GISTENTRY * gbt_num_compress(GISTENTRY *retval, GISTENTRY *entry, const gbtree_ninfo *tinfo) { if (entry->leafkey) { union { int16 i2; int32 i4; int64 i8; float4 f4; float8 f8; DateADT dt; TimeADT tm; Timestamp ts; Cash ch; } v; GBT_NUMKEY *r = (GBT_NUMKEY *) palloc0(2 * tinfo->size); void *leaf = NULL; switch (tinfo->t) { case gbt_t_int2: v.i2 = DatumGetInt16(entry->key); leaf = &v.i2; break; case gbt_t_int4: v.i4 = DatumGetInt32(entry->key); leaf = &v.i4; break; case gbt_t_int8: v.i8 = DatumGetInt64(entry->key); leaf = &v.i8; break; case gbt_t_oid: v.i4 = DatumGetObjectId(entry->key); leaf = &v.i4; break; case gbt_t_float4: v.f4 = DatumGetFloat4(entry->key); leaf = &v.f4; break; case gbt_t_float8: v.f8 = DatumGetFloat8(entry->key); leaf = &v.f8; break; case gbt_t_date: v.dt = DatumGetDateADT(entry->key); leaf = &v.dt; break; case gbt_t_time: v.tm = DatumGetTimeADT(entry->key); leaf = &v.tm; break; case gbt_t_ts: v.ts = DatumGetTimestamp(entry->key); leaf = &v.ts; break; case gbt_t_cash: v.ch = DatumGetCash(entry->key); leaf = &v.ch; break; default: leaf = DatumGetPointer(entry->key); } memcpy((void *) &r[0], leaf, tinfo->size); memcpy((void *) &r[tinfo->size], leaf, tinfo->size); retval = palloc(sizeof(GISTENTRY)); gistentryinit(*retval, PointerGetDatum(r), entry->rel, entry->page, entry->offset, FALSE); } else retval = entry; return retval; }
/* * Add an attribute to the hash calculation. * **IMPORTANT: any new hard coded support for a data type in here * must be added to isGreenplumDbHashable() below! * * Note that the caller should provide the base type if the datum is * of a domain type. It is quite expensive to call get_typtype() and * getBaseType() here since this function gets called a lot for the * same set of Datums. * * @param hashFn called to update the hash value. * @param clientData passed to hashFn. */ void hashDatum(Datum datum, Oid type, datumHashFunction hashFn, void *clientData) { void *buf = NULL; /* pointer to the data */ size_t len = 0; /* length for the data buffer */ int64 intbuf; /* an 8 byte buffer for all integer sizes */ float4 buf_f4; float8 buf_f8; Timestamp tsbuf; /* timestamp data dype is either a double or * int8 (determined in compile time) */ TimestampTz tstzbuf; DateADT datebuf; TimeADT timebuf; TimeTzADT *timetzptr; Interval *intervalptr; AbsoluteTime abstime_buf; RelativeTime reltime_buf; TimeInterval tinterval; AbsoluteTime tinterval_len; Numeric num; bool bool_buf; char char_buf; Name namebuf; ArrayType *arrbuf; inet *inetptr; /* inet/cidr */ unsigned char inet_hkey[sizeof(inet_struct)]; macaddr *macptr; /* MAC address */ VarBit *vbitptr; oidvector *oidvec_buf; Complex *complex_ptr; Complex complex_buf; double complex_real; double complex_imag; Cash cash_buf; pg_uuid_t *uuid_buf; /* * special case buffers */ uint32 nanbuf; uint32 invalidbuf; void *tofree = NULL; if (typeIsEnumType(type)) type = ANYENUMOID; /* * Select the hash to be performed according to the field type we are * adding to the hash. */ switch (type) { /* * ======= NUMERIC TYPES ======== */ case INT2OID: /* -32 thousand to 32 thousand, 2-byte storage */ intbuf = (int64) DatumGetInt16(datum); /* cast to 8 byte before * hashing */ buf = &intbuf; len = sizeof(intbuf); break; case INT4OID: /* -2 billion to 2 billion integer, 4-byte * storage */ intbuf = (int64) DatumGetInt32(datum); /* cast to 8 byte before * hashing */ buf = &intbuf; len = sizeof(intbuf); break; case INT8OID: /* ~18 digit integer, 8-byte storage */ intbuf = DatumGetInt64(datum); /* cast to 8 byte before hashing */ buf = &intbuf; len = sizeof(intbuf); break; case FLOAT4OID: /* single-precision floating point number, * 4-byte storage */ buf_f4 = DatumGetFloat4(datum); /* * On IEEE-float machines, minus zero and zero have different bit * patterns but should compare as equal. We must ensure that they * have the same hash value, which is most easily done this way: */ if (buf_f4 == (float4) 0) buf_f4 = 0.0; buf = &buf_f4; len = sizeof(buf_f4); break; case FLOAT8OID: /* double-precision floating point number, * 8-byte storage */ buf_f8 = DatumGetFloat8(datum); /* * On IEEE-float machines, minus zero and zero have different bit * patterns but should compare as equal. We must ensure that they * have the same hash value, which is most easily done this way: */ if (buf_f8 == (float8) 0) buf_f8 = 0.0; buf = &buf_f8; len = sizeof(buf_f8); break; case NUMERICOID: num = DatumGetNumeric(datum); if (NUMERIC_IS_NAN(num)) { nanbuf = NAN_VAL; buf = &nanbuf; len = sizeof(nanbuf); } else /* not a nan */ { buf = num->n_data; len = (VARSIZE(num) - NUMERIC_HDRSZ); } /* * If we did a pg_detoast_datum, we need to remember to pfree, or * we will leak memory. Because of the 1-byte varlena header * stuff. */ if (num != (Numeric) DatumGetPointer(datum)) tofree = num; break; /* * ====== CHARACTER TYPES ======= */ case CHAROID: /* char(1), single character */ char_buf = DatumGetChar(datum); buf = &char_buf; len = 1; break; case BPCHAROID: /* char(n), blank-padded string, fixed storage */ case TEXTOID: /* text */ case VARCHAROID: /* varchar */ case BYTEAOID: /* bytea */ { int tmplen; varattrib_untoast_ptr_len(datum, (char **) &buf, &tmplen, &tofree); /* adjust length to not include trailing blanks */ if (type != BYTEAOID && tmplen > 1) tmplen = ignoreblanks((char *) buf, tmplen); len = tmplen; break; } case NAMEOID: namebuf = DatumGetName(datum); len = NAMEDATALEN; buf = NameStr(*namebuf); /* adjust length to not include trailing blanks */ if (len > 1) len = ignoreblanks((char *) buf, len); break; /* * ====== OBJECT IDENTIFIER TYPES ====== */ case OIDOID: /* object identifier(oid), maximum 4 billion */ case REGPROCOID: /* function name */ case REGPROCEDUREOID: /* function name with argument types */ case REGOPEROID: /* operator name */ case REGOPERATOROID: /* operator with argument types */ case REGCLASSOID: /* relation name */ case REGTYPEOID: /* data type name */ case ANYENUMOID: /* enum type name */ intbuf = (int64) DatumGetUInt32(datum); /* cast to 8 byte before * hashing */ buf = &intbuf; len = sizeof(intbuf); break; case TIDOID: /* tuple id (6 bytes) */ buf = DatumGetPointer(datum); len = SizeOfIptrData; break; /* * ====== DATE/TIME TYPES ====== */ case TIMESTAMPOID: /* date and time */ tsbuf = DatumGetTimestamp(datum); buf = &tsbuf; len = sizeof(tsbuf); break; case TIMESTAMPTZOID: /* date and time with time zone */ tstzbuf = DatumGetTimestampTz(datum); buf = &tstzbuf; len = sizeof(tstzbuf); break; case DATEOID: /* ANSI SQL date */ datebuf = DatumGetDateADT(datum); buf = &datebuf; len = sizeof(datebuf); break; case TIMEOID: /* hh:mm:ss, ANSI SQL time */ timebuf = DatumGetTimeADT(datum); buf = &timebuf; len = sizeof(timebuf); break; case TIMETZOID: /* time with time zone */ /* * will not compare to TIMEOID on equal values. Postgres never * attempts to compare the two as well. */ timetzptr = DatumGetTimeTzADTP(datum); buf = (unsigned char *) timetzptr; /* * Specify hash length as sizeof(double) + sizeof(int4), not as * sizeof(TimeTzADT), so that any garbage pad bytes in the * structure won't be included in the hash! */ len = sizeof(timetzptr->time) + sizeof(timetzptr->zone); break; case INTERVALOID: /* @ <number> <units>, time interval */ intervalptr = DatumGetIntervalP(datum); buf = (unsigned char *) intervalptr; /* * Specify hash length as sizeof(double) + sizeof(int4), not as * sizeof(Interval), so that any garbage pad bytes in the * structure won't be included in the hash! */ len = sizeof(intervalptr->time) + sizeof(intervalptr->month); break; case ABSTIMEOID: abstime_buf = DatumGetAbsoluteTime(datum); if (abstime_buf == INVALID_ABSTIME) { /* hash to a constant value */ invalidbuf = INVALID_VAL; len = sizeof(invalidbuf); buf = &invalidbuf; } else { len = sizeof(abstime_buf); buf = &abstime_buf; } break; case RELTIMEOID: reltime_buf = DatumGetRelativeTime(datum); if (reltime_buf == INVALID_RELTIME) { /* hash to a constant value */ invalidbuf = INVALID_VAL; len = sizeof(invalidbuf); buf = &invalidbuf; } else { len = sizeof(reltime_buf); buf = &reltime_buf; } break; case TINTERVALOID: tinterval = DatumGetTimeInterval(datum); /* * check if a valid interval. the '0' status code stands for * T_INTERVAL_INVAL which is defined in nabstime.c. We use the * actual value instead of defining it again here. */ if (tinterval->status == 0 || tinterval->data[0] == INVALID_ABSTIME || tinterval->data[1] == INVALID_ABSTIME) { /* hash to a constant value */ invalidbuf = INVALID_VAL; len = sizeof(invalidbuf); buf = &invalidbuf; } else { /* normalize on length of the time interval */ tinterval_len = tinterval->data[1] - tinterval->data[0]; len = sizeof(tinterval_len); buf = &tinterval_len; } break; /* * ======= NETWORK TYPES ======== */ case INETOID: case CIDROID: inetptr = DatumGetInetP(datum); len = inet_getkey(inetptr, inet_hkey, sizeof(inet_hkey)); /* fill-in inet_key & * get len */ buf = inet_hkey; break; case MACADDROID: macptr = DatumGetMacaddrP(datum); len = sizeof(macaddr); buf = (unsigned char *) macptr; break; /* * ======== BIT STRINGS ======== */ case BITOID: case VARBITOID: /* * Note that these are essentially strings. we don't need to worry * about '10' and '010' to compare, b/c they will not, by design. * (see SQL standard, and varbit.c) */ vbitptr = DatumGetVarBitP(datum); len = VARBITBYTES(vbitptr); buf = (char *) VARBITS(vbitptr); break; /* * ======= other types ======= */ case BOOLOID: /* boolean, 'true'/'false' */ bool_buf = DatumGetBool(datum); buf = &bool_buf; len = sizeof(bool_buf); break; /* * ANYARRAY is a pseudo-type. We use it to include any of the * array types (OIDs 1007-1033 in pg_type.h). caller needs to be * sure the type is ANYARRAYOID before calling cdbhash on an array * (INSERT and COPY do so). */ case ANYARRAYOID: arrbuf = DatumGetArrayTypeP(datum); len = VARSIZE(arrbuf) - VARHDRSZ; buf = VARDATA(arrbuf); break; case OIDVECTOROID: oidvec_buf = (oidvector *) DatumGetPointer(datum); len = oidvec_buf->dim1 * sizeof(Oid); buf = oidvec_buf->values; break; case CASHOID: /* cash is stored in int64 internally */ cash_buf = DatumGetCash(datum); len = sizeof(Cash); buf = &cash_buf; break; /* pg_uuid_t is defined as a char array of size UUID_LEN in uuid.c */ case UUIDOID: uuid_buf = DatumGetUUIDP(datum); len = UUID_LEN; buf = (char *) uuid_buf; break; case COMPLEXOID: complex_ptr = DatumGetComplexP(datum); complex_real = re(complex_ptr); complex_imag = im(complex_ptr); /* * On IEEE-float machines, minus zero and zero have different bit * patterns but should compare as equal. We must ensure that they * have the same hash value, which is most easily done this way: */ if (complex_real == (float8) 0) { complex_real = 0.0; } if (complex_imag == (float8) 0) { complex_imag = 0.0; } INIT_COMPLEX(&complex_buf, complex_real, complex_imag); len = sizeof(Complex); buf = (unsigned char *) &complex_buf; break; default: ereport(ERROR, (errcode(ERRCODE_GP_FEATURE_NOT_YET), errmsg("Type %u is not hashable.", type))); } /* switch(type) */ /* do the hash using the selected algorithm */ hashFn(clientData, buf, len); if (tofree) pfree(tofree); }