Datum
spg_text_choose(PG_FUNCTION_ARGS)
{
spgChooseIn *in = (spgChooseIn *) PG_GETARG_POINTER(0);
spgChooseOut *out = (spgChooseOut *) PG_GETARG_POINTER(1);
text *inText = DatumGetTextPP(in->datum);
char *inStr = VARDATA_ANY(inText);
int inSize = VARSIZE_ANY_EXHDR(inText);
uint8 nodeChar = '\0';
int i = 0;
int commonLen = 0;
/* Check for prefix match, set nodeChar to first byte after prefix */
if (in->hasPrefix)
{
text *prefixText = DatumGetTextPP(in->prefixDatum);
char *prefixStr = VARDATA_ANY(prefixText);
int prefixSize = VARSIZE_ANY_EXHDR(prefixText);
commonLen = commonPrefix(inStr + in->level,
prefixStr,
inSize - in->level,
prefixSize);
if (commonLen == prefixSize)
{
if (inSize - in->level > commonLen)
nodeChar = *(uint8 *) (inStr + in->level + commonLen);
else
nodeChar = '\0';
}
else
{
/* Must split tuple because incoming value doesn't match prefix */
out->resultType = spgSplitTuple;
if (commonLen == 0)
{
out->result.splitTuple.prefixHasPrefix = false;
}
else
{
out->result.splitTuple.prefixHasPrefix = true;
out->result.splitTuple.prefixPrefixDatum =
formTextDatum(prefixStr, commonLen);
}
out->result.splitTuple.nodeLabel =
UInt8GetDatum(*(prefixStr + commonLen));
if (prefixSize - commonLen == 1)
{
out->result.splitTuple.postfixHasPrefix = false;
}
else
{
out->result.splitTuple.postfixHasPrefix = true;
out->result.splitTuple.postfixPrefixDatum =
formTextDatum(prefixStr + commonLen + 1,
prefixSize - commonLen - 1);
}
PG_RETURN_VOID();
}
}
else if (inSize > in->level)
{
nodeChar = *(uint8 *) (inStr + in->level);
}
else
{
nodeChar = '\0';
}
/* Look up nodeChar in the node label array */
if (searchChar(in->nodeLabels, in->nNodes, nodeChar, &i))
{
/*
* Descend to existing node. (If in->allTheSame, the core code will
* ignore our nodeN specification here, but that's OK. We still
* have to provide the correct levelAdd and restDatum values, and
* those are the same regardless of which node gets chosen by core.)
*/
out->resultType = spgMatchNode;
out->result.matchNode.nodeN = i;
out->result.matchNode.levelAdd = commonLen + 1;
if (inSize - in->level - commonLen - 1 > 0)
out->result.matchNode.restDatum =
formTextDatum(inStr + in->level + commonLen + 1,
inSize - in->level - commonLen - 1);
else
out->result.matchNode.restDatum =
formTextDatum(NULL, 0);
}
else if (in->allTheSame)
{
/*
* Can't use AddNode action, so split the tuple. The upper tuple
* has the same prefix as before and uses an empty node label for
* the lower tuple. The lower tuple has no prefix and the same
* node labels as the original tuple.
*/
out->resultType = spgSplitTuple;
out->result.splitTuple.prefixHasPrefix = in->hasPrefix;
out->result.splitTuple.prefixPrefixDatum = in->prefixDatum;
out->result.splitTuple.nodeLabel = UInt8GetDatum('\0');
out->result.splitTuple.postfixHasPrefix = false;
}
else
{
/* Add a node for the not-previously-seen nodeChar value */
out->resultType = spgAddNode;
out->result.addNode.nodeLabel = UInt8GetDatum(nodeChar);
out->result.addNode.nodeN = i;
}
PG_RETURN_VOID();
}
Datum
spg_text_picksplit(PG_FUNCTION_ARGS)
{
spgPickSplitIn *in = (spgPickSplitIn *) PG_GETARG_POINTER(0);
spgPickSplitOut *out = (spgPickSplitOut *) PG_GETARG_POINTER(1);
text *text0 = DatumGetTextPP(in->datums[0]);
int i,
commonLen;
spgNodePtr *nodes;
/* Identify longest common prefix, if any */
commonLen = VARSIZE_ANY_EXHDR(text0);
for (i = 1; i < in->nTuples && commonLen > 0; i++)
{
text *texti = DatumGetTextPP(in->datums[i]);
int tmp = commonPrefix(VARDATA_ANY(text0),
VARDATA_ANY(texti),
VARSIZE_ANY_EXHDR(text0),
VARSIZE_ANY_EXHDR(texti));
if (tmp < commonLen)
commonLen = tmp;
}
/*
* Limit the prefix length, if necessary, to ensure that the resulting
* inner tuple will fit on a page.
*/
commonLen = Min(commonLen, SPGIST_MAX_PREFIX_LENGTH);
/* Set node prefix to be that string, if it's not empty */
if (commonLen == 0)
{
out->hasPrefix = false;
}
else
{
out->hasPrefix = true;
out->prefixDatum = formTextDatum(VARDATA_ANY(text0), commonLen);
}
/* Extract the node label (first non-common byte) from each value */
nodes = (spgNodePtr *) palloc(sizeof(spgNodePtr) * in->nTuples);
for (i = 0; i < in->nTuples; i++)
{
text *texti = DatumGetTextPP(in->datums[i]);
if (commonLen < VARSIZE_ANY_EXHDR(texti))
nodes[i].c = *(uint8 *) (VARDATA_ANY(texti) + commonLen);
else
nodes[i].c = '\0'; /* use \0 if string is all common */
nodes[i].i = i;
nodes[i].d = in->datums[i];
}
/*
* Sort by label bytes so that we can group the values into nodes. This
* also ensures that the nodes are ordered by label value, allowing the
* use of binary search in searchChar.
*/
qsort(nodes, in->nTuples, sizeof(*nodes), cmpNodePtr);
/* And emit results */
out->nNodes = 0;
out->nodeLabels = (Datum *) palloc(sizeof(Datum) * in->nTuples);
out->mapTuplesToNodes = (int *) palloc(sizeof(int) * in->nTuples);
out->leafTupleDatums = (Datum *) palloc(sizeof(Datum) * in->nTuples);
for (i = 0; i < in->nTuples; i++)
{
text *texti = DatumGetTextPP(nodes[i].d);
Datum leafD;
if (i == 0 || nodes[i].c != nodes[i - 1].c)
{
out->nodeLabels[out->nNodes] = UInt8GetDatum(nodes[i].c);
out->nNodes++;
}
if (commonLen < VARSIZE_ANY_EXHDR(texti))
leafD = formTextDatum(VARDATA_ANY(texti) + commonLen + 1,
VARSIZE_ANY_EXHDR(texti) - commonLen - 1);
else
leafD = formTextDatum(NULL, 0);
out->leafTupleDatums[nodes[i].i] = leafD;
out->mapTuplesToNodes[nodes[i].i] = out->nNodes - 1;
}
PG_RETURN_VOID();
}
Datum
spg_text_choose(PG_FUNCTION_ARGS)
{
spgChooseIn *in = (spgChooseIn *) PG_GETARG_POINTER(0);
spgChooseOut *out = (spgChooseOut *) PG_GETARG_POINTER(1);
text *inText = DatumGetTextPP(in->datum);
char *inStr = VARDATA_ANY(inText);
int inSize = VARSIZE_ANY_EXHDR(inText);
char *prefixStr = NULL;
int prefixSize = 0;
int commonLen = 0;
int16 nodeChar = 0;
int i = 0;
/* Check for prefix match, set nodeChar to first byte after prefix */
if (in->hasPrefix)
{
text *prefixText = DatumGetTextPP(in->prefixDatum);
prefixStr = VARDATA_ANY(prefixText);
prefixSize = VARSIZE_ANY_EXHDR(prefixText);
commonLen = commonPrefix(inStr + in->level,
prefixStr,
inSize - in->level,
prefixSize);
if (commonLen == prefixSize)
{
if (inSize - in->level > commonLen)
nodeChar = *(unsigned char *) (inStr + in->level + commonLen);
else
nodeChar = -1;
}
else
{
/* Must split tuple because incoming value doesn't match prefix */
out->resultType = spgSplitTuple;
if (commonLen == 0)
{
out->result.splitTuple.prefixHasPrefix = false;
}
else
{
out->result.splitTuple.prefixHasPrefix = true;
out->result.splitTuple.prefixPrefixDatum =
formTextDatum(prefixStr, commonLen);
}
out->result.splitTuple.nodeLabel =
Int16GetDatum(*(unsigned char *) (prefixStr + commonLen));
if (prefixSize - commonLen == 1)
{
out->result.splitTuple.postfixHasPrefix = false;
}
else
{
out->result.splitTuple.postfixHasPrefix = true;
out->result.splitTuple.postfixPrefixDatum =
formTextDatum(prefixStr + commonLen + 1,
prefixSize - commonLen - 1);
}
PG_RETURN_VOID();
}
}
else if (inSize > in->level)
{
nodeChar = *(unsigned char *) (inStr + in->level);
}
else
{
nodeChar = -1;
}
/* Look up nodeChar in the node label array */
if (searchChar(in->nodeLabels, in->nNodes, nodeChar, &i))
{
/*
* Descend to existing node. (If in->allTheSame, the core code will
* ignore our nodeN specification here, but that's OK. We still have
* to provide the correct levelAdd and restDatum values, and those are
* the same regardless of which node gets chosen by core.)
*/
int levelAdd;
out->resultType = spgMatchNode;
out->result.matchNode.nodeN = i;
levelAdd = commonLen;
if (nodeChar >= 0)
levelAdd++;
out->result.matchNode.levelAdd = levelAdd;
if (inSize - in->level - levelAdd > 0)
out->result.matchNode.restDatum =
formTextDatum(inStr + in->level + levelAdd,
inSize - in->level - levelAdd);
else
out->result.matchNode.restDatum =
formTextDatum(NULL, 0);
}
else if (in->allTheSame)
{
/*
* Can't use AddNode action, so split the tuple. The upper tuple has
* the same prefix as before and uses a dummy node label -2 for the
* lower tuple. The lower tuple has no prefix and the same node
* labels as the original tuple.
*
* Note: it might seem tempting to shorten the upper tuple's prefix,
* if it has one, then use its last byte as label for the lower tuple.
* But that doesn't win since we know the incoming value matches the
* whole prefix: we'd just end up splitting the lower tuple again.
*/
out->resultType = spgSplitTuple;
out->result.splitTuple.prefixHasPrefix = in->hasPrefix;
out->result.splitTuple.prefixPrefixDatum = in->prefixDatum;
out->result.splitTuple.nodeLabel = Int16GetDatum(-2);
out->result.splitTuple.postfixHasPrefix = false;
}
else
{
/* Add a node for the not-previously-seen nodeChar value */
out->resultType = spgAddNode;
out->result.addNode.nodeLabel = Int16GetDatum(nodeChar);
out->result.addNode.nodeN = i;
}
PG_RETURN_VOID();
}
