ICU-5208 Make it easier for dumb compilers to optimize the hashtable.
X-SVN-Rev: 19704
This commit is contained in:
parent
5e15aecbe2
commit
290c62fdcd
@ -133,50 +133,396 @@ static const float RESIZE_POLICY_RATIO_TABLE[6] = {
|
||||
#define HINT_VALUE_POINTER (2)
|
||||
|
||||
/********************************************************************
|
||||
* Debugging
|
||||
* PRIVATE Implementation
|
||||
********************************************************************/
|
||||
|
||||
static UHashTok
|
||||
_uhash_setElement(UHashtable *hash, UHashElement* e,
|
||||
int32_t hashcode,
|
||||
UHashTok key, UHashTok value, int8_t hint) {
|
||||
|
||||
/********************************************************************
|
||||
* PRIVATE Prototypes
|
||||
********************************************************************/
|
||||
UHashTok oldValue = e->value;
|
||||
if (hash->keyDeleter != NULL && e->key.pointer != NULL &&
|
||||
e->key.pointer != key.pointer) { /* Avoid double deletion */
|
||||
(*hash->keyDeleter)(e->key.pointer);
|
||||
}
|
||||
if (hash->valueDeleter != NULL) {
|
||||
if (oldValue.pointer != NULL &&
|
||||
oldValue.pointer != value.pointer) { /* Avoid double deletion */
|
||||
(*hash->valueDeleter)(oldValue.pointer);
|
||||
}
|
||||
oldValue.pointer = NULL;
|
||||
}
|
||||
/* Compilers should copy the UHashTok union correctly, but even if
|
||||
* they do, memory heap tools (e.g. BoundsChecker) can get
|
||||
* confused when a pointer is cloaked in a union and then copied.
|
||||
* TO ALLEVIATE THIS, we use hints (based on what API the user is
|
||||
* calling) to copy pointers when we know the user thinks
|
||||
* something is a pointer. */
|
||||
if (hint & HINT_KEY_POINTER) {
|
||||
e->key.pointer = key.pointer;
|
||||
} else {
|
||||
e->key = key;
|
||||
}
|
||||
if (hint & HINT_VALUE_POINTER) {
|
||||
e->value.pointer = value.pointer;
|
||||
} else {
|
||||
e->value = value;
|
||||
}
|
||||
e->hashcode = hashcode;
|
||||
return oldValue;
|
||||
}
|
||||
|
||||
static UHashtable* _uhash_init(UHashtable *fillinResult,
|
||||
UHashFunction *keyHash,
|
||||
UKeyComparator *keyComp,
|
||||
UValueComparator *valueComp,
|
||||
int32_t primeIndex, UErrorCode *status);
|
||||
/**
|
||||
* 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);
|
||||
}
|
||||
|
||||
static UHashtable* _uhash_create(UHashFunction *keyHash,
|
||||
UKeyComparator *keyComp,
|
||||
UValueComparator *valueComp,
|
||||
int32_t primeIndex, UErrorCode *status);
|
||||
static void
|
||||
_uhash_internalSetResizePolicy(UHashtable *hash, enum UHashResizePolicy policy) {
|
||||
U_ASSERT(hash != NULL);
|
||||
U_ASSERT(((int32_t)policy) >= 0);
|
||||
U_ASSERT(((int32_t)policy) < 3);
|
||||
hash->lowWaterRatio = RESIZE_POLICY_RATIO_TABLE[policy * 2];
|
||||
hash->highWaterRatio = RESIZE_POLICY_RATIO_TABLE[policy * 2 + 1];
|
||||
}
|
||||
|
||||
static void _uhash_allocate(UHashtable *hash, int32_t primeIndex,
|
||||
UErrorCode *status);
|
||||
/**
|
||||
* Allocate internal data array of a size determined by the given
|
||||
* prime index. If the index is out of range it is pinned into range.
|
||||
* If the allocation fails the status is set to
|
||||
* U_MEMORY_ALLOCATION_ERROR and all array storage is freed. In
|
||||
* either case the previous array pointer is overwritten.
|
||||
*
|
||||
* Caller must ensure primeIndex is in range 0..PRIME_LENGTH-1.
|
||||
*/
|
||||
static void
|
||||
_uhash_allocate(UHashtable *hash,
|
||||
int32_t primeIndex,
|
||||
UErrorCode *status) {
|
||||
|
||||
static void _uhash_rehash(UHashtable *hash);
|
||||
UHashElement *p, *limit;
|
||||
UHashTok emptytok;
|
||||
|
||||
static UHashElement* _uhash_find(const UHashtable *hash, UHashTok key,
|
||||
int32_t hashcode);
|
||||
if (U_FAILURE(*status)) return;
|
||||
|
||||
static UHashTok _uhash_put(UHashtable *hash,
|
||||
UHashTok key,
|
||||
UHashTok value,
|
||||
int8_t hint,
|
||||
UErrorCode *status);
|
||||
U_ASSERT(primeIndex >= 0 && primeIndex < PRIMES_LENGTH);
|
||||
|
||||
static UHashTok _uhash_remove(UHashtable *hash,
|
||||
UHashTok key);
|
||||
hash->primeIndex = primeIndex;
|
||||
hash->length = PRIMES[primeIndex];
|
||||
|
||||
static UHashTok _uhash_internalRemoveElement(UHashtable *hash, UHashElement* e);
|
||||
p = hash->elements = (UHashElement*)
|
||||
uprv_malloc(sizeof(UHashElement) * hash->length);
|
||||
|
||||
static UHashTok _uhash_setElement(UHashtable* hash, UHashElement* e,
|
||||
int32_t hashcode,
|
||||
UHashTok key, UHashTok value,
|
||||
int8_t hint);
|
||||
if (hash->elements == NULL) {
|
||||
*status = U_MEMORY_ALLOCATION_ERROR;
|
||||
return;
|
||||
}
|
||||
|
||||
emptytok.pointer = NULL; /* Only one of these two is needed */
|
||||
emptytok.integer = 0; /* but we don't know which one. */
|
||||
|
||||
limit = p + hash->length;
|
||||
while (p < limit) {
|
||||
p->key = emptytok;
|
||||
p->value = emptytok;
|
||||
p->hashcode = HASH_EMPTY;
|
||||
++p;
|
||||
}
|
||||
|
||||
hash->count = 0;
|
||||
hash->lowWaterMark = (int32_t)(hash->length * hash->lowWaterRatio);
|
||||
hash->highWaterMark = (int32_t)(hash->length * hash->highWaterRatio);
|
||||
}
|
||||
|
||||
static UHashtable*
|
||||
_uhash_init(UHashtable *result,
|
||||
UHashFunction *keyHash,
|
||||
UKeyComparator *keyComp,
|
||||
UValueComparator *valueComp,
|
||||
int32_t primeIndex,
|
||||
UErrorCode *status)
|
||||
{
|
||||
if (U_FAILURE(*status)) return NULL;
|
||||
U_ASSERT(keyHash != NULL);
|
||||
U_ASSERT(keyComp != NULL);
|
||||
|
||||
result->keyHasher = keyHash;
|
||||
result->keyComparator = keyComp;
|
||||
result->valueComparator = valueComp;
|
||||
result->keyDeleter = NULL;
|
||||
result->valueDeleter = NULL;
|
||||
result->allocated = FALSE;
|
||||
_uhash_internalSetResizePolicy(result, U_GROW);
|
||||
|
||||
_uhash_allocate(result, primeIndex, status);
|
||||
|
||||
if (U_FAILURE(*status)) {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
static UHashtable*
|
||||
_uhash_create(UHashFunction *keyHash,
|
||||
UKeyComparator *keyComp,
|
||||
UValueComparator *valueComp,
|
||||
int32_t primeIndex,
|
||||
UErrorCode *status) {
|
||||
UHashtable *result;
|
||||
|
||||
if (U_FAILURE(*status)) return NULL;
|
||||
|
||||
result = (UHashtable*) uprv_malloc(sizeof(UHashtable));
|
||||
if (result == NULL) {
|
||||
*status = U_MEMORY_ALLOCATION_ERROR;
|
||||
return NULL;
|
||||
}
|
||||
|
||||
_uhash_init(result, keyHash, keyComp, valueComp, primeIndex, status);
|
||||
result->allocated = TRUE;
|
||||
|
||||
if (U_FAILURE(*status)) {
|
||||
uprv_free(result);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/**
|
||||
* 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]);
|
||||
}
|
||||
|
||||
/**
|
||||
* 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;
|
||||
}
|
||||
|
||||
static void _uhash_internalSetResizePolicy(UHashtable *hash, enum UHashResizePolicy policy);
|
||||
|
||||
/********************************************************************
|
||||
* PUBLIC API
|
||||
@ -642,392 +988,3 @@ uhash_freeBlock(void *obj) {
|
||||
uprv_free(obj);
|
||||
}
|
||||
|
||||
/********************************************************************
|
||||
* PRIVATE Implementation
|
||||
********************************************************************/
|
||||
|
||||
static UHashtable*
|
||||
_uhash_init(UHashtable *result,
|
||||
UHashFunction *keyHash,
|
||||
UKeyComparator *keyComp,
|
||||
UValueComparator *valueComp,
|
||||
int32_t primeIndex,
|
||||
UErrorCode *status)
|
||||
{
|
||||
if (U_FAILURE(*status)) return NULL;
|
||||
U_ASSERT(keyHash != NULL);
|
||||
U_ASSERT(keyComp != NULL);
|
||||
|
||||
result->keyHasher = keyHash;
|
||||
result->keyComparator = keyComp;
|
||||
result->valueComparator = valueComp;
|
||||
result->keyDeleter = NULL;
|
||||
result->valueDeleter = NULL;
|
||||
result->allocated = FALSE;
|
||||
_uhash_internalSetResizePolicy(result, U_GROW);
|
||||
|
||||
_uhash_allocate(result, primeIndex, status);
|
||||
|
||||
if (U_FAILURE(*status)) {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
static UHashtable*
|
||||
_uhash_create(UHashFunction *keyHash,
|
||||
UKeyComparator *keyComp,
|
||||
UValueComparator *valueComp,
|
||||
int32_t primeIndex,
|
||||
UErrorCode *status) {
|
||||
UHashtable *result;
|
||||
|
||||
if (U_FAILURE(*status)) return NULL;
|
||||
|
||||
result = (UHashtable*) uprv_malloc(sizeof(UHashtable));
|
||||
if (result == NULL) {
|
||||
*status = U_MEMORY_ALLOCATION_ERROR;
|
||||
return NULL;
|
||||
}
|
||||
|
||||
_uhash_init(result, keyHash, keyComp, valueComp, primeIndex, status);
|
||||
result->allocated = TRUE;
|
||||
|
||||
if (U_FAILURE(*status)) {
|
||||
uprv_free(result);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/**
|
||||
* Allocate internal data array of a size determined by the given
|
||||
* prime index. If the index is out of range it is pinned into range.
|
||||
* If the allocation fails the status is set to
|
||||
* U_MEMORY_ALLOCATION_ERROR and all array storage is freed. In
|
||||
* either case the previous array pointer is overwritten.
|
||||
*
|
||||
* Caller must ensure primeIndex is in range 0..PRIME_LENGTH-1.
|
||||
*/
|
||||
static void
|
||||
_uhash_allocate(UHashtable *hash,
|
||||
int32_t primeIndex,
|
||||
UErrorCode *status) {
|
||||
|
||||
UHashElement *p, *limit;
|
||||
UHashTok emptytok;
|
||||
|
||||
if (U_FAILURE(*status)) return;
|
||||
|
||||
U_ASSERT(primeIndex >= 0 && primeIndex < PRIMES_LENGTH);
|
||||
|
||||
hash->primeIndex = primeIndex;
|
||||
hash->length = PRIMES[primeIndex];
|
||||
|
||||
p = hash->elements = (UHashElement*)
|
||||
uprv_malloc(sizeof(UHashElement) * hash->length);
|
||||
|
||||
if (hash->elements == NULL) {
|
||||
*status = U_MEMORY_ALLOCATION_ERROR;
|
||||
return;
|
||||
}
|
||||
|
||||
emptytok.pointer = NULL; /* Only one of these two is needed */
|
||||
emptytok.integer = 0; /* but we don't know which one. */
|
||||
|
||||
limit = p + hash->length;
|
||||
while (p < limit) {
|
||||
p->key = emptytok;
|
||||
p->value = emptytok;
|
||||
p->hashcode = HASH_EMPTY;
|
||||
++p;
|
||||
}
|
||||
|
||||
hash->count = 0;
|
||||
hash->lowWaterMark = (int32_t)(hash->length * hash->lowWaterRatio);
|
||||
hash->highWaterMark = (int32_t)(hash->length * hash->highWaterRatio);
|
||||
}
|
||||
|
||||
/**
|
||||
* 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);
|
||||
}
|
||||
|
||||
/**
|
||||
* 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;
|
||||
|
||||
hashcode &= 0x7FFFFFFF; /* must be positive */
|
||||
startIndex = theIndex = (hashcode ^ 0x4000000) % hash->length;
|
||||
|
||||
do {
|
||||
tableHash = hash->elements[theIndex].hashcode;
|
||||
if (tableHash == hashcode) { /* quick check */
|
||||
if ((*hash->keyComparator)(key, hash->elements[theIndex].key)) {
|
||||
return &(hash->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 &(hash->elements[theIndex]);
|
||||
}
|
||||
|
||||
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;
|
||||
}
|
||||
|
||||
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_setElement(UHashtable *hash, UHashElement* e,
|
||||
int32_t hashcode,
|
||||
UHashTok key, UHashTok value, int8_t hint) {
|
||||
|
||||
UHashTok oldValue = e->value;
|
||||
if (hash->keyDeleter != NULL && e->key.pointer != NULL &&
|
||||
e->key.pointer != key.pointer) { /* Avoid double deletion */
|
||||
(*hash->keyDeleter)(e->key.pointer);
|
||||
}
|
||||
if (hash->valueDeleter != NULL) {
|
||||
if (oldValue.pointer != NULL &&
|
||||
oldValue.pointer != value.pointer) { /* Avoid double deletion */
|
||||
(*hash->valueDeleter)(oldValue.pointer);
|
||||
}
|
||||
oldValue.pointer = NULL;
|
||||
}
|
||||
/* Compilers should copy the UHashTok union correctly, but even if
|
||||
* they do, memory heap tools (e.g. BoundsChecker) can get
|
||||
* confused when a pointer is cloaked in a union and then copied.
|
||||
* TO ALLEVIATE THIS, we use hints (based on what API the user is
|
||||
* calling) to copy pointers when we know the user thinks
|
||||
* something is a pointer. */
|
||||
if (hint & HINT_KEY_POINTER) {
|
||||
e->key.pointer = key.pointer;
|
||||
} else {
|
||||
e->key = key;
|
||||
}
|
||||
if (hint & HINT_VALUE_POINTER) {
|
||||
e->value.pointer = value.pointer;
|
||||
} else {
|
||||
e->value = value;
|
||||
}
|
||||
e->hashcode = hashcode;
|
||||
return oldValue;
|
||||
}
|
||||
|
||||
/**
|
||||
* 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);
|
||||
}
|
||||
|
||||
static void
|
||||
_uhash_internalSetResizePolicy(UHashtable *hash, enum UHashResizePolicy policy) {
|
||||
U_ASSERT(hash != NULL);
|
||||
U_ASSERT(((int32_t)policy) >= 0);
|
||||
U_ASSERT(((int32_t)policy) < 3);
|
||||
hash->lowWaterRatio = RESIZE_POLICY_RATIO_TABLE[policy * 2];
|
||||
hash->highWaterRatio = RESIZE_POLICY_RATIO_TABLE[policy * 2 + 1];
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user