/* This hashtable is implemented as a double hash. All elements are * stored in a single array with no secondary storage for collision * resolution (no linked list, etc.). When there is a hash collision * (when two unequal keys have the same hashcode) we resolve this by * using a secondary hash. The secondary hash is an increment * computed as a hash function (a different one) of the primary * hashcode. This increment is added to the initial hash value to * obtain further slots assigned to the same hash code. For this to * work, the length of the array and the increment must be relatively * prime. The easiest way to achieve this is to have the length of * the array be prime, and the increment be any value from * 1..length-1. * * Hashcodes are 32-bit integers. We make sure all hashcodes are * non-negative by masking off the top bit. This has two effects: (1) * modulo arithmetic is simplified. If we allowed negative hashcodes, * then when we computed hashcode % length, we could get a negative * result, which we would then have to adjust back into range. It's * simpler to just make hashcodes non-negative. (2) It makes it easy * to check for empty vs. occupied slots in the table. We just mark * empty or deleted slots with a negative hashcode. * * The central function is _uhash_find(). This function looks for a * slot matching the given key and hashcode. If one is found, it * returns a pointer to that slot. If the table is full, and no match * is found, it returns nullptr -- in theory. This would make the code * more complicated, since all callers of _uhash_find() would then * have to check for a nullptr result. To keep this from happening, we * don't allow the table to fill. When there is only one * empty/deleted slot left, uhash_put() will refuse to increase the * count, and fail. This simplifies the code. In practice, one will * seldom encounter this using default UHashtables. However, if a * hashtable is set to a U_FIXED resize policy, or if memory is * exhausted, then the table may fill. * * High and low water ratios control rehashing. They establish levels * of fullness (from 0 to 1) outside of which the data array is * reallocated and repopulated. Setting the low water ratio to zero * means the table will never shrink. Setting the high water ratio to * one means the table will never grow. The ratios should be * coordinated with the ratio between successive elements of the * PRIMES table, so that when the primeIndex is incremented or * decremented during rehashing, it brings the ratio of count / length * back into the desired range (between low and high water ratios).
*/
/* This is a list of non-consecutive primes chosen such that * PRIMES[i+1] ~ 2*PRIMES[i]. (Currently, the ratio ranges from 1.81 * to 2.18; the inverse ratio ranges from 0.459 to 0.552.) If this * ratio is changed, the low and high water ratios should also be * adjusted to suit. * * These prime numbers were also chosen so that they are the largest * prime number while being less than a power of two.
*/ staticconst int32_t PRIMES[] = {
7, 13, 31, 61, 127, 251, 509, 1021, 2039, 4093, 8191, 16381, 32749,
65521, 131071, 262139, 524287, 1048573, 2097143, 4194301, 8388593,
16777213, 33554393, 67108859, 134217689, 268435399, 536870909,
1073741789, 2147483647 /*, 4294967291 */
};
/* These ratios are tuned to the PRIMES array such that a resize * places the table back into the zone of non-resizing. That is, * after a call to _uhash_rehash(), a subsequent call to * _uhash_rehash() should do nothing (should not churn). This is only * a potential problem with U_GROW_AND_SHRINK.
*/ staticconstfloat RESIZE_POLICY_RATIO_TABLE[6] = { /* low, high water ratio */
0.0F, 0.5F, /* U_GROW: Grow on demand, do not shrink */
0.1F, 0.5F, /* U_GROW_AND_SHRINK: Grow and shrink on demand */
0.0F, 1.0F /* U_FIXED: Never change size */
};
/* Invariants for hashcode values:
* DELETED < 0 * EMPTY < 0 * Real hashes >= 0
Hashcodes may not start out this way, but internally they are adjusted so that they are always positive. We assume 32-bit hashcodes; adjust these constants for other hashcode sizes.
*/ #define HASH_DELETED ((int32_t) 0x80000000) #define HASH_EMPTY ((int32_t) HASH_DELETED + 1)
#define IS_EMPTY_OR_DELETED(x) ((x) < 0)
/* This macro expects a UHashTok.pointer as its keypointer and
valuepointer parameters */ #define HASH_DELETE_KEY_VALUE(hash, keypointer, valuepointer) UPRV_BLOCK_MACRO_BEGIN { \ if (hash->keyDeleter != nullptr && keypointer != nullptr) { \
(*hash->keyDeleter)(keypointer); \
} \ if (hash->valueDeleter != nullptr && valuepointer != nullptr) { \
(*hash->valueDeleter)(valuepointer); \
} \
} UPRV_BLOCK_MACRO_END
/* * Constants for hinting whether a key or value is an integer * or a pointer. If a hint bit is zero, then the associated * token is assumed to be an integer.
*/ #define HINT_BOTH_INTEGERS (0) #define HINT_KEY_POINTER (1) #define HINT_VALUE_POINTER (2) #define HINT_ALLOW_ZERO (4)
UHashTok oldValue = e->value; if (hash->keyDeleter != nullptr && e->key.pointer != nullptr &&
e->key.pointer != key.pointer) { /* Avoid double deletion */
(*hash->keyDeleter)(e->key.pointer);
} if (hash->valueDeleter != nullptr) { if (oldValue.pointer != nullptr &&
oldValue.pointer != value.pointer) { /* Avoid double deletion */
(*hash->valueDeleter)(oldValue.pointer);
}
oldValue.pointer = nullptr;
} /* 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 = nullptr; empty.integer = 0; return _uhash_setElement(hash, e, HASH_DELETED, empty, empty, 0);
}
/** * 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.
*/ staticvoid
_uhash_allocate(UHashtable *hash,
int32_t primeIndex,
UErrorCode *status) {
if (U_FAILURE(*status)) {
uprv_free(result); return nullptr;
}
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 searches stop within a reasonable amount of time. * This can be changed by changing the high/low water marks. * * In theory, this function can return nullptr, 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) {
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]);
}
} elseif (!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.
*/
} elseif (tableHash == HASH_EMPTY) { /* empty, end o' the line */ break;
} elseif (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 */
} elseif (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.
*/
UPRV_UNREACHABLE_EXIT;
} 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.
*/ staticvoid
_uhash_rehash(UHashtable *hash, UErrorCode *status) {
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 != nullptr);
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 != nullptr);
result.pointer = nullptr;
result.integer = 0; if (!IS_EMPTY_OR_DELETED(e->hashcode)) {
result = _uhash_internalRemoveElement(hash, e); if (hash->count < hash->lowWaterMark) {
UErrorCode status = U_ZERO_ERROR;
_uhash_rehash(hash, &status);
}
} return result;
}
/* 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-nullptr 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 != nullptr); if ((hint & HINT_VALUE_POINTER) ?
value.pointer == nullptr :
value.integer == 0 && (hint & HINT_ALLOW_ZERO) == 0) { /* Disallow storage of nullptr values, since nullptr is returned by * get() to indicate an absent key. Storing nullptr == removing.
*/ return _uhash_remove(hash, key);
} if (hash->count > hash->highWaterMark) {
_uhash_rehash(hash, status); if (U_FAILURE(*status)) { goto err;
}
}
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 nullptr, and we'll have * to check for nullptr 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 != nullptr). * Make hashcodes stored in table positive.
*/ return _uhash_setElement(hash, e, hashcode & 0x7FFFFFFF, key, value, hint);
err: /* If the deleters are non-nullptr, 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 = nullptr; emptytok.integer = 0; return emptytok;
}
/******************************************************************** * PUBLIC API
********************************************************************/
// Find the smallest index i for which PRIMES[i] >= size.
int32_t i = 0; while (i<(PRIMES_LENGTH-1) && PRIMES[i]<size) {
++i;
} return _uhash_init(fillinResult, keyHash, keyComp, valueComp, i, status);
}
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 != nullptr); for (i = *pos + 1; i < hash->length; ++i) { if (!IS_EMPTY_OR_DELETED(hash->elements[i].hashcode)) {
*pos = i; return &(hash->elements[i]);
}
}
/* * Make sure that we are comparing 2 valid hashes of the same type * with valid comparison functions. * Without valid comparison functions, a binary comparison * of the hash values will yield random results on machines * with 64-bit pointers and 32-bit integer hashes. * A valueComparator is normally optional.
*/ if (hash1==nullptr || hash2==nullptr ||
hash1->keyComparator != hash2->keyComparator ||
hash1->valueComparator != hash2->valueComparator ||
hash1->valueComparator == nullptr)
{ /* Normally we would return an error here about incompatible hash tables, but we return false instead.
*/ returnfalse;
}
pos=UHASH_FIRST; for(i=0; i<count1; i++){ const UHashElement* elem1 = uhash_nextElement(hash1, &pos); const UHashTok key1 = elem1->key; const UHashTok val1 = elem1->value; /* here the keys are not compared, instead the key form hash1 is used to fetch * value from hash2. If the hashes are equal then then both hashes should * contain equal values for the same key!
*/ const UHashElement* elem2 = _uhash_find(hash2, key1, hash2->keyHasher(key1)); const UHashTok val2 = elem2->value; if(hash1->valueComparator(val1, val2)==false){ returnfalse;
}
} returntrue;
}
/******************************************************************** * PUBLIC Comparator Functions
********************************************************************/
/******************************************************************** * PUBLIC int32_t Support Functions
********************************************************************/
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.
