2011-12-02 18:58:12 +01:00
|
|
|
/*
|
|
|
|
* Copyright (c) 2010, Swedish Institute of Computer Science
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* 3. Neither the name of the Institute nor the names of its contributors
|
|
|
|
* may be used to endorse or promote products derived from this software
|
|
|
|
* without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE INSTITUTE AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE INSTITUTE OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* \file
|
|
|
|
* Logic for relational databases.
|
|
|
|
* \author
|
|
|
|
* Nicolas Tsiftes <nvt@sics.se>
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <limits.h>
|
|
|
|
#include <string.h>
|
|
|
|
|
|
|
|
#include "lib/crc16.h"
|
|
|
|
#include "lib/list.h"
|
|
|
|
#include "lib/memb.h"
|
|
|
|
|
|
|
|
#define DEBUG DEBUG_NONE
|
2013-11-22 09:17:54 +01:00
|
|
|
#include "net/ip/uip-debug.h"
|
2011-12-02 18:58:12 +01:00
|
|
|
|
|
|
|
#include "db-options.h"
|
|
|
|
#include "index.h"
|
|
|
|
#include "lvm.h"
|
|
|
|
#include "relation.h"
|
|
|
|
#include "result.h"
|
|
|
|
#include "storage.h"
|
|
|
|
#include "aql.h"
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The source_dest_map structure is used for mapping the pointers to
|
|
|
|
* data in a source row and in the corresponding destination row. The
|
|
|
|
* structure is calculated just before processing a relational
|
|
|
|
* selection, and then used to improve the performance when processing
|
|
|
|
* each row.
|
|
|
|
*/
|
|
|
|
struct source_dest_map {
|
|
|
|
attribute_t *from_attr;
|
|
|
|
attribute_t *to_attr;
|
|
|
|
unsigned from_offset;
|
|
|
|
unsigned to_offset;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct source_dest_map attr_map[AQL_ATTRIBUTE_LIMIT];
|
|
|
|
|
|
|
|
#if DB_FEATURE_JOIN
|
|
|
|
/*
|
|
|
|
* The source_map structure is used for mapping attributes to
|
|
|
|
* their offsets in rows.
|
|
|
|
*/
|
|
|
|
struct source_map {
|
|
|
|
attribute_t *attr;
|
|
|
|
unsigned char *from_ptr;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct source_map source_map[AQL_ATTRIBUTE_LIMIT];
|
|
|
|
#endif /* DB_FEATURE_JOIN */
|
|
|
|
|
|
|
|
static unsigned char row[DB_MAX_ATTRIBUTES_PER_RELATION * DB_MAX_ELEMENT_SIZE];
|
|
|
|
static unsigned char extra_row[DB_MAX_ATTRIBUTES_PER_RELATION * DB_MAX_ELEMENT_SIZE];
|
|
|
|
static unsigned char result_row[AQL_ATTRIBUTE_LIMIT * DB_MAX_ELEMENT_SIZE];
|
|
|
|
static unsigned char * const left_row = row;
|
|
|
|
static unsigned char * const right_row = extra_row;
|
|
|
|
static unsigned char * const join_row = result_row;
|
|
|
|
|
|
|
|
LIST(relations);
|
|
|
|
MEMB(relations_memb, relation_t, DB_RELATION_POOL_SIZE);
|
|
|
|
MEMB(attributes_memb, attribute_t, DB_ATTRIBUTE_POOL_SIZE);
|
|
|
|
|
|
|
|
static relation_t *relation_find(char *);
|
|
|
|
static attribute_t *attribute_find(relation_t *, char *);
|
|
|
|
static int get_attribute_value_offset(relation_t *, attribute_t *);
|
|
|
|
static void attribute_free(relation_t *, attribute_t *);
|
|
|
|
static void purge_relations(void);
|
|
|
|
static void relation_clear(relation_t *);
|
|
|
|
static relation_t *relation_allocate(void);
|
|
|
|
static void relation_free(relation_t *);
|
|
|
|
|
|
|
|
static relation_t *
|
|
|
|
relation_find(char *name)
|
|
|
|
{
|
|
|
|
relation_t *rel;
|
|
|
|
|
|
|
|
for(rel = list_head(relations); rel != NULL; rel = rel->next) {
|
|
|
|
if(strcmp(rel->name, name) == 0) {
|
|
|
|
return rel;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static attribute_t *
|
|
|
|
attribute_find(relation_t *rel, char *name)
|
|
|
|
{
|
|
|
|
attribute_t *attr;
|
|
|
|
|
|
|
|
for(attr = list_head(rel->attributes); attr != NULL; attr = attr->next) {
|
|
|
|
if(strcmp(attr->name, name) == 0) {
|
|
|
|
return attr;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
get_attribute_value_offset(relation_t *rel, attribute_t *attr)
|
|
|
|
{
|
|
|
|
attribute_t *ptr;
|
|
|
|
int offset;
|
|
|
|
|
|
|
|
for(offset = 0, ptr = list_head(rel->attributes);
|
|
|
|
ptr != NULL;
|
|
|
|
ptr = ptr->next) {
|
|
|
|
if(ptr == attr) {
|
|
|
|
return offset;
|
|
|
|
}
|
|
|
|
offset += ptr->element_size;
|
|
|
|
}
|
|
|
|
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
attribute_free(relation_t *rel, attribute_t *attr)
|
|
|
|
{
|
|
|
|
if(attr->index != NULL) {
|
|
|
|
index_release(attr->index);
|
|
|
|
}
|
|
|
|
memb_free(&attributes_memb, attr);
|
|
|
|
rel->attribute_count--;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
purge_relations(void)
|
|
|
|
{
|
|
|
|
relation_t *rel;
|
|
|
|
relation_t *next;
|
|
|
|
|
|
|
|
for(rel = list_head(relations); rel != NULL;) {
|
|
|
|
next = rel->next;
|
|
|
|
if(rel->references == 0) {
|
|
|
|
relation_free(rel);
|
|
|
|
}
|
|
|
|
rel = next;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
relation_clear(relation_t *rel)
|
|
|
|
{
|
|
|
|
memset(rel, 0, sizeof(*rel));
|
|
|
|
rel->tuple_storage = -1;
|
|
|
|
rel->cardinality = INVALID_TUPLE;
|
|
|
|
rel->dir = DB_STORAGE;
|
|
|
|
LIST_STRUCT_INIT(rel, attributes);
|
|
|
|
}
|
|
|
|
|
|
|
|
static relation_t *
|
|
|
|
relation_allocate(void)
|
|
|
|
{
|
|
|
|
relation_t *rel;
|
|
|
|
|
|
|
|
rel = memb_alloc(&relations_memb);
|
|
|
|
if(rel == NULL) {
|
|
|
|
purge_relations();
|
|
|
|
rel = memb_alloc(&relations_memb);
|
|
|
|
if(rel == NULL) {
|
|
|
|
PRINTF("DB: Failed to allocate a relation\n");
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
relation_clear(rel);
|
|
|
|
return rel;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
relation_free(relation_t *rel)
|
|
|
|
{
|
|
|
|
attribute_t *attr;
|
|
|
|
|
|
|
|
while((attr = list_pop(rel->attributes)) != NULL) {
|
|
|
|
attribute_free(rel, attr);
|
|
|
|
}
|
|
|
|
|
|
|
|
list_remove(relations, rel);
|
|
|
|
memb_free(&relations_memb, rel);
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_init(void)
|
|
|
|
{
|
|
|
|
list_init(relations);
|
|
|
|
memb_init(&relations_memb);
|
|
|
|
memb_init(&attributes_memb);
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
relation_t *
|
|
|
|
relation_load(char *name)
|
|
|
|
{
|
|
|
|
relation_t *rel;
|
|
|
|
|
|
|
|
rel = relation_find(name);
|
|
|
|
if(rel != NULL) {
|
|
|
|
rel->references++;
|
|
|
|
goto end;
|
|
|
|
}
|
|
|
|
|
|
|
|
rel = relation_allocate();
|
|
|
|
if(rel == NULL) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(DB_ERROR(storage_get_relation(rel, name))) {
|
|
|
|
memb_free(&relations_memb, rel);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
memcpy(rel->name, name, sizeof(rel->name));
|
|
|
|
rel->name[sizeof(rel->name) - 1] = '\0';
|
|
|
|
rel->references = 1;
|
|
|
|
list_add(relations, rel);
|
|
|
|
|
|
|
|
end:
|
|
|
|
if(rel->dir == DB_STORAGE && DB_ERROR(storage_load(rel))) {
|
|
|
|
relation_release(rel);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return rel;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_release(relation_t *rel)
|
|
|
|
{
|
|
|
|
if(rel->references > 0) {
|
|
|
|
rel->references--;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(rel->references == 0) {
|
|
|
|
storage_unload(rel);
|
|
|
|
}
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
relation_t *
|
|
|
|
relation_create(char *name, db_direction_t dir)
|
|
|
|
{
|
|
|
|
relation_t old_rel;
|
|
|
|
relation_t *rel;
|
|
|
|
|
|
|
|
if(*name != '\0') {
|
|
|
|
relation_clear(&old_rel);
|
|
|
|
|
|
|
|
if(storage_get_relation(&old_rel, name) == DB_OK) {
|
|
|
|
/* Reject a creation request if the relation already exists. */
|
|
|
|
PRINTF("DB: Attempted to create a relation that already exists (%s)\n",
|
|
|
|
name);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
rel = relation_allocate();
|
|
|
|
if(rel == NULL) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
rel->cardinality = 0;
|
|
|
|
|
|
|
|
strncpy(rel->name, name, sizeof(rel->name) - 1);
|
|
|
|
rel->name[sizeof(rel->name) - 1] = '\0';
|
|
|
|
rel->dir = dir;
|
|
|
|
|
|
|
|
if(dir == DB_STORAGE) {
|
|
|
|
storage_drop_relation(rel, 1);
|
|
|
|
|
|
|
|
if(storage_put_relation(rel) == DB_OK) {
|
|
|
|
list_add(relations, rel);
|
|
|
|
return rel;
|
|
|
|
}
|
|
|
|
memb_free(&relations_memb, rel);
|
|
|
|
} else {
|
|
|
|
list_add(relations, rel);
|
|
|
|
return rel;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
#if DB_FEATURE_REMOVE
|
|
|
|
db_result_t
|
|
|
|
relation_rename(char *old_name, char *new_name)
|
|
|
|
{
|
|
|
|
if(DB_ERROR(relation_remove(new_name, 0)) ||
|
|
|
|
DB_ERROR(storage_rename_relation(old_name, new_name))) {
|
|
|
|
return DB_STORAGE_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
#endif /* DB_FEATURE_REMOVE */
|
|
|
|
|
|
|
|
attribute_t *
|
|
|
|
relation_attribute_add(relation_t *rel, db_direction_t dir, char *name,
|
|
|
|
domain_t domain, size_t element_size)
|
|
|
|
{
|
|
|
|
attribute_t *attribute;
|
|
|
|
tuple_id_t cardinality;
|
|
|
|
|
|
|
|
cardinality = relation_cardinality(rel);
|
|
|
|
if(cardinality != INVALID_TUPLE && cardinality > 0) {
|
|
|
|
PRINTF("DB: Attempt to create an attribute in a non-empty relation\n");
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(element_size == 0 || element_size > DB_MAX_ELEMENT_SIZE) {
|
|
|
|
PRINTF("DB: Unacceptable element size: %u\n", element_size);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
attribute = memb_alloc(&attributes_memb);
|
|
|
|
if(attribute == NULL) {
|
|
|
|
PRINTF("DB: Failed to allocate attribute \"%s\"!\n", name);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
strncpy(attribute->name, name, sizeof(attribute->name) - 1);
|
|
|
|
attribute->name[sizeof(attribute->name) - 1] = '\0';
|
|
|
|
attribute->domain = domain;
|
|
|
|
attribute->element_size = element_size;
|
|
|
|
attribute->aggregator = 0;
|
|
|
|
attribute->index = NULL;
|
|
|
|
attribute->flags = 0 /*ATTRIBUTE_FLAG_UNIQUE*/;
|
|
|
|
|
|
|
|
rel->row_length += element_size;
|
|
|
|
|
|
|
|
list_add(rel->attributes, attribute);
|
|
|
|
rel->attribute_count++;
|
|
|
|
|
|
|
|
if(dir == DB_STORAGE) {
|
|
|
|
if(DB_ERROR(storage_put_attribute(rel, attribute))) {
|
|
|
|
PRINTF("DB: Failed to store attribute %s\n", attribute->name);
|
|
|
|
memb_free(&attributes_memb, attribute);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
index_load(rel, attribute);
|
|
|
|
}
|
|
|
|
|
|
|
|
return attribute;
|
|
|
|
}
|
|
|
|
|
|
|
|
attribute_t *
|
|
|
|
relation_attribute_get(relation_t *rel, char *name)
|
|
|
|
{
|
|
|
|
attribute_t *attr;
|
|
|
|
|
|
|
|
attr = attribute_find(rel, name);
|
|
|
|
if(attr != NULL && !(attr->flags & ATTRIBUTE_FLAG_INVALID)) {
|
|
|
|
return attr;
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_attribute_remove(relation_t *rel, char *name)
|
|
|
|
{
|
|
|
|
/* Not implemented completely. */
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
#if 0
|
|
|
|
attribute_t *attr;
|
|
|
|
|
|
|
|
if(rel->references > 1) {
|
|
|
|
return DB_BUSY_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
attr = relation_attribute_get(rel, name);
|
|
|
|
if(attr == NULL) {
|
|
|
|
return DB_NAME_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_remove(rel->attributes, attr);
|
|
|
|
attribute_free(rel, attr);
|
|
|
|
return DB_OK;
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_get_value(relation_t *rel, attribute_t *attr,
|
|
|
|
unsigned char *row_ptr, attribute_value_t *value)
|
|
|
|
{
|
|
|
|
int offset;
|
|
|
|
unsigned char *from_ptr;
|
|
|
|
|
|
|
|
offset = get_attribute_value_offset(rel, attr);
|
|
|
|
if(offset < 0) {
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
}
|
|
|
|
from_ptr = row_ptr + offset;
|
|
|
|
|
|
|
|
return db_phy_to_value(value, attr, from_ptr);
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_set_primary_key(relation_t *rel, char *name)
|
|
|
|
{
|
|
|
|
attribute_t *attribute;
|
|
|
|
|
|
|
|
attribute = relation_attribute_get(rel, name);
|
|
|
|
if(attribute == NULL) {
|
|
|
|
return DB_NAME_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
attribute->flags = ATTRIBUTE_FLAG_PRIMARY_KEY;
|
|
|
|
PRINTF("DB: New primary key: %s\n", attribute->name);
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_remove(char *name, int remove_tuples)
|
|
|
|
{
|
|
|
|
relation_t *rel;
|
|
|
|
db_result_t result;
|
|
|
|
|
|
|
|
rel = relation_load(name);
|
|
|
|
if(rel == NULL) {
|
|
|
|
/*
|
|
|
|
* Attempt to remove an inexistent relation. To allow for this
|
|
|
|
* operation to be used for setting up repeatable tests and
|
|
|
|
* experiments, we do not signal an error.
|
|
|
|
*/
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(rel->references > 1) {
|
|
|
|
return DB_BUSY_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
result = storage_drop_relation(rel, remove_tuples);
|
|
|
|
relation_free(rel);
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_insert(relation_t *rel, attribute_value_t *values)
|
|
|
|
{
|
|
|
|
attribute_t *attr;
|
2012-01-10 11:38:33 +01:00
|
|
|
unsigned char record[rel->row_length];
|
2011-12-02 18:58:12 +01:00
|
|
|
unsigned char *ptr;
|
|
|
|
attribute_value_t *value;
|
|
|
|
db_result_t result;
|
|
|
|
|
|
|
|
value = values;
|
|
|
|
|
2012-01-10 11:38:33 +01:00
|
|
|
PRINTF("DB: Relation %s has a record size of %u bytes\n",
|
|
|
|
rel->name, (unsigned)rel->row_length);
|
|
|
|
ptr = record;
|
2011-12-02 18:58:12 +01:00
|
|
|
|
|
|
|
PRINTF("DB: Insert (");
|
|
|
|
|
|
|
|
for(attr = list_head(rel->attributes); attr != NULL; attr = attr->next, value++) {
|
|
|
|
/* Verify that the value is in the expected domain. An exception
|
|
|
|
to this rule is that INT may be promoted to LONG. */
|
|
|
|
if(attr->domain != value->domain &&
|
|
|
|
!(attr->domain == DOMAIN_LONG && value->domain == DOMAIN_INT)) {
|
|
|
|
PRINTF("DB: The value domain %d does not match the domain %d of attribute %s\n",
|
|
|
|
value->domain, attr->domain, attr->name);
|
|
|
|
return DB_RELATIONAL_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Set the data area for removed attributes to 0. */
|
|
|
|
if(attr->flags & ATTRIBUTE_FLAG_INVALID) {
|
|
|
|
memset(ptr, 0, attr->element_size);
|
|
|
|
ptr += attr->element_size;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
result = db_value_to_phy((unsigned char *)ptr, attr, value);
|
|
|
|
if(DB_ERROR(result)) {
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
#if DEBUG
|
|
|
|
switch(attr->domain) {
|
|
|
|
case DOMAIN_INT:
|
|
|
|
PRINTF("%s=%d", attr->name, VALUE_INT(value));
|
|
|
|
break;
|
|
|
|
case DOMAIN_LONG:
|
|
|
|
PRINTF("%s=%ld", attr->name, VALUE_LONG(value));
|
|
|
|
break;
|
|
|
|
case DOMAIN_STRING:
|
|
|
|
PRINTF("%s='%s", attr->name, VALUE_STRING(value));
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
PRINTF(")\nDB: Unhandled attribute domain: %d\n", attr->domain);
|
|
|
|
return DB_TYPE_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(attr->next != NULL) {
|
|
|
|
PRINTF(", ");
|
|
|
|
}
|
|
|
|
#endif /* DEBUG */
|
|
|
|
|
|
|
|
ptr += attr->element_size;
|
|
|
|
if(attr->index != NULL) {
|
|
|
|
if(DB_ERROR(index_insert(attr->index, value, rel->next_row))) {
|
|
|
|
return DB_INDEX_ERROR;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
PRINTF(")\n");
|
|
|
|
|
|
|
|
rel->cardinality++;
|
|
|
|
rel->next_row++;
|
|
|
|
return storage_put_row(rel, record);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
aggregate(attribute_t *attr, attribute_value_t *value)
|
|
|
|
{
|
|
|
|
long long_value;
|
|
|
|
|
|
|
|
switch(value->domain) {
|
|
|
|
case DOMAIN_INT:
|
|
|
|
long_value = VALUE_INT(value);
|
|
|
|
break;
|
|
|
|
case DOMAIN_LONG:
|
|
|
|
long_value = VALUE_LONG(value);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch(attr->aggregator) {
|
|
|
|
case AQL_COUNT:
|
|
|
|
attr->aggregation_value++;
|
|
|
|
break;
|
|
|
|
case AQL_SUM:
|
|
|
|
attr->aggregation_value += long_value;
|
|
|
|
break;
|
|
|
|
case AQL_MEAN:
|
|
|
|
break;
|
|
|
|
case AQL_MEDIAN:
|
|
|
|
break;
|
|
|
|
case AQL_MAX:
|
|
|
|
if(long_value > attr->aggregation_value) {
|
|
|
|
attr->aggregation_value = long_value;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case AQL_MIN:
|
|
|
|
if(long_value < attr->aggregation_value) {
|
|
|
|
attr->aggregation_value = long_value;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static db_result_t
|
|
|
|
generate_attribute_map(struct source_dest_map *attr_map, unsigned attribute_count,
|
|
|
|
relation_t *from_rel, relation_t *to_rel,
|
|
|
|
unsigned char *from_row, unsigned char *to_row)
|
|
|
|
{
|
|
|
|
attribute_t *from_attr;
|
|
|
|
attribute_t *to_attr;
|
|
|
|
unsigned size_sum;
|
|
|
|
struct source_dest_map *attr_map_ptr;
|
|
|
|
int offset;
|
|
|
|
|
|
|
|
attr_map_ptr = attr_map;
|
|
|
|
for(size_sum = 0, to_attr = list_head(to_rel->attributes);
|
|
|
|
to_attr != NULL;
|
|
|
|
to_attr = to_attr->next) {
|
|
|
|
from_attr = attribute_find(from_rel, to_attr->name);
|
|
|
|
if(from_attr == NULL) {
|
|
|
|
PRINTF("DB: Invalid attribute in the result relation: %s\n",
|
|
|
|
to_attr->name);
|
|
|
|
return DB_NAME_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
attr_map_ptr->from_attr = from_attr;
|
|
|
|
attr_map_ptr->to_attr = to_attr;
|
|
|
|
offset = get_attribute_value_offset(from_rel, from_attr);
|
|
|
|
if(offset < 0) {
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
}
|
|
|
|
attr_map_ptr->from_offset = offset;
|
|
|
|
attr_map_ptr->to_offset = size_sum;
|
|
|
|
|
|
|
|
size_sum += to_attr->element_size;
|
|
|
|
attr_map_ptr++;
|
|
|
|
}
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
select_index(db_handle_t *handle, lvm_instance_t *lvm_instance)
|
|
|
|
{
|
|
|
|
index_t *index;
|
|
|
|
attribute_t *attr;
|
|
|
|
operand_value_t min;
|
|
|
|
operand_value_t max;
|
|
|
|
attribute_value_t av_min;
|
|
|
|
attribute_value_t av_max;
|
|
|
|
long range;
|
|
|
|
unsigned long min_range;
|
|
|
|
|
|
|
|
index = NULL;
|
|
|
|
min_range = ULONG_MAX;
|
|
|
|
|
|
|
|
/* Find all indexed and derived attributes, and select the index of
|
|
|
|
the attribute with the smallest range. */
|
|
|
|
for(attr = list_head(handle->rel->attributes);
|
|
|
|
attr != NULL;
|
|
|
|
attr = attr->next) {
|
|
|
|
if(attr->index != NULL &&
|
|
|
|
!LVM_ERROR(lvm_get_derived_range(lvm_instance, attr->name, &min, &max))) {
|
|
|
|
range = (unsigned long)max.l - (unsigned long)min.l;
|
|
|
|
PRINTF("DB: The search range for attribute \"%s\" comprises %ld values\n",
|
|
|
|
attr->name, range + 1);
|
|
|
|
|
|
|
|
if(range <= min_range) {
|
|
|
|
index = attr->index;
|
|
|
|
av_min.domain = av_max.domain = DOMAIN_INT;
|
|
|
|
VALUE_LONG(&av_min) = min.l;
|
|
|
|
VALUE_LONG(&av_max) = max.l;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if(index != NULL) {
|
|
|
|
/* We found a suitable index; get an iterator for it. */
|
|
|
|
if(index_get_iterator(&handle->index_iterator, index,
|
|
|
|
&av_min, &av_max) == DB_OK) {
|
|
|
|
handle->flags |= DB_HANDLE_FLAG_SEARCH_INDEX;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static db_result_t
|
|
|
|
generate_selection_result(db_handle_t *handle, relation_t *rel, aql_adt_t *adt)
|
|
|
|
{
|
|
|
|
relation_t *result_rel;
|
|
|
|
unsigned attribute_count;
|
|
|
|
attribute_t *attr;
|
|
|
|
|
|
|
|
result_rel = handle->result_rel;
|
|
|
|
|
|
|
|
handle->current_row = 0;
|
|
|
|
handle->ncolumns = 0;
|
|
|
|
handle->tuple_id = 0;
|
|
|
|
for(attr = list_head(result_rel->attributes); attr != NULL; attr = attr->next) {
|
|
|
|
if(attr->flags & ATTRIBUTE_FLAG_NO_STORE) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
handle->ncolumns++;
|
|
|
|
}
|
|
|
|
handle->tuple = (tuple_t)result_row;
|
|
|
|
|
|
|
|
attribute_count = result_rel->attribute_count;
|
|
|
|
if(DB_ERROR(generate_attribute_map(attr_map, attribute_count, rel, result_rel, row, result_row))) {
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(adt->lvm_instance != NULL) {
|
|
|
|
/* Try to establish acceptable ranges for the attribute values. */
|
|
|
|
if(!LVM_ERROR(lvm_derive(adt->lvm_instance))) {
|
|
|
|
select_index(handle, adt->lvm_instance);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->flags |= DB_HANDLE_FLAG_PROCESSING;
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
#if DB_FEATURE_REMOVE
|
|
|
|
db_result_t
|
|
|
|
relation_process_remove(void *handle_ptr)
|
|
|
|
{
|
|
|
|
db_handle_t *handle;
|
|
|
|
aql_adt_t *adt;
|
|
|
|
db_result_t result;
|
|
|
|
|
|
|
|
handle = (db_handle_t *)handle_ptr;
|
|
|
|
adt = handle->adt;
|
|
|
|
|
|
|
|
result = relation_process_select(handle_ptr);
|
|
|
|
if(result == DB_FINISHED) {
|
|
|
|
PRINTF("DB: Finished removing tuples. Overwriting relation %s with the result\n",
|
|
|
|
adt->relations[1]);
|
|
|
|
relation_release(handle->rel);
|
|
|
|
relation_rename(adt->relations[0], adt->relations[1]);
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_process_select(void *handle_ptr)
|
|
|
|
{
|
|
|
|
db_handle_t *handle;
|
|
|
|
aql_adt_t *adt;
|
|
|
|
db_result_t result;
|
|
|
|
unsigned attribute_count;
|
|
|
|
struct source_dest_map *attr_map_ptr, *attr_map_end;
|
|
|
|
attribute_t *result_attr;
|
|
|
|
unsigned char *from_ptr;
|
|
|
|
unsigned char *to_ptr;
|
|
|
|
operand_value_t operand_value;
|
|
|
|
uint8_t intbuf[2];
|
|
|
|
attribute_value_t value;
|
|
|
|
lvm_status_t wanted_result;
|
|
|
|
|
|
|
|
handle = (db_handle_t *)handle_ptr;
|
|
|
|
adt = (aql_adt_t *)handle->adt;
|
|
|
|
|
|
|
|
attribute_count = handle->result_rel->attribute_count;
|
|
|
|
attr_map_end = attr_map + attribute_count;
|
|
|
|
|
|
|
|
if(handle->flags & DB_HANDLE_FLAG_SEARCH_INDEX) {
|
|
|
|
handle->tuple_id = index_get_next(&handle->index_iterator);
|
|
|
|
if(handle->tuple_id == INVALID_TUPLE) {
|
|
|
|
PRINTF("DB: An attribute value could not be found in the index\n");
|
|
|
|
if(handle->index_iterator.next_item_no == 0) {
|
|
|
|
return DB_INDEX_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(adt->flags & AQL_FLAG_AGGREGATE) {
|
|
|
|
goto end_aggregation;
|
|
|
|
}
|
|
|
|
|
|
|
|
return DB_FINISHED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Put the tuples fulfilling the given condition into a new relation.
|
|
|
|
The tuples may be projected. */
|
|
|
|
result = storage_get_row(handle->rel, &handle->tuple_id, row);
|
|
|
|
handle->tuple_id++;
|
|
|
|
if(DB_ERROR(result)) {
|
|
|
|
PRINTF("DB: Failed to get a row in relation %s!\n", handle->rel->name);
|
|
|
|
return result;
|
|
|
|
} else if(result == DB_FINISHED) {
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_AGGREGATE) {
|
|
|
|
goto end_aggregation;
|
|
|
|
}
|
|
|
|
return DB_FINISHED;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Process the attributes in the result relation. */
|
|
|
|
for(attr_map_ptr = attr_map; attr_map_ptr < attr_map_end; attr_map_ptr++) {
|
|
|
|
from_ptr = row + attr_map_ptr->from_offset;
|
|
|
|
result_attr = attr_map_ptr->to_attr;
|
|
|
|
|
|
|
|
/* Update the internal state of the PLE. */
|
|
|
|
if(result_attr->domain == DOMAIN_INT) {
|
|
|
|
operand_value.l = from_ptr[0] << 8 | from_ptr[1];
|
|
|
|
lvm_set_variable_value(result_attr->name, operand_value);
|
|
|
|
} else if(result_attr->domain == DOMAIN_LONG) {
|
|
|
|
operand_value.l = (uint32_t)from_ptr[0] << 24 |
|
|
|
|
(uint32_t)from_ptr[1] << 16 |
|
|
|
|
(uint32_t)from_ptr[2] << 8 |
|
|
|
|
from_ptr[3];
|
|
|
|
lvm_set_variable_value(result_attr->name, operand_value);
|
|
|
|
}
|
|
|
|
|
|
|
|
if(result_attr->flags & ATTRIBUTE_FLAG_NO_STORE) {
|
|
|
|
/* The attribute is used just for the predicate,
|
|
|
|
so do not copy the current value into the result. */
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(!(AQL_GET_FLAGS(adt) & AQL_FLAG_AGGREGATE)) {
|
|
|
|
/* No aggregators. Copy the original value into the resulting tuple. */
|
|
|
|
memcpy(result_row + attr_map_ptr->to_offset, from_ptr,
|
|
|
|
result_attr->element_size);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
wanted_result = TRUE;
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_INVERSE_LOGIC) {
|
|
|
|
wanted_result = FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Check whether the given predicate is true for this tuple. */
|
|
|
|
if(adt->lvm_instance == NULL ||
|
|
|
|
lvm_execute(adt->lvm_instance) == wanted_result) {
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_AGGREGATE) {
|
|
|
|
for(attr_map_ptr = attr_map; attr_map_ptr < attr_map_end; attr_map_ptr++) {
|
|
|
|
from_ptr = row + attr_map_ptr->from_offset;
|
|
|
|
result = db_phy_to_value(&value, attr_map_ptr->to_attr, from_ptr);
|
|
|
|
if(DB_ERROR(result)) {
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
aggregate(attr_map_ptr->to_attr, &value);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_ASSIGN) {
|
|
|
|
if(DB_ERROR(storage_put_row(handle->result_rel, result_row))) {
|
|
|
|
PRINTF("DB: Failed to store a row in the result relation!\n");
|
|
|
|
return DB_STORAGE_ERROR;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
handle->current_row++;
|
|
|
|
return DB_GOT_ROW;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
|
|
|
|
end_aggregation:
|
|
|
|
/* Generate aggregated result if requested. */
|
|
|
|
for(attr_map_ptr = attr_map; attr_map_ptr < attr_map_end; attr_map_ptr++) {
|
|
|
|
result_attr = attr_map_ptr->to_attr;
|
|
|
|
to_ptr = result_row + attr_map_ptr->to_offset;
|
|
|
|
|
|
|
|
intbuf[0] = result_attr->aggregation_value >> 8;
|
|
|
|
intbuf[1] = result_attr->aggregation_value & 0xff;
|
|
|
|
from_ptr = intbuf;
|
|
|
|
memcpy(to_ptr, from_ptr, result_attr->element_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_ASSIGN) {
|
|
|
|
if(DB_ERROR(storage_put_row(handle->result_rel, result_row))) {
|
|
|
|
PRINTF("DB: Failed to store a row in the result relation!\n");
|
|
|
|
return DB_STORAGE_ERROR;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->current_row = 1;
|
|
|
|
AQL_GET_FLAGS(adt) &= ~AQL_FLAG_AGGREGATE; /* Stop the aggregation. */
|
|
|
|
|
|
|
|
return DB_GOT_ROW;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_select(void *handle_ptr, relation_t *rel, void *adt_ptr)
|
|
|
|
{
|
|
|
|
aql_adt_t *adt;
|
|
|
|
db_handle_t *handle;
|
|
|
|
char *name;
|
|
|
|
db_direction_t dir;
|
|
|
|
char *attribute_name;
|
|
|
|
attribute_t *attr;
|
|
|
|
int i;
|
|
|
|
int normal_attributes;
|
|
|
|
|
|
|
|
adt = (aql_adt_t *)adt_ptr;
|
|
|
|
|
|
|
|
handle = (db_handle_t *)handle_ptr;
|
|
|
|
handle->rel = rel;
|
|
|
|
handle->adt = adt;
|
|
|
|
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_ASSIGN) {
|
|
|
|
name = adt->relations[0];
|
|
|
|
dir = DB_STORAGE;
|
|
|
|
} else {
|
|
|
|
name = RESULT_RELATION;
|
|
|
|
dir = DB_MEMORY;
|
|
|
|
}
|
|
|
|
relation_remove(name, 1);
|
|
|
|
relation_create(name, dir);
|
|
|
|
handle->result_rel = relation_load(name);
|
|
|
|
|
|
|
|
if(handle->result_rel == NULL) {
|
|
|
|
PRINTF("DB: Failed to load a relation for the query result\n");
|
|
|
|
return DB_ALLOCATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
for(i = normal_attributes = 0; i < AQL_ATTRIBUTE_COUNT(adt); i++) {
|
|
|
|
attribute_name = adt->attributes[i].name;
|
|
|
|
|
|
|
|
attr = relation_attribute_get(rel, attribute_name);
|
|
|
|
if(attr == NULL) {
|
|
|
|
PRINTF("DB: Select for invalid attribute %s in relation %s!\n",
|
|
|
|
attribute_name, rel->name);
|
|
|
|
return DB_NAME_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
PRINTF("DB: Found attribute %s in relation %s\n",
|
|
|
|
attribute_name, rel->name);
|
|
|
|
|
|
|
|
attr = relation_attribute_add(handle->result_rel, dir,
|
|
|
|
attribute_name,
|
|
|
|
adt->aggregators[i] ? DOMAIN_INT : attr->domain,
|
|
|
|
attr->element_size);
|
|
|
|
if(attr == NULL) {
|
|
|
|
PRINTF("DB: Failed to add a result attribute\n");
|
|
|
|
relation_release(handle->result_rel);
|
|
|
|
return DB_ALLOCATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
attr->aggregator = adt->aggregators[i];
|
|
|
|
switch(attr->aggregator) {
|
|
|
|
case AQL_NONE:
|
|
|
|
if(!(adt->attributes[i].flags & ATTRIBUTE_FLAG_NO_STORE)) {
|
|
|
|
/* Only count attributes projected into the result set. */
|
|
|
|
normal_attributes++;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case AQL_MAX:
|
|
|
|
attr->aggregation_value = LONG_MIN;
|
|
|
|
break;
|
|
|
|
case AQL_MIN:
|
|
|
|
attr->aggregation_value = LONG_MAX;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
attr->aggregation_value = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
attr->flags = adt->attributes[i].flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Preclude mixes of normal attributes and aggregated ones in
|
|
|
|
selection results. */
|
|
|
|
if(normal_attributes > 0 &&
|
|
|
|
handle->result_rel->attribute_count > normal_attributes) {
|
|
|
|
return DB_RELATIONAL_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
return generate_selection_result(handle, rel, adt);
|
|
|
|
}
|
|
|
|
|
|
|
|
#if DB_FEATURE_JOIN
|
|
|
|
db_result_t
|
|
|
|
relation_process_join(void *handle_ptr)
|
|
|
|
{
|
|
|
|
db_handle_t *handle;
|
|
|
|
db_result_t result;
|
|
|
|
relation_t *left_rel;
|
|
|
|
relation_t *right_rel;
|
|
|
|
relation_t *join_rel;
|
|
|
|
unsigned char *join_next_attribute_ptr;
|
|
|
|
size_t element_size;
|
|
|
|
tuple_id_t right_tuple_id;
|
|
|
|
attribute_value_t value;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
handle = (db_handle_t *)handle_ptr;
|
|
|
|
left_rel = handle->left_rel;
|
|
|
|
right_rel = handle->right_rel;
|
|
|
|
join_rel = handle->join_rel;
|
|
|
|
|
|
|
|
if(!(handle->flags & DB_HANDLE_FLAG_INDEX_STEP)) {
|
|
|
|
goto inner_loop;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Equi-join for indexed attributes only. In the outer loop, we iterate over
|
|
|
|
each tuple in the left relation. */
|
|
|
|
for(handle->tuple_id = 0;; handle->tuple_id++) {
|
|
|
|
result = storage_get_row(left_rel, &handle->tuple_id, left_row);
|
|
|
|
if(DB_ERROR(result)) {
|
|
|
|
PRINTF("DB: Failed to get a row in left relation %s!\n", left_rel->name);
|
|
|
|
return result;
|
|
|
|
} else if(result == DB_FINISHED) {
|
|
|
|
return DB_FINISHED;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(DB_ERROR(relation_get_value(left_rel, handle->left_join_attr, left_row, &value))) {
|
|
|
|
PRINTF("DB: Failed to get a value of the attribute \"%s\" to join on\n",
|
|
|
|
handle->left_join_attr->name);
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(DB_ERROR(index_get_iterator(&handle->index_iterator,
|
|
|
|
handle->right_join_attr->index,
|
|
|
|
&value, &value))) {
|
|
|
|
PRINTF("DB: Failed to get an index iterator\n");
|
|
|
|
return DB_INDEX_ERROR;
|
|
|
|
}
|
|
|
|
handle->flags &= ~DB_HANDLE_FLAG_INDEX_STEP;
|
|
|
|
|
|
|
|
/* In the inner loop, we iterate over all rows with a matching value for the
|
|
|
|
join attribute. The index component provides an iterator for this purpose. */
|
|
|
|
inner_loop:
|
|
|
|
for(;;) {
|
|
|
|
/* Get all rows matching the attribute value in the right relation. */
|
|
|
|
right_tuple_id = index_get_next(&handle->index_iterator);
|
|
|
|
if(right_tuple_id == INVALID_TUPLE) {
|
|
|
|
/* Exclude this row from the left relation in the result,
|
|
|
|
and step to the next value in the index iteration. */
|
|
|
|
handle->flags |= DB_HANDLE_FLAG_INDEX_STEP;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
result = storage_get_row(right_rel, &right_tuple_id, right_row);
|
|
|
|
if(DB_ERROR(result)) {
|
|
|
|
PRINTF("DB: Failed to get a row in right relation %s!\n", right_rel->name);
|
|
|
|
return result;
|
|
|
|
} else if(result == DB_FINISHED) {
|
|
|
|
PRINTF("DB: The index refers to an invalid row: %lu\n",
|
|
|
|
(unsigned long)right_tuple_id);
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Use the source attribute map to fill in the physical representation
|
|
|
|
of the resulting tuple. */
|
|
|
|
join_next_attribute_ptr = join_row;
|
|
|
|
|
|
|
|
for(i = 0; i < join_rel->attribute_count; i++) {
|
|
|
|
element_size = source_map[i].attr->element_size;
|
|
|
|
|
|
|
|
memcpy(join_next_attribute_ptr, source_map[i].from_ptr, element_size);
|
|
|
|
join_next_attribute_ptr += element_size;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(((aql_adt_t *)handle->adt)->flags & AQL_FLAG_ASSIGN) {
|
|
|
|
if(DB_ERROR(storage_put_row(join_rel, join_row))) {
|
|
|
|
return DB_STORAGE_ERROR;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->current_row++;
|
|
|
|
return DB_GOT_ROW;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
static db_result_t
|
|
|
|
generate_join_result(db_handle_t *handle)
|
|
|
|
{
|
|
|
|
relation_t *left_rel;
|
|
|
|
relation_t *right_rel;
|
|
|
|
relation_t *join_rel;
|
|
|
|
attribute_t *attr;
|
|
|
|
attribute_t *result_attr;
|
|
|
|
struct source_map *source_pair;
|
|
|
|
int i;
|
|
|
|
int offset;
|
|
|
|
unsigned char *from_ptr;
|
|
|
|
|
|
|
|
handle->tuple = (tuple_t)join_row;
|
|
|
|
handle->tuple_id = 0;
|
|
|
|
|
|
|
|
left_rel = handle->left_rel;
|
|
|
|
right_rel = handle->right_rel;
|
|
|
|
join_rel = handle->join_rel;
|
|
|
|
|
|
|
|
/* Generate a map over the source attributes for each
|
|
|
|
attribute in the join relation. */
|
|
|
|
for(i = 0, result_attr = list_head(join_rel->attributes);
|
|
|
|
result_attr != NULL;
|
|
|
|
result_attr = result_attr->next, i++) {
|
|
|
|
source_pair = &source_map[i];
|
|
|
|
attr = attribute_find(left_rel, result_attr->name);
|
|
|
|
if(attr != NULL) {
|
|
|
|
offset = get_attribute_value_offset(left_rel, attr);
|
|
|
|
from_ptr = left_row + offset;
|
|
|
|
} else if((attr = attribute_find(right_rel, result_attr->name)) != NULL) {
|
|
|
|
offset = get_attribute_value_offset(right_rel, attr);
|
|
|
|
from_ptr = right_row + offset;
|
|
|
|
} else {
|
|
|
|
PRINTF("DB: The attribute %s could not be found\n", result_attr->name);
|
|
|
|
return DB_NAME_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(offset < 0) {
|
|
|
|
PRINTF("DB: Unable to retrieve attribute values for the JOIN result\n");
|
|
|
|
return DB_IMPLEMENTATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
source_pair->attr = attr;
|
|
|
|
source_pair->from_ptr = from_ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->flags |= DB_HANDLE_FLAG_PROCESSING;
|
|
|
|
|
|
|
|
return DB_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_result_t
|
|
|
|
relation_join(void *query_result, void *adt_ptr)
|
|
|
|
{
|
|
|
|
aql_adt_t *adt;
|
|
|
|
db_handle_t *handle;
|
|
|
|
relation_t *left_rel;
|
|
|
|
relation_t *right_rel;
|
|
|
|
relation_t *join_rel;
|
|
|
|
char *name;
|
|
|
|
db_direction_t dir;
|
|
|
|
int i;
|
|
|
|
char *attribute_name;
|
|
|
|
attribute_t *attr;
|
|
|
|
|
|
|
|
adt = (aql_adt_t *)adt_ptr;
|
|
|
|
|
|
|
|
handle = (db_handle_t *)query_result;
|
|
|
|
handle->current_row = 0;
|
|
|
|
handle->ncolumns = 0;
|
|
|
|
handle->adt = adt;
|
|
|
|
handle->flags = DB_HANDLE_FLAG_INDEX_STEP;
|
|
|
|
|
|
|
|
if(AQL_GET_FLAGS(adt) & AQL_FLAG_ASSIGN) {
|
|
|
|
name = adt->relations[0];
|
|
|
|
dir = DB_STORAGE;
|
|
|
|
} else {
|
|
|
|
name = RESULT_RELATION;
|
|
|
|
dir = DB_MEMORY;
|
|
|
|
}
|
|
|
|
relation_remove(name, 1);
|
|
|
|
relation_create(name, dir);
|
|
|
|
join_rel = relation_load(name);
|
|
|
|
handle->result_rel = join_rel;
|
|
|
|
|
|
|
|
if(join_rel == NULL) {
|
|
|
|
PRINTF("DB: Failed to create a join relation!\n");
|
|
|
|
return DB_ALLOCATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->join_rel = handle->result_rel = join_rel;
|
|
|
|
left_rel = handle->left_rel;
|
|
|
|
right_rel = handle->right_rel;
|
|
|
|
|
|
|
|
handle->left_join_attr = relation_attribute_get(left_rel, adt->attributes[0].name);
|
|
|
|
handle->right_join_attr = relation_attribute_get(right_rel, adt->attributes[0].name);
|
|
|
|
if(handle->left_join_attr == NULL || handle->right_join_attr == NULL) {
|
|
|
|
PRINTF("DB: The attribute (\"%s\") to join on does not exist in both relations\n",
|
|
|
|
adt->attributes[0].name);
|
|
|
|
return DB_RELATIONAL_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(!index_exists(handle->right_join_attr)) {
|
|
|
|
PRINTF("DB: The attribute to join on is not indexed\n");
|
|
|
|
return DB_INDEX_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Define the resulting relation. We start from 1 when counting attributes
|
|
|
|
* because the first attribute is only the one to join, and is not included
|
|
|
|
* by default in the projected attributes.
|
|
|
|
*/
|
|
|
|
for(i = 1; i < AQL_ATTRIBUTE_COUNT(adt); i++) {
|
|
|
|
attribute_name = adt->attributes[i].name;
|
|
|
|
attr = relation_attribute_get(left_rel, attribute_name);
|
|
|
|
if(attr == NULL) {
|
|
|
|
attr = relation_attribute_get(right_rel, attribute_name);
|
|
|
|
if(attr == NULL) {
|
|
|
|
PRINTF("DB: The projection attribute \"%s\" does not exist in any of the relations to join\n",
|
|
|
|
attribute_name);
|
|
|
|
return DB_RELATIONAL_ERROR;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if(relation_attribute_add(join_rel, dir, attr->name, attr->domain,
|
|
|
|
attr->element_size) == NULL) {
|
|
|
|
PRINTF("DB: Failed to add an attribute to the join relation\n");
|
|
|
|
return DB_ALLOCATION_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->ncolumns++;
|
|
|
|
}
|
|
|
|
|
|
|
|
return generate_join_result(handle);
|
|
|
|
}
|
|
|
|
#endif /* DB_FEATURE_JOIN */
|
|
|
|
|
|
|
|
tuple_id_t
|
|
|
|
relation_cardinality(relation_t *rel)
|
|
|
|
{
|
|
|
|
tuple_id_t tuple_id;
|
|
|
|
|
|
|
|
|
|
|
|
if(rel->cardinality != INVALID_TUPLE) {
|
|
|
|
return rel->cardinality;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(!RELATION_HAS_TUPLES(rel)) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(DB_ERROR(storage_get_row_amount(rel, &tuple_id))) {
|
|
|
|
return INVALID_TUPLE;
|
|
|
|
}
|
|
|
|
|
|
|
|
rel->cardinality = tuple_id;
|
|
|
|
|
|
|
|
PRINTF("DB: Relation %s has cardinality %lu\n", rel->name,
|
|
|
|
(unsigned long)tuple_id);
|
|
|
|
|
|
|
|
return tuple_id;
|
|
|
|
}
|