You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

530 lines
12 KiB

// Copyright 2017 The Bazel Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Package syntax provides a Starlark parser and abstract syntax tree.
package syntax // import "go.starlark.net/syntax"
// A Node is a node in a Starlark syntax tree.
type Node interface {
// Span returns the start and end position of the expression.
Span() (start, end Position)
// Comments returns the comments associated with this node.
// It returns nil if RetainComments was not specified during parsing,
// or if AllocComments was not called.
Comments() *Comments
// AllocComments allocates a new Comments node if there was none.
// This makes possible to add new comments using Comments() method.
AllocComments()
}
// A Comment represents a single # comment.
type Comment struct {
Start Position
Text string // without trailing newline
}
// Comments collects the comments associated with an expression.
type Comments struct {
Before []Comment // whole-line comments before this expression
Suffix []Comment // end-of-line comments after this expression (up to 1)
// For top-level expressions only, After lists whole-line
// comments following the expression.
After []Comment
}
// A commentsRef is a possibly-nil reference to a set of comments.
// A commentsRef is embedded in each type of syntax node,
// and provides its Comments and AllocComments methods.
type commentsRef struct{ ref *Comments }
// Comments returns the comments associated with a syntax node,
// or nil if AllocComments has not yet been called.
func (cr commentsRef) Comments() *Comments { return cr.ref }
// AllocComments enables comments to be associated with a syntax node.
func (cr *commentsRef) AllocComments() {
if cr.ref == nil {
cr.ref = new(Comments)
}
}
// Start returns the start position of the expression.
func Start(n Node) Position {
start, _ := n.Span()
return start
}
// End returns the end position of the expression.
func End(n Node) Position {
_, end := n.Span()
return end
}
// A File represents a Starlark file.
type File struct {
commentsRef
Path string
Stmts []Stmt
Module interface{} // a *resolve.Module, set by resolver
}
func (x *File) Span() (start, end Position) {
if len(x.Stmts) == 0 {
return
}
start, _ = x.Stmts[0].Span()
_, end = x.Stmts[len(x.Stmts)-1].Span()
return start, end
}
// A Stmt is a Starlark statement.
type Stmt interface {
Node
stmt()
}
func (*AssignStmt) stmt() {}
func (*BranchStmt) stmt() {}
func (*DefStmt) stmt() {}
func (*ExprStmt) stmt() {}
func (*ForStmt) stmt() {}
func (*WhileStmt) stmt() {}
func (*IfStmt) stmt() {}
func (*LoadStmt) stmt() {}
func (*ReturnStmt) stmt() {}
// An AssignStmt represents an assignment:
// x = 0
// x, y = y, x
// x += 1
type AssignStmt struct {
commentsRef
OpPos Position
Op Token // = EQ | {PLUS,MINUS,STAR,PERCENT}_EQ
LHS Expr
RHS Expr
}
func (x *AssignStmt) Span() (start, end Position) {
start, _ = x.LHS.Span()
_, end = x.RHS.Span()
return
}
// A DefStmt represents a function definition.
type DefStmt struct {
commentsRef
Def Position
Name *Ident
Params []Expr // param = ident | ident=expr | * | *ident | **ident
Body []Stmt
Function interface{} // a *resolve.Function, set by resolver
}
func (x *DefStmt) Span() (start, end Position) {
_, end = x.Body[len(x.Body)-1].Span()
return x.Def, end
}
// An ExprStmt is an expression evaluated for side effects.
type ExprStmt struct {
commentsRef
X Expr
}
func (x *ExprStmt) Span() (start, end Position) {
return x.X.Span()
}
// An IfStmt is a conditional: If Cond: True; else: False.
// 'elseif' is desugared into a chain of IfStmts.
type IfStmt struct {
commentsRef
If Position // IF or ELIF
Cond Expr
True []Stmt
ElsePos Position // ELSE or ELIF
False []Stmt // optional
}
func (x *IfStmt) Span() (start, end Position) {
body := x.False
if body == nil {
body = x.True
}
_, end = body[len(body)-1].Span()
return x.If, end
}
// A LoadStmt loads another module and binds names from it:
// load(Module, "x", y="foo").
//
// The AST is slightly unfaithful to the concrete syntax here because
// Starlark's load statement, so that it can be implemented in Python,
// binds some names (like y above) with an identifier and some (like x)
// without. For consistency we create fake identifiers for all the
// strings.
type LoadStmt struct {
commentsRef
Load Position
Module *Literal // a string
From []*Ident // name defined in loading module
To []*Ident // name in loaded module
Rparen Position
}
func (x *LoadStmt) Span() (start, end Position) {
return x.Load, x.Rparen
}
// ModuleName returns the name of the module loaded by this statement.
func (x *LoadStmt) ModuleName() string { return x.Module.Value.(string) }
// A BranchStmt changes the flow of control: break, continue, pass.
type BranchStmt struct {
commentsRef
Token Token // = BREAK | CONTINUE | PASS
TokenPos Position
}
func (x *BranchStmt) Span() (start, end Position) {
return x.TokenPos, x.TokenPos.add(x.Token.String())
}
// A ReturnStmt returns from a function.
type ReturnStmt struct {
commentsRef
Return Position
Result Expr // may be nil
}
func (x *ReturnStmt) Span() (start, end Position) {
if x.Result == nil {
return x.Return, x.Return.add("return")
}
_, end = x.Result.Span()
return x.Return, end
}
// An Expr is a Starlark expression.
type Expr interface {
Node
expr()
}
func (*BinaryExpr) expr() {}
func (*CallExpr) expr() {}
func (*Comprehension) expr() {}
func (*CondExpr) expr() {}
func (*DictEntry) expr() {}
func (*DictExpr) expr() {}
func (*DotExpr) expr() {}
func (*Ident) expr() {}
func (*IndexExpr) expr() {}
func (*LambdaExpr) expr() {}
func (*ListExpr) expr() {}
func (*Literal) expr() {}
func (*ParenExpr) expr() {}
func (*SliceExpr) expr() {}
func (*TupleExpr) expr() {}
func (*UnaryExpr) expr() {}
// An Ident represents an identifier.
type Ident struct {
commentsRef
NamePos Position
Name string
Binding interface{} // a *resolver.Binding, set by resolver
}
func (x *Ident) Span() (start, end Position) {
return x.NamePos, x.NamePos.add(x.Name)
}
// A Literal represents a literal string or number.
type Literal struct {
commentsRef
Token Token // = STRING | BYTES | INT | FLOAT
TokenPos Position
Raw string // uninterpreted text
Value interface{} // = string | int64 | *big.Int | float64
}
func (x *Literal) Span() (start, end Position) {
return x.TokenPos, x.TokenPos.add(x.Raw)
}
// A ParenExpr represents a parenthesized expression: (X).
type ParenExpr struct {
commentsRef
Lparen Position
X Expr
Rparen Position
}
func (x *ParenExpr) Span() (start, end Position) {
return x.Lparen, x.Rparen.add(")")
}
// A CallExpr represents a function call expression: Fn(Args).
type CallExpr struct {
commentsRef
Fn Expr
Lparen Position
Args []Expr // arg = expr | ident=expr | *expr | **expr
Rparen Position
}
func (x *CallExpr) Span() (start, end Position) {
start, _ = x.Fn.Span()
return start, x.Rparen.add(")")
}
// A DotExpr represents a field or method selector: X.Name.
type DotExpr struct {
commentsRef
X Expr
Dot Position
NamePos Position
Name *Ident
}
func (x *DotExpr) Span() (start, end Position) {
start, _ = x.X.Span()
_, end = x.Name.Span()
return
}
// A Comprehension represents a list or dict comprehension:
// [Body for ... if ...] or {Body for ... if ...}
type Comprehension struct {
commentsRef
Curly bool // {x:y for ...} or {x for ...}, not [x for ...]
Lbrack Position
Body Expr
Clauses []Node // = *ForClause | *IfClause
Rbrack Position
}
func (x *Comprehension) Span() (start, end Position) {
return x.Lbrack, x.Rbrack.add("]")
}
// A ForStmt represents a loop: for Vars in X: Body.
type ForStmt struct {
commentsRef
For Position
Vars Expr // name, or tuple of names
X Expr
Body []Stmt
}
func (x *ForStmt) Span() (start, end Position) {
_, end = x.Body[len(x.Body)-1].Span()
return x.For, end
}
// A WhileStmt represents a while loop: while X: Body.
type WhileStmt struct {
commentsRef
While Position
Cond Expr
Body []Stmt
}
func (x *WhileStmt) Span() (start, end Position) {
_, end = x.Body[len(x.Body)-1].Span()
return x.While, end
}
// A ForClause represents a for clause in a list comprehension: for Vars in X.
type ForClause struct {
commentsRef
For Position
Vars Expr // name, or tuple of names
In Position
X Expr
}
func (x *ForClause) Span() (start, end Position) {
_, end = x.X.Span()
return x.For, end
}
// An IfClause represents an if clause in a list comprehension: if Cond.
type IfClause struct {
commentsRef
If Position
Cond Expr
}
func (x *IfClause) Span() (start, end Position) {
_, end = x.Cond.Span()
return x.If, end
}
// A DictExpr represents a dictionary literal: { List }.
type DictExpr struct {
commentsRef
Lbrace Position
List []Expr // all *DictEntrys
Rbrace Position
}
func (x *DictExpr) Span() (start, end Position) {
return x.Lbrace, x.Rbrace.add("}")
}
// A DictEntry represents a dictionary entry: Key: Value.
// Used only within a DictExpr.
type DictEntry struct {
commentsRef
Key Expr
Colon Position
Value Expr
}
func (x *DictEntry) Span() (start, end Position) {
start, _ = x.Key.Span()
_, end = x.Value.Span()
return start, end
}
// A LambdaExpr represents an inline function abstraction.
//
// Although they may be added in future, lambda expressions are not
// currently part of the Starlark spec, so their use is controlled by the
// resolver.AllowLambda flag.
type LambdaExpr struct {
commentsRef
Lambda Position
Params []Expr // param = ident | ident=expr | * | *ident | **ident
Body Expr
Function interface{} // a *resolve.Function, set by resolver
}
func (x *LambdaExpr) Span() (start, end Position) {
_, end = x.Body.Span()
return x.Lambda, end
}
// A ListExpr represents a list literal: [ List ].
type ListExpr struct {
commentsRef
Lbrack Position
List []Expr
Rbrack Position
}
func (x *ListExpr) Span() (start, end Position) {
return x.Lbrack, x.Rbrack.add("]")
}
// CondExpr represents the conditional: X if COND else ELSE.
type CondExpr struct {
commentsRef
If Position
Cond Expr
True Expr
ElsePos Position
False Expr
}
func (x *CondExpr) Span() (start, end Position) {
start, _ = x.True.Span()
_, end = x.False.Span()
return start, end
}
// A TupleExpr represents a tuple literal: (List).
type TupleExpr struct {
commentsRef
Lparen Position // optional (e.g. in x, y = 0, 1), but required if List is empty
List []Expr
Rparen Position
}
func (x *TupleExpr) Span() (start, end Position) {
if x.Lparen.IsValid() {
return x.Lparen, x.Rparen
} else {
return Start(x.List[0]), End(x.List[len(x.List)-1])
}
}
// A UnaryExpr represents a unary expression: Op X.
//
// As a special case, UnaryOp{Op:Star} may also represent
// the star parameter in def f(*args) or def f(*, x).
type UnaryExpr struct {
commentsRef
OpPos Position
Op Token
X Expr // may be nil if Op==STAR
}
func (x *UnaryExpr) Span() (start, end Position) {
if x.X != nil {
_, end = x.X.Span()
} else {
end = x.OpPos.add("*")
}
return x.OpPos, end
}
// A BinaryExpr represents a binary expression: X Op Y.
//
// As a special case, BinaryExpr{Op:EQ} may also
// represent a named argument in a call f(k=v)
// or a named parameter in a function declaration
// def f(param=default).
type BinaryExpr struct {
commentsRef
X Expr
OpPos Position
Op Token
Y Expr
}
func (x *BinaryExpr) Span() (start, end Position) {
start, _ = x.X.Span()
_, end = x.Y.Span()
return start, end
}
// A SliceExpr represents a slice or substring expression: X[Lo:Hi:Step].
type SliceExpr struct {
commentsRef
X Expr
Lbrack Position
Lo, Hi, Step Expr // all optional
Rbrack Position
}
func (x *SliceExpr) Span() (start, end Position) {
start, _ = x.X.Span()
return start, x.Rbrack
}
// An IndexExpr represents an index expression: X[Y].
type IndexExpr struct {
commentsRef
X Expr
Lbrack Position
Y Expr
Rbrack Position
}
func (x *IndexExpr) Span() (start, end Position) {
start, _ = x.X.Span()
return start, x.Rbrack
}