// UpdateAggFields adds aggregate function resultfield to select result field list.
func (s *SelectList) UpdateAggFields(expr expression.Expression, tableFields []*field.ResultField) (expression.Expression, error) {
// For aggregate function, the name can be in table or select list.
names := expression.MentionedColumns(expr)
for _, name := range names {
if field.ContainFieldName(name, tableFields, field.DefaultFieldFlag) {
continue
}
if field.ContainFieldName(name, s.ResultFields, field.DefaultFieldFlag) {
continue
}
return nil, errors.Errorf("Unknown column '%s'", name)
}
// We must add aggregate function to hidden select list
// and use a position expression to fetch its value later.
exprName := expr.String()
idx := field.GetResultFieldIndex(exprName, s.ResultFields, field.CheckFieldFlag)
if len(idx) == 0 {
f := &field.Field{Expr: expr}
resultField := &field.ResultField{Name: exprName}
s.AddField(f, resultField)
return &expression.Position{N: len(s.Fields), Name: exprName}, nil
}
// select list has this field, use it directly.
return &expression.Position{N: idx[0] + 1, Name: exprName}, nil
}
func (r *TableDefaultPlan) filter(ctx context.Context, expr expression.Expression, checkColumns bool) (plan.Plan, bool, error) {
if checkColumns {
colNames := expression.MentionedColumns(expr)
// make sure all mentioned column names are in Fields
// if not, e.g. the expr has two table like t1.c1 = t2.c2, we can't use filter
if !field.ContainAllFieldNames(colNames, r.Fields) {
return r, false, nil
}
}
switch x := expr.(type) {
case *expression.BinaryOperation:
return r.filterBinOp(ctx, x)
case *expression.Ident:
return r.filterIdent(ctx, x, true)
case *expression.IsNull:
return r.filterIsNull(ctx, x)
case *expression.UnaryOperation:
if x.Op != '!' {
break
}
if operand, ok := x.V.(*expression.Ident); ok {
return r.filterIdent(ctx, operand, false)
}
}
return r, false, nil
}
// CheckAndUpdateSelectList checks order by fields validity and set hidden fields to selectList.
func (r *OrderByRset) CheckAndUpdateSelectList(selectList *plans.SelectList, tableFields []*field.ResultField) error {
for i, v := range r.By {
if expression.ContainAggregateFunc(v.Expr) {
expr, err := selectList.UpdateAggFields(v.Expr, tableFields)
if err != nil {
return errors.Errorf("%s in 'order clause'", err.Error())
}
r.By[i].Expr = expr
} else {
names := expression.MentionedColumns(v.Expr)
for _, name := range names {
// try to find in select list
// TODO: mysql has confused result for this, see #555.
// now we use select list then order by, later we should make it easier.
if field.ContainFieldName(name, selectList.ResultFields, field.CheckFieldFlag) {
// check ambiguous fields, like `select c1 as c2, c2 from t order by c2`.
if err := field.CheckAmbiguousField(name, selectList.ResultFields, field.DefaultFieldFlag); err != nil {
return errors.Errorf("Column '%s' in order statement is ambiguous", name)
}
continue
}
if !selectList.CloneHiddenField(name, tableFields) {
return errors.Errorf("Unknown column '%s' in 'order clause'", name)
}
}
}
}
return nil
}
func (r *TableDefaultPlan) filterIsNull(ctx context.Context, x *expression.IsNull) (plan.Plan, bool, error) {
if _, ok := x.Expr.(*expression.Ident); !ok {
// if expression is not Ident expression, we cannot use index
// e.g, "(x > null) is not null", (x > null) is a binary expression, we must evaluate it first
return r, false, nil
}
cns := expression.MentionedColumns(x.Expr)
if len(cns) == 0 {
return r, false, nil
}
cn := cns[0]
t := r.T
ix := t.FindIndexByColName(cn)
if ix == nil { // Column cn has no index.
return r, false, nil
}
var spans []*indexSpan
if x.Not {
spans = toSpans(opcode.GE, minNotNullVal)
} else {
spans = toSpans(opcode.EQ, nil)
}
return &indexPlan{
src: t,
colName: cn,
idxName: ix.Name.L,
idx: ix.X,
spans: spans,
}, true, nil
}
// Plan gets SelectExprPlan.
func (r *SelectFromDualRset) Plan(ctx context.Context) (plan.Plan, error) {
// field cannot contain identifier
for _, f := range r.Fields {
if cs := expression.MentionedColumns(f.Expr); len(cs) > 0 {
return nil, errors.Errorf("Unknown column '%s' in 'field list'", cs[0])
}
}
return &plans.SelectFromDualPlan{Fields: r.Fields}, nil
}
// Plan get SrcPlan/OrderByDefaultPlan.
// If src is NullPlan or order by fields are empty, then gets SrcPlan.
// Default gets OrderByDefaultPlan.
func (r *OrderByRset) Plan(ctx context.Context) (plan.Plan, error) {
if _, ok := r.Src.(*plans.NullPlan); ok {
return r.Src, nil
}
var (
by []expression.Expression
ascs []bool
)
fields := r.Src.GetFields()
for i := range r.By {
e := r.By[i].Expr
if v, ok := e.(expression.Value); ok {
var (
position int
isPosition = true
)
switch u := v.Val.(type) {
case int64:
position = int(u)
case uint64:
position = int(u)
default:
isPosition = false
// only const value
}
if isPosition {
if position < 1 || position > len(fields) {
return nil, errors.Errorf("Unknown column '%d' in 'order clause'", position)
}
// use Position expression for the associated field.
r.By[i].Expr = &expression.Position{N: position}
}
} else {
colNames := expression.MentionedColumns(e)
if err := field.CheckAllFieldNames(colNames, fields, field.CheckFieldFlag); err != nil {
return nil, errors.Trace(err)
}
}
by = append(by, r.By[i].Expr)
ascs = append(ascs, r.By[i].Asc)
}
// if no order by key, use src plan instead.
if len(by) == 0 {
return r.Src, nil
}
return &plans.OrderByDefaultPlan{By: by, Ascs: ascs, Src: r.Src,
SelectList: r.SelectList}, nil
}
// Plan gets SelectExprPlan.
func (r *SelectFromDualRset) Plan(ctx context.Context) (plan.Plan, error) {
// field cannot contain identifier
for _, f := range r.Fields {
if cs := expression.MentionedColumns(f.Expr); len(cs) > 0 {
// TODO: check in outer query, like select * from t where t.c = (select c limit 1);
return nil, errors.Errorf("Unknown column '%s' in 'field list'", cs[0])
}
}
return &plans.SelectFromDualPlan{Fields: r.Fields}, nil
}
// CheckAndUpdateSelectList checks having fields validity and set hidden fields to selectList.
func (r *HavingRset) CheckAndUpdateSelectList(selectList *plans.SelectList, groupBy []expression.Expression, tableFields []*field.ResultField) error {
if expression.ContainAggregateFunc(r.Expr) {
expr, err := selectList.UpdateAggFields(r.Expr, tableFields)
if err != nil {
return errors.Errorf("%s in 'having clause'", err.Error())
}
r.Expr = expr
} else {
// having can only contain group by column and select list, e.g,
// `select c1 from t group by c2 having c3 > 0` is invalid,
// because c3 is not in group by and select list.
names := expression.MentionedColumns(r.Expr)
for _, name := range names {
found := false
// check name whether in select list.
// notice that `select c1 as c2 from t group by c1, c2, c3 having c2 > c3`,
// will use t.c2 not t.c1 here.
if field.ContainFieldName(name, selectList.ResultFields, field.OrgFieldNameFlag) {
continue
}
if field.ContainFieldName(name, selectList.ResultFields, field.FieldNameFlag) {
if field.ContainFieldName(name, tableFields, field.OrgFieldNameFlag) {
selectList.CloneHiddenField(name, tableFields)
}
continue
}
// check name whether in group by.
// group by must only have column name, e.g,
// `select c1 from t group by c2 having c2 > 0` is valid,
// but `select c1 from t group by c2 + 1 having c2 > 0` is invalid.
for _, by := range groupBy {
if !field.CheckFieldsEqual(name, by.String()) {
continue
}
// if name is not in table fields, it will get an unknown field error in GroupByRset,
// so no need to check return value.
selectList.CloneHiddenField(name, tableFields)
found = true
break
}
if !found {
return errors.Errorf("Unknown column '%s' in 'having clause'", name)
}
}
}
return nil
}
func (r *TableDefaultPlan) filterIsNull(ctx context.Context, x *expression.IsNull) (plan.Plan, bool, error) {
if _, ok := x.Expr.(*expression.Ident); !ok {
// if expression is not Ident expression, we cannot use index
// e.g, "(x > null) is not null", (x > null) is a binary expression, we must evaluate it first
return r, false, nil
}
cns := expression.MentionedColumns(x.Expr)
if len(cns) == 0 {
return r, false, nil
}
cn := cns[0]
t := r.T
col := column.FindCol(t.Cols(), cn)
if col == nil {
return r, false, nil
}
if col.IsPKHandleColumn(t.Meta()) {
if x.Not {
// PK handle column never be null.
return r, false, nil
}
return &NullPlan{
Fields: r.Fields,
}, true, nil
} else if r.rangeScan {
return r, false, nil
}
var spans []*indexSpan
if x.Not {
spans = toSpans(opcode.GE, minNotNullVal, nil)
} else {
spans = toSpans(opcode.EQ, nil, nil)
}
ix := t.FindIndexByColName(cn)
if ix == nil { // Column cn has no index.
return r, false, nil
}
return &indexPlan{
src: t,
col: col,
unique: ix.Unique,
idxName: ix.Name.L,
idx: ix.X,
spans: spans,
}, true, nil
}
func (r *WhereRset) planIsNull(ctx context.Context, x *expression.IsNull) (plan.Plan, error) {
p := r.Src
cns := expression.MentionedColumns(x.Expr)
if len(cns) == 0 {
return &plans.FilterDefaultPlan{Plan: p, Expr: x}, nil
}
p2, filtered, err := p.Filter(ctx, x)
if err != nil {
return nil, err
}
if filtered {
return p2, nil
}
return &plans.FilterDefaultPlan{Plan: p, Expr: x}, nil
}
// CheckAndUpdateSelectList checks order by fields validity and set hidden fields to selectList.
func (r *OrderByRset) CheckAndUpdateSelectList(selectList *plans.SelectList, tableFields []*field.ResultField) error {
for i, v := range r.By {
if expression.ContainAggregateFunc(v.Expr) {
expr, err := selectList.UpdateAggFields(v.Expr, tableFields)
if err != nil {
return errors.Errorf("%s in 'order clause'", err.Error())
}
r.By[i].Expr = expr
} else {
if _, err := selectList.CheckReferAmbiguous(v.Expr); err != nil {
return errors.Errorf("Column '%s' in order statement is ambiguous", v.Expr)
}
// TODO: check more ambiguous case
// Order by ambiguous rule:
// select c1 as a, c2 as a from t order by a is ambiguous
// select c1 as a, c2 as a from t order by a + 1 is ambiguous
// select c1 as c2, c2 from t order by c2 is ambiguous
// select c1 as c2, c2 from t order by c2 + 1 is ambiguous
// TODO: use vistor to refactor all and combine following plan check.
names := expression.MentionedColumns(v.Expr)
for _, name := range names {
// try to find in select list
// TODO: mysql has confused result for this, see #555.
// now we use select list then order by, later we should make it easier.
if field.ContainFieldName(name, selectList.ResultFields, field.CheckFieldFlag) {
continue
}
if !selectList.CloneHiddenField(name, tableFields) {
return errors.Errorf("Unknown column '%s' in 'order clause'", name)
}
}
}
}
return nil
}
// ResolveSelectList gets fields and result fields from selectFields and srcFields,
// including field validity check and wildcard field processing.
func ResolveSelectList(selectFields []*field.Field, srcFields []*field.ResultField) (*SelectList, error) {
selectList := &SelectList{
Fields: make([]*field.Field, 0, len(selectFields)),
ResultFields: make([]*field.ResultField, 0, len(selectFields)),
AggFields: make(map[int]struct{}),
FromFields: srcFields,
}
wildcardNum := 0
for _, v := range selectFields {
// Check metioned field.
names := expression.MentionedColumns(v.Expr)
if len(names) == 0 {
selectList.AddField(v, nil)
continue
}
// Check wildcard field.
name := names[0]
table, ok, err := field.CheckWildcardField(name)
if err != nil {
return nil, err
}
if ok {
// Check unqualified wildcard field number,
// like `select *, * from t`.
if table == "" {
wildcardNum++
if wildcardNum > 1 {
return nil, errors.Errorf("wildcard field exist more than once")
}
}
selectList.updateFields(table, srcFields)
continue
}
var result *field.ResultField
for _, name := range names {
idx := field.GetResultFieldIndex(name, srcFields, field.DefaultFieldFlag)
if len(idx) > 1 {
return nil, errors.Errorf("ambiguous field %s", name)
}
// TODO: must check in outer query too.
if len(idx) == 0 {
return nil, errors.Errorf("unknown field %s", name)
}
}
if _, ok := v.Expr.(*expression.Ident); ok {
// Field is ident.
if result, err = field.CloneFieldByName(name, srcFields, field.DefaultFieldFlag); err != nil {
return nil, errors.Trace(err)
}
// Use alias name
if len(v.AsName) > 0 {
result.Name = v.AsName
} else {
// use field identifier name directly, but not contain qualified name.
// e.g, select t.c will only return c as the column name.
s := v.Expr.String()
n := strings.LastIndexByte(s, '.')
if n == -1 {
result.Name = s
} else {
result.Name = s[n+1:]
}
}
}
selectList.AddField(v, result)
}
selectList.HiddenFieldOffset = len(selectList.Fields)
selectList.resolveAggFields()
if selectList.HiddenFieldOffset == 0 {
return nil, errors.Errorf("invalid empty select fields")
}
return selectList, nil
}
// Plan get SrcPlan/OrderByDefaultPlan.
// If src is NullPlan or order by fields are empty, then gets SrcPlan.
// Default gets OrderByDefaultPlan.
func (r *OrderByRset) Plan(ctx context.Context) (plan.Plan, error) {
if _, ok := r.Src.(*plans.NullPlan); ok {
return r.Src, nil
}
var (
by []expression.Expression
ascs []bool
)
fields := r.Src.GetFields()
for i := range r.By {
e := r.By[i].Expr
if v, ok := e.(expression.Value); ok {
var (
position int
isPosition = true
)
switch u := v.Val.(type) {
case int64:
position = int(u)
case uint64:
position = int(u)
default:
isPosition = false
// only const value
}
if isPosition {
if position < 1 || position > len(fields) {
return nil, errors.Errorf("Unknown column '%d' in 'order clause'", position)
}
// use Position expression for the associated field.
r.By[i].Expr = &expression.Position{N: position}
}
} else {
// Don't check ambiguous here, only check field exists or not.
// TODO: use visitor to refactor.
colNames := expression.MentionedColumns(e)
for _, name := range colNames {
if idx := field.GetResultFieldIndex(name, r.SelectList.ResultFields, field.DefaultFieldFlag); len(idx) == 0 {
// find in from
if idx = field.GetResultFieldIndex(name, r.SelectList.FromFields, field.DefaultFieldFlag); len(idx) == 0 {
return nil, errors.Errorf("unknown field %s", name)
}
}
}
}
by = append(by, r.By[i].Expr)
ascs = append(ascs, r.By[i].Asc)
}
// if no order by key, use src plan instead.
if len(by) == 0 {
return r.Src, nil
}
return &plans.OrderByDefaultPlan{By: by, Ascs: ascs, Src: r.Src,
SelectList: r.SelectList}, nil
}
// ResolveSelectList gets fields and result fields from selectFields and srcFields,
// including field validity check and wildcard field processing.
func ResolveSelectList(selectFields []*field.Field, srcFields []*field.ResultField) (*SelectList, error) {
selectList := &SelectList{
Fields: make([]*field.Field, 0, len(selectFields)),
ResultFields: make([]*field.ResultField, 0, len(selectFields)),
AggFields: make(map[int]struct{}),
FromFields: srcFields,
}
wildcardNum := 0
for _, v := range selectFields {
// Check metioned field.
names := expression.MentionedColumns(v.Expr)
if len(names) == 0 {
selectList.AddField(v, nil)
continue
}
// Check wildcard field.
name := names[0]
table, ok, err := field.CheckWildcardField(name)
if err != nil {
return nil, err
}
if ok {
// Check unqualified wildcard field number,
// like `select *, * from t`.
if table == "" {
wildcardNum++
if wildcardNum > 1 {
return nil, errors.Errorf("wildcard field exist more than once")
}
}
selectList.updateFields(table, srcFields)
continue
}
var result *field.ResultField
if err = field.CheckAllFieldNames(names, srcFields, field.DefaultFieldFlag); err != nil {
return nil, errors.Trace(err)
}
if _, ok := v.Expr.(*expression.Ident); ok {
// Field is ident.
if result, err = field.CloneFieldByName(name, srcFields, field.DefaultFieldFlag); err != nil {
return nil, errors.Trace(err)
}
// Maybe alias name or only column name.
if !expression.IsQualified(v.Name) {
result.Name = v.Name
}
} else {
// The field is not an ident, maybe binary expression,
// like `select c1 + c2`, or `select c1 + 10`, etc.
result = &field.ResultField{Name: v.Name}
}
selectList.AddField(v, result)
}
selectList.HiddenFieldOffset = len(selectList.Fields)
selectList.resolveAggFields()
if selectList.HiddenFieldOffset == 0 {
return nil, errors.Errorf("invalid empty select fields")
}
return selectList, nil
}
// Plan gets GroupByDefaultPlan.
func (r *GroupByRset) Plan(ctx context.Context) (plan.Plan, error) {
fields := r.SelectList.Fields
r.SelectList.AggFields = GetAggFields(fields)
aggFields := r.SelectList.AggFields
for i, e := range r.By {
if v, ok := e.(expression.Value); ok {
var position int
switch u := v.Val.(type) {
case int64:
position = int(u)
case uint64:
position = int(u)
default:
continue
}
if position < 1 || position > len(fields) {
return nil, errors.Errorf("Unknown column '%d' in 'group statement'", position)
}
index := position - 1
if _, ok := aggFields[index]; ok {
return nil, errors.Errorf("Can't group on '%s'", fields[index])
}
// use Position expression for the associated field.
r.By[i] = &expression.Position{N: position}
} else {
index, err := r.SelectList.CheckReferAmbiguous(e)
if err != nil {
return nil, errors.Errorf("Column '%s' in group statement is ambiguous", e)
} else if _, ok := aggFields[index]; ok {
return nil, errors.Errorf("Can't group on '%s'", e)
}
// TODO: check more ambiguous case
// Group by ambiguous rule:
// select c1 as a, c2 as a from t group by a is ambiguous
// select c1 as a, c2 as a from t group by a + 1 is ambiguous
// select c1 as c2, c2 from t group by c2 is ambiguous
// select c1 as c2, c2 from t group by c2 + 1 is ambiguous
// TODO: use visitor to check aggregate function
names := expression.MentionedColumns(e)
for _, name := range names {
indices := field.GetFieldIndex(name, fields[0:r.SelectList.HiddenFieldOffset], field.DefaultFieldFlag)
if len(indices) == 1 {
// check reference to aggregate function, like `select c1, count(c1) as b from t group by b + 1`.
index := indices[0]
if _, ok := aggFields[index]; ok {
return nil, errors.Errorf("Reference '%s' not supported (reference to group function)", name)
}
}
}
// group by should be an expression, a qualified field name or a select field position,
// but can not contain any aggregate function.
if e := r.By[i]; expression.ContainAggregateFunc(e) {
return nil, errors.Errorf("group by cannot contain aggregate function %s", e.String())
}
}
}
return &plans.GroupByDefaultPlan{By: r.By, Src: r.Src,
SelectList: r.SelectList}, nil
}
// Plan gets GroupByDefaultPlan.
func (r *GroupByRset) Plan(ctx context.Context) (plan.Plan, error) {
fields := r.SelectList.Fields
resultfields := r.SelectList.ResultFields
srcFields := r.Src.GetFields()
r.SelectList.AggFields = GetAggFields(fields)
aggFields := r.SelectList.AggFields
for i, e := range r.By {
if v, ok := e.(expression.Value); ok {
var position int
switch u := v.Val.(type) {
case int64:
position = int(u)
case uint64:
position = int(u)
default:
continue
}
if position < 1 || position > len(fields) {
return nil, errors.Errorf("Unknown column '%d' in 'group statement'", position)
}
index := position - 1
if _, ok := aggFields[index]; ok {
return nil, errors.Errorf("Can't group on '%s'", fields[index].Name)
}
// use Position expression for the associated field.
r.By[i] = &expression.Position{N: position}
} else {
names := expression.MentionedColumns(e)
for _, name := range names {
if field.ContainFieldName(name, srcFields, field.DefaultFieldFlag) {
// check whether column is qualified, like `select t.c1 c1, t.c2 from t group by t.c1, t.c2`
// no need to check ambiguous field.
if expression.IsQualified(name) {
continue
}
// check ambiguous fields, like `select c1 as c2, c2 from t group by c2`.
if err := field.CheckAmbiguousField(name, resultfields, field.DefaultFieldFlag); err == nil {
continue
}
}
// check reference to group function name
indices := field.GetFieldIndex(name, fields[0:r.SelectList.HiddenFieldOffset], field.CheckFieldFlag)
if len(indices) > 1 {
// check ambiguous fields, like `select c1 as a, c2 as a from t group by a`,
// notice that `select c2 as c2, c2 as c2 from t group by c2;` is valid.
if r.HasAmbiguousField(indices, fields[0:r.SelectList.HiddenFieldOffset]) {
return nil, errors.Errorf("Column '%s' in group statement is ambiguous", name)
}
} else if len(indices) == 1 {
// check reference to aggregate function, like `select c1, count(c1) as b from t group by b + 1`.
index := indices[0]
if _, ok := aggFields[index]; ok {
return nil, errors.Errorf("Reference '%s' not supported (reference to group function)", name)
}
}
}
// group by should be an expression, a qualified field name or a select field position,
// but can not contain any aggregate function.
if e := r.By[i]; expression.ContainAggregateFunc(e) {
return nil, errors.Errorf("group by cannot contain aggregate function %s", e.String())
}
}
}
return &plans.GroupByDefaultPlan{By: r.By, Src: r.Src,
SelectList: r.SelectList}, nil
}
