In Tengo, all object types (both runtime types and user types) must implement Object interface.
TypeName() string
TypeName method should return the name of the type. Type names are not directly used by the runtime (except when it reports a run-time error), but, it is generally a good idea to keep it short but distinguishable from other types.
String() string
String method should return a string representation of the underlying value. The value returned by String method will be used whenever string formatting for the value is required, most commonly when being converted into String value.
BinaryOp(op token.Token, rhs Object) (res Object, err error)
In Tengo, a type can overload binary operators
(+
, -
, *
, /
, %
, &
, |
, ^
, &^
, >>
, <<
, >
, >=
; note
that <
and <=
operators are not overloadable as they're simply implemented
by switching left-hand side and right-hand side of >
/>=
operator) by
implementing BinaryOp method. BinaryOp method takes the operator op
and the
right-hand side object rhs
, and, should return a resulting value res
.
Error value vs runtime error
If BinaryOp method returns an error err
(the second return value), it will be
treated as a run-time error, which will halt the execution (VM.Run() error
)
and will return the error to the user. All runtime type implementations, for
example, will return an ErrInvalidOperator
error when the given operator is
not supported by the type.
Alternatively the method can return an Error
value as its result res
(the first return value), which will not halt the runtime and will be treated
like any other values. As a dynamically typed language, the receiver (another
expression or statement) can determine how to translate Error
value returned
from binary operator expression.
IsFalsy() bool
IsFalsy method should return true if the underlying value is considered to be falsy.
Equals(o Object) bool
Equals method should return true if the underlying value is considered to be
equal to the underlying value of another object o
. When comparing values of
different types, the runtime does not guarantee or force anything, but, it's
generally a good idea to make the result consistent. For example, a custom
integer type may return true when comparing against String value, but, it
should return the same result for the same inputs.
Copy() Object
Copy method should return a new copy of the object. Builtin function copy
uses this method to copy values. Default implementation of all runtime types
return a deep-copy values, but, it's not a requirement by the runtime.
IndexGet(index Object) (value Object, err error)
IndexGet should take an index Object and return a result Object or an error for indexable objects. Indexable is an object that can take an index and return an object. If a type is indexable, its values support dot selector (value = object.index) and indexer (value = object[index]) syntax.
If Object is not indexable, ErrNotIndexable should be returned as error. If nil is returned as value, it will be converted to Undefined value by the runtime.
If IndexGet
returns an error (err
), the VM will treat it as a run-time
error and ignore the returned value.
Array and Map implementation forces the type of index Object to be Int and String respectively, but, it's not a required behavior of the VM. It is completely okay to take various index types as long as it is consistent.
By convention, Array or Array-like types and Map or Map-like types return
Undefined
value when the key does not exist. But, again, this is not a
required behavior.
IndexSet(index, value Object) error
IndexSet should take an index Object and a value Object for index assignable
objects. Index assignable is an object that can take an index and a value on
the left-hand side of the assignment statement. If a type is index assignable,
its values support assignment using dot selector (object.index = value
) and
indexer (object[index] = value
) in the assignment statements.
If Object is not index assignable, ErrNotIndexAssignable should be returned as error. If an error is returned, it will be treated as a run-time error.
Array and Map implementation forces the type of index Object to be Int and String respectively, but, it's not a required behavior of the VM. It is completely okay to take various index types as long as it is consistent.
If the type is Callable, its values can be invoked as if they were functions. Two functions need to be implemented for Callable objects.
CanCall() bool
CanCall should return whether the Object can be called. When this function returns true, the Object is considered Callable.
Call(args ...Object) (ret Object, err error)
Call should take an arbitrary number of arguments and return a return value and/or an error, which the VM will consider as a run-time error.
If a type is iterable, its values can be used in for-in
statements
(for key, value in object { ... }
). Two functions need to be implemented
for Iterable Objects
CanIterate() bool
CanIterate should return whether the Object can be Iterated.
Iterate() Iterator
The Iterate method should return another object that implements Iterator interface.
Next() bool
Next method should return true if there are more elements to iterate. When used
with for-in
statements, the compiler uses Key and Value methods to populate
the current element's key (or index) and value from the object that this
iterator represents. The runtime will stop iterating in for-in
statement
when this method returns false.
Key() Object
Key method should return a key (or an index) Object for the current element of the underlying object. It should return the same value until Next method is called again. By convention, iterators for the map or map-like objects returns the String key, and, iterators for array or array-like objects returns the Int ndex. But, it's not a requirement by the VM.
Value() Object
Value method should return a value Object for the current element of the underlying object. It should return the same value until Next method is called again.
These are the basic types Tengo runtime supports out of the box:
See Runtime Types for more details on these runtime types.
Users can easily extend and add their own types by implementing the same
Object interface and the
default ObjectImpl
implementation. Tengo runtime will treat them in the
same way as its runtime types with no performance overhead.
Here's an example user type implementation, StringArray
:
type StringArray struct {
tengo.ObjectImpl
Value []string
}
func (o *StringArray) String() string {
return strings.Join(o.Value, ", ")
}
func (o *StringArray) BinaryOp(op token.Token, rhs tengo.Object) (tengo.Object, error) {
if rhs, ok := rhs.(*StringArray); ok {
switch op {
case token.Add:
if len(rhs.Value) == 0 {
return o, nil
}
return &StringArray{Value: append(o.Value, rhs.Value...)}, nil
}
}
return nil, tengo.ErrInvalidOperator
}
func (o *StringArray) IsFalsy() bool {
return len(o.Value) == 0
}
func (o *StringArray) Equals(x tengo.Object) bool {
if x, ok := x.(*StringArray); ok {
if len(o.Value) != len(x.Value) {
return false
}
for i, v := range o.Value {
if v != x.Value[i] {
return false
}
}
return true
}
return false
}
func (o *StringArray) Copy() tengo.Object {
return &StringArray{
Value: append([]string{}, o.Value...),
}
}
func (o *StringArray) TypeName() string {
return "string-array"
}
You can use a user type via either
Script.Add or by directly
manipulating the symbol table and the global variables. Here's an example code
to add StringArray
to the script:
// script that uses 'my_list'
s := tengo.NewScript([]byte(`
fmt := import("fmt")
fmt.println(my_list, ", three")
`))
s.SetImports(stdlib.GetModuleMap("fmt"))
myList := &StringArray{Value: []string{"one", "two"}}
s.Add("my_list", myList) // add StringArray value 'my_list'
s.Run() // prints "one, two, three"
It can also implement IndexGet
and IndexSet
:
func (o *StringArray) IndexGet(index tengo.Object) (tengo.Object, error) {
intIdx, ok := index.(*tengo.Int)
if ok {
if intIdx.Value >= 0 && intIdx.Value < int64(len(o.Value)) {
return &tengo.String{Value: o.Value[intIdx.Value]}, nil
}
return nil, tengo.ErrIndexOutOfBounds
}
strIdx, ok := index.(*tengo.String)
if ok {
for vidx, str := range o.Value {
if strIdx.Value == str {
return &tengo.Int{Value: int64(vidx)}, nil
}
}
return tengo.UndefinedValue, nil
}
return nil, tengo.ErrInvalidIndexType
}
func (o *StringArray) IndexSet(index, value tengo.Object) error {
strVal, ok := tengo.ToString(value)
if !ok {
return tengo.ErrInvalidIndexValueType
}
intIdx, ok := index.(*tengo.Int)
if ok {
if intIdx.Value >= 0 && intIdx.Value < int64(len(o.Value)) {
o.Value[intIdx.Value] = strVal
return nil
}
return tengo.ErrIndexOutOfBounds
}
return tengo.ErrInvalidIndexType
}
If we implement CanCall
and Call
:
func (o *StringArray) CanCall() bool {
return true
}
func (o *StringArray) Call(args ...tengo.Object) (ret tengo.Object, err error) {
if len(args) != 1 {
return nil, tengo.ErrWrongNumArguments
}
s1, ok := tengo.ToString(args[0])
if !ok {
return nil, tengo.ErrInvalidArgumentType{
Name: "first",
Expected: "string",
Found: args[0].TypeName(),
}
}
for i, v := range o.Value {
if v == s1 {
return &tengo.Int{Value: int64(i)}, nil
}
}
return tengo.UndefinedValue, nil
}
Then it can be "invoked":
s := tengo.NewScript([]byte(`
print(my_list("two"))
`))
myList := &StringArray{Value: []string{"one", "two", "three"}}
s.Add("my_list", myList) // add StringArray value 'my_list'
s.Run() // prints "1" (index of "two")
We can also make StringArray
iterable:
func (o *StringArray) CanIterate() bool {
return true
}
func (o *StringArray) Iterate() tengo.Iterator {
return &StringArrayIterator{
strArr: o,
}
}
type StringArrayIterator struct {
tengo.ObjectImpl
strArr *StringArray
idx int
}
func (i *StringArrayIterator) TypeName() string {
return "string-array-iterator"
}
func (i *StringArrayIterator) Next() bool {
i.idx++
return i.idx <= len(i.strArr.Value)
}
func (i *StringArrayIterator) Key() tengo.Object {
return &tengo.Int{Value: int64(i.idx - 1)}
}
func (i *StringArrayIterator) Value() tengo.Object {
return &tengo.String{Value: i.strArr.Value[i.idx-1]}
}
ObjectImpl represents a default Object Implementation. To defined a new value type, one can embed ObjectImpl in their type declarations to avoid implementing all non-significant methods. TypeName() and String() methods still need to be implemented.