mirror of https://github.com/minio/minio.git
3451 lines
74 KiB
Go
3451 lines
74 KiB
Go
/*
|
|
Copyright 2017 Google Inc.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
*/
|
|
|
|
package sqlparser
|
|
|
|
import (
|
|
"bytes"
|
|
"encoding/hex"
|
|
"encoding/json"
|
|
"fmt"
|
|
"io"
|
|
"log"
|
|
"strings"
|
|
|
|
"github.com/xwb1989/sqlparser/dependency/querypb"
|
|
"github.com/xwb1989/sqlparser/dependency/sqltypes"
|
|
)
|
|
|
|
// Instructions for creating new types: If a type
|
|
// needs to satisfy an interface, declare that function
|
|
// along with that interface. This will help users
|
|
// identify the list of types to which they can assert
|
|
// those interfaces.
|
|
// If the member of a type has a string with a predefined
|
|
// list of values, declare those values as const following
|
|
// the type.
|
|
// For interfaces that define dummy functions to consolidate
|
|
// a set of types, define the function as iTypeName.
|
|
// This will help avoid name collisions.
|
|
|
|
// Parse parses the SQL in full and returns a Statement, which
|
|
// is the AST representation of the query. If a DDL statement
|
|
// is partially parsed but still contains a syntax error, the
|
|
// error is ignored and the DDL is returned anyway.
|
|
func Parse(sql string) (Statement, error) {
|
|
tokenizer := NewStringTokenizer(sql)
|
|
if yyParse(tokenizer) != 0 {
|
|
if tokenizer.partialDDL != nil {
|
|
log.Printf("ignoring error parsing DDL '%s': %v", sql, tokenizer.LastError)
|
|
tokenizer.ParseTree = tokenizer.partialDDL
|
|
return tokenizer.ParseTree, nil
|
|
}
|
|
return nil, tokenizer.LastError
|
|
}
|
|
return tokenizer.ParseTree, nil
|
|
}
|
|
|
|
// ParseStrictDDL is the same as Parse except it errors on
|
|
// partially parsed DDL statements.
|
|
func ParseStrictDDL(sql string) (Statement, error) {
|
|
tokenizer := NewStringTokenizer(sql)
|
|
if yyParse(tokenizer) != 0 {
|
|
return nil, tokenizer.LastError
|
|
}
|
|
return tokenizer.ParseTree, nil
|
|
}
|
|
|
|
// ParseNext parses a single SQL statement from the tokenizer
|
|
// returning a Statement which is the AST representation of the query.
|
|
// The tokenizer will always read up to the end of the statement, allowing for
|
|
// the next call to ParseNext to parse any subsequent SQL statements. When
|
|
// there are no more statements to parse, a error of io.EOF is returned.
|
|
func ParseNext(tokenizer *Tokenizer) (Statement, error) {
|
|
if tokenizer.lastChar == ';' {
|
|
tokenizer.next()
|
|
tokenizer.skipBlank()
|
|
}
|
|
if tokenizer.lastChar == eofChar {
|
|
return nil, io.EOF
|
|
}
|
|
|
|
tokenizer.reset()
|
|
tokenizer.multi = true
|
|
if yyParse(tokenizer) != 0 {
|
|
if tokenizer.partialDDL != nil {
|
|
tokenizer.ParseTree = tokenizer.partialDDL
|
|
return tokenizer.ParseTree, nil
|
|
}
|
|
return nil, tokenizer.LastError
|
|
}
|
|
return tokenizer.ParseTree, nil
|
|
}
|
|
|
|
// SplitStatement returns the first sql statement up to either a ; or EOF
|
|
// and the remainder from the given buffer
|
|
func SplitStatement(blob string) (string, string, error) {
|
|
tokenizer := NewStringTokenizer(blob)
|
|
tkn := 0
|
|
for {
|
|
tkn, _ = tokenizer.Scan()
|
|
if tkn == 0 || tkn == ';' || tkn == eofChar {
|
|
break
|
|
}
|
|
}
|
|
if tokenizer.LastError != nil {
|
|
return "", "", tokenizer.LastError
|
|
}
|
|
if tkn == ';' {
|
|
return blob[:tokenizer.Position-2], blob[tokenizer.Position-1:], nil
|
|
}
|
|
return blob, "", nil
|
|
}
|
|
|
|
// SplitStatementToPieces split raw sql statement that may have multi sql pieces to sql pieces
|
|
// returns the sql pieces blob contains; or error if sql cannot be parsed
|
|
func SplitStatementToPieces(blob string) (pieces []string, err error) {
|
|
pieces = make([]string, 0, 16)
|
|
tokenizer := NewStringTokenizer(blob)
|
|
|
|
tkn := 0
|
|
var stmt string
|
|
stmtBegin := 0
|
|
for {
|
|
tkn, _ = tokenizer.Scan()
|
|
if tkn == ';' {
|
|
stmt = blob[stmtBegin : tokenizer.Position-2]
|
|
pieces = append(pieces, stmt)
|
|
stmtBegin = tokenizer.Position - 1
|
|
|
|
} else if tkn == 0 || tkn == eofChar {
|
|
blobTail := tokenizer.Position - 2
|
|
|
|
if stmtBegin < blobTail {
|
|
stmt = blob[stmtBegin : blobTail+1]
|
|
pieces = append(pieces, stmt)
|
|
}
|
|
break
|
|
}
|
|
}
|
|
|
|
err = tokenizer.LastError
|
|
return
|
|
}
|
|
|
|
// SQLNode defines the interface for all nodes
|
|
// generated by the parser.
|
|
type SQLNode interface {
|
|
Format(buf *TrackedBuffer)
|
|
// walkSubtree calls visit on all underlying nodes
|
|
// of the subtree, but not the current one. Walking
|
|
// must be interrupted if visit returns an error.
|
|
walkSubtree(visit Visit) error
|
|
}
|
|
|
|
// Visit defines the signature of a function that
|
|
// can be used to visit all nodes of a parse tree.
|
|
type Visit func(node SQLNode) (kontinue bool, err error)
|
|
|
|
// Walk calls visit on every node.
|
|
// If visit returns true, the underlying nodes
|
|
// are also visited. If it returns an error, walking
|
|
// is interrupted, and the error is returned.
|
|
func Walk(visit Visit, nodes ...SQLNode) error {
|
|
for _, node := range nodes {
|
|
if node == nil {
|
|
continue
|
|
}
|
|
kontinue, err := visit(node)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if kontinue {
|
|
err = node.walkSubtree(visit)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// String returns a string representation of an SQLNode.
|
|
func String(node SQLNode) string {
|
|
if node == nil {
|
|
return "<nil>"
|
|
}
|
|
|
|
buf := NewTrackedBuffer(nil)
|
|
buf.Myprintf("%v", node)
|
|
return buf.String()
|
|
}
|
|
|
|
// Append appends the SQLNode to the buffer.
|
|
func Append(buf *bytes.Buffer, node SQLNode) {
|
|
tbuf := &TrackedBuffer{
|
|
Buffer: buf,
|
|
}
|
|
node.Format(tbuf)
|
|
}
|
|
|
|
// Statement represents a statement.
|
|
type Statement interface {
|
|
iStatement()
|
|
SQLNode
|
|
}
|
|
|
|
func (*Union) iStatement() {}
|
|
func (*Select) iStatement() {}
|
|
func (*Stream) iStatement() {}
|
|
func (*Insert) iStatement() {}
|
|
func (*Update) iStatement() {}
|
|
func (*Delete) iStatement() {}
|
|
func (*Set) iStatement() {}
|
|
func (*DBDDL) iStatement() {}
|
|
func (*DDL) iStatement() {}
|
|
func (*Show) iStatement() {}
|
|
func (*Use) iStatement() {}
|
|
func (*Begin) iStatement() {}
|
|
func (*Commit) iStatement() {}
|
|
func (*Rollback) iStatement() {}
|
|
func (*OtherRead) iStatement() {}
|
|
func (*OtherAdmin) iStatement() {}
|
|
|
|
// ParenSelect can actually not be a top level statement,
|
|
// but we have to allow it because it's a requirement
|
|
// of SelectStatement.
|
|
func (*ParenSelect) iStatement() {}
|
|
|
|
// SelectStatement any SELECT statement.
|
|
type SelectStatement interface {
|
|
iSelectStatement()
|
|
iStatement()
|
|
iInsertRows()
|
|
AddOrder(*Order)
|
|
SetLimit(*Limit)
|
|
SQLNode
|
|
}
|
|
|
|
func (*Select) iSelectStatement() {}
|
|
func (*Union) iSelectStatement() {}
|
|
func (*ParenSelect) iSelectStatement() {}
|
|
|
|
// Select represents a SELECT statement.
|
|
type Select struct {
|
|
Cache string
|
|
Comments Comments
|
|
Distinct string
|
|
Hints string
|
|
SelectExprs SelectExprs
|
|
From TableExprs
|
|
Where *Where
|
|
GroupBy GroupBy
|
|
Having *Where
|
|
OrderBy OrderBy
|
|
Limit *Limit
|
|
Lock string
|
|
}
|
|
|
|
// Select.Distinct
|
|
const (
|
|
DistinctStr = "distinct "
|
|
StraightJoinHint = "straight_join "
|
|
)
|
|
|
|
// Select.Lock
|
|
const (
|
|
ForUpdateStr = " for update"
|
|
ShareModeStr = " lock in share mode"
|
|
)
|
|
|
|
// Select.Cache
|
|
const (
|
|
SQLCacheStr = "sql_cache "
|
|
SQLNoCacheStr = "sql_no_cache "
|
|
)
|
|
|
|
// AddOrder adds an order by element
|
|
func (node *Select) AddOrder(order *Order) {
|
|
node.OrderBy = append(node.OrderBy, order)
|
|
}
|
|
|
|
// SetLimit sets the limit clause
|
|
func (node *Select) SetLimit(limit *Limit) {
|
|
node.Limit = limit
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Select) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("select %v%s%s%s%v from %v%v%v%v%v%v%s",
|
|
node.Comments, node.Cache, node.Distinct, node.Hints, node.SelectExprs,
|
|
node.From, node.Where,
|
|
node.GroupBy, node.Having, node.OrderBy,
|
|
node.Limit, node.Lock)
|
|
}
|
|
|
|
func (node *Select) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Comments,
|
|
node.SelectExprs,
|
|
node.From,
|
|
node.Where,
|
|
node.GroupBy,
|
|
node.Having,
|
|
node.OrderBy,
|
|
node.Limit,
|
|
)
|
|
}
|
|
|
|
// AddWhere adds the boolean expression to the
|
|
// WHERE clause as an AND condition. If the expression
|
|
// is an OR clause, it parenthesizes it. Currently,
|
|
// the OR operator is the only one that's lower precedence
|
|
// than AND.
|
|
func (node *Select) AddWhere(expr Expr) {
|
|
if _, ok := expr.(*OrExpr); ok {
|
|
expr = &ParenExpr{Expr: expr}
|
|
}
|
|
if node.Where == nil {
|
|
node.Where = &Where{
|
|
Type: WhereStr,
|
|
Expr: expr,
|
|
}
|
|
return
|
|
}
|
|
node.Where.Expr = &AndExpr{
|
|
Left: node.Where.Expr,
|
|
Right: expr,
|
|
}
|
|
return
|
|
}
|
|
|
|
// AddHaving adds the boolean expression to the
|
|
// HAVING clause as an AND condition. If the expression
|
|
// is an OR clause, it parenthesizes it. Currently,
|
|
// the OR operator is the only one that's lower precedence
|
|
// than AND.
|
|
func (node *Select) AddHaving(expr Expr) {
|
|
if _, ok := expr.(*OrExpr); ok {
|
|
expr = &ParenExpr{Expr: expr}
|
|
}
|
|
if node.Having == nil {
|
|
node.Having = &Where{
|
|
Type: HavingStr,
|
|
Expr: expr,
|
|
}
|
|
return
|
|
}
|
|
node.Having.Expr = &AndExpr{
|
|
Left: node.Having.Expr,
|
|
Right: expr,
|
|
}
|
|
return
|
|
}
|
|
|
|
// ParenSelect is a parenthesized SELECT statement.
|
|
type ParenSelect struct {
|
|
Select SelectStatement
|
|
}
|
|
|
|
// AddOrder adds an order by element
|
|
func (node *ParenSelect) AddOrder(order *Order) {
|
|
panic("unreachable")
|
|
}
|
|
|
|
// SetLimit sets the limit clause
|
|
func (node *ParenSelect) SetLimit(limit *Limit) {
|
|
panic("unreachable")
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ParenSelect) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("(%v)", node.Select)
|
|
}
|
|
|
|
func (node *ParenSelect) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Select,
|
|
)
|
|
}
|
|
|
|
// Union represents a UNION statement.
|
|
type Union struct {
|
|
Type string
|
|
Left, Right SelectStatement
|
|
OrderBy OrderBy
|
|
Limit *Limit
|
|
Lock string
|
|
}
|
|
|
|
// Union.Type
|
|
const (
|
|
UnionStr = "union"
|
|
UnionAllStr = "union all"
|
|
UnionDistinctStr = "union distinct"
|
|
)
|
|
|
|
// AddOrder adds an order by element
|
|
func (node *Union) AddOrder(order *Order) {
|
|
node.OrderBy = append(node.OrderBy, order)
|
|
}
|
|
|
|
// SetLimit sets the limit clause
|
|
func (node *Union) SetLimit(limit *Limit) {
|
|
node.Limit = limit
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Union) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %s %v%v%v%s", node.Left, node.Type, node.Right,
|
|
node.OrderBy, node.Limit, node.Lock)
|
|
}
|
|
|
|
func (node *Union) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Left,
|
|
node.Right,
|
|
)
|
|
}
|
|
|
|
// Stream represents a SELECT statement.
|
|
type Stream struct {
|
|
Comments Comments
|
|
SelectExpr SelectExpr
|
|
Table TableName
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Stream) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("stream %v%v from %v",
|
|
node.Comments, node.SelectExpr, node.Table)
|
|
}
|
|
|
|
func (node *Stream) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Comments,
|
|
node.SelectExpr,
|
|
node.Table,
|
|
)
|
|
}
|
|
|
|
// Insert represents an INSERT or REPLACE statement.
|
|
// Per the MySQL docs, http://dev.mysql.com/doc/refman/5.7/en/replace.html
|
|
// Replace is the counterpart to `INSERT IGNORE`, and works exactly like a
|
|
// normal INSERT except if the row exists. In that case it first deletes
|
|
// the row and re-inserts with new values. For that reason we keep it as an Insert struct.
|
|
// Replaces are currently disallowed in sharded schemas because
|
|
// of the implications the deletion part may have on vindexes.
|
|
// If you add fields here, consider adding them to calls to validateSubquerySamePlan.
|
|
type Insert struct {
|
|
Action string
|
|
Comments Comments
|
|
Ignore string
|
|
Table TableName
|
|
Partitions Partitions
|
|
Columns Columns
|
|
Rows InsertRows
|
|
OnDup OnDup
|
|
}
|
|
|
|
// DDL strings.
|
|
const (
|
|
InsertStr = "insert"
|
|
ReplaceStr = "replace"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *Insert) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%s %v%sinto %v%v%v %v%v",
|
|
node.Action,
|
|
node.Comments, node.Ignore,
|
|
node.Table, node.Partitions, node.Columns, node.Rows, node.OnDup)
|
|
}
|
|
|
|
func (node *Insert) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Comments,
|
|
node.Table,
|
|
node.Columns,
|
|
node.Rows,
|
|
node.OnDup,
|
|
)
|
|
}
|
|
|
|
// InsertRows represents the rows for an INSERT statement.
|
|
type InsertRows interface {
|
|
iInsertRows()
|
|
SQLNode
|
|
}
|
|
|
|
func (*Select) iInsertRows() {}
|
|
func (*Union) iInsertRows() {}
|
|
func (Values) iInsertRows() {}
|
|
func (*ParenSelect) iInsertRows() {}
|
|
|
|
// Update represents an UPDATE statement.
|
|
// If you add fields here, consider adding them to calls to validateSubquerySamePlan.
|
|
type Update struct {
|
|
Comments Comments
|
|
TableExprs TableExprs
|
|
Exprs UpdateExprs
|
|
Where *Where
|
|
OrderBy OrderBy
|
|
Limit *Limit
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Update) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("update %v%v set %v%v%v%v",
|
|
node.Comments, node.TableExprs,
|
|
node.Exprs, node.Where, node.OrderBy, node.Limit)
|
|
}
|
|
|
|
func (node *Update) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Comments,
|
|
node.TableExprs,
|
|
node.Exprs,
|
|
node.Where,
|
|
node.OrderBy,
|
|
node.Limit,
|
|
)
|
|
}
|
|
|
|
// Delete represents a DELETE statement.
|
|
// If you add fields here, consider adding them to calls to validateSubquerySamePlan.
|
|
type Delete struct {
|
|
Comments Comments
|
|
Targets TableNames
|
|
TableExprs TableExprs
|
|
Partitions Partitions
|
|
Where *Where
|
|
OrderBy OrderBy
|
|
Limit *Limit
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Delete) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("delete %v", node.Comments)
|
|
if node.Targets != nil {
|
|
buf.Myprintf("%v ", node.Targets)
|
|
}
|
|
buf.Myprintf("from %v%v%v%v%v", node.TableExprs, node.Partitions, node.Where, node.OrderBy, node.Limit)
|
|
}
|
|
|
|
func (node *Delete) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Comments,
|
|
node.Targets,
|
|
node.TableExprs,
|
|
node.Where,
|
|
node.OrderBy,
|
|
node.Limit,
|
|
)
|
|
}
|
|
|
|
// Set represents a SET statement.
|
|
type Set struct {
|
|
Comments Comments
|
|
Exprs SetExprs
|
|
Scope string
|
|
}
|
|
|
|
// Set.Scope or Show.Scope
|
|
const (
|
|
SessionStr = "session"
|
|
GlobalStr = "global"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *Set) Format(buf *TrackedBuffer) {
|
|
if node.Scope == "" {
|
|
buf.Myprintf("set %v%v", node.Comments, node.Exprs)
|
|
} else {
|
|
buf.Myprintf("set %v%s %v", node.Comments, node.Scope, node.Exprs)
|
|
}
|
|
}
|
|
|
|
func (node *Set) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Comments,
|
|
node.Exprs,
|
|
)
|
|
}
|
|
|
|
// DBDDL represents a CREATE, DROP database statement.
|
|
type DBDDL struct {
|
|
Action string
|
|
DBName string
|
|
IfExists bool
|
|
Collate string
|
|
Charset string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *DBDDL) Format(buf *TrackedBuffer) {
|
|
switch node.Action {
|
|
case CreateStr:
|
|
buf.WriteString(fmt.Sprintf("%s database %s", node.Action, node.DBName))
|
|
case DropStr:
|
|
exists := ""
|
|
if node.IfExists {
|
|
exists = " if exists"
|
|
}
|
|
buf.WriteString(fmt.Sprintf("%s database%s %v", node.Action, exists, node.DBName))
|
|
}
|
|
}
|
|
|
|
// walkSubtree walks the nodes of the subtree.
|
|
func (node *DBDDL) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// DDL represents a CREATE, ALTER, DROP, RENAME or TRUNCATE statement.
|
|
// Table is set for AlterStr, DropStr, RenameStr, TruncateStr
|
|
// NewName is set for AlterStr, CreateStr, RenameStr.
|
|
// VindexSpec is set for CreateVindexStr, DropVindexStr, AddColVindexStr, DropColVindexStr
|
|
// VindexCols is set for AddColVindexStr
|
|
type DDL struct {
|
|
Action string
|
|
Table TableName
|
|
NewName TableName
|
|
IfExists bool
|
|
TableSpec *TableSpec
|
|
PartitionSpec *PartitionSpec
|
|
VindexSpec *VindexSpec
|
|
VindexCols []ColIdent
|
|
}
|
|
|
|
// DDL strings.
|
|
const (
|
|
CreateStr = "create"
|
|
AlterStr = "alter"
|
|
DropStr = "drop"
|
|
RenameStr = "rename"
|
|
TruncateStr = "truncate"
|
|
CreateVindexStr = "create vindex"
|
|
AddColVindexStr = "add vindex"
|
|
DropColVindexStr = "drop vindex"
|
|
|
|
// Vindex DDL param to specify the owner of a vindex
|
|
VindexOwnerStr = "owner"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *DDL) Format(buf *TrackedBuffer) {
|
|
switch node.Action {
|
|
case CreateStr:
|
|
if node.TableSpec == nil {
|
|
buf.Myprintf("%s table %v", node.Action, node.NewName)
|
|
} else {
|
|
buf.Myprintf("%s table %v %v", node.Action, node.NewName, node.TableSpec)
|
|
}
|
|
case DropStr:
|
|
exists := ""
|
|
if node.IfExists {
|
|
exists = " if exists"
|
|
}
|
|
buf.Myprintf("%s table%s %v", node.Action, exists, node.Table)
|
|
case RenameStr:
|
|
buf.Myprintf("%s table %v to %v", node.Action, node.Table, node.NewName)
|
|
case AlterStr:
|
|
if node.PartitionSpec != nil {
|
|
buf.Myprintf("%s table %v %v", node.Action, node.Table, node.PartitionSpec)
|
|
} else {
|
|
buf.Myprintf("%s table %v", node.Action, node.Table)
|
|
}
|
|
case CreateVindexStr:
|
|
buf.Myprintf("%s %v %v", node.Action, node.VindexSpec.Name, node.VindexSpec)
|
|
case AddColVindexStr:
|
|
buf.Myprintf("alter table %v %s %v (", node.Table, node.Action, node.VindexSpec.Name)
|
|
for i, col := range node.VindexCols {
|
|
if i != 0 {
|
|
buf.Myprintf(", %v", col)
|
|
} else {
|
|
buf.Myprintf("%v", col)
|
|
}
|
|
}
|
|
buf.Myprintf(")")
|
|
if node.VindexSpec.Type.String() != "" {
|
|
buf.Myprintf(" %v", node.VindexSpec)
|
|
}
|
|
case DropColVindexStr:
|
|
buf.Myprintf("alter table %v %s %v", node.Table, node.Action, node.VindexSpec.Name)
|
|
default:
|
|
buf.Myprintf("%s table %v", node.Action, node.Table)
|
|
}
|
|
}
|
|
|
|
func (node *DDL) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Table,
|
|
node.NewName,
|
|
)
|
|
}
|
|
|
|
// Partition strings
|
|
const (
|
|
ReorganizeStr = "reorganize partition"
|
|
)
|
|
|
|
// PartitionSpec describe partition actions (for alter and create)
|
|
type PartitionSpec struct {
|
|
Action string
|
|
Name ColIdent
|
|
Definitions []*PartitionDefinition
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *PartitionSpec) Format(buf *TrackedBuffer) {
|
|
switch node.Action {
|
|
case ReorganizeStr:
|
|
buf.Myprintf("%s %v into (", node.Action, node.Name)
|
|
var prefix string
|
|
for _, pd := range node.Definitions {
|
|
buf.Myprintf("%s%v", prefix, pd)
|
|
prefix = ", "
|
|
}
|
|
buf.Myprintf(")")
|
|
default:
|
|
panic("unimplemented")
|
|
}
|
|
}
|
|
|
|
func (node *PartitionSpec) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
if err := Walk(visit, node.Name); err != nil {
|
|
return err
|
|
}
|
|
for _, def := range node.Definitions {
|
|
if err := Walk(visit, def); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// PartitionDefinition describes a very minimal partition definition
|
|
type PartitionDefinition struct {
|
|
Name ColIdent
|
|
Limit Expr
|
|
Maxvalue bool
|
|
}
|
|
|
|
// Format formats the node
|
|
func (node *PartitionDefinition) Format(buf *TrackedBuffer) {
|
|
if !node.Maxvalue {
|
|
buf.Myprintf("partition %v values less than (%v)", node.Name, node.Limit)
|
|
} else {
|
|
buf.Myprintf("partition %v values less than (maxvalue)", node.Name)
|
|
}
|
|
}
|
|
|
|
func (node *PartitionDefinition) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
node.Limit,
|
|
)
|
|
}
|
|
|
|
// TableSpec describes the structure of a table from a CREATE TABLE statement
|
|
type TableSpec struct {
|
|
Columns []*ColumnDefinition
|
|
Indexes []*IndexDefinition
|
|
Options string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (ts *TableSpec) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("(\n")
|
|
for i, col := range ts.Columns {
|
|
if i == 0 {
|
|
buf.Myprintf("\t%v", col)
|
|
} else {
|
|
buf.Myprintf(",\n\t%v", col)
|
|
}
|
|
}
|
|
for _, idx := range ts.Indexes {
|
|
buf.Myprintf(",\n\t%v", idx)
|
|
}
|
|
|
|
buf.Myprintf("\n)%s", strings.Replace(ts.Options, ", ", ",\n ", -1))
|
|
}
|
|
|
|
// AddColumn appends the given column to the list in the spec
|
|
func (ts *TableSpec) AddColumn(cd *ColumnDefinition) {
|
|
ts.Columns = append(ts.Columns, cd)
|
|
}
|
|
|
|
// AddIndex appends the given index to the list in the spec
|
|
func (ts *TableSpec) AddIndex(id *IndexDefinition) {
|
|
ts.Indexes = append(ts.Indexes, id)
|
|
}
|
|
|
|
func (ts *TableSpec) walkSubtree(visit Visit) error {
|
|
if ts == nil {
|
|
return nil
|
|
}
|
|
|
|
for _, n := range ts.Columns {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
for _, n := range ts.Indexes {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// ColumnDefinition describes a column in a CREATE TABLE statement
|
|
type ColumnDefinition struct {
|
|
Name ColIdent
|
|
Type ColumnType
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (col *ColumnDefinition) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %v", col.Name, &col.Type)
|
|
}
|
|
|
|
func (col *ColumnDefinition) walkSubtree(visit Visit) error {
|
|
if col == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
col.Name,
|
|
&col.Type,
|
|
)
|
|
}
|
|
|
|
// ColumnType represents a sql type in a CREATE TABLE statement
|
|
// All optional fields are nil if not specified
|
|
type ColumnType struct {
|
|
// The base type string
|
|
Type string
|
|
|
|
// Generic field options.
|
|
NotNull BoolVal
|
|
Autoincrement BoolVal
|
|
Default *SQLVal
|
|
OnUpdate *SQLVal
|
|
Comment *SQLVal
|
|
|
|
// Numeric field options
|
|
Length *SQLVal
|
|
Unsigned BoolVal
|
|
Zerofill BoolVal
|
|
Scale *SQLVal
|
|
|
|
// Text field options
|
|
Charset string
|
|
Collate string
|
|
|
|
// Enum values
|
|
EnumValues []string
|
|
|
|
// Key specification
|
|
KeyOpt ColumnKeyOption
|
|
}
|
|
|
|
// Format returns a canonical string representation of the type and all relevant options
|
|
func (ct *ColumnType) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%s", ct.Type)
|
|
|
|
if ct.Length != nil && ct.Scale != nil {
|
|
buf.Myprintf("(%v,%v)", ct.Length, ct.Scale)
|
|
|
|
} else if ct.Length != nil {
|
|
buf.Myprintf("(%v)", ct.Length)
|
|
}
|
|
|
|
if ct.EnumValues != nil {
|
|
buf.Myprintf("(%s)", strings.Join(ct.EnumValues, ", "))
|
|
}
|
|
|
|
opts := make([]string, 0, 16)
|
|
if ct.Unsigned {
|
|
opts = append(opts, keywordStrings[UNSIGNED])
|
|
}
|
|
if ct.Zerofill {
|
|
opts = append(opts, keywordStrings[ZEROFILL])
|
|
}
|
|
if ct.Charset != "" {
|
|
opts = append(opts, keywordStrings[CHARACTER], keywordStrings[SET], ct.Charset)
|
|
}
|
|
if ct.Collate != "" {
|
|
opts = append(opts, keywordStrings[COLLATE], ct.Collate)
|
|
}
|
|
if ct.NotNull {
|
|
opts = append(opts, keywordStrings[NOT], keywordStrings[NULL])
|
|
}
|
|
if ct.Default != nil {
|
|
opts = append(opts, keywordStrings[DEFAULT], String(ct.Default))
|
|
}
|
|
if ct.OnUpdate != nil {
|
|
opts = append(opts, keywordStrings[ON], keywordStrings[UPDATE], String(ct.OnUpdate))
|
|
}
|
|
if ct.Autoincrement {
|
|
opts = append(opts, keywordStrings[AUTO_INCREMENT])
|
|
}
|
|
if ct.Comment != nil {
|
|
opts = append(opts, keywordStrings[COMMENT_KEYWORD], String(ct.Comment))
|
|
}
|
|
if ct.KeyOpt == colKeyPrimary {
|
|
opts = append(opts, keywordStrings[PRIMARY], keywordStrings[KEY])
|
|
}
|
|
if ct.KeyOpt == colKeyUnique {
|
|
opts = append(opts, keywordStrings[UNIQUE])
|
|
}
|
|
if ct.KeyOpt == colKeyUniqueKey {
|
|
opts = append(opts, keywordStrings[UNIQUE], keywordStrings[KEY])
|
|
}
|
|
if ct.KeyOpt == colKeySpatialKey {
|
|
opts = append(opts, keywordStrings[SPATIAL], keywordStrings[KEY])
|
|
}
|
|
if ct.KeyOpt == colKey {
|
|
opts = append(opts, keywordStrings[KEY])
|
|
}
|
|
|
|
if len(opts) != 0 {
|
|
buf.Myprintf(" %s", strings.Join(opts, " "))
|
|
}
|
|
}
|
|
|
|
// DescribeType returns the abbreviated type information as required for
|
|
// describe table
|
|
func (ct *ColumnType) DescribeType() string {
|
|
buf := NewTrackedBuffer(nil)
|
|
buf.Myprintf("%s", ct.Type)
|
|
if ct.Length != nil && ct.Scale != nil {
|
|
buf.Myprintf("(%v,%v)", ct.Length, ct.Scale)
|
|
} else if ct.Length != nil {
|
|
buf.Myprintf("(%v)", ct.Length)
|
|
}
|
|
|
|
opts := make([]string, 0, 16)
|
|
if ct.Unsigned {
|
|
opts = append(opts, keywordStrings[UNSIGNED])
|
|
}
|
|
if ct.Zerofill {
|
|
opts = append(opts, keywordStrings[ZEROFILL])
|
|
}
|
|
if len(opts) != 0 {
|
|
buf.Myprintf(" %s", strings.Join(opts, " "))
|
|
}
|
|
return buf.String()
|
|
}
|
|
|
|
// SQLType returns the sqltypes type code for the given column
|
|
func (ct *ColumnType) SQLType() querypb.Type {
|
|
switch ct.Type {
|
|
case keywordStrings[TINYINT]:
|
|
if ct.Unsigned {
|
|
return sqltypes.Uint8
|
|
}
|
|
return sqltypes.Int8
|
|
case keywordStrings[SMALLINT]:
|
|
if ct.Unsigned {
|
|
return sqltypes.Uint16
|
|
}
|
|
return sqltypes.Int16
|
|
case keywordStrings[MEDIUMINT]:
|
|
if ct.Unsigned {
|
|
return sqltypes.Uint24
|
|
}
|
|
return sqltypes.Int24
|
|
case keywordStrings[INT]:
|
|
fallthrough
|
|
case keywordStrings[INTEGER]:
|
|
if ct.Unsigned {
|
|
return sqltypes.Uint32
|
|
}
|
|
return sqltypes.Int32
|
|
case keywordStrings[BIGINT]:
|
|
if ct.Unsigned {
|
|
return sqltypes.Uint64
|
|
}
|
|
return sqltypes.Int64
|
|
case keywordStrings[TEXT]:
|
|
return sqltypes.Text
|
|
case keywordStrings[TINYTEXT]:
|
|
return sqltypes.Text
|
|
case keywordStrings[MEDIUMTEXT]:
|
|
return sqltypes.Text
|
|
case keywordStrings[LONGTEXT]:
|
|
return sqltypes.Text
|
|
case keywordStrings[BLOB]:
|
|
return sqltypes.Blob
|
|
case keywordStrings[TINYBLOB]:
|
|
return sqltypes.Blob
|
|
case keywordStrings[MEDIUMBLOB]:
|
|
return sqltypes.Blob
|
|
case keywordStrings[LONGBLOB]:
|
|
return sqltypes.Blob
|
|
case keywordStrings[CHAR]:
|
|
return sqltypes.Char
|
|
case keywordStrings[VARCHAR]:
|
|
return sqltypes.VarChar
|
|
case keywordStrings[BINARY]:
|
|
return sqltypes.Binary
|
|
case keywordStrings[VARBINARY]:
|
|
return sqltypes.VarBinary
|
|
case keywordStrings[DATE]:
|
|
return sqltypes.Date
|
|
case keywordStrings[TIME]:
|
|
return sqltypes.Time
|
|
case keywordStrings[DATETIME]:
|
|
return sqltypes.Datetime
|
|
case keywordStrings[TIMESTAMP]:
|
|
return sqltypes.Timestamp
|
|
case keywordStrings[YEAR]:
|
|
return sqltypes.Year
|
|
case keywordStrings[FLOAT_TYPE]:
|
|
return sqltypes.Float32
|
|
case keywordStrings[DOUBLE]:
|
|
return sqltypes.Float64
|
|
case keywordStrings[DECIMAL]:
|
|
return sqltypes.Decimal
|
|
case keywordStrings[BIT]:
|
|
return sqltypes.Bit
|
|
case keywordStrings[ENUM]:
|
|
return sqltypes.Enum
|
|
case keywordStrings[SET]:
|
|
return sqltypes.Set
|
|
case keywordStrings[JSON]:
|
|
return sqltypes.TypeJSON
|
|
case keywordStrings[GEOMETRY]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[POINT]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[LINESTRING]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[POLYGON]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[GEOMETRYCOLLECTION]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[MULTIPOINT]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[MULTILINESTRING]:
|
|
return sqltypes.Geometry
|
|
case keywordStrings[MULTIPOLYGON]:
|
|
return sqltypes.Geometry
|
|
}
|
|
panic("unimplemented type " + ct.Type)
|
|
}
|
|
|
|
func (ct *ColumnType) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// IndexDefinition describes an index in a CREATE TABLE statement
|
|
type IndexDefinition struct {
|
|
Info *IndexInfo
|
|
Columns []*IndexColumn
|
|
Options []*IndexOption
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (idx *IndexDefinition) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v (", idx.Info)
|
|
for i, col := range idx.Columns {
|
|
if i != 0 {
|
|
buf.Myprintf(", %v", col.Column)
|
|
} else {
|
|
buf.Myprintf("%v", col.Column)
|
|
}
|
|
if col.Length != nil {
|
|
buf.Myprintf("(%v)", col.Length)
|
|
}
|
|
}
|
|
buf.Myprintf(")")
|
|
|
|
for _, opt := range idx.Options {
|
|
buf.Myprintf(" %s", opt.Name)
|
|
if opt.Using != "" {
|
|
buf.Myprintf(" %s", opt.Using)
|
|
} else {
|
|
buf.Myprintf(" %v", opt.Value)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (idx *IndexDefinition) walkSubtree(visit Visit) error {
|
|
if idx == nil {
|
|
return nil
|
|
}
|
|
|
|
for _, n := range idx.Columns {
|
|
if err := Walk(visit, n.Column); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// IndexInfo describes the name and type of an index in a CREATE TABLE statement
|
|
type IndexInfo struct {
|
|
Type string
|
|
Name ColIdent
|
|
Primary bool
|
|
Spatial bool
|
|
Unique bool
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (ii *IndexInfo) Format(buf *TrackedBuffer) {
|
|
if ii.Primary {
|
|
buf.Myprintf("%s", ii.Type)
|
|
} else {
|
|
buf.Myprintf("%s %v", ii.Type, ii.Name)
|
|
}
|
|
}
|
|
|
|
func (ii *IndexInfo) walkSubtree(visit Visit) error {
|
|
return Walk(visit, ii.Name)
|
|
}
|
|
|
|
// IndexColumn describes a column in an index definition with optional length
|
|
type IndexColumn struct {
|
|
Column ColIdent
|
|
Length *SQLVal
|
|
}
|
|
|
|
// LengthScaleOption is used for types that have an optional length
|
|
// and scale
|
|
type LengthScaleOption struct {
|
|
Length *SQLVal
|
|
Scale *SQLVal
|
|
}
|
|
|
|
// IndexOption is used for trailing options for indexes: COMMENT, KEY_BLOCK_SIZE, USING
|
|
type IndexOption struct {
|
|
Name string
|
|
Value *SQLVal
|
|
Using string
|
|
}
|
|
|
|
// ColumnKeyOption indicates whether or not the given column is defined as an
|
|
// index element and contains the type of the option
|
|
type ColumnKeyOption int
|
|
|
|
const (
|
|
colKeyNone ColumnKeyOption = iota
|
|
colKeyPrimary
|
|
colKeySpatialKey
|
|
colKeyUnique
|
|
colKeyUniqueKey
|
|
colKey
|
|
)
|
|
|
|
// VindexSpec defines a vindex for a CREATE VINDEX or DROP VINDEX statement
|
|
type VindexSpec struct {
|
|
Name ColIdent
|
|
Type ColIdent
|
|
Params []VindexParam
|
|
}
|
|
|
|
// ParseParams parses the vindex parameter list, pulling out the special-case
|
|
// "owner" parameter
|
|
func (node *VindexSpec) ParseParams() (string, map[string]string) {
|
|
var owner string
|
|
params := map[string]string{}
|
|
for _, p := range node.Params {
|
|
if p.Key.Lowered() == VindexOwnerStr {
|
|
owner = p.Val
|
|
} else {
|
|
params[p.Key.String()] = p.Val
|
|
}
|
|
}
|
|
return owner, params
|
|
}
|
|
|
|
// Format formats the node. The "CREATE VINDEX" preamble was formatted in
|
|
// the containing DDL node Format, so this just prints the type, any
|
|
// parameters, and optionally the owner
|
|
func (node *VindexSpec) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("using %v", node.Type)
|
|
|
|
numParams := len(node.Params)
|
|
if numParams != 0 {
|
|
buf.Myprintf(" with ")
|
|
for i, p := range node.Params {
|
|
if i != 0 {
|
|
buf.Myprintf(", ")
|
|
}
|
|
buf.Myprintf("%v", p)
|
|
}
|
|
}
|
|
}
|
|
|
|
func (node *VindexSpec) walkSubtree(visit Visit) error {
|
|
err := Walk(visit,
|
|
node.Name,
|
|
)
|
|
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
for _, p := range node.Params {
|
|
err := Walk(visit, p)
|
|
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// VindexParam defines a key/value parameter for a CREATE VINDEX statement
|
|
type VindexParam struct {
|
|
Key ColIdent
|
|
Val string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node VindexParam) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%s=%s", node.Key.String(), node.Val)
|
|
}
|
|
|
|
func (node VindexParam) walkSubtree(visit Visit) error {
|
|
return Walk(visit,
|
|
node.Key,
|
|
)
|
|
}
|
|
|
|
// Show represents a show statement.
|
|
type Show struct {
|
|
Type string
|
|
OnTable TableName
|
|
ShowTablesOpt *ShowTablesOpt
|
|
Scope string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Show) Format(buf *TrackedBuffer) {
|
|
if node.Type == "tables" && node.ShowTablesOpt != nil {
|
|
opt := node.ShowTablesOpt
|
|
if opt.DbName != "" {
|
|
if opt.Filter != nil {
|
|
buf.Myprintf("show %s%stables from %s %v", opt.Extended, opt.Full, opt.DbName, opt.Filter)
|
|
} else {
|
|
buf.Myprintf("show %s%stables from %s", opt.Extended, opt.Full, opt.DbName)
|
|
}
|
|
} else {
|
|
if opt.Filter != nil {
|
|
buf.Myprintf("show %s%stables %v", opt.Extended, opt.Full, opt.Filter)
|
|
} else {
|
|
buf.Myprintf("show %s%stables", opt.Extended, opt.Full)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
if node.Scope == "" {
|
|
buf.Myprintf("show %s", node.Type)
|
|
} else {
|
|
buf.Myprintf("show %s %s", node.Scope, node.Type)
|
|
}
|
|
if node.HasOnTable() {
|
|
buf.Myprintf(" on %v", node.OnTable)
|
|
}
|
|
}
|
|
|
|
// HasOnTable returns true if the show statement has an "on" clause
|
|
func (node *Show) HasOnTable() bool {
|
|
return node.OnTable.Name.v != ""
|
|
}
|
|
|
|
func (node *Show) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// ShowTablesOpt is show tables option
|
|
type ShowTablesOpt struct {
|
|
Extended string
|
|
Full string
|
|
DbName string
|
|
Filter *ShowFilter
|
|
}
|
|
|
|
// ShowFilter is show tables filter
|
|
type ShowFilter struct {
|
|
Like string
|
|
Filter Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ShowFilter) Format(buf *TrackedBuffer) {
|
|
if node.Like != "" {
|
|
buf.Myprintf("like '%s'", node.Like)
|
|
} else {
|
|
buf.Myprintf("where %v", node.Filter)
|
|
}
|
|
}
|
|
|
|
func (node *ShowFilter) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// Use represents a use statement.
|
|
type Use struct {
|
|
DBName TableIdent
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Use) Format(buf *TrackedBuffer) {
|
|
if node.DBName.v != "" {
|
|
buf.Myprintf("use %v", node.DBName)
|
|
} else {
|
|
buf.Myprintf("use")
|
|
}
|
|
}
|
|
|
|
func (node *Use) walkSubtree(visit Visit) error {
|
|
return Walk(visit, node.DBName)
|
|
}
|
|
|
|
// Begin represents a Begin statement.
|
|
type Begin struct{}
|
|
|
|
// Format formats the node.
|
|
func (node *Begin) Format(buf *TrackedBuffer) {
|
|
buf.WriteString("begin")
|
|
}
|
|
|
|
func (node *Begin) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// Commit represents a Commit statement.
|
|
type Commit struct{}
|
|
|
|
// Format formats the node.
|
|
func (node *Commit) Format(buf *TrackedBuffer) {
|
|
buf.WriteString("commit")
|
|
}
|
|
|
|
func (node *Commit) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// Rollback represents a Rollback statement.
|
|
type Rollback struct{}
|
|
|
|
// Format formats the node.
|
|
func (node *Rollback) Format(buf *TrackedBuffer) {
|
|
buf.WriteString("rollback")
|
|
}
|
|
|
|
func (node *Rollback) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// OtherRead represents a DESCRIBE, or EXPLAIN statement.
|
|
// It should be used only as an indicator. It does not contain
|
|
// the full AST for the statement.
|
|
type OtherRead struct{}
|
|
|
|
// Format formats the node.
|
|
func (node *OtherRead) Format(buf *TrackedBuffer) {
|
|
buf.WriteString("otherread")
|
|
}
|
|
|
|
func (node *OtherRead) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// OtherAdmin represents a misc statement that relies on ADMIN privileges,
|
|
// such as REPAIR, OPTIMIZE, or TRUNCATE statement.
|
|
// It should be used only as an indicator. It does not contain
|
|
// the full AST for the statement.
|
|
type OtherAdmin struct{}
|
|
|
|
// Format formats the node.
|
|
func (node *OtherAdmin) Format(buf *TrackedBuffer) {
|
|
buf.WriteString("otheradmin")
|
|
}
|
|
|
|
func (node *OtherAdmin) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// Comments represents a list of comments.
|
|
type Comments [][]byte
|
|
|
|
// Format formats the node.
|
|
func (node Comments) Format(buf *TrackedBuffer) {
|
|
for _, c := range node {
|
|
buf.Myprintf("%s ", c)
|
|
}
|
|
}
|
|
|
|
func (node Comments) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// SelectExprs represents SELECT expressions.
|
|
type SelectExprs []SelectExpr
|
|
|
|
// Format formats the node.
|
|
func (node SelectExprs) Format(buf *TrackedBuffer) {
|
|
var prefix string
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node SelectExprs) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// SelectExpr represents a SELECT expression.
|
|
type SelectExpr interface {
|
|
iSelectExpr()
|
|
SQLNode
|
|
}
|
|
|
|
func (*StarExpr) iSelectExpr() {}
|
|
func (*AliasedExpr) iSelectExpr() {}
|
|
func (Nextval) iSelectExpr() {}
|
|
|
|
// StarExpr defines a '*' or 'table.*' expression.
|
|
type StarExpr struct {
|
|
TableName TableName
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *StarExpr) Format(buf *TrackedBuffer) {
|
|
if !node.TableName.IsEmpty() {
|
|
buf.Myprintf("%v.", node.TableName)
|
|
}
|
|
buf.Myprintf("*")
|
|
}
|
|
|
|
func (node *StarExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.TableName,
|
|
)
|
|
}
|
|
|
|
// AliasedExpr defines an aliased SELECT expression.
|
|
type AliasedExpr struct {
|
|
Expr Expr
|
|
As ColIdent
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *AliasedExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v", node.Expr)
|
|
if !node.As.IsEmpty() {
|
|
buf.Myprintf(" as %v", node.As)
|
|
}
|
|
}
|
|
|
|
func (node *AliasedExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
node.As,
|
|
)
|
|
}
|
|
|
|
// Nextval defines the NEXT VALUE expression.
|
|
type Nextval struct {
|
|
Expr Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node Nextval) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("next %v values", node.Expr)
|
|
}
|
|
|
|
func (node Nextval) walkSubtree(visit Visit) error {
|
|
return Walk(visit, node.Expr)
|
|
}
|
|
|
|
// Columns represents an insert column list.
|
|
type Columns []ColIdent
|
|
|
|
// Format formats the node.
|
|
func (node Columns) Format(buf *TrackedBuffer) {
|
|
if node == nil {
|
|
return
|
|
}
|
|
prefix := "("
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
buf.WriteString(")")
|
|
}
|
|
|
|
func (node Columns) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// FindColumn finds a column in the column list, returning
|
|
// the index if it exists or -1 otherwise
|
|
func (node Columns) FindColumn(col ColIdent) int {
|
|
for i, colName := range node {
|
|
if colName.Equal(col) {
|
|
return i
|
|
}
|
|
}
|
|
return -1
|
|
}
|
|
|
|
// Partitions is a type alias for Columns so we can handle printing efficiently
|
|
type Partitions Columns
|
|
|
|
// Format formats the node
|
|
func (node Partitions) Format(buf *TrackedBuffer) {
|
|
if node == nil {
|
|
return
|
|
}
|
|
prefix := " partition ("
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
buf.WriteString(")")
|
|
}
|
|
|
|
func (node Partitions) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// TableExprs represents a list of table expressions.
|
|
type TableExprs []TableExpr
|
|
|
|
// Format formats the node.
|
|
func (node TableExprs) Format(buf *TrackedBuffer) {
|
|
var prefix string
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node TableExprs) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// TableExpr represents a table expression.
|
|
type TableExpr interface {
|
|
iTableExpr()
|
|
SQLNode
|
|
}
|
|
|
|
func (*AliasedTableExpr) iTableExpr() {}
|
|
func (*ParenTableExpr) iTableExpr() {}
|
|
func (*JoinTableExpr) iTableExpr() {}
|
|
|
|
// AliasedTableExpr represents a table expression
|
|
// coupled with an optional alias or index hint.
|
|
// If As is empty, no alias was used.
|
|
type AliasedTableExpr struct {
|
|
Expr SimpleTableExpr
|
|
Partitions Partitions
|
|
As TableIdent
|
|
Hints *IndexHints
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *AliasedTableExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v%v", node.Expr, node.Partitions)
|
|
if !node.As.IsEmpty() {
|
|
buf.Myprintf(" as %v", node.As)
|
|
}
|
|
if node.Hints != nil {
|
|
// Hint node provides the space padding.
|
|
buf.Myprintf("%v", node.Hints)
|
|
}
|
|
}
|
|
|
|
func (node *AliasedTableExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
node.As,
|
|
node.Hints,
|
|
)
|
|
}
|
|
|
|
// RemoveHints returns a new AliasedTableExpr with the hints removed.
|
|
func (node *AliasedTableExpr) RemoveHints() *AliasedTableExpr {
|
|
noHints := *node
|
|
noHints.Hints = nil
|
|
return &noHints
|
|
}
|
|
|
|
// SimpleTableExpr represents a simple table expression.
|
|
type SimpleTableExpr interface {
|
|
iSimpleTableExpr()
|
|
SQLNode
|
|
}
|
|
|
|
func (TableName) iSimpleTableExpr() {}
|
|
func (*Subquery) iSimpleTableExpr() {}
|
|
|
|
// TableNames is a list of TableName.
|
|
type TableNames []TableName
|
|
|
|
// Format formats the node.
|
|
func (node TableNames) Format(buf *TrackedBuffer) {
|
|
var prefix string
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node TableNames) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// TableName represents a table name.
|
|
// Qualifier, if specified, represents a database or keyspace.
|
|
// TableName is a value struct whose fields are case sensitive.
|
|
// This means two TableName vars can be compared for equality
|
|
// and a TableName can also be used as key in a map.
|
|
type TableName struct {
|
|
Name, Qualifier TableIdent
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node TableName) Format(buf *TrackedBuffer) {
|
|
if node.IsEmpty() {
|
|
return
|
|
}
|
|
if !node.Qualifier.IsEmpty() {
|
|
buf.Myprintf("%v.", node.Qualifier)
|
|
}
|
|
buf.Myprintf("%v", node.Name)
|
|
}
|
|
|
|
func (node TableName) walkSubtree(visit Visit) error {
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
node.Qualifier,
|
|
)
|
|
}
|
|
|
|
// IsEmpty returns true if TableName is nil or empty.
|
|
func (node TableName) IsEmpty() bool {
|
|
// If Name is empty, Qualifer is also empty.
|
|
return node.Name.IsEmpty()
|
|
}
|
|
|
|
// ToViewName returns a TableName acceptable for use as a VIEW. VIEW names are
|
|
// always lowercase, so ToViewName lowercasese the name. Databases are case-sensitive
|
|
// so Qualifier is left untouched.
|
|
func (node TableName) ToViewName() TableName {
|
|
return TableName{
|
|
Qualifier: node.Qualifier,
|
|
Name: NewTableIdent(strings.ToLower(node.Name.v)),
|
|
}
|
|
}
|
|
|
|
// ParenTableExpr represents a parenthesized list of TableExpr.
|
|
type ParenTableExpr struct {
|
|
Exprs TableExprs
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ParenTableExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("(%v)", node.Exprs)
|
|
}
|
|
|
|
func (node *ParenTableExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Exprs,
|
|
)
|
|
}
|
|
|
|
// JoinCondition represents the join conditions (either a ON or USING clause)
|
|
// of a JoinTableExpr.
|
|
type JoinCondition struct {
|
|
On Expr
|
|
Using Columns
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node JoinCondition) Format(buf *TrackedBuffer) {
|
|
if node.On != nil {
|
|
buf.Myprintf(" on %v", node.On)
|
|
}
|
|
if node.Using != nil {
|
|
buf.Myprintf(" using %v", node.Using)
|
|
}
|
|
}
|
|
|
|
func (node JoinCondition) walkSubtree(visit Visit) error {
|
|
return Walk(
|
|
visit,
|
|
node.On,
|
|
node.Using,
|
|
)
|
|
}
|
|
|
|
// JoinTableExpr represents a TableExpr that's a JOIN operation.
|
|
type JoinTableExpr struct {
|
|
LeftExpr TableExpr
|
|
Join string
|
|
RightExpr TableExpr
|
|
Condition JoinCondition
|
|
}
|
|
|
|
// JoinTableExpr.Join
|
|
const (
|
|
JoinStr = "join"
|
|
StraightJoinStr = "straight_join"
|
|
LeftJoinStr = "left join"
|
|
RightJoinStr = "right join"
|
|
NaturalJoinStr = "natural join"
|
|
NaturalLeftJoinStr = "natural left join"
|
|
NaturalRightJoinStr = "natural right join"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *JoinTableExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %s %v%v", node.LeftExpr, node.Join, node.RightExpr, node.Condition)
|
|
}
|
|
|
|
func (node *JoinTableExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.LeftExpr,
|
|
node.RightExpr,
|
|
node.Condition,
|
|
)
|
|
}
|
|
|
|
// IndexHints represents a list of index hints.
|
|
type IndexHints struct {
|
|
Type string
|
|
Indexes []ColIdent
|
|
}
|
|
|
|
// Index hints.
|
|
const (
|
|
UseStr = "use "
|
|
IgnoreStr = "ignore "
|
|
ForceStr = "force "
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *IndexHints) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf(" %sindex ", node.Type)
|
|
prefix := "("
|
|
for _, n := range node.Indexes {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
buf.Myprintf(")")
|
|
}
|
|
|
|
func (node *IndexHints) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
for _, n := range node.Indexes {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// Where represents a WHERE or HAVING clause.
|
|
type Where struct {
|
|
Type string
|
|
Expr Expr
|
|
}
|
|
|
|
// Where.Type
|
|
const (
|
|
WhereStr = "where"
|
|
HavingStr = "having"
|
|
)
|
|
|
|
// NewWhere creates a WHERE or HAVING clause out
|
|
// of a Expr. If the expression is nil, it returns nil.
|
|
func NewWhere(typ string, expr Expr) *Where {
|
|
if expr == nil {
|
|
return nil
|
|
}
|
|
return &Where{Type: typ, Expr: expr}
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Where) Format(buf *TrackedBuffer) {
|
|
if node == nil || node.Expr == nil {
|
|
return
|
|
}
|
|
buf.Myprintf(" %s %v", node.Type, node.Expr)
|
|
}
|
|
|
|
func (node *Where) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
// Expr represents an expression.
|
|
type Expr interface {
|
|
iExpr()
|
|
// replace replaces any subexpression that matches
|
|
// from with to. The implementation can use the
|
|
// replaceExprs convenience function.
|
|
replace(from, to Expr) bool
|
|
SQLNode
|
|
}
|
|
|
|
func (*AndExpr) iExpr() {}
|
|
func (*OrExpr) iExpr() {}
|
|
func (*NotExpr) iExpr() {}
|
|
func (*ParenExpr) iExpr() {}
|
|
func (*ComparisonExpr) iExpr() {}
|
|
func (*RangeCond) iExpr() {}
|
|
func (*IsExpr) iExpr() {}
|
|
func (*ExistsExpr) iExpr() {}
|
|
func (*SQLVal) iExpr() {}
|
|
func (*NullVal) iExpr() {}
|
|
func (BoolVal) iExpr() {}
|
|
func (*ColName) iExpr() {}
|
|
func (ValTuple) iExpr() {}
|
|
func (*Subquery) iExpr() {}
|
|
func (ListArg) iExpr() {}
|
|
func (*BinaryExpr) iExpr() {}
|
|
func (*UnaryExpr) iExpr() {}
|
|
func (*IntervalExpr) iExpr() {}
|
|
func (*CollateExpr) iExpr() {}
|
|
func (*FuncExpr) iExpr() {}
|
|
func (*CaseExpr) iExpr() {}
|
|
func (*ValuesFuncExpr) iExpr() {}
|
|
func (*ConvertExpr) iExpr() {}
|
|
func (*SubstrExpr) iExpr() {}
|
|
func (*ConvertUsingExpr) iExpr() {}
|
|
func (*MatchExpr) iExpr() {}
|
|
func (*GroupConcatExpr) iExpr() {}
|
|
func (*Default) iExpr() {}
|
|
|
|
// ReplaceExpr finds the from expression from root
|
|
// and replaces it with to. If from matches root,
|
|
// then to is returned.
|
|
func ReplaceExpr(root, from, to Expr) Expr {
|
|
if root == from {
|
|
return to
|
|
}
|
|
root.replace(from, to)
|
|
return root
|
|
}
|
|
|
|
// replaceExprs is a convenience function used by implementors
|
|
// of the replace method.
|
|
func replaceExprs(from, to Expr, exprs ...*Expr) bool {
|
|
for _, expr := range exprs {
|
|
if *expr == nil {
|
|
continue
|
|
}
|
|
if *expr == from {
|
|
*expr = to
|
|
return true
|
|
}
|
|
if (*expr).replace(from, to) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// Exprs represents a list of value expressions.
|
|
// It's not a valid expression because it's not parenthesized.
|
|
type Exprs []Expr
|
|
|
|
// Format formats the node.
|
|
func (node Exprs) Format(buf *TrackedBuffer) {
|
|
var prefix string
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node Exprs) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// AndExpr represents an AND expression.
|
|
type AndExpr struct {
|
|
Left, Right Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *AndExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v and %v", node.Left, node.Right)
|
|
}
|
|
|
|
func (node *AndExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Left,
|
|
node.Right,
|
|
)
|
|
}
|
|
|
|
func (node *AndExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Left, &node.Right)
|
|
}
|
|
|
|
// OrExpr represents an OR expression.
|
|
type OrExpr struct {
|
|
Left, Right Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *OrExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v or %v", node.Left, node.Right)
|
|
}
|
|
|
|
func (node *OrExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Left,
|
|
node.Right,
|
|
)
|
|
}
|
|
|
|
func (node *OrExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Left, &node.Right)
|
|
}
|
|
|
|
// NotExpr represents a NOT expression.
|
|
type NotExpr struct {
|
|
Expr Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *NotExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("not %v", node.Expr)
|
|
}
|
|
|
|
func (node *NotExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *NotExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// ParenExpr represents a parenthesized boolean expression.
|
|
type ParenExpr struct {
|
|
Expr Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ParenExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("(%v)", node.Expr)
|
|
}
|
|
|
|
func (node *ParenExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *ParenExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// ComparisonExpr represents a two-value comparison expression.
|
|
type ComparisonExpr struct {
|
|
Operator string
|
|
Left, Right Expr
|
|
Escape Expr
|
|
}
|
|
|
|
// ComparisonExpr.Operator
|
|
const (
|
|
EqualStr = "="
|
|
LessThanStr = "<"
|
|
GreaterThanStr = ">"
|
|
LessEqualStr = "<="
|
|
GreaterEqualStr = ">="
|
|
NotEqualStr = "!="
|
|
NullSafeEqualStr = "<=>"
|
|
InStr = "in"
|
|
NotInStr = "not in"
|
|
LikeStr = "like"
|
|
NotLikeStr = "not like"
|
|
RegexpStr = "regexp"
|
|
NotRegexpStr = "not regexp"
|
|
JSONExtractOp = "->"
|
|
JSONUnquoteExtractOp = "->>"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *ComparisonExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %s %v", node.Left, node.Operator, node.Right)
|
|
if node.Escape != nil {
|
|
buf.Myprintf(" escape %v", node.Escape)
|
|
}
|
|
}
|
|
|
|
func (node *ComparisonExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Left,
|
|
node.Right,
|
|
node.Escape,
|
|
)
|
|
}
|
|
|
|
func (node *ComparisonExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Left, &node.Right, &node.Escape)
|
|
}
|
|
|
|
// RangeCond represents a BETWEEN or a NOT BETWEEN expression.
|
|
type RangeCond struct {
|
|
Operator string
|
|
Left Expr
|
|
From, To Expr
|
|
}
|
|
|
|
// RangeCond.Operator
|
|
const (
|
|
BetweenStr = "between"
|
|
NotBetweenStr = "not between"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *RangeCond) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %s %v and %v", node.Left, node.Operator, node.From, node.To)
|
|
}
|
|
|
|
func (node *RangeCond) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Left,
|
|
node.From,
|
|
node.To,
|
|
)
|
|
}
|
|
|
|
func (node *RangeCond) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Left, &node.From, &node.To)
|
|
}
|
|
|
|
// IsExpr represents an IS ... or an IS NOT ... expression.
|
|
type IsExpr struct {
|
|
Operator string
|
|
Expr Expr
|
|
}
|
|
|
|
// IsExpr.Operator
|
|
const (
|
|
IsNullStr = "is null"
|
|
IsNotNullStr = "is not null"
|
|
IsTrueStr = "is true"
|
|
IsNotTrueStr = "is not true"
|
|
IsFalseStr = "is false"
|
|
IsNotFalseStr = "is not false"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *IsExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %s", node.Expr, node.Operator)
|
|
}
|
|
|
|
func (node *IsExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *IsExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// ExistsExpr represents an EXISTS expression.
|
|
type ExistsExpr struct {
|
|
Subquery *Subquery
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ExistsExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("exists %v", node.Subquery)
|
|
}
|
|
|
|
func (node *ExistsExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Subquery,
|
|
)
|
|
}
|
|
|
|
func (node *ExistsExpr) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// ExprFromValue converts the given Value into an Expr or returns an error.
|
|
func ExprFromValue(value sqltypes.Value) (Expr, error) {
|
|
// The type checks here follow the rules defined in sqltypes/types.go.
|
|
switch {
|
|
case value.Type() == sqltypes.Null:
|
|
return &NullVal{}, nil
|
|
case value.IsIntegral():
|
|
return NewIntVal(value.ToBytes()), nil
|
|
case value.IsFloat() || value.Type() == sqltypes.Decimal:
|
|
return NewFloatVal(value.ToBytes()), nil
|
|
case value.IsQuoted():
|
|
return NewStrVal(value.ToBytes()), nil
|
|
default:
|
|
// We cannot support sqltypes.Expression, or any other invalid type.
|
|
return nil, fmt.Errorf("cannot convert value %v to AST", value)
|
|
}
|
|
}
|
|
|
|
// ValType specifies the type for SQLVal.
|
|
type ValType int
|
|
|
|
// These are the possible Valtype values.
|
|
// HexNum represents a 0x... value. It cannot
|
|
// be treated as a simple value because it can
|
|
// be interpreted differently depending on the
|
|
// context.
|
|
const (
|
|
StrVal = ValType(iota)
|
|
IntVal
|
|
FloatVal
|
|
HexNum
|
|
HexVal
|
|
ValArg
|
|
BitVal
|
|
)
|
|
|
|
// SQLVal represents a single value.
|
|
type SQLVal struct {
|
|
Type ValType
|
|
Val []byte
|
|
}
|
|
|
|
// NewStrVal builds a new StrVal.
|
|
func NewStrVal(in []byte) *SQLVal {
|
|
return &SQLVal{Type: StrVal, Val: in}
|
|
}
|
|
|
|
// NewIntVal builds a new IntVal.
|
|
func NewIntVal(in []byte) *SQLVal {
|
|
return &SQLVal{Type: IntVal, Val: in}
|
|
}
|
|
|
|
// NewFloatVal builds a new FloatVal.
|
|
func NewFloatVal(in []byte) *SQLVal {
|
|
return &SQLVal{Type: FloatVal, Val: in}
|
|
}
|
|
|
|
// NewHexNum builds a new HexNum.
|
|
func NewHexNum(in []byte) *SQLVal {
|
|
return &SQLVal{Type: HexNum, Val: in}
|
|
}
|
|
|
|
// NewHexVal builds a new HexVal.
|
|
func NewHexVal(in []byte) *SQLVal {
|
|
return &SQLVal{Type: HexVal, Val: in}
|
|
}
|
|
|
|
// NewBitVal builds a new BitVal containing a bit literal.
|
|
func NewBitVal(in []byte) *SQLVal {
|
|
return &SQLVal{Type: BitVal, Val: in}
|
|
}
|
|
|
|
// NewValArg builds a new ValArg.
|
|
func NewValArg(in []byte) *SQLVal {
|
|
return &SQLVal{Type: ValArg, Val: in}
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *SQLVal) Format(buf *TrackedBuffer) {
|
|
switch node.Type {
|
|
case StrVal:
|
|
sqltypes.MakeTrusted(sqltypes.VarBinary, node.Val).EncodeSQL(buf)
|
|
case IntVal, FloatVal, HexNum:
|
|
buf.Myprintf("%s", []byte(node.Val))
|
|
case HexVal:
|
|
buf.Myprintf("X'%s'", []byte(node.Val))
|
|
case BitVal:
|
|
buf.Myprintf("B'%s'", []byte(node.Val))
|
|
case ValArg:
|
|
buf.WriteArg(string(node.Val))
|
|
default:
|
|
panic("unexpected")
|
|
}
|
|
}
|
|
|
|
func (node *SQLVal) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
func (node *SQLVal) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// HexDecode decodes the hexval into bytes.
|
|
func (node *SQLVal) HexDecode() ([]byte, error) {
|
|
dst := make([]byte, hex.DecodedLen(len([]byte(node.Val))))
|
|
_, err := hex.Decode(dst, []byte(node.Val))
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
return dst, err
|
|
}
|
|
|
|
// NullVal represents a NULL value.
|
|
type NullVal struct{}
|
|
|
|
// Format formats the node.
|
|
func (node *NullVal) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("null")
|
|
}
|
|
|
|
func (node *NullVal) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
func (node *NullVal) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// BoolVal is true or false.
|
|
type BoolVal bool
|
|
|
|
// Format formats the node.
|
|
func (node BoolVal) Format(buf *TrackedBuffer) {
|
|
if node {
|
|
buf.Myprintf("true")
|
|
} else {
|
|
buf.Myprintf("false")
|
|
}
|
|
}
|
|
|
|
func (node BoolVal) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
func (node BoolVal) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// ColName represents a column name.
|
|
type ColName struct {
|
|
// Metadata is not populated by the parser.
|
|
// It's a placeholder for analyzers to store
|
|
// additional data, typically info about which
|
|
// table or column this node references.
|
|
Metadata interface{}
|
|
Name ColIdent
|
|
Qualifier TableName
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ColName) Format(buf *TrackedBuffer) {
|
|
if !node.Qualifier.IsEmpty() {
|
|
buf.Myprintf("%v.", node.Qualifier)
|
|
}
|
|
buf.Myprintf("%v", node.Name)
|
|
}
|
|
|
|
func (node *ColName) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
node.Qualifier,
|
|
)
|
|
}
|
|
|
|
func (node *ColName) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// Equal returns true if the column names match.
|
|
func (node *ColName) Equal(c *ColName) bool {
|
|
// Failsafe: ColName should not be empty.
|
|
if node == nil || c == nil {
|
|
return false
|
|
}
|
|
return node.Name.Equal(c.Name) && node.Qualifier == c.Qualifier
|
|
}
|
|
|
|
// ColTuple represents a list of column values.
|
|
// It can be ValTuple, Subquery, ListArg.
|
|
type ColTuple interface {
|
|
iColTuple()
|
|
Expr
|
|
}
|
|
|
|
func (ValTuple) iColTuple() {}
|
|
func (*Subquery) iColTuple() {}
|
|
func (ListArg) iColTuple() {}
|
|
|
|
// ValTuple represents a tuple of actual values.
|
|
type ValTuple Exprs
|
|
|
|
// Format formats the node.
|
|
func (node ValTuple) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("(%v)", Exprs(node))
|
|
}
|
|
|
|
func (node ValTuple) walkSubtree(visit Visit) error {
|
|
return Walk(visit, Exprs(node))
|
|
}
|
|
|
|
func (node ValTuple) replace(from, to Expr) bool {
|
|
for i := range node {
|
|
if replaceExprs(from, to, &node[i]) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// Subquery represents a subquery.
|
|
type Subquery struct {
|
|
Select SelectStatement
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Subquery) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("(%v)", node.Select)
|
|
}
|
|
|
|
func (node *Subquery) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Select,
|
|
)
|
|
}
|
|
|
|
func (node *Subquery) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// ListArg represents a named list argument.
|
|
type ListArg []byte
|
|
|
|
// Format formats the node.
|
|
func (node ListArg) Format(buf *TrackedBuffer) {
|
|
buf.WriteArg(string(node))
|
|
}
|
|
|
|
func (node ListArg) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
func (node ListArg) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// BinaryExpr represents a binary value expression.
|
|
type BinaryExpr struct {
|
|
Operator string
|
|
Left, Right Expr
|
|
}
|
|
|
|
// BinaryExpr.Operator
|
|
const (
|
|
BitAndStr = "&"
|
|
BitOrStr = "|"
|
|
BitXorStr = "^"
|
|
PlusStr = "+"
|
|
MinusStr = "-"
|
|
MultStr = "*"
|
|
DivStr = "/"
|
|
IntDivStr = "div"
|
|
ModStr = "%"
|
|
ShiftLeftStr = "<<"
|
|
ShiftRightStr = ">>"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *BinaryExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v %s %v", node.Left, node.Operator, node.Right)
|
|
}
|
|
|
|
func (node *BinaryExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Left,
|
|
node.Right,
|
|
)
|
|
}
|
|
|
|
func (node *BinaryExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Left, &node.Right)
|
|
}
|
|
|
|
// UnaryExpr represents a unary value expression.
|
|
type UnaryExpr struct {
|
|
Operator string
|
|
Expr Expr
|
|
}
|
|
|
|
// UnaryExpr.Operator
|
|
const (
|
|
UPlusStr = "+"
|
|
UMinusStr = "-"
|
|
TildaStr = "~"
|
|
BangStr = "!"
|
|
BinaryStr = "binary "
|
|
UBinaryStr = "_binary "
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *UnaryExpr) Format(buf *TrackedBuffer) {
|
|
if _, unary := node.Expr.(*UnaryExpr); unary {
|
|
buf.Myprintf("%s %v", node.Operator, node.Expr)
|
|
return
|
|
}
|
|
buf.Myprintf("%s%v", node.Operator, node.Expr)
|
|
}
|
|
|
|
func (node *UnaryExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *UnaryExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// IntervalExpr represents a date-time INTERVAL expression.
|
|
type IntervalExpr struct {
|
|
Expr Expr
|
|
Unit string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *IntervalExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("interval %v %s", node.Expr, node.Unit)
|
|
}
|
|
|
|
func (node *IntervalExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *IntervalExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// CollateExpr represents dynamic collate operator.
|
|
type CollateExpr struct {
|
|
Expr Expr
|
|
Charset string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *CollateExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v collate %s", node.Expr, node.Charset)
|
|
}
|
|
|
|
func (node *CollateExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *CollateExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// FuncExpr represents a function call.
|
|
type FuncExpr struct {
|
|
Qualifier TableIdent
|
|
Name ColIdent
|
|
Distinct bool
|
|
Exprs SelectExprs
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *FuncExpr) Format(buf *TrackedBuffer) {
|
|
var distinct string
|
|
if node.Distinct {
|
|
distinct = "distinct "
|
|
}
|
|
if !node.Qualifier.IsEmpty() {
|
|
buf.Myprintf("%v.", node.Qualifier)
|
|
}
|
|
// Function names should not be back-quoted even
|
|
// if they match a reserved word. So, print the
|
|
// name as is.
|
|
buf.Myprintf("%s(%s%v)", node.Name.String(), distinct, node.Exprs)
|
|
}
|
|
|
|
func (node *FuncExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Qualifier,
|
|
node.Name,
|
|
node.Exprs,
|
|
)
|
|
}
|
|
|
|
func (node *FuncExpr) replace(from, to Expr) bool {
|
|
for _, sel := range node.Exprs {
|
|
aliased, ok := sel.(*AliasedExpr)
|
|
if !ok {
|
|
continue
|
|
}
|
|
if replaceExprs(from, to, &aliased.Expr) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// Aggregates is a map of all aggregate functions.
|
|
var Aggregates = map[string]bool{
|
|
"avg": true,
|
|
"bit_and": true,
|
|
"bit_or": true,
|
|
"bit_xor": true,
|
|
"count": true,
|
|
"group_concat": true,
|
|
"max": true,
|
|
"min": true,
|
|
"std": true,
|
|
"stddev_pop": true,
|
|
"stddev_samp": true,
|
|
"stddev": true,
|
|
"sum": true,
|
|
"var_pop": true,
|
|
"var_samp": true,
|
|
"variance": true,
|
|
}
|
|
|
|
// IsAggregate returns true if the function is an aggregate.
|
|
func (node *FuncExpr) IsAggregate() bool {
|
|
return Aggregates[node.Name.Lowered()]
|
|
}
|
|
|
|
// GroupConcatExpr represents a call to GROUP_CONCAT
|
|
type GroupConcatExpr struct {
|
|
Distinct string
|
|
Exprs SelectExprs
|
|
OrderBy OrderBy
|
|
Separator string
|
|
}
|
|
|
|
// Format formats the node
|
|
func (node *GroupConcatExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("group_concat(%s%v%v%s)", node.Distinct, node.Exprs, node.OrderBy, node.Separator)
|
|
}
|
|
|
|
func (node *GroupConcatExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Exprs,
|
|
node.OrderBy,
|
|
)
|
|
}
|
|
|
|
func (node *GroupConcatExpr) replace(from, to Expr) bool {
|
|
for _, sel := range node.Exprs {
|
|
aliased, ok := sel.(*AliasedExpr)
|
|
if !ok {
|
|
continue
|
|
}
|
|
if replaceExprs(from, to, &aliased.Expr) {
|
|
return true
|
|
}
|
|
}
|
|
for _, order := range node.OrderBy {
|
|
if replaceExprs(from, to, &order.Expr) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// ValuesFuncExpr represents a function call.
|
|
type ValuesFuncExpr struct {
|
|
Name *ColName
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ValuesFuncExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("values(%v)", node.Name)
|
|
}
|
|
|
|
func (node *ValuesFuncExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
)
|
|
}
|
|
|
|
func (node *ValuesFuncExpr) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// SubstrExpr represents a call to SubstrExpr(column, value_expression) or SubstrExpr(column, value_expression,value_expression)
|
|
// also supported syntax SubstrExpr(column from value_expression for value_expression)
|
|
type SubstrExpr struct {
|
|
Name *ColName
|
|
From Expr
|
|
To Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *SubstrExpr) Format(buf *TrackedBuffer) {
|
|
|
|
if node.To == nil {
|
|
buf.Myprintf("substr(%v, %v)", node.Name, node.From)
|
|
} else {
|
|
buf.Myprintf("substr(%v, %v, %v)", node.Name, node.From, node.To)
|
|
}
|
|
}
|
|
|
|
func (node *SubstrExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.From, &node.To)
|
|
}
|
|
|
|
func (node *SubstrExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
node.From,
|
|
node.To,
|
|
)
|
|
}
|
|
|
|
// ConvertExpr represents a call to CONVERT(expr, type)
|
|
// or it's equivalent CAST(expr AS type). Both are rewritten to the former.
|
|
type ConvertExpr struct {
|
|
Expr Expr
|
|
Type *ConvertType
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ConvertExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("convert(%v, %v)", node.Expr, node.Type)
|
|
}
|
|
|
|
func (node *ConvertExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
node.Type,
|
|
)
|
|
}
|
|
|
|
func (node *ConvertExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// ConvertUsingExpr represents a call to CONVERT(expr USING charset).
|
|
type ConvertUsingExpr struct {
|
|
Expr Expr
|
|
Type string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *ConvertUsingExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("convert(%v using %s)", node.Expr, node.Type)
|
|
}
|
|
|
|
func (node *ConvertUsingExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *ConvertUsingExpr) replace(from, to Expr) bool {
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// ConvertType represents the type in call to CONVERT(expr, type)
|
|
type ConvertType struct {
|
|
Type string
|
|
Length *SQLVal
|
|
Scale *SQLVal
|
|
Operator string
|
|
Charset string
|
|
}
|
|
|
|
// this string is "character set" and this comment is required
|
|
const (
|
|
CharacterSetStr = " character set"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *ConvertType) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%s", node.Type)
|
|
if node.Length != nil {
|
|
buf.Myprintf("(%v", node.Length)
|
|
if node.Scale != nil {
|
|
buf.Myprintf(", %v", node.Scale)
|
|
}
|
|
buf.Myprintf(")")
|
|
}
|
|
if node.Charset != "" {
|
|
buf.Myprintf("%s %s", node.Operator, node.Charset)
|
|
}
|
|
}
|
|
|
|
func (node *ConvertType) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// MatchExpr represents a call to the MATCH function
|
|
type MatchExpr struct {
|
|
Columns SelectExprs
|
|
Expr Expr
|
|
Option string
|
|
}
|
|
|
|
// MatchExpr.Option
|
|
const (
|
|
BooleanModeStr = " in boolean mode"
|
|
NaturalLanguageModeStr = " in natural language mode"
|
|
NaturalLanguageModeWithQueryExpansionStr = " in natural language mode with query expansion"
|
|
QueryExpansionStr = " with query expansion"
|
|
)
|
|
|
|
// Format formats the node
|
|
func (node *MatchExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("match(%v) against (%v%s)", node.Columns, node.Expr, node.Option)
|
|
}
|
|
|
|
func (node *MatchExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Columns,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
func (node *MatchExpr) replace(from, to Expr) bool {
|
|
for _, sel := range node.Columns {
|
|
aliased, ok := sel.(*AliasedExpr)
|
|
if !ok {
|
|
continue
|
|
}
|
|
if replaceExprs(from, to, &aliased.Expr) {
|
|
return true
|
|
}
|
|
}
|
|
return replaceExprs(from, to, &node.Expr)
|
|
}
|
|
|
|
// CaseExpr represents a CASE expression.
|
|
type CaseExpr struct {
|
|
Expr Expr
|
|
Whens []*When
|
|
Else Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *CaseExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("case ")
|
|
if node.Expr != nil {
|
|
buf.Myprintf("%v ", node.Expr)
|
|
}
|
|
for _, when := range node.Whens {
|
|
buf.Myprintf("%v ", when)
|
|
}
|
|
if node.Else != nil {
|
|
buf.Myprintf("else %v ", node.Else)
|
|
}
|
|
buf.Myprintf("end")
|
|
}
|
|
|
|
func (node *CaseExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
if err := Walk(visit, node.Expr); err != nil {
|
|
return err
|
|
}
|
|
for _, n := range node.Whens {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return Walk(visit, node.Else)
|
|
}
|
|
|
|
func (node *CaseExpr) replace(from, to Expr) bool {
|
|
for _, when := range node.Whens {
|
|
if replaceExprs(from, to, &when.Cond, &when.Val) {
|
|
return true
|
|
}
|
|
}
|
|
return replaceExprs(from, to, &node.Expr, &node.Else)
|
|
}
|
|
|
|
// Default represents a DEFAULT expression.
|
|
type Default struct {
|
|
ColName string
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Default) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("default")
|
|
if node.ColName != "" {
|
|
buf.Myprintf("(%s)", node.ColName)
|
|
}
|
|
}
|
|
|
|
func (node *Default) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
func (node *Default) replace(from, to Expr) bool {
|
|
return false
|
|
}
|
|
|
|
// When represents a WHEN sub-expression.
|
|
type When struct {
|
|
Cond Expr
|
|
Val Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *When) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("when %v then %v", node.Cond, node.Val)
|
|
}
|
|
|
|
func (node *When) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Cond,
|
|
node.Val,
|
|
)
|
|
}
|
|
|
|
// GroupBy represents a GROUP BY clause.
|
|
type GroupBy []Expr
|
|
|
|
// Format formats the node.
|
|
func (node GroupBy) Format(buf *TrackedBuffer) {
|
|
prefix := " group by "
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node GroupBy) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// OrderBy represents an ORDER By clause.
|
|
type OrderBy []*Order
|
|
|
|
// Format formats the node.
|
|
func (node OrderBy) Format(buf *TrackedBuffer) {
|
|
prefix := " order by "
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node OrderBy) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// Order represents an ordering expression.
|
|
type Order struct {
|
|
Expr Expr
|
|
Direction string
|
|
}
|
|
|
|
// Order.Direction
|
|
const (
|
|
AscScr = "asc"
|
|
DescScr = "desc"
|
|
)
|
|
|
|
// Format formats the node.
|
|
func (node *Order) Format(buf *TrackedBuffer) {
|
|
if node, ok := node.Expr.(*NullVal); ok {
|
|
buf.Myprintf("%v", node)
|
|
return
|
|
}
|
|
if node, ok := node.Expr.(*FuncExpr); ok {
|
|
if node.Name.Lowered() == "rand" {
|
|
buf.Myprintf("%v", node)
|
|
return
|
|
}
|
|
}
|
|
|
|
buf.Myprintf("%v %s", node.Expr, node.Direction)
|
|
}
|
|
|
|
func (node *Order) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
// Limit represents a LIMIT clause.
|
|
type Limit struct {
|
|
Offset, Rowcount Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *Limit) Format(buf *TrackedBuffer) {
|
|
if node == nil {
|
|
return
|
|
}
|
|
buf.Myprintf(" limit ")
|
|
if node.Offset != nil {
|
|
buf.Myprintf("%v, ", node.Offset)
|
|
}
|
|
buf.Myprintf("%v", node.Rowcount)
|
|
}
|
|
|
|
func (node *Limit) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Offset,
|
|
node.Rowcount,
|
|
)
|
|
}
|
|
|
|
// Values represents a VALUES clause.
|
|
type Values []ValTuple
|
|
|
|
// Format formats the node.
|
|
func (node Values) Format(buf *TrackedBuffer) {
|
|
prefix := "values "
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node Values) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// UpdateExprs represents a list of update expressions.
|
|
type UpdateExprs []*UpdateExpr
|
|
|
|
// Format formats the node.
|
|
func (node UpdateExprs) Format(buf *TrackedBuffer) {
|
|
var prefix string
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node UpdateExprs) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// UpdateExpr represents an update expression.
|
|
type UpdateExpr struct {
|
|
Name *ColName
|
|
Expr Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *UpdateExpr) Format(buf *TrackedBuffer) {
|
|
buf.Myprintf("%v = %v", node.Name, node.Expr)
|
|
}
|
|
|
|
func (node *UpdateExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
// SetExprs represents a list of set expressions.
|
|
type SetExprs []*SetExpr
|
|
|
|
// Format formats the node.
|
|
func (node SetExprs) Format(buf *TrackedBuffer) {
|
|
var prefix string
|
|
for _, n := range node {
|
|
buf.Myprintf("%s%v", prefix, n)
|
|
prefix = ", "
|
|
}
|
|
}
|
|
|
|
func (node SetExprs) walkSubtree(visit Visit) error {
|
|
for _, n := range node {
|
|
if err := Walk(visit, n); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// SetExpr represents a set expression.
|
|
type SetExpr struct {
|
|
Name ColIdent
|
|
Expr Expr
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node *SetExpr) Format(buf *TrackedBuffer) {
|
|
// We don't have to backtick set variable names.
|
|
if node.Name.EqualString("charset") || node.Name.EqualString("names") {
|
|
buf.Myprintf("%s %v", node.Name.String(), node.Expr)
|
|
} else {
|
|
buf.Myprintf("%s = %v", node.Name.String(), node.Expr)
|
|
}
|
|
}
|
|
|
|
func (node *SetExpr) walkSubtree(visit Visit) error {
|
|
if node == nil {
|
|
return nil
|
|
}
|
|
return Walk(
|
|
visit,
|
|
node.Name,
|
|
node.Expr,
|
|
)
|
|
}
|
|
|
|
// OnDup represents an ON DUPLICATE KEY clause.
|
|
type OnDup UpdateExprs
|
|
|
|
// Format formats the node.
|
|
func (node OnDup) Format(buf *TrackedBuffer) {
|
|
if node == nil {
|
|
return
|
|
}
|
|
buf.Myprintf(" on duplicate key update %v", UpdateExprs(node))
|
|
}
|
|
|
|
func (node OnDup) walkSubtree(visit Visit) error {
|
|
return Walk(visit, UpdateExprs(node))
|
|
}
|
|
|
|
// ColIdent is a case insensitive SQL identifier. It will be escaped with
|
|
// backquotes if necessary.
|
|
type ColIdent struct {
|
|
// This artifact prevents this struct from being compared
|
|
// with itself. It consumes no space as long as it's not the
|
|
// last field in the struct.
|
|
_ [0]struct{ _ []byte }
|
|
val, lowered string
|
|
}
|
|
|
|
// NewColIdent makes a new ColIdent.
|
|
func NewColIdent(str string) ColIdent {
|
|
return ColIdent{
|
|
val: str,
|
|
}
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node ColIdent) Format(buf *TrackedBuffer) {
|
|
formatID(buf, node.val, node.Lowered())
|
|
}
|
|
|
|
func (node ColIdent) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// IsEmpty returns true if the name is empty.
|
|
func (node ColIdent) IsEmpty() bool {
|
|
return node.val == ""
|
|
}
|
|
|
|
// String returns the unescaped column name. It must
|
|
// not be used for SQL generation. Use sqlparser.String
|
|
// instead. The Stringer conformance is for usage
|
|
// in templates.
|
|
func (node ColIdent) String() string {
|
|
return node.val
|
|
}
|
|
|
|
// CompliantName returns a compliant id name
|
|
// that can be used for a bind var.
|
|
func (node ColIdent) CompliantName() string {
|
|
return compliantName(node.val)
|
|
}
|
|
|
|
// Lowered returns a lower-cased column name.
|
|
// This function should generally be used only for optimizing
|
|
// comparisons.
|
|
func (node ColIdent) Lowered() string {
|
|
if node.val == "" {
|
|
return ""
|
|
}
|
|
if node.lowered == "" {
|
|
node.lowered = strings.ToLower(node.val)
|
|
}
|
|
return node.lowered
|
|
}
|
|
|
|
// Equal performs a case-insensitive compare.
|
|
func (node ColIdent) Equal(in ColIdent) bool {
|
|
return node.Lowered() == in.Lowered()
|
|
}
|
|
|
|
// EqualString performs a case-insensitive compare with str.
|
|
func (node ColIdent) EqualString(str string) bool {
|
|
return node.Lowered() == strings.ToLower(str)
|
|
}
|
|
|
|
// MarshalJSON marshals into JSON.
|
|
func (node ColIdent) MarshalJSON() ([]byte, error) {
|
|
return json.Marshal(node.val)
|
|
}
|
|
|
|
// UnmarshalJSON unmarshals from JSON.
|
|
func (node *ColIdent) UnmarshalJSON(b []byte) error {
|
|
var result string
|
|
err := json.Unmarshal(b, &result)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
node.val = result
|
|
return nil
|
|
}
|
|
|
|
// TableIdent is a case sensitive SQL identifier. It will be escaped with
|
|
// backquotes if necessary.
|
|
type TableIdent struct {
|
|
v string
|
|
}
|
|
|
|
// NewTableIdent creates a new TableIdent.
|
|
func NewTableIdent(str string) TableIdent {
|
|
return TableIdent{v: str}
|
|
}
|
|
|
|
// Format formats the node.
|
|
func (node TableIdent) Format(buf *TrackedBuffer) {
|
|
formatID(buf, node.v, strings.ToLower(node.v))
|
|
}
|
|
|
|
func (node TableIdent) walkSubtree(visit Visit) error {
|
|
return nil
|
|
}
|
|
|
|
// IsEmpty returns true if TabIdent is empty.
|
|
func (node TableIdent) IsEmpty() bool {
|
|
return node.v == ""
|
|
}
|
|
|
|
// String returns the unescaped table name. It must
|
|
// not be used for SQL generation. Use sqlparser.String
|
|
// instead. The Stringer conformance is for usage
|
|
// in templates.
|
|
func (node TableIdent) String() string {
|
|
return node.v
|
|
}
|
|
|
|
// CompliantName returns a compliant id name
|
|
// that can be used for a bind var.
|
|
func (node TableIdent) CompliantName() string {
|
|
return compliantName(node.v)
|
|
}
|
|
|
|
// MarshalJSON marshals into JSON.
|
|
func (node TableIdent) MarshalJSON() ([]byte, error) {
|
|
return json.Marshal(node.v)
|
|
}
|
|
|
|
// UnmarshalJSON unmarshals from JSON.
|
|
func (node *TableIdent) UnmarshalJSON(b []byte) error {
|
|
var result string
|
|
err := json.Unmarshal(b, &result)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
node.v = result
|
|
return nil
|
|
}
|
|
|
|
// Backtick produces a backticked literal given an input string.
|
|
func Backtick(in string) string {
|
|
var buf bytes.Buffer
|
|
buf.WriteByte('`')
|
|
for _, c := range in {
|
|
buf.WriteRune(c)
|
|
if c == '`' {
|
|
buf.WriteByte('`')
|
|
}
|
|
}
|
|
buf.WriteByte('`')
|
|
return buf.String()
|
|
}
|
|
|
|
func formatID(buf *TrackedBuffer, original, lowered string) {
|
|
isDbSystemVariable := false
|
|
if len(original) > 1 && original[:2] == "@@" {
|
|
isDbSystemVariable = true
|
|
}
|
|
|
|
for i, c := range original {
|
|
if !isLetter(uint16(c)) && (!isDbSystemVariable || !isCarat(uint16(c))) {
|
|
if i == 0 || !isDigit(uint16(c)) {
|
|
goto mustEscape
|
|
}
|
|
}
|
|
}
|
|
if _, ok := keywords[lowered]; ok {
|
|
goto mustEscape
|
|
}
|
|
buf.Myprintf("%s", original)
|
|
return
|
|
|
|
mustEscape:
|
|
buf.WriteByte('`')
|
|
for _, c := range original {
|
|
buf.WriteRune(c)
|
|
if c == '`' {
|
|
buf.WriteByte('`')
|
|
}
|
|
}
|
|
buf.WriteByte('`')
|
|
}
|
|
|
|
func compliantName(in string) string {
|
|
var buf bytes.Buffer
|
|
for i, c := range in {
|
|
if !isLetter(uint16(c)) {
|
|
if i == 0 || !isDigit(uint16(c)) {
|
|
buf.WriteByte('_')
|
|
continue
|
|
}
|
|
}
|
|
buf.WriteRune(c)
|
|
}
|
|
return buf.String()
|
|
}
|