row0sel.c

这是linux下运行的mysql软件包,可用于linux 下安装 php + mysql + apach 的网络配置

/*******************************************************
Select

(c) 1997 Innobase Oy

Created 12/19/1997 Heikki Tuuri
*******************************************************/

#include "row0sel.h"

#ifdef UNIV_NONINL
#include "row0sel.ic"
#endif

#include "dict0dict.h"
#include "dict0boot.h"
#include "trx0undo.h"
#include "trx0trx.h"
#include "btr0btr.h"
#include "btr0cur.h"
#include "btr0sea.h"
#include "mach0data.h"
#include "que0que.h"
#include "row0upd.h"
#include "row0row.h"
#include "row0vers.h"
#include "rem0cmp.h"
#include "lock0lock.h"
#include "eval0eval.h"
#include "pars0sym.h"
#include "pars0pars.h"
#include "row0mysql.h"
#include "read0read.h"
#include "buf0lru.h"

/* Maximum number of rows to prefetch; MySQL interface has another parameter */
#define SEL_MAX_N_PREFETCH	16

/* Number of rows fetched, after which to start prefetching; MySQL interface
has another parameter */
#define SEL_PREFETCH_LIMIT	1

/* When a select has accessed about this many pages, it returns control back
to que_run_threads: this is to allow canceling runaway queries */

#define SEL_COST_LIMIT	100

/* Flags for search shortcut */
#define SEL_FOUND	0
#define	SEL_EXHAUSTED	1
#define SEL_RETRY	2

/************************************************************************
Returns TRUE if the user-defined column values in a secondary index record
are alphabetically the same as the corresponding columns in the clustered
index record.
NOTE: the comparison is NOT done as a binary comparison, but character
fields are compared with collation! */
static
ibool
row_sel_sec_rec_is_for_clust_rec(
/*=============================*/
					/* out: TRUE if the secondary
					record is equal to the corresponding
					fields in the clustered record,
					when compared with collation */
	rec_t*		sec_rec,	/* in: secondary index record */
	dict_index_t*	sec_index,	/* in: secondary index */
	rec_t*		clust_rec,	/* in: clustered index record */
	dict_index_t*	clust_index)	/* in: clustered index */
{
	dict_field_t*	ifield;
        dict_col_t*     col;
        byte*           sec_field;
        ulint           sec_len;
        byte*           clust_field;
        ulint           clust_len;
        ulint           n;
        ulint           i;
	dtype_t*	cur_type;
	mem_heap_t*	heap		= NULL;
	ulint		clust_offsets_[REC_OFFS_NORMAL_SIZE];
	ulint		sec_offsets_[REC_OFFS_SMALL_SIZE];
	ulint*		clust_offs	= clust_offsets_;
	ulint*		sec_offs	= sec_offsets_;
	ibool		is_equal	= TRUE;

	*clust_offsets_ = (sizeof clust_offsets_) / sizeof *clust_offsets_;
	*sec_offsets_ = (sizeof sec_offsets_) / sizeof *sec_offsets_;

	clust_offs = rec_get_offsets(clust_rec, clust_index, clust_offs,
						ULINT_UNDEFINED, &heap);
	sec_offs = rec_get_offsets(sec_rec, sec_index, sec_offs,
						ULINT_UNDEFINED, &heap);

        n = dict_index_get_n_ordering_defined_by_user(sec_index);

        for (i = 0; i < n; i++) {
		ifield = dict_index_get_nth_field(sec_index, i);
                col = dict_field_get_col(ifield);
                
		clust_field = rec_get_nth_field(clust_rec, clust_offs,
                                                dict_col_get_clust_pos(col),
                                                &clust_len);
		sec_field = rec_get_nth_field(sec_rec, sec_offs, i, &sec_len);

		if (ifield->prefix_len > 0
		    && clust_len != UNIV_SQL_NULL) {

			cur_type = dict_col_get_type(
				dict_field_get_col(ifield));

			clust_len = dtype_get_at_most_n_mbchars(
				cur_type,
				ifield->prefix_len,
				clust_len, (char*) clust_field);
		}

                if (0 != cmp_data_data(dict_col_get_type(col),
                                        clust_field, clust_len,
                                        sec_field, sec_len)) {
			is_equal = FALSE;
			goto func_exit;
                }
        }

func_exit:
	if (UNIV_LIKELY_NULL(heap)) {
		mem_heap_free(heap);
	}
	return(is_equal);
}

/*************************************************************************
Creates a select node struct. */

sel_node_t*
sel_node_create(
/*============*/
				/* out, own: select node struct */
	mem_heap_t*	heap)	/* in: memory heap where created */
{
	sel_node_t*	node;

	node = mem_heap_alloc(heap, sizeof(sel_node_t));
	node->common.type = QUE_NODE_SELECT;
	node->state = SEL_NODE_OPEN;

	node->select_will_do_update = FALSE;
	node->latch_mode = BTR_SEARCH_LEAF;

	node->plans = NULL;
	
	return(node);
}

/*************************************************************************
Frees the memory private to a select node when a query graph is freed,
does not free the heap where the node was originally created. */

void
sel_node_free_private(
/*==================*/
	sel_node_t*	node)	/* in: select node struct */
{
	ulint	i;
	plan_t*	plan;

	if (node->plans != NULL) {
		for (i = 0; i < node->n_tables; i++) {
			plan = sel_node_get_nth_plan(node, i);

			btr_pcur_close(&(plan->pcur));
			btr_pcur_close(&(plan->clust_pcur));

			if (plan->old_vers_heap) {
				mem_heap_free(plan->old_vers_heap);
			}
		}
	}
}

/*************************************************************************
Evaluates the values in a select list. If there are aggregate functions,
their argument value is added to the aggregate total. */
UNIV_INLINE
void
sel_eval_select_list(
/*=================*/
	sel_node_t*	node)	/* in: select node */
{
	que_node_t*	exp;

	exp = node->select_list;

	while (exp) {
		eval_exp(exp);

		exp = que_node_get_next(exp);
	}
}

/*************************************************************************
Assigns the values in the select list to the possible into-variables in
SELECT ... INTO ... */
UNIV_INLINE
void
sel_assign_into_var_values(
/*=======================*/
	sym_node_t*	var,	/* in: first variable in a list of variables */
	sel_node_t*	node)	/* in: select node */
{
	que_node_t*	exp;

	if (var == NULL) {

		return;
	}

	exp = node->select_list;

	while (var) {
		ut_ad(exp);

		eval_node_copy_val(var->alias, exp);

		exp = que_node_get_next(exp);
		var = que_node_get_next(var);
	}
}

/*************************************************************************
Resets the aggregate value totals in the select list of an aggregate type
query. */
UNIV_INLINE
void
sel_reset_aggregate_vals(
/*=====================*/
	sel_node_t*	node)	/* in: select node */
{
	func_node_t*	func_node;

	ut_ad(node->is_aggregate);

	func_node = node->select_list;

	while (func_node) {
		eval_node_set_int_val(func_node, 0);

		func_node = que_node_get_next(func_node);
	}	

	node->aggregate_already_fetched = FALSE;
}

/*************************************************************************
Copies the input variable values when an explicit cursor is opened. */
UNIV_INLINE
void
row_sel_copy_input_variable_vals(
/*=============================*/
	sel_node_t*	node)	/* in: select node */
{
	sym_node_t*	var;

	var = UT_LIST_GET_FIRST(node->copy_variables);

	while (var) {
		eval_node_copy_val(var, var->alias);

		var->indirection = NULL;

		var = UT_LIST_GET_NEXT(col_var_list, var);
	}
}

/*************************************************************************
Fetches the column values from a record. */
static
void
row_sel_fetch_columns(
/*==================*/
	dict_index_t*	index,	/* in: record index */
	rec_t*		rec,	/* in: record in a clustered or non-clustered
				index */
	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
	sym_node_t*	column)	/* in: first column in a column list, or
				NULL */
{
	dfield_t*	val;
	ulint		index_type;
	ulint		field_no;
	byte*		data;
	ulint		len;
	
	ut_ad(rec_offs_validate(rec, index, offsets));

	if (index->type & DICT_CLUSTERED) {
		index_type = SYM_CLUST_FIELD_NO;
	} else {
		index_type = SYM_SEC_FIELD_NO;
	}

	while (column) {
		field_no = column->field_nos[index_type];

		if (field_no != ULINT_UNDEFINED) {
	
			data = rec_get_nth_field(rec, offsets, field_no, &len);
			
			if (column->copy_val) {
				eval_node_copy_and_alloc_val(column, data,
									len);
			} else {
				val = que_node_get_val(column);
				dfield_set_data(val, data, len);
			}
		}

		column = UT_LIST_GET_NEXT(col_var_list, column);
	}
}

/*************************************************************************
Allocates a prefetch buffer for a column when prefetch is first time done. */
static
void
sel_col_prefetch_buf_alloc(
/*=======================*/
	sym_node_t*	column)	/* in: symbol table node for a column */
{
	sel_buf_t*	sel_buf;
	ulint		i;

	ut_ad(que_node_get_type(column) == QUE_NODE_SYMBOL);
	
	column->prefetch_buf = mem_alloc(SEL_MAX_N_PREFETCH
							* sizeof(sel_buf_t));
	for (i = 0; i < SEL_MAX_N_PREFETCH; i++) {
		sel_buf = column->prefetch_buf + i;

		sel_buf->data = NULL;

		sel_buf->val_buf_size = 0;
	}
}

/*************************************************************************
Frees a prefetch buffer for a column, including the dynamically allocated
memory for data stored there. */

void
sel_col_prefetch_buf_free(
/*======================*/
	sel_buf_t*	prefetch_buf)	/* in, own: prefetch buffer */
{
	sel_buf_t*	sel_buf;
	ulint		i;

	for (i = 0; i < SEL_MAX_N_PREFETCH; i++) {
		sel_buf = prefetch_buf + i;

		if (sel_buf->val_buf_size > 0) {

			mem_free(sel_buf->data);
		}
	}
}

/*************************************************************************
Pops the column values for a prefetched, cached row from the column prefetch
buffers and places them to the val fields in the column nodes. */
static
void
sel_pop_prefetched_row(
/*===================*/
	plan_t*	plan)	/* in: plan node for a table */
{
	sym_node_t*	column;
	sel_buf_t*	sel_buf;
	dfield_t*	val;
	byte*		data;
	ulint		len;
	ulint		val_buf_size;
	
	ut_ad(plan->n_rows_prefetched > 0);

	column = UT_LIST_GET_FIRST(plan->columns);

	while (column) {
		val = que_node_get_val(column);

		if (!column->copy_val) {
			/* We did not really push any value for the
			column */

			ut_ad(!column->prefetch_buf);
			ut_ad(que_node_get_val_buf_size(column) == 0);
#ifdef UNIV_DEBUG
			dfield_set_data(val, NULL, 0);
#endif
			goto next_col;
		}

		ut_ad(column->prefetch_buf);

		sel_buf = column->prefetch_buf + plan->first_prefetched;

		data = sel_buf->data;
		len = sel_buf->len;
		val_buf_size = sel_buf->val_buf_size;

		/* We must keep track of the allocated memory for
		column values to be able to free it later: therefore
		we swap the values for sel_buf and val */

		sel_buf->data = dfield_get_data(val);
		sel_buf->len = dfield_get_len(val);
		sel_buf->val_buf_size = que_node_get_val_buf_size(column);
		
		dfield_set_data(val, data, len);
		que_node_set_val_buf_size(column, val_buf_size);
next_col:
		column = UT_LIST_GET_NEXT(col_var_list, column);
	}

	plan->n_rows_prefetched--;

	plan->first_prefetched++;
}

/*************************************************************************
Pushes the column values for a prefetched, cached row to the column prefetch
buffers from the val fields in the column nodes. */
UNIV_INLINE
void
sel_push_prefetched_row(
/*====================*/
	plan_t*	plan)	/* in: plan node for a table */
{
	sym_node_t*	column;
	sel_buf_t*	sel_buf;
	dfield_t*	val;
	byte*		data;
	ulint		len;
	ulint		pos;
	ulint		val_buf_size;

	if (plan->n_rows_prefetched == 0) {
		pos = 0;
		plan->first_prefetched = 0;
	} else {
		pos = plan->n_rows_prefetched;

		/* We have the convention that pushing new rows starts only
		after the prefetch stack has been emptied: */
		
		ut_ad(plan->first_prefetched == 0);
	}

	plan->n_rows_prefetched++;
	
	ut_ad(pos < SEL_MAX_N_PREFETCH);
	
	column = UT_LIST_GET_FIRST(plan->columns);

	while (column) {
		if (!column->copy_val) {
			/* There is no sense to push pointers to database
			page fields when we do not keep latch on the page! */

			goto next_col;
		}
		
		if (!column->prefetch_buf) {
			/* Allocate a new prefetch buffer */

			sel_col_prefetch_buf_alloc(column);
		}

		sel_buf = column->prefetch_buf + pos;

		val = que_node_get_val(column);

		data = dfield_get_data(val);
		len = dfield_get_len(val);
		val_buf_size = que_node_get_val_buf_size(column);

		/* We must keep track of the allocated memory for
		column values to be able to free it later: therefore
		we swap the values for sel_buf and val */

		dfield_set_data(val, sel_buf->data, sel_buf->len);
		que_node_set_val_buf_size(column, sel_buf->val_buf_size);
		
		sel_buf->data = data;
		sel_buf->len = len;
		sel_buf->val_buf_size = val_buf_size;
next_col:		
		column = UT_LIST_GET_NEXT(col_var_list, column);
	}
}

/*************************************************************************
Builds a previous version of a clustered index record for a consistent read */
static
ulint