0;
if (!IS_EMPTY_OR_DELETED(e->hashcode)) {
result = _uhash_internalRemoveElement(hash, e);
if (hash->count < hash->lowWaterMark) {
_uhash_rehash(hash);
}
}
return result;
}
static UHashTok
_uhash_put(UHashtable *hash,
UHashTok key,
UHashTok value,
int8_t hint,
UErrorCode *status) {
/* Put finds the position in the table for the new value. If the
* key is already in the table, it is deleted, if there is a
* non-NULL keyDeleter. Then the key, the hash and the value are
* all put at the position in their respective arrays.
*/
int32_t hashcode;
UHashElement* e;
UHashTok emptytok;
if (U_FAILURE(*status)) {
goto err;
}
U_ASSERT(hash != NULL);
/* Cannot always check pointer here or iSeries sees NULL every time. */
if ((hint & HINT_VALUE_POINTER) && value.pointer == NULL) {
/* Disallow storage of NULL values, since NULL is returned by
* get() to indicate an absent key. Storing NULL == removing.
*/
return _uhash_remove(hash, key);
}
if (hash->count > hash->highWaterMark) {
_uhash_rehash(hash);
}
hashcode = (*hash->keyHasher)(key);
e = _uhash_find(hash, key, hashcode);
U_ASSERT(e != NULL);
if (IS_EMPTY_OR_DELETED(e->hashcode)) {
/* Important: We must never actually fill the table up. If we
* do so, then _uhash_find() will return NULL, and we'll have
* to check for NULL after every call to _uhash_find(). To
* avoid this we make sure there is always at least one empty
* or deleted slot in the table. This only is a problem if we
* are out of memory and rehash isn't working.
*/
++hash->count;
if (hash->count == hash->length) {
/* Don't allow count to reach length */
--hash->count;
*status = U_MEMORY_ALLOCATION_ERROR;
goto err;
}
}
/* We must in all cases handle storage properly. If there was an
* old key, then it must be deleted (if the deleter != NULL).
* Make hashcodes stored in table positive.
*/
return _uhash_setElement(hash, e, hashcode & 0x7FFFFFFF, key, value, hint);
err:
/* If the deleters are non-NULL, this method adopts its key and/or
* value arguments, and we must be sure to delete the key and/or
* value in all cases, even upon failure.
*/
HASH_DELETE_KEY_VALUE(hash, key.pointer, value.pointer);
emptytok.pointer = NULL; emptytok.integer = 0;
return emptytok;
}
