QUERY EXECUTION in Font

Print QR Code in Font QUERY EXECUTION

is set to TRUE during the prepare() method prior to executing the query tree. This allows the join method to detect when the first call is made so that the temporary buffers can be created. In this way, the traversal of the tree is preempted until the join operation completes for the first match (or the end of the file if no matches). Notice that the algorithm uses two buffers to store the ordered rows from the incoming tables. These buffers are used to read records for the join operation and are represented using a record pointer for each buffer. If a match is found, both record pointers are set to NULL, which forces the code to read the next record. If the evaluation of the join condition indicates that the join value from the left table is less than the right, the left record pointer is set to NULL so that on the next call to the do_join() method, the next record is read from the left record buffer. Similarly, if the left join value is greater than the right, the right record pointer is set to NULL and on the next call a new record is read from the right record buffer. Now that the basics of the do_join() method have been explained, take a look at the source code. Listing 12-23 shows the code for the do_join() method.

Note I chose to not use a helper function to create the temporary buffers for the first step of the join operation so that I could keep the code together for easier debugging. Thus, the decision was purely for convenience. You can save a bit of code if you want by making this part of the code a helper function.

Listing 12-23. DBXP Join Method /* Perform join operation. SYNOPSIS do_join() query_node *qn IN the operational node in the query tree. READ_RECORD *t -- the tuple to apply the operation to. DESCRIPTION This method performs the relational model operation entitled "join". This operation is the combination of two relations to form a composite view. This algorithm implements ALL variants of the join operation. NOTES Returns 0 (null) if no tuple satisfies child operation (does NOT indicate the end of the file or end of query operation. Use Eof() to verify. RETURN VALUE Success = new tuple with correct attributes Failed = NULL */

READ_RECORD *Query_tree::do_join(query_node *qn) { READ_RECORD *next_tup; int i; TABLE *ltable = NULL; TABLE *rtable = NULL; Field *fright = NULL; Field *fleft = NULL; record_buff *lprev; record_buff *rprev; expr_node *expr; DBUG_ENTER("do_join"); if (qn == NULL) DBUG_RETURN(NULL); /* check join type because some joins require other processing */ switch (qn->join_type) { case (jnINNER) : case (jnLEFTOUTER) : case (jnRIGHTOUTER) : case (jnFULLOUTER) : { /* preempt_pipeline == true means we need to stop the pipeline and sort the incoming rows. We do that by making an in-memory copy of the record buffers stored in left_record_buff and right_record_buff */ if (qn->preempt_pipeline) { left_record_buff = NULL; right_record_buff = NULL; next_tup = NULL; /* Build buffer for tuples from left child. */ do { /* if left child exists, get row from it */ if (qn->left != NULL) lbuff = get_next(qn->left); /* else, read the row from the table (the storage handler */ else {

/* Create space for the record buffer and store pointer in lbuff */ lbuff = (READ_RECORD *) my_malloc(sizeof(READ_RECORD), MYF(MY_ZEROFILL | MY_WME)); lbuff->rec_buf = (byte *) my_malloc(qn->relations[0]->table->s->rec_buff_length, MYF(MY_ZEROFILL | MY_WME)); /* check for end of file. Store result in eof array */ qn->eof[0] = qn->relations[0]->table->file->rnd_next(lbuff->rec_buf); if (qn->eof[0] != HA_ERR_END_OF_FILE) qn->eof[0] = false; else { lbuff = NULL; qn->eof[0] = true; } } /* if the left buffer is not null, get a new row from table */ if (lbuff != NULL) { /* we need the table information for processing fields */ if (qn->left == NULL) ltable = qn->relations[0]->table; else ltable = get_table(qn->left); if (ltable != NULL) memcpy((byte *)ltable->record[0], (byte *)lbuff->rec_buf, ltable->s->rec_buff_length); /* get the join expression */ expr = qn->join_expr->get_expression(0); Field *cur_field = (Field *)expr->left_op; for (Field **field = ltable->field; *field; field++) if (strcasecmp((*field)->field_name, cur_field->field_name)==0) fleft = (*field); /* If field was found, add the row to the in-memory buffer ordered by the join column. */ if ((fleft != NULL) && (!fleft->is_null())) insertion_sort(true, fleft, lbuff); }