neo-go/pkg/compiler/codegen.go

2338 lines
63 KiB
Go
Raw Normal View History

package compiler
import (
"encoding/binary"
"errors"
2019-11-22 14:16:52 +00:00
"fmt"
"go/ast"
"go/constant"
"go/token"
"go/types"
"math"
"sort"
"strings"
"github.com/nspcc-dev/neo-go/pkg/encoding/address"
"github.com/nspcc-dev/neo-go/pkg/io"
"github.com/nspcc-dev/neo-go/pkg/smartcontract"
"github.com/nspcc-dev/neo-go/pkg/util"
"github.com/nspcc-dev/neo-go/pkg/util/bitfield"
"github.com/nspcc-dev/neo-go/pkg/vm"
"github.com/nspcc-dev/neo-go/pkg/vm/emit"
"github.com/nspcc-dev/neo-go/pkg/vm/opcode"
"github.com/nspcc-dev/neo-go/pkg/vm/stackitem"
"golang.org/x/tools/go/loader" //nolint:staticcheck // SA1019: package golang.org/x/tools/go/loader is deprecated
)
type codegen struct {
// Information about the program with all its dependencies.
buildInfo *buildInfo
2019-10-22 14:56:03 +00:00
// prog holds the output buffer.
prog *io.BufBinWriter
2019-10-22 14:56:03 +00:00
// Type information.
typeInfo *types.Info
2021-02-04 12:41:00 +00:00
// pkgInfoInline is stack of type information for packages containing inline functions.
pkgInfoInline []*loader.PackageInfo
// A mapping of func identifiers with their scope.
funcs map[string]*funcScope
// A mapping of lambda functions into their scope.
lambda map[string]*funcScope
// reverseOffsetMap maps function offsets to a local variable count.
reverseOffsetMap map[int]nameWithLocals
// Current funcScope being converted.
scope *funcScope
globals map[string]int
// staticVariables contains global (static in NDX-DN11) variable names and types.
staticVariables []string
// initVariables contains variables local to `_initialize` method.
initVariables []string
// deployVariables contains variables local to `_initialize` method.
deployVariables []string
// A mapping from label's names to their ids.
labels map[labelWithType]uint16
// A list of nested label names together with evaluation stack depth.
labelList []labelWithStackSize
// inlineLabelOffsets contains size of labelList at the start of inline call processing.
// For such calls we need to drop only newly created part of stack.
inlineLabelOffsets []int
// globalInlineCount contains amount of auxiliary variables introduced by
// function inlining during global variables initialization.
globalInlineCount int
// A label for the for-loop being currently visited.
currentFor string
// A label for the switch statement being visited.
currentSwitch string
// A label to be used in the next statement.
nextLabel string
// sequencePoints is mapping from method name to a slice
// containing info about mapping from opcode's offset
// to a text span in the source file.
sequencePoints map[string][]DebugSeqPoint
// initEndOffset specifies the end of the initialization method.
initEndOffset int
2020-10-02 10:16:02 +00:00
// deployEndOffset specifies the end of the deployment method.
deployEndOffset int
// importMap contains mapping from package aliases to full package names for the current file.
importMap map[string]string
// constMap contains constants from foreign packages.
constMap map[string]types.TypeAndValue
// currPkg is current package being processed.
currPkg *types.Package
// mainPkg is a main package metadata.
mainPkg *loader.PackageInfo
// packages contains packages in the order they were loaded.
packages []string
2020-08-21 12:37:46 +00:00
// exceptionIndex is the index of static slot where exception is stored.
exceptionIndex int
// documents contains paths to all files used by the program.
documents []string
// docIndex maps file path to an index in documents array.
docIndex map[string]int
// emittedEvents contains all events emitted by contract.
emittedEvents map[string][][]string
// invokedContracts contains invoked methods of other contracts.
invokedContracts map[util.Uint160][]string
// Label table for recording jump destinations.
l []int
}
type labelOffsetType byte
const (
labelStart labelOffsetType = iota // labelStart is a default label type
labelEnd // labelEnd is a type for labels that are targets for break
labelPost // labelPost is a type for labels that are targets for continue
)
type labelWithType struct {
name string
typ labelOffsetType
}
type labelWithStackSize struct {
name string
sz int
}
type nameWithLocals struct {
name string
count int
}
type varType int
const (
varGlobal varType = iota
varLocal
varArgument
)
// newLabel creates a new label to jump to.
func (c *codegen) newLabel() (l uint16) {
li := len(c.l)
if li > math.MaxUint16 {
c.prog.Err = errors.New("label number is too big")
return
}
l = uint16(li)
c.l = append(c.l, -1)
return
}
// newNamedLabel creates a new label with a specified name.
func (c *codegen) newNamedLabel(typ labelOffsetType, name string) (l uint16) {
l = c.newLabel()
lt := labelWithType{name: name, typ: typ}
c.labels[lt] = l
return
}
func (c *codegen) setLabel(l uint16) {
c.l[l] = c.pc() + 1
}
2019-10-22 14:56:03 +00:00
// pc returns the program offset off the last instruction.
func (c *codegen) pc() int {
return c.prog.Len() - 1
}
func (c *codegen) emitLoadConst(t types.TypeAndValue) {
if c.prog.Err != nil {
return
}
2020-05-13 15:09:55 +00:00
typ, ok := t.Type.Underlying().(*types.Basic)
if !ok {
2019-11-22 14:16:52 +00:00
c.prog.Err = fmt.Errorf("compiler doesn't know how to convert this constant: %v", t)
return
}
switch typ.Kind() {
case types.Int, types.UntypedInt, types.Uint,
2020-05-13 07:53:11 +00:00
types.Int8, types.Uint8,
types.Int16, types.Uint16,
types.Int32, types.Uint32, types.Int64, types.Uint64:
val, _ := constant.Int64Val(t.Value)
emit.Int(c.prog.BinWriter, val)
case types.String, types.UntypedString:
val := constant.StringVal(t.Value)
emit.String(c.prog.BinWriter, val)
case types.Bool, types.UntypedBool:
val := constant.BoolVal(t.Value)
emit.Bool(c.prog.BinWriter, val)
default:
c.prog.Err = fmt.Errorf("compiler doesn't know how to convert this basic type: %v", t)
return
}
}
func (c *codegen) emitLoadField(i int) {
emit.Int(c.prog.BinWriter, int64(i))
emit.Opcodes(c.prog.BinWriter, opcode.PICKITEM)
}
func (c *codegen) emitStoreStructField(i int) {
emit.Int(c.prog.BinWriter, int64(i))
emit.Opcodes(c.prog.BinWriter, opcode.ROT, opcode.SETITEM)
}
// getVarIndex returns variable type and position in corresponding slot,
// according to current scope.
func (c *codegen) getVarIndex(pkg string, name string) *varInfo {
if pkg == "" {
if c.scope != nil {
vi := c.scope.vars.getVarInfo(name)
if vi != nil {
return vi
}
}
}
if i, ok := c.globals[c.getIdentName(pkg, name)]; ok {
return &varInfo{refType: varGlobal, index: i}
}
c.scope.newVariable(varLocal, name)
return c.scope.vars.getVarInfo(name)
}
func getBaseOpcode(t varType) (opcode.Opcode, opcode.Opcode) {
switch t {
case varGlobal:
return opcode.LDSFLD0, opcode.STSFLD0
case varLocal:
return opcode.LDLOC0, opcode.STLOC0
case varArgument:
return opcode.LDARG0, opcode.STARG0
default:
panic("invalid type")
}
}
// emitLoadVar loads specified variable to the evaluation stack.
func (c *codegen) emitLoadVar(pkg string, name string) {
vi := c.getVarIndex(pkg, name)
if vi.ctx != nil && c.typeAndValueOf(vi.ctx.expr).Value != nil {
c.emitLoadConst(c.typeAndValueOf(vi.ctx.expr))
return
} else if vi.ctx != nil {
var oldScope []map[string]varInfo
oldMap := c.importMap
c.importMap = vi.ctx.importMap
if c.scope != nil {
oldScope = c.scope.vars.locals
c.scope.vars.locals = vi.ctx.scope
}
ast.Walk(c, vi.ctx.expr)
if c.scope != nil {
c.scope.vars.locals = oldScope
}
c.importMap = oldMap
return
} else if vi.index == unspecifiedVarIndex {
emit.Opcodes(c.prog.BinWriter, opcode.PUSHNULL)
return
}
c.emitLoadByIndex(vi.refType, vi.index)
2020-08-21 12:37:46 +00:00
}
// emitLoadByIndex loads specified variable type with index i.
func (c *codegen) emitLoadByIndex(t varType, i int) {
base, _ := getBaseOpcode(t)
if i < 7 {
emit.Opcodes(c.prog.BinWriter, base+opcode.Opcode(i))
} else {
emit.Instruction(c.prog.BinWriter, base+7, []byte{byte(i)})
}
}
// emitStoreVar stores top value from the evaluation stack in the specified variable.
func (c *codegen) emitStoreVar(pkg string, name string) {
if name == "_" {
emit.Opcodes(c.prog.BinWriter, opcode.DROP)
return
}
vi := c.getVarIndex(pkg, name)
c.emitStoreByIndex(vi.refType, vi.index)
2020-08-21 12:37:46 +00:00
}
// emitLoadByIndex stores top value in the specified variable type with index i.
func (c *codegen) emitStoreByIndex(t varType, i int) {
_, base := getBaseOpcode(t)
if i < 7 {
emit.Opcodes(c.prog.BinWriter, base+opcode.Opcode(i))
} else {
emit.Instruction(c.prog.BinWriter, base+7, []byte{byte(i)})
}
}
func (c *codegen) emitDefault(t types.Type) {
switch t := t.Underlying().(type) {
case *types.Basic:
info := t.Info()
switch {
case info&types.IsInteger != 0:
emit.Int(c.prog.BinWriter, 0)
case info&types.IsString != 0:
emit.Bytes(c.prog.BinWriter, []byte{})
case info&types.IsBoolean != 0:
emit.Bool(c.prog.BinWriter, false)
default:
emit.Opcodes(c.prog.BinWriter, opcode.PUSHNULL)
}
case *types.Struct:
num := t.NumFields()
for i := num - 1; i >= 0; i-- {
c.emitDefault(t.Field(i).Type())
}
emit.Int(c.prog.BinWriter, int64(num))
emit.Opcodes(c.prog.BinWriter, opcode.PACKSTRUCT)
default:
emit.Opcodes(c.prog.BinWriter, opcode.PUSHNULL)
}
}
2019-10-22 14:56:03 +00:00
// convertGlobals traverses the AST and only converts global declarations.
// If we call this in convertFuncDecl then it will load all global variables
// into the scope of the function.
func (c *codegen) convertGlobals(f *ast.File, _ *types.Package) {
ast.Inspect(f, func(node ast.Node) bool {
switch n := node.(type) {
case *ast.FuncDecl:
return false
case *ast.GenDecl:
ast.Walk(c, n)
}
return true
})
}
func isInitFunc(decl *ast.FuncDecl) bool {
return decl.Name.Name == "init" && decl.Recv == nil &&
decl.Type.Params.NumFields() == 0 &&
decl.Type.Results.NumFields() == 0
}
func (c *codegen) isVerifyFunc(decl *ast.FuncDecl) bool {
return decl.Name.Name == "Verify" && decl.Recv == nil &&
decl.Type.Results.NumFields() == 1 &&
isBool(c.typeOf(decl.Type.Results.List[0].Type))
}
func (c *codegen) clearSlots(n int) {
for i := 0; i < n; i++ {
emit.Opcodes(c.prog.BinWriter, opcode.PUSHNULL)
c.emitStoreByIndex(varLocal, i)
}
}
// convertInitFuncs converts `init()` functions in file f and returns
// number of locals in last processed definition as well as maximum locals number encountered.
func (c *codegen) convertInitFuncs(f *ast.File, pkg *types.Package, lastCount int) (int, int) {
maxCount := -1
ast.Inspect(f, func(node ast.Node) bool {
switch n := node.(type) {
case *ast.FuncDecl:
if isInitFunc(n) {
if lastCount != -1 {
c.clearSlots(lastCount)
}
f := c.convertFuncDecl(f, n, pkg)
lastCount = f.vars.localsCnt
if lastCount > maxCount {
maxCount = lastCount
}
}
case *ast.GenDecl:
return false
}
return true
})
return lastCount, maxCount
}
2020-10-02 10:16:02 +00:00
func isDeployFunc(decl *ast.FuncDecl) bool {
if decl.Name.Name != "_deploy" || decl.Recv != nil ||
decl.Type.Params.NumFields() != 2 ||
2020-10-02 10:16:02 +00:00
decl.Type.Results.NumFields() != 0 {
return false
}
typ, ok := decl.Type.Params.List[1].Type.(*ast.Ident)
2020-10-02 10:16:02 +00:00
return ok && typ.Name == "bool"
}
func (c *codegen) convertDeployFuncs() int {
maxCount, lastCount := 0, -1
2020-10-02 10:16:02 +00:00
c.ForEachFile(func(f *ast.File, pkg *types.Package) {
ast.Inspect(f, func(node ast.Node) bool {
switch n := node.(type) {
case *ast.FuncDecl:
if isDeployFunc(n) {
if lastCount != -1 {
c.clearSlots(lastCount)
}
f := c.convertFuncDecl(f, n, pkg)
lastCount = f.vars.localsCnt
if lastCount > maxCount {
maxCount = lastCount
2020-10-02 10:16:02 +00:00
}
}
case *ast.GenDecl:
return false
}
return true
})
})
return maxCount
2020-10-02 10:16:02 +00:00
}
func (c *codegen) convertFuncDecl(file ast.Node, decl *ast.FuncDecl, pkg *types.Package) *funcScope {
var (
f *funcScope
ok, isLambda bool
)
isInit := isInitFunc(decl)
2020-10-02 10:16:02 +00:00
isDeploy := isDeployFunc(decl)
if isInit || isDeploy {
f = c.newFuncScope(decl, c.newLabel())
} else {
f, ok = c.funcs[c.getFuncNameFromDecl("", decl)]
if ok {
// If this function is a syscall or builtin we will not convert it to bytecode.
if isSyscall(f) || isCustomBuiltin(f) {
return f
}
c.setLabel(f.label)
} else if f, ok = c.lambda[c.getIdentName("", decl.Name.Name)]; ok {
isLambda = ok
c.setLabel(f.label)
} else {
f = c.newFunc(decl)
}
}
f.rng.Start = uint16(c.prog.Len())
c.scope = f
ast.Inspect(decl, c.scope.analyzeVoidCalls) // @OPTIMIZE
// All globals copied into the scope of the function need to be added
// to the stack size of the function.
2020-10-02 10:16:02 +00:00
if !isInit && !isDeploy {
sizeArg := f.countArgs()
if sizeArg > 255 {
c.prog.Err = errors.New("maximum of 255 local variables is allowed")
}
emit.Instruction(c.prog.BinWriter, opcode.INITSLOT, []byte{byte(0), byte(sizeArg)})
}
f.vars.newScope()
defer f.vars.dropScope()
// We need to handle methods, which in Go, is just syntactic sugar.
// The method receiver will be passed in as first argument.
// We check if this declaration has a receiver and load it into scope.
//
// FIXME: For now we will hard cast this to a struct. We can later fine tune this
// to support other types.
if decl.Recv != nil {
for _, arg := range decl.Recv.List {
// only create an argument here, it will be stored via INITSLOT
c.scope.newVariable(varArgument, arg.Names[0].Name)
}
}
// Load the arguments in scope.
for _, arg := range decl.Type.Params.List {
for _, id := range arg.Names {
// only create an argument here, it will be stored via INITSLOT
c.scope.newVariable(varArgument, id.Name)
}
}
ast.Walk(c, decl.Body)
// If we have reached the end of the function without encountering `return` statement,
// we should clean alt.stack manually.
// This can be the case with void and named-return functions.
if !isInit && !isDeploy && !lastStmtIsReturn(decl.Body) {
c.processDefers()
c.saveSequencePoint(decl.Body)
emit.Opcodes(c.prog.BinWriter, opcode.RET)
}
if isInit {
c.initVariables = append(c.initVariables, f.variables...)
} else if isDeploy {
c.deployVariables = append(c.deployVariables, f.variables...)
}
f.rng.End = uint16(c.prog.Len() - 1)
if !isLambda {
for _, f := range c.lambda {
if _, ok := c.lambda[c.getIdentName("", f.decl.Name.Name)]; !ok {
panic("ICE: lambda name doesn't match map key")
}
c.convertFuncDecl(file, f.decl, pkg)
}
c.lambda = make(map[string]*funcScope)
}
if !isInit && !isDeploy {
c.reverseOffsetMap[int(f.rng.Start)] = nameWithLocals{
name: f.name,
count: f.vars.localsCnt,
}
}
return f
}
func (c *codegen) Visit(node ast.Node) ast.Visitor {
if c.prog.Err != nil {
return nil
}
switch n := node.(type) {
// General declarations.
// var (
// x = 2
// )
case *ast.GenDecl:
if n.Tok == token.VAR || n.Tok == token.CONST {
c.saveSequencePoint(n)
}
if n.Tok == token.CONST {
for _, spec := range n.Specs {
vs := spec.(*ast.ValueSpec)
for i := range vs.Names {
2020-08-25 07:22:58 +00:00
info := c.buildInfo.program.Package(c.currPkg.Path())
obj := info.Defs[vs.Names[i]]
c.constMap[c.getIdentName("", vs.Names[i].Name)] = types.TypeAndValue{
Type: obj.Type(),
Value: obj.(*types.Const).Val(),
}
}
}
return nil
}
for _, spec := range n.Specs {
switch t := spec.(type) {
case *ast.ValueSpec:
for _, id := range t.Names {
if id.Name != "_" {
if c.scope == nil {
// it is a global declaration
c.newGlobal("", id.Name)
} else {
c.scope.newLocal(id.Name)
}
c.registerDebugVariable(id.Name, t.Type)
}
}
for i := range t.Names {
if len(t.Values) != 0 {
ast.Walk(c, t.Values[i])
2020-05-13 15:09:55 +00:00
} else {
c.emitDefault(c.typeOf(t.Type))
}
c.emitStoreVar("", t.Names[i].Name)
}
}
}
return nil
case *ast.AssignStmt:
multiRet := len(n.Rhs) != len(n.Lhs)
c.saveSequencePoint(n)
// Assign operations are grouped https://github.com/golang/go/blob/master/src/go/types/stmt.go#L160
isAssignOp := token.ADD_ASSIGN <= n.Tok && n.Tok <= token.AND_NOT_ASSIGN
if isAssignOp {
// RHS can contain exactly one expression, thus there is no need to iterate.
ast.Walk(c, n.Lhs[0])
ast.Walk(c, n.Rhs[0])
c.emitToken(n.Tok, c.typeOf(n.Rhs[0]))
}
for i := 0; i < len(n.Lhs); i++ {
switch t := n.Lhs[i].(type) {
case *ast.Ident:
if n.Tok == token.DEFINE {
if !multiRet {
c.registerDebugVariable(t.Name, n.Rhs[i])
}
if t.Name != "_" {
c.scope.newLocal(t.Name)
}
}
if !isAssignOp && (i == 0 || !multiRet) {
ast.Walk(c, n.Rhs[i])
}
c.emitStoreVar("", t.Name)
case *ast.SelectorExpr:
if !isAssignOp {
ast.Walk(c, n.Rhs[i])
}
typ := c.typeOf(t.X)
if typ == nil {
// Store to other package global variable.
c.emitStoreVar(t.X.(*ast.Ident).Name, t.Sel.Name)
return nil
}
strct, ok := c.getStruct(typ)
if !ok {
2019-11-22 14:16:52 +00:00
c.prog.Err = fmt.Errorf("nested selector assigns not supported yet")
return nil
}
ast.Walk(c, t.X) // load the struct
i := indexOfStruct(strct, t.Sel.Name) // get the index of the field
c.emitStoreStructField(i) // store the field
// Assignments to index expressions.
// slice[0] = 10
case *ast.IndexExpr:
if !isAssignOp {
ast.Walk(c, n.Rhs[i])
}
ast.Walk(c, t.X)
ast.Walk(c, t.Index)
emit.Opcodes(c.prog.BinWriter, opcode.ROT, opcode.SETITEM)
}
}
return nil
2020-02-10 15:14:34 +00:00
case *ast.SliceExpr:
if isCompoundSlice(c.typeOf(n.X).Underlying()) {
c.prog.Err = errors.New("subslices are supported only for []byte")
return nil
}
ast.Walk(c, n.X)
2020-02-10 15:14:34 +00:00
if n.Low != nil {
ast.Walk(c, n.Low)
} else {
emit.Opcodes(c.prog.BinWriter, opcode.PUSH0)
2020-02-10 15:14:34 +00:00
}
if n.High != nil {
ast.Walk(c, n.High)
} else {
emit.Opcodes(c.prog.BinWriter, opcode.OVER, opcode.SIZE)
2020-02-10 15:14:34 +00:00
}
emit.Opcodes(c.prog.BinWriter, opcode.OVER, opcode.SUB, opcode.SUBSTR)
2020-02-10 15:14:34 +00:00
return nil
case *ast.ReturnStmt:
l := c.newLabel()
c.setLabel(l)
cnt := 0
start := 0
if len(c.inlineLabelOffsets) > 0 {
start = c.inlineLabelOffsets[len(c.inlineLabelOffsets)-1]
}
for i := start; i < len(c.labelList); i++ {
cnt += c.labelList[i].sz
}
c.dropItems(cnt)
2020-05-06 14:24:32 +00:00
if len(n.Results) == 0 {
results := c.scope.decl.Type.Results
if results.NumFields() != 0 {
// function with named returns
for i := len(results.List) - 1; i >= 0; i-- {
names := results.List[i].Names
for j := len(names) - 1; j >= 0; j-- {
c.emitLoadVar("", names[j].Name)
2020-05-06 14:24:32 +00:00
}
}
}
} else {
// first result should be on top of the stack
for i := len(n.Results) - 1; i >= 0; i-- {
ast.Walk(c, n.Results[i])
}
}
c.processDefers()
c.saveSequencePoint(n)
2021-02-04 12:41:00 +00:00
if len(c.pkgInfoInline) == 0 {
emit.Opcodes(c.prog.BinWriter, opcode.RET)
}
return nil
case *ast.IfStmt:
c.scope.vars.newScope()
defer c.scope.vars.dropScope()
lIf := c.newLabel()
lElse := c.newLabel()
lElseEnd := c.newLabel()
2020-08-21 08:41:10 +00:00
if n.Init != nil {
ast.Walk(c, n.Init)
}
if n.Cond != nil {
c.emitBoolExpr(n.Cond, true, false, lElse)
}
c.setLabel(lIf)
ast.Walk(c, n.Body)
if n.Else != nil {
emit.Jmp(c.prog.BinWriter, opcode.JMPL, lElseEnd)
}
c.setLabel(lElse)
if n.Else != nil {
ast.Walk(c, n.Else)
}
c.setLabel(lElseEnd)
return nil
case *ast.SwitchStmt:
eqOpcode := opcode.EQUAL
if n.Tag != nil {
ast.Walk(c, n.Tag)
eqOpcode, _ = convertToken(token.EQL, c.typeOf(n.Tag))
} else {
emit.Bool(c.prog.BinWriter, true)
}
switchEnd, label := c.generateLabel(labelEnd)
lastSwitch := c.currentSwitch
c.currentSwitch = label
c.pushStackLabel(label, 1)
startLabels := make([]uint16, len(n.Body.List))
for i := range startLabels {
startLabels[i] = c.newLabel()
}
for i := range n.Body.List {
lEnd := c.newLabel()
lStart := startLabels[i]
cc := n.Body.List[i].(*ast.CaseClause)
if l := len(cc.List); l != 0 { // if not `default`
for j := range cc.List {
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
ast.Walk(c, cc.List[j])
emit.Opcodes(c.prog.BinWriter, eqOpcode)
if j == l-1 {
emit.Jmp(c.prog.BinWriter, opcode.JMPIFNOTL, lEnd)
} else {
emit.Jmp(c.prog.BinWriter, opcode.JMPIFL, lStart)
}
}
}
c.scope.vars.newScope()
c.setLabel(lStart)
last := len(cc.Body) - 1
for j, stmt := range cc.Body {
if j == last && isFallthroughStmt(stmt) {
emit.Jmp(c.prog.BinWriter, opcode.JMPL, startLabels[i+1])
break
}
ast.Walk(c, stmt)
}
emit.Jmp(c.prog.BinWriter, opcode.JMPL, switchEnd)
c.setLabel(lEnd)
c.scope.vars.dropScope()
}
c.setLabel(switchEnd)
c.dropStackLabel()
c.currentSwitch = lastSwitch
return nil
case *ast.FuncLit:
l := c.newLabel()
c.newLambda(l, n)
buf := make([]byte, 4)
binary.LittleEndian.PutUint16(buf, l)
emit.Instruction(c.prog.BinWriter, opcode.PUSHA, buf)
return nil
case *ast.BasicLit:
2020-05-13 15:09:55 +00:00
c.emitLoadConst(c.typeAndValueOf(n))
return nil
case *ast.StarExpr:
_, ok := c.getStruct(c.typeOf(n.X))
if !ok {
c.prog.Err = errors.New("dereferencing is only supported on structs")
return nil
}
ast.Walk(c, n.X)
c.emitConvert(stackitem.StructT)
return nil
case *ast.Ident:
if tv := c.typeAndValueOf(n); tv.Value != nil {
c.emitLoadConst(tv)
} else if n.Name == "nil" {
emit.Opcodes(c.prog.BinWriter, opcode.PUSHNULL)
} else {
c.emitLoadVar("", n.Name)
}
return nil
case *ast.CompositeLit:
t := c.typeOf(n)
switch typ := t.Underlying().(type) {
case *types.Struct:
c.convertStruct(n, false)
case *types.Map:
c.convertMap(n)
default:
if tn, ok := t.(*types.Named); ok && isInteropPath(tn.String()) {
st, _ := scAndVMInteropTypeFromExpr(tn)
expectedLen := -1
switch st {
case smartcontract.Hash160Type:
expectedLen = 20
case smartcontract.Hash256Type:
expectedLen = 32
}
if expectedLen != -1 && expectedLen != len(n.Elts) {
c.prog.Err = fmt.Errorf("%s type must have size %d", tn.Obj().Name(), expectedLen)
return nil
}
}
ln := len(n.Elts)
2019-10-22 14:56:03 +00:00
// ByteArrays needs a different approach than normal arrays.
2020-05-13 15:09:55 +00:00
if isByteSlice(typ) {
c.convertByteArray(n.Elts)
return nil
}
for i := ln - 1; i >= 0; i-- {
ast.Walk(c, n.Elts[i])
}
emit.Int(c.prog.BinWriter, int64(ln))
emit.Opcodes(c.prog.BinWriter, opcode.PACK)
}
return nil
case *ast.BinaryExpr:
c.emitBinaryExpr(n, false, false, 0)
return nil
case *ast.CallExpr:
var (
f *funcScope
ok bool
name string
numArgs = len(n.Args)
isBuiltin bool
isFunc bool
isLiteral bool
)
switch fun := n.Fun.(type) {
case *ast.Ident:
f, ok = c.getFuncFromIdent(fun)
isBuiltin = isGoBuiltin(fun.Name)
if !ok && !isBuiltin {
name = fun.Name
}
// distinguish lambda invocations from type conversions
if fun.Obj != nil && fun.Obj.Kind == ast.Var {
isFunc = true
}
if ok && canInline(f.pkg.Path(), f.decl.Name.Name) {
2021-02-04 12:41:00 +00:00
c.inlineCall(f, n)
return nil
}
case *ast.SelectorExpr:
// If this is a method call we need to walk the AST to load the struct locally.
// Otherwise this is a function call from a imported package and we can call it
// directly.
name, isMethod := c.getFuncNameFromSelector(fun)
if isMethod {
ast.Walk(c, fun.X)
// Dont forget to add 1 extra argument when its a method.
numArgs++
}
f, ok = c.funcs[name]
if ok {
f.selector = fun.X.(*ast.Ident)
isBuiltin = isCustomBuiltin(f)
if canInline(f.pkg.Path(), f.decl.Name.Name) {
2021-02-04 12:41:00 +00:00
c.inlineCall(f, n)
return nil
}
} else {
typ := c.typeOf(fun)
if _, ok := typ.(*types.Signature); ok {
c.prog.Err = fmt.Errorf("could not resolve function %s", fun.Sel.Name)
return nil
}
ast.Walk(c, n.Args[0])
c.emitExplicitConvert(c.typeOf(n.Args[0]), typ)
2019-11-22 14:16:52 +00:00
return nil
}
case *ast.ArrayType:
// For now we will assume that there are only byte slice conversions.
// E.g. []byte("foobar") or []byte(scriptHash).
ast.Walk(c, n.Args[0])
c.emitConvert(stackitem.BufferT)
return nil
case *ast.FuncLit:
isLiteral = true
}
c.saveSequencePoint(n)
args := transformArgs(f, n.Fun, n.Args)
// Handle the arguments
for _, arg := range args {
ast.Walk(c, arg)
typ := c.typeOf(arg)
_, ok := typ.Underlying().(*types.Struct)
if ok && !isInteropPath(typ.String()) {
// To clone struct fields we create a new array and append struct to it.
// This way even non-pointer struct fields will be copied.
emit.Opcodes(c.prog.BinWriter, opcode.NEWARRAY0,
opcode.DUP, opcode.ROT, opcode.APPEND,
opcode.POPITEM)
}
}
// Do not swap for builtin functions.
if !isBuiltin && (f != nil && !isSyscall(f)) {
2020-06-27 07:39:30 +00:00
typ, ok := c.typeOf(n.Fun).(*types.Signature)
if ok && typ.Variadic() && !n.Ellipsis.IsValid() {
// pack variadic args into an array only if last argument is not of form `...`
varSize := c.packVarArgs(n, typ)
numArgs -= varSize - 1
}
c.emitReverse(numArgs)
}
// Check builtin first to avoid nil pointer on funcScope!
switch {
case isBuiltin:
// Use the ident to check, builtins are not in func scopes.
// We can be sure builtins are of type *ast.Ident.
2018-04-22 18:11:37 +00:00
c.convertBuiltin(n)
case name != "":
// Function was not found thus is can be only an invocation of func-typed variable or type conversion.
// We care only about string conversions because all others are effectively no-op in NeoVM.
// E.g. one cannot write `bool(int(a))`, only `int32(int(a))`.
if isString(c.typeOf(n.Fun)) {
c.emitConvert(stackitem.ByteArrayT)
} else if isFunc {
c.emitLoadVar("", name)
emit.Opcodes(c.prog.BinWriter, opcode.CALLA)
}
case isLiteral:
ast.Walk(c, n.Fun)
emit.Opcodes(c.prog.BinWriter, opcode.CALLA)
case isSyscall(f):
c.convertSyscall(f, n)
default:
emit.Call(c.prog.BinWriter, opcode.CALLL, f.label)
}
if c.scope != nil && c.scope.voidCalls[n] {
var sz int
if f != nil {
sz = f.decl.Type.Results.NumFields()
} else if !isBuiltin {
// lambda invocation
f := c.typeOf(n.Fun).Underlying().(*types.Signature)
sz = f.Results().Len()
}
for i := 0; i < sz; i++ {
emit.Opcodes(c.prog.BinWriter, opcode.DROP)
}
}
return nil
case *ast.DeferStmt:
2020-08-21 12:37:46 +00:00
catch := c.newLabel()
finally := c.newLabel()
param := make([]byte, 8)
2020-08-21 12:37:46 +00:00
binary.LittleEndian.PutUint16(param[0:], catch)
binary.LittleEndian.PutUint16(param[4:], finally)
emit.Instruction(c.prog.BinWriter, opcode.TRYL, param)
c.scope.deferStack = append(c.scope.deferStack, deferInfo{
2020-08-21 12:37:46 +00:00
catchLabel: catch,
finallyLabel: finally,
expr: n.Call,
})
return nil
case *ast.SelectorExpr:
typ := c.typeOf(n.X)
if typ == nil {
// This is a global variable from a package.
pkgAlias := n.X.(*ast.Ident).Name
name := c.getIdentName(pkgAlias, n.Sel.Name)
if tv, ok := c.constMap[name]; ok {
c.emitLoadConst(tv)
} else {
c.emitLoadVar(pkgAlias, n.Sel.Name)
}
return nil
}
strct, ok := c.getStruct(typ)
if !ok {
c.prog.Err = fmt.Errorf("selectors are supported only on structs")
2019-11-22 14:16:52 +00:00
return nil
}
ast.Walk(c, n.X) // load the struct
i := indexOfStruct(strct, n.Sel.Name)
c.emitLoadField(i) // load the field
return nil
case *ast.UnaryExpr:
if n.Op == token.AND {
// We support only taking address from struct literals.
// For identifiers we can't support "taking address" in a general way
// because both struct and array are reference types.
lit, ok := n.X.(*ast.CompositeLit)
if ok {
c.convertStruct(lit, true)
return nil
}
c.prog.Err = fmt.Errorf("'&' can be used only with struct literals")
return nil
}
ast.Walk(c, n.X)
// From https://golang.org/ref/spec#Operators
// there can be only following unary operators
// "+" | "-" | "!" | "^" | "*" | "&" | "<-" .
// of which last three are not used in SC
switch n.Op {
case token.ADD:
// +10 == 10, no need to do anything in this case
case token.SUB:
emit.Opcodes(c.prog.BinWriter, opcode.NEGATE)
case token.NOT:
emit.Opcodes(c.prog.BinWriter, opcode.NOT)
case token.XOR:
emit.Opcodes(c.prog.BinWriter, opcode.INVERT)
default:
2019-11-22 14:16:52 +00:00
c.prog.Err = fmt.Errorf("invalid unary operator: %s", n.Op)
return nil
}
return nil
case *ast.IncDecStmt:
ast.Walk(c, n.X)
c.emitToken(n.Tok, c.typeOf(n.X))
// For now only identifiers are supported for (post) for stmts.
// for i := 0; i < 10; i++ {}
// Where the post stmt is ( i++ )
if ident, ok := n.X.(*ast.Ident); ok {
c.emitStoreVar("", ident.Name)
}
return nil
case *ast.IndexExpr:
// Walk the expression, this could be either an Ident or SelectorExpr.
// This will load local whatever X is.
ast.Walk(c, n.X)
ast.Walk(c, n.Index)
emit.Opcodes(c.prog.BinWriter, opcode.PICKITEM) // just pickitem here
return nil
case *ast.BranchStmt:
var label string
if n.Label != nil {
label = n.Label.Name
} else if n.Tok == token.BREAK {
label = c.currentSwitch
} else if n.Tok == token.CONTINUE {
label = c.currentFor
}
cnt := 0
for i := len(c.labelList) - 1; i >= 0 && c.labelList[i].name != label; i-- {
cnt += c.labelList[i].sz
}
c.dropItems(cnt)
switch n.Tok {
case token.BREAK:
end := c.getLabelOffset(labelEnd, label)
emit.Jmp(c.prog.BinWriter, opcode.JMPL, end)
case token.CONTINUE:
post := c.getLabelOffset(labelPost, label)
emit.Jmp(c.prog.BinWriter, opcode.JMPL, post)
}
return nil
case *ast.LabeledStmt:
c.nextLabel = n.Label.Name
ast.Walk(c, n.Stmt)
return nil
case *ast.BlockStmt:
c.scope.vars.newScope()
defer c.scope.vars.dropScope()
for i := range n.List {
ast.Walk(c, n.List[i])
}
return nil
case *ast.ForStmt:
c.scope.vars.newScope()
defer c.scope.vars.dropScope()
fstart, label := c.generateLabel(labelStart)
fend := c.newNamedLabel(labelEnd, label)
fpost := c.newNamedLabel(labelPost, label)
lastLabel := c.currentFor
lastSwitch := c.currentSwitch
c.currentFor = label
c.currentSwitch = label
// Walk the initializer and condition.
if n.Init != nil {
ast.Walk(c, n.Init)
}
// Set label and walk the condition.
c.pushStackLabel(label, 0)
c.setLabel(fstart)
if n.Cond != nil {
ast.Walk(c, n.Cond)
// Jump if the condition is false
emit.Jmp(c.prog.BinWriter, opcode.JMPIFNOTL, fend)
}
// Walk body followed by the iterator (post stmt).
ast.Walk(c, n.Body)
c.setLabel(fpost)
if n.Post != nil {
ast.Walk(c, n.Post)
}
// Jump back to condition.
emit.Jmp(c.prog.BinWriter, opcode.JMPL, fstart)
c.setLabel(fend)
c.dropStackLabel()
c.currentFor = lastLabel
c.currentSwitch = lastSwitch
return nil
2020-02-11 12:26:11 +00:00
case *ast.RangeStmt:
c.scope.vars.newScope()
defer c.scope.vars.dropScope()
start, label := c.generateLabel(labelStart)
end := c.newNamedLabel(labelEnd, label)
post := c.newNamedLabel(labelPost, label)
lastFor := c.currentFor
lastSwitch := c.currentSwitch
c.currentFor = label
c.currentSwitch = label
2020-02-11 12:26:11 +00:00
ast.Walk(c, n.X)
// Implementation is a bit different for slices and maps:
// For slices we iterate index from 0 to len-1, storing array, len and index on stack.
// For maps we iterate index from 0 to len-1, storing map, keyarray, size and index on stack.
_, isMap := c.typeOf(n.X).Underlying().(*types.Map)
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
if isMap {
emit.Opcodes(c.prog.BinWriter, opcode.KEYS, opcode.DUP)
}
emit.Opcodes(c.prog.BinWriter, opcode.SIZE, opcode.PUSH0)
stackSize := 3 // slice, len(slice), index
if isMap {
stackSize++ // map, keys, len(keys), index in keys
}
c.pushStackLabel(label, stackSize)
2020-02-11 12:26:11 +00:00
c.setLabel(start)
emit.Opcodes(c.prog.BinWriter, opcode.OVER, opcode.OVER)
emit.Jmp(c.prog.BinWriter, opcode.JMPLEL, end)
2020-02-11 12:26:11 +00:00
var keyLoaded bool
needValue := n.Value != nil && n.Value.(*ast.Ident).Name != "_"
if n.Key != nil && n.Key.(*ast.Ident).Name != "_" {
if isMap {
c.rangeLoadKey()
if needValue {
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
keyLoaded = true
}
} else {
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
}
c.emitStoreVar("", n.Key.(*ast.Ident).Name)
2020-02-11 12:26:11 +00:00
}
if needValue {
if !isMap || !keyLoaded {
c.rangeLoadKey()
}
if isMap {
// we have loaded only key from key array, now load value
emit.Int(c.prog.BinWriter, 4)
emit.Opcodes(c.prog.BinWriter,
opcode.PICK, // load map itself (+1 because key was pushed)
opcode.SWAP, // key should be on top
opcode.PICKITEM)
}
c.emitStoreVar("", n.Value.(*ast.Ident).Name)
}
2020-02-11 12:26:11 +00:00
ast.Walk(c, n.Body)
c.setLabel(post)
emit.Opcodes(c.prog.BinWriter, opcode.INC)
emit.Jmp(c.prog.BinWriter, opcode.JMPL, start)
2020-02-11 12:26:11 +00:00
c.setLabel(end)
c.dropStackLabel()
2020-02-11 12:26:11 +00:00
c.currentFor = lastFor
c.currentSwitch = lastSwitch
2020-02-11 12:26:11 +00:00
return nil
// We dont really care about assertions for the core logic.
// The only thing we need is to please the compiler type checking.
// For this to work properly, we only need to walk the expression
// not the assertion type.
case *ast.TypeAssertExpr:
ast.Walk(c, n.X)
if c.isCallExprSyscall(n.X) {
return nil
}
goTyp := c.typeOf(n.Type)
if canConvert(goTyp.String()) {
typ := toNeoType(goTyp)
c.emitConvert(typ)
}
return nil
}
return c
}
// packVarArgs packs variadic arguments into an array
// and returns amount of arguments packed.
func (c *codegen) packVarArgs(n *ast.CallExpr, typ *types.Signature) int {
varSize := len(n.Args) - typ.Params().Len() + 1
c.emitReverse(varSize)
emit.Int(c.prog.BinWriter, int64(varSize))
emit.Opcodes(c.prog.BinWriter, opcode.PACK)
return varSize
}
func (c *codegen) isCallExprSyscall(e ast.Expr) bool {
ce, ok := e.(*ast.CallExpr)
if !ok {
return false
}
sel, ok := ce.Fun.(*ast.SelectorExpr)
if !ok {
return false
}
name, _ := c.getFuncNameFromSelector(sel)
f, ok := c.funcs[name]
return ok && isSyscall(f)
}
// processDefers emits code for `defer` statements.
2020-08-21 12:37:46 +00:00
// TRY-related opcodes handle exception as follows:
// 1. CATCH block is executed only if exception has occurred.
2020-08-21 12:37:46 +00:00
// 2. FINALLY block is always executed, but after catch block.
// Go `defer` statements are a bit different:
// 1. `defer` is always executed irregardless of whether an exception has occurred.
2020-08-21 12:37:46 +00:00
// 2. `recover` can or can not handle a possible exception.
// Thus we use the following approach:
// 1. Throwed exception is saved in a static field X, static fields Y and is set to true.
// 2. CATCH and FINALLY blocks are the same, and both contain the same CALLs.
// 3. In CATCH block we set Y to true and emit default return values if it is the last defer.
// 4. Execute FINALLY block only if Y is false.
func (c *codegen) processDefers() {
for i := len(c.scope.deferStack) - 1; i >= 0; i-- {
stmt := c.scope.deferStack[i]
after := c.newLabel()
emit.Jmp(c.prog.BinWriter, opcode.ENDTRYL, after)
2020-08-21 12:37:46 +00:00
c.setLabel(stmt.catchLabel)
c.emitStoreByIndex(varGlobal, c.exceptionIndex)
emit.Int(c.prog.BinWriter, 1)
finalIndex := c.getVarIndex("", finallyVarName).index
c.emitStoreByIndex(varLocal, finalIndex)
2020-08-21 12:37:46 +00:00
ast.Walk(c, stmt.expr)
if i == 0 {
results := c.scope.decl.Type.Results
if results.NumFields() != 0 {
// After panic, default values must be returns, except for named returns,
// which we don't support here for now.
for i := len(results.List) - 1; i >= 0; i-- {
c.emitDefault(c.typeOf(results.List[i].Type))
}
2020-08-21 12:37:46 +00:00
}
}
emit.Jmp(c.prog.BinWriter, opcode.ENDTRYL, after)
c.setLabel(stmt.finallyLabel)
2020-08-21 12:37:46 +00:00
before := c.newLabel()
c.emitLoadByIndex(varLocal, finalIndex)
2020-08-21 12:37:46 +00:00
emit.Jmp(c.prog.BinWriter, opcode.JMPIFL, before)
ast.Walk(c, stmt.expr)
2020-08-21 12:37:46 +00:00
c.setLabel(before)
emit.Int(c.prog.BinWriter, 0)
c.emitStoreByIndex(varLocal, finalIndex)
emit.Opcodes(c.prog.BinWriter, opcode.ENDFINALLY)
c.setLabel(after)
}
}
// emitExplicitConvert handles `someType(someValue)` conversions between string/[]byte.
// Rules for conversion:
// 1. interop.* types are converted to ByteArray if not already.
// 2. Otherwise convert between ByteArray/Buffer.
// 3. Rules for types which are not string/[]byte should already
// be enforced by go parser.
func (c *codegen) emitExplicitConvert(from, to types.Type) {
if isInteropPath(to.String()) {
if isByteSlice(from) && !isString(from) {
c.emitConvert(stackitem.ByteArrayT)
}
} else if isByteSlice(to) && !isByteSlice(from) {
c.emitConvert(stackitem.BufferT)
} else if isString(to) && !isString(from) {
c.emitConvert(stackitem.ByteArrayT)
}
}
func (c *codegen) rangeLoadKey() {
emit.Int(c.prog.BinWriter, 2)
emit.Opcodes(c.prog.BinWriter,
opcode.PICK, // load keys
opcode.OVER, // load index in key array
opcode.PICKITEM)
}
func isFallthroughStmt(c ast.Node) bool {
s, ok := c.(*ast.BranchStmt)
return ok && s.Tok == token.FALLTHROUGH
}
func (c *codegen) getCompareWithNilArg(n *ast.BinaryExpr) ast.Expr {
if isExprNil(n.X) {
return n.Y
} else if isExprNil(n.Y) {
return n.X
}
return nil
}
func (c *codegen) emitJumpOnCondition(cond bool, jmpLabel uint16) {
if cond {
emit.Jmp(c.prog.BinWriter, opcode.JMPIFL, jmpLabel)
} else {
emit.Jmp(c.prog.BinWriter, opcode.JMPIFNOTL, jmpLabel)
}
}
// emitBoolExpr emits boolean expression. If needJump is true and expression evaluates to `cond`,
// jump to jmpLabel is performed and no item is left on stack.
func (c *codegen) emitBoolExpr(n ast.Expr, needJump bool, cond bool, jmpLabel uint16) {
if be, ok := n.(*ast.BinaryExpr); ok {
c.emitBinaryExpr(be, needJump, cond, jmpLabel)
} else {
ast.Walk(c, n)
if needJump {
c.emitJumpOnCondition(cond, jmpLabel)
}
}
}
// emitBinaryExpr emits binary expression. If needJump is true and expression evaluates to `cond`,
// jump to jmpLabel is performed and no item is left on stack.
func (c *codegen) emitBinaryExpr(n *ast.BinaryExpr, needJump bool, cond bool, jmpLabel uint16) {
// The AST package will try to resolve all basic literals for us.
// If the typeinfo.Value is not nil we know that the expr is resolved
// and needs no further action. e.g. x := 2 + 2 + 2 will be resolved to 6.
// NOTE: Constants will also be automatically resolved be the AST parser.
// example:
// const x = 10
// x + 2 will results into 12
tinfo := c.typeAndValueOf(n)
if tinfo.Value != nil {
c.emitLoadConst(tinfo)
if needJump && isBool(tinfo.Type) {
c.emitJumpOnCondition(cond, jmpLabel)
}
return
} else if arg := c.getCompareWithNilArg(n); arg != nil {
ast.Walk(c, arg)
emit.Opcodes(c.prog.BinWriter, opcode.ISNULL)
if needJump {
c.emitJumpOnCondition(cond == (n.Op == token.EQL), jmpLabel)
} else if n.Op == token.NEQ {
emit.Opcodes(c.prog.BinWriter, opcode.NOT)
}
return
}
switch n.Op {
case token.LAND, token.LOR:
end := c.newLabel()
// true || .. == true, false && .. == false
condShort := n.Op == token.LOR
if needJump {
l := end
if cond == condShort {
l = jmpLabel
}
c.emitBoolExpr(n.X, true, condShort, l)
c.emitBoolExpr(n.Y, true, cond, jmpLabel)
} else {
push := c.newLabel()
c.emitBoolExpr(n.X, true, condShort, push)
c.emitBoolExpr(n.Y, false, false, 0)
emit.Jmp(c.prog.BinWriter, opcode.JMPL, end)
c.setLabel(push)
emit.Bool(c.prog.BinWriter, condShort)
}
c.setLabel(end)
default:
ast.Walk(c, n.X)
ast.Walk(c, n.Y)
typ := c.typeOf(n.X)
if !needJump {
c.emitToken(n.Op, typ)
return
}
op, ok := getJumpForToken(n.Op, typ)
if !ok {
c.emitToken(n.Op, typ)
c.emitJumpOnCondition(cond, jmpLabel)
return
}
if !cond {
op = negateJmp(op)
}
emit.Jmp(c.prog.BinWriter, op, jmpLabel)
}
}
func (c *codegen) pushStackLabel(name string, size int) {
c.labelList = append(c.labelList, labelWithStackSize{
name: name,
sz: size,
})
}
func (c *codegen) dropStackLabel() {
last := len(c.labelList) - 1
c.dropItems(c.labelList[last].sz)
c.labelList = c.labelList[:last]
}
func (c *codegen) dropItems(n int) {
if n < 4 {
for i := 0; i < n; i++ {
emit.Opcodes(c.prog.BinWriter, opcode.DROP)
}
return
}
emit.Int(c.prog.BinWriter, int64(n))
emit.Opcodes(c.prog.BinWriter, opcode.PACK, opcode.DROP)
}
// emitReverse reverses top num items of the stack.
func (c *codegen) emitReverse(num int) {
switch num {
case 0, 1:
case 2:
emit.Opcodes(c.prog.BinWriter, opcode.SWAP)
case 3:
emit.Opcodes(c.prog.BinWriter, opcode.REVERSE3)
case 4:
emit.Opcodes(c.prog.BinWriter, opcode.REVERSE4)
default:
emit.Int(c.prog.BinWriter, int64(num))
emit.Opcodes(c.prog.BinWriter, opcode.REVERSEN)
}
}
// generateLabel returns a new label.
func (c *codegen) generateLabel(typ labelOffsetType) (uint16, string) {
name := c.nextLabel
if name == "" {
name = fmt.Sprintf("@%d", len(c.l))
}
c.nextLabel = ""
return c.newNamedLabel(typ, name), name
}
func (c *codegen) getLabelOffset(typ labelOffsetType, name string) uint16 {
return c.labels[labelWithType{name: name, typ: typ}]
}
// For `&&` and `||` it return an opcode which jumps only if result is known:
// false && .. == false, true || .. = true.
func getJumpForToken(tok token.Token, typ types.Type) (opcode.Opcode, bool) {
switch tok {
case token.GTR:
return opcode.JMPGTL, true
case token.GEQ:
return opcode.JMPGEL, true
case token.LSS:
return opcode.JMPLTL, true
case token.LEQ:
return opcode.JMPLEL, true
case token.EQL, token.NEQ:
if isNumber(typ) {
if tok == token.EQL {
return opcode.JMPEQL, true
}
return opcode.JMPNEL, true
}
}
return 0, false
}
func (c *codegen) convertSyscall(f *funcScope, expr *ast.CallExpr) {
for _, arg := range expr.Args[1:] {
ast.Walk(c, arg)
}
tv := c.typeAndValueOf(expr.Args[0])
name := constant.StringVal(tv.Value)
if strings.HasPrefix(f.name, "Syscall") {
c.emitReverse(len(expr.Args) - 1)
emit.Syscall(c.prog.BinWriter, name)
} else {
op, err := opcode.FromString(name)
if err != nil {
c.prog.Err = fmt.Errorf("invalid opcode: %s", op)
return
}
emit.Opcodes(c.prog.BinWriter, op)
}
}
2020-08-24 09:54:01 +00:00
// emitSliceHelper emits 3 items on stack: slice, its first index, and its size.
func (c *codegen) emitSliceHelper(e ast.Expr) {
if !isByteSlice(c.typeOf(e)) {
c.prog.Err = fmt.Errorf("copy is supported only for byte-slices")
return
}
var hasLowIndex bool
switch src := e.(type) {
case *ast.SliceExpr:
ast.Walk(c, src.X)
if src.High != nil {
ast.Walk(c, src.High)
} else {
emit.Opcodes(c.prog.BinWriter, opcode.DUP, opcode.SIZE)
2020-08-24 09:54:01 +00:00
}
if src.Low != nil {
ast.Walk(c, src.Low)
hasLowIndex = true
} else {
emit.Int(c.prog.BinWriter, 0)
}
default:
ast.Walk(c, src)
emit.Opcodes(c.prog.BinWriter, opcode.DUP, opcode.SIZE)
2020-08-24 09:54:01 +00:00
emit.Int(c.prog.BinWriter, 0)
}
if !hasLowIndex {
emit.Opcodes(c.prog.BinWriter, opcode.SWAP)
2020-08-24 09:54:01 +00:00
} else {
emit.Opcodes(c.prog.BinWriter, opcode.DUP, opcode.ROT, opcode.SWAP, opcode.SUB)
2020-08-24 09:54:01 +00:00
}
}
2018-04-22 18:11:37 +00:00
func (c *codegen) convertBuiltin(expr *ast.CallExpr) {
var name string
switch t := expr.Fun.(type) {
case *ast.Ident:
name = t.Name
case *ast.SelectorExpr:
name = t.Sel.Name
}
switch name {
2020-08-24 09:54:01 +00:00
case "copy":
// stack for MEMCPY is: dst, dst_index, src, src_index, count
c.emitSliceHelper(expr.Args[0])
c.emitSliceHelper(expr.Args[1])
emit.Int(c.prog.BinWriter, 3)
emit.Opcodes(c.prog.BinWriter, opcode.ROLL, opcode.MIN)
if !c.scope.voidCalls[expr] {
// insert top item to the bottom of MEMCPY args, so that it is left on stack
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
emit.Int(c.prog.BinWriter, 6)
emit.Opcodes(c.prog.BinWriter, opcode.REVERSEN)
emit.Int(c.prog.BinWriter, 5)
emit.Opcodes(c.prog.BinWriter, opcode.REVERSEN)
}
emit.Opcodes(c.prog.BinWriter, opcode.MEMCPY)
2020-08-24 08:58:29 +00:00
case "make":
typ := c.typeOf(expr.Args[0])
switch {
case isMap(typ):
emit.Opcodes(c.prog.BinWriter, opcode.NEWMAP)
2020-08-24 08:58:29 +00:00
default:
if len(expr.Args) == 3 {
c.prog.Err = fmt.Errorf("`make()` with a capacity argument is not supported")
return
}
ast.Walk(c, expr.Args[1])
if isByteSlice(typ) {
emit.Opcodes(c.prog.BinWriter, opcode.NEWBUFFER)
2020-08-24 08:58:29 +00:00
} else {
neoT := toNeoType(typ.(*types.Slice).Elem())
emit.Instruction(c.prog.BinWriter, opcode.NEWARRAYT, []byte{byte(neoT)})
}
}
case "len":
emit.Opcodes(c.prog.BinWriter, opcode.DUP, opcode.ISNULL)
emit.Instruction(c.prog.BinWriter, opcode.JMPIF, []byte{2 + 1 + 2})
emit.Opcodes(c.prog.BinWriter, opcode.SIZE)
emit.Instruction(c.prog.BinWriter, opcode.JMP, []byte{2 + 1 + 1})
emit.Opcodes(c.prog.BinWriter, opcode.DROP, opcode.PUSH0)
case "append":
2019-09-12 14:39:23 +00:00
arg := expr.Args[0]
typ := c.typeInfo.Types[arg].Type
ast.Walk(c, arg)
emit.Opcodes(c.prog.BinWriter, opcode.DUP, opcode.ISNULL)
2020-05-13 15:09:55 +00:00
if isByteSlice(typ) {
emit.Instruction(c.prog.BinWriter, opcode.JMPIFNOT, []byte{2 + 3})
emit.Opcodes(c.prog.BinWriter, opcode.DROP, opcode.PUSHDATA1, 0)
if expr.Ellipsis.IsValid() {
ast.Walk(c, expr.Args[1])
} else {
c.convertByteArray(expr.Args[1:])
}
emit.Opcodes(c.prog.BinWriter, opcode.CAT)
} else {
emit.Instruction(c.prog.BinWriter, opcode.JMPIFNOT, []byte{2 + 2})
emit.Opcodes(c.prog.BinWriter, opcode.DROP, opcode.NEWARRAY0)
if expr.Ellipsis.IsValid() {
ast.Walk(c, expr.Args[1]) // x y
emit.Opcodes(c.prog.BinWriter, opcode.PUSH0) // x y cnt=0
start := c.newLabel()
c.setLabel(start)
emit.Opcodes(c.prog.BinWriter, opcode.PUSH2, opcode.PICK) // x y cnt x
emit.Opcodes(c.prog.BinWriter, opcode.PUSH2, opcode.PICK) // x y cnt x y
emit.Opcodes(c.prog.BinWriter, opcode.DUP, opcode.SIZE) // x y cnt x y len(y)
emit.Opcodes(c.prog.BinWriter, opcode.PUSH3, opcode.PICK) // x y cnt x y len(y) cnt
after := c.newLabel()
emit.Jmp(c.prog.BinWriter, opcode.JMPEQL, after) // x y cnt x y
emit.Opcodes(c.prog.BinWriter, opcode.PUSH2, opcode.PICK, // x y cnt x y cnt
opcode.PICKITEM, // x y cnt x y[cnt]
opcode.APPEND, // x=append(x, y[cnt]) y cnt
opcode.INC) // x y cnt+1
emit.Jmp(c.prog.BinWriter, opcode.JMPL, start)
c.setLabel(after)
for i := 0; i < 4; i++ { // leave x on stack
emit.Opcodes(c.prog.BinWriter, opcode.DROP)
}
} else {
for _, e := range expr.Args[1:] {
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
ast.Walk(c, e)
emit.Opcodes(c.prog.BinWriter, opcode.APPEND)
}
}
2019-09-12 14:39:23 +00:00
}
case "panic":
emit.Opcodes(c.prog.BinWriter, opcode.THROW)
2020-08-21 12:37:46 +00:00
case "recover":
if !c.scope.voidCalls[expr] {
c.emitLoadByIndex(varGlobal, c.exceptionIndex)
}
emit.Opcodes(c.prog.BinWriter, opcode.PUSHNULL)
2020-08-21 12:37:46 +00:00
c.emitStoreByIndex(varGlobal, c.exceptionIndex)
2020-09-06 12:49:41 +00:00
case "delete":
emit.Opcodes(c.prog.BinWriter, opcode.REMOVE)
case "FromAddress":
// We can be sure that this is a ast.BasicLit just containing a simple
// address string. Note that the string returned from calling Value will
// contain double quotes that need to be stripped.
addressStr := expr.Args[0].(*ast.BasicLit).Value
addressStr = strings.Replace(addressStr, "\"", "", 2)
uint160, err := address.StringToUint160(addressStr)
if err != nil {
2019-11-22 14:16:52 +00:00
c.prog.Err = err
return
}
2019-11-27 09:20:31 +00:00
bytes := uint160.BytesBE()
emit.Bytes(c.prog.BinWriter, bytes)
c.emitConvert(stackitem.BufferT)
}
}
// transformArgs returns a list of function arguments
// which should be put on stack.
// There are special cases for builtins:
// 1. With FromAddress, parameter conversion is happening at compile-time
// so there is no need to push parameters on stack and perform an actual call
// 2. With panic, generated code depends on if argument was nil or a string so
// it should be handled accordingly.
func transformArgs(fs *funcScope, fun ast.Expr, args []ast.Expr) []ast.Expr {
switch f := fun.(type) {
case *ast.SelectorExpr:
if f.Sel.Name == "FromAddress" {
return args[1:]
}
if fs != nil && isSyscall(fs) {
return nil
}
case *ast.Ident:
2020-08-24 09:54:01 +00:00
switch f.Name {
case "make", "copy", "append":
2020-08-24 08:58:29 +00:00
return nil
}
}
return args
}
// emitConvert converts top stack item to the specified type.
func (c *codegen) emitConvert(typ stackitem.Type) {
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
emit.Instruction(c.prog.BinWriter, opcode.ISTYPE, []byte{byte(typ)})
emit.Instruction(c.prog.BinWriter, opcode.JMPIF, []byte{2 + 2}) // After CONVERT.
emit.Instruction(c.prog.BinWriter, opcode.CONVERT, []byte{byte(typ)})
}
func (c *codegen) convertByteArray(elems []ast.Expr) {
buf := make([]byte, len(elems))
varIndices := []int{}
for i := 0; i < len(elems); i++ {
t := c.typeAndValueOf(elems[i])
if t.Value != nil {
val, _ := constant.Int64Val(t.Value)
buf[i] = byte(val)
} else {
varIndices = append(varIndices, i)
}
}
emit.Bytes(c.prog.BinWriter, buf)
c.emitConvert(stackitem.BufferT)
for _, i := range varIndices {
emit.Opcodes(c.prog.BinWriter, opcode.DUP)
emit.Int(c.prog.BinWriter, int64(i))
ast.Walk(c, elems[i])
emit.Opcodes(c.prog.BinWriter, opcode.SETITEM)
}
}
2020-01-23 14:06:15 +00:00
func (c *codegen) convertMap(lit *ast.CompositeLit) {
l := len(lit.Elts)
for i := l - 1; i >= 0; i-- {
2020-01-23 14:06:15 +00:00
elem := lit.Elts[i].(*ast.KeyValueExpr)
ast.Walk(c, elem.Value)
ast.Walk(c, elem.Key)
2020-01-23 14:06:15 +00:00
}
emit.Int(c.prog.BinWriter, int64(l))
emit.Opcodes(c.prog.BinWriter, opcode.PACKMAP)
2020-01-23 14:06:15 +00:00
}
func (c *codegen) getStruct(typ types.Type) (*types.Struct, bool) {
switch t := typ.Underlying().(type) {
case *types.Struct:
return t, true
case *types.Pointer:
strct, ok := t.Elem().Underlying().(*types.Struct)
return strct, ok
default:
return nil, false
}
}
func (c *codegen) convertStruct(lit *ast.CompositeLit, ptr bool) {
// Create a new structScope to initialize and store
// the positions of its variables.
2020-05-13 15:09:55 +00:00
strct, ok := c.typeOf(lit).Underlying().(*types.Struct)
if !ok {
2019-11-22 14:16:52 +00:00
c.prog.Err = fmt.Errorf("the given literal is not of type struct: %v", lit)
return
}
keyedLit := len(lit.Elts) > 0
if keyedLit {
_, ok := lit.Elts[0].(*ast.KeyValueExpr)
keyedLit = keyedLit && ok
}
// We need to locally store all the fields, even if they are not initialized.
// We will initialize all fields to their "zero" value.
for i := strct.NumFields() - 1; i >= 0; i-- {
sField := strct.Field(i)
var initialized bool
if !keyedLit {
if len(lit.Elts) > i {
ast.Walk(c, lit.Elts[i])
initialized = true
}
} else {
// Fields initialized by the program.
for _, field := range lit.Elts {
f := field.(*ast.KeyValueExpr)
fieldName := f.Key.(*ast.Ident).Name
if sField.Name() == fieldName {
ast.Walk(c, f.Value)
initialized = true
break
}
}
}
if !initialized {
c.emitDefault(sField.Type())
}
}
emit.Int(c.prog.BinWriter, int64(strct.NumFields()))
if ptr {
emit.Opcodes(c.prog.BinWriter, opcode.PACK)
} else {
emit.Opcodes(c.prog.BinWriter, opcode.PACKSTRUCT)
}
}
func (c *codegen) emitToken(tok token.Token, typ types.Type) {
op, err := convertToken(tok, typ)
if err != nil {
c.prog.Err = err
return
}
emit.Opcodes(c.prog.BinWriter, op)
}
func convertToken(tok token.Token, typ types.Type) (opcode.Opcode, error) {
switch tok {
2020-08-23 10:12:10 +00:00
case token.ADD_ASSIGN, token.ADD:
// VM has separate opcodes for number and string concatenation
if isString(typ) {
return opcode.CAT, nil
}
return opcode.ADD, nil
case token.SUB_ASSIGN:
return opcode.SUB, nil
case token.MUL_ASSIGN:
return opcode.MUL, nil
case token.QUO_ASSIGN:
return opcode.DIV, nil
2019-12-19 10:11:05 +00:00
case token.REM_ASSIGN:
return opcode.MOD, nil
case token.SUB:
return opcode.SUB, nil
case token.MUL:
return opcode.MUL, nil
case token.QUO:
return opcode.DIV, nil
2019-12-19 10:11:05 +00:00
case token.REM:
return opcode.MOD, nil
case token.LSS:
return opcode.LT, nil
case token.LEQ:
return opcode.LE, nil
case token.GTR:
return opcode.GT, nil
case token.GEQ:
return opcode.GE, nil
case token.EQL:
// VM has separate opcodes for number and string equality
if isNumber(typ) {
return opcode.NUMEQUAL, nil
}
return opcode.EQUAL, nil
case token.NEQ:
// VM has separate opcodes for number and string equality
if isNumber(typ) {
return opcode.NUMNOTEQUAL, nil
}
return opcode.NOTEQUAL, nil
case token.DEC:
return opcode.DEC, nil
case token.INC:
return opcode.INC, nil
case token.NOT:
return opcode.NOT, nil
case token.AND:
return opcode.AND, nil
case token.OR:
return opcode.OR, nil
case token.SHL:
return opcode.SHL, nil
case token.SHR:
return opcode.SHR, nil
case token.XOR:
return opcode.XOR, nil
default:
return 0, fmt.Errorf("compiler could not convert token: %s", tok)
}
}
func (c *codegen) newFunc(decl *ast.FuncDecl) *funcScope {
f := c.newFuncScope(decl, c.newLabel())
c.funcs[c.getFuncNameFromDecl("", decl)] = f
return f
}
func (c *codegen) getFuncFromIdent(fun *ast.Ident) (*funcScope, bool) {
var pkgName string
if len(c.pkgInfoInline) != 0 {
pkgName = c.pkgInfoInline[len(c.pkgInfoInline)-1].Pkg.Path()
}
f, ok := c.funcs[c.getIdentName(pkgName, fun.Name)]
return f, ok
}
// getFuncNameFromSelector returns fully-qualified function name from the selector expression.
// Second return value is true iff this was a method call, not foreign package call.
func (c *codegen) getFuncNameFromSelector(e *ast.SelectorExpr) (string, bool) {
ident := e.X.(*ast.Ident)
if c.typeInfo.Selections[e] != nil {
typ := c.typeInfo.Types[ident].Type.String()
return c.getIdentName(typ, e.Sel.Name), true
}
return c.getIdentName(ident.Name, e.Sel.Name), false
}
func (c *codegen) newLambda(u uint16, lit *ast.FuncLit) {
name := fmt.Sprintf("lambda@%d", u)
f := c.newFuncScope(&ast.FuncDecl{
Name: ast.NewIdent(name),
Type: lit.Type,
Body: lit.Body,
}, u)
c.lambda[c.getFuncNameFromDecl("", f.decl)] = f
}
func (c *codegen) compile(info *buildInfo, pkg *loader.PackageInfo) error {
c.mainPkg = pkg
c.analyzePkgOrder()
c.fillDocumentInfo()
funUsage := c.analyzeFuncUsage()
2019-10-22 14:56:03 +00:00
// Bring all imported functions into scope.
c.ForEachFile(c.resolveFuncDecls)
hasDeploy := c.traverseGlobals()
2020-10-02 10:16:02 +00:00
if hasDeploy {
deployOffset := c.prog.Len()
emit.Instruction(c.prog.BinWriter, opcode.INITSLOT, []byte{0, 2})
locCount := c.convertDeployFuncs()
c.reverseOffsetMap[deployOffset] = nameWithLocals{
name: "_deploy",
count: locCount,
}
2020-10-02 10:16:02 +00:00
c.deployEndOffset = c.prog.Len()
emit.Opcodes(c.prog.BinWriter, opcode.RET)
}
2019-10-22 14:56:03 +00:00
// sort map keys to generate code deterministically.
keys := make([]*types.Package, 0, len(info.program.AllPackages))
for p := range info.program.AllPackages {
keys = append(keys, p)
}
sort.Slice(keys, func(i, j int) bool { return keys[i].Path() < keys[j].Path() })
2019-10-22 14:56:03 +00:00
// Generate the code for the program.
c.ForEachFile(func(f *ast.File, pkg *types.Package) {
for _, decl := range f.Decls {
switch n := decl.(type) {
case *ast.FuncDecl:
// Don't convert the function if it's not used. This will save a lot
// of bytecode space.
2021-03-04 11:36:15 +00:00
pkgPath := ""
if pkg != c.mainPkg.Pkg { // not a main package
pkgPath = pkg.Path()
}
name := c.getFuncNameFromDecl(pkgPath, n)
if !isInitFunc(n) && !isDeployFunc(n) && funUsage.funcUsed(name) &&
(!isInteropPath(pkg.Path()) && !canInline(pkg.Path(), n.Name.Name)) {
c.convertFuncDecl(f, n, pkg)
}
}
}
})
return c.prog.Err
}
func newCodegen(info *buildInfo, pkg *loader.PackageInfo) *codegen {
return &codegen{
buildInfo: info,
prog: io.NewBufBinWriter(),
l: []int{},
funcs: map[string]*funcScope{},
lambda: map[string]*funcScope{},
reverseOffsetMap: map[int]nameWithLocals{},
globals: map[string]int{},
labels: map[labelWithType]uint16{},
typeInfo: &pkg.Info,
constMap: map[string]types.TypeAndValue{},
docIndex: map[string]int{},
2020-10-02 10:16:02 +00:00
initEndOffset: -1,
deployEndOffset: -1,
emittedEvents: make(map[string][][]string),
invokedContracts: make(map[util.Uint160][]string),
sequencePoints: make(map[string][]DebugSeqPoint),
}
}
// CodeGen compiles the program to bytecode.
func CodeGen(info *buildInfo) ([]byte, *DebugInfo, error) {
pkg := info.program.Package(info.initialPackage)
c := newCodegen(info, pkg)
if err := c.compile(info, pkg); err != nil {
return nil, nil, err
}
buf, err := c.writeJumps(c.prog.Bytes())
if err != nil {
return nil, nil, err
}
methods := bitfield.New(len(buf))
di := c.emitDebugInfo(buf)
for i := range di.Methods {
methods.Set(int(di.Methods[i].Range.Start))
}
return buf, di, vm.IsScriptCorrect(buf, methods)
}
func (c *codegen) resolveFuncDecls(f *ast.File, pkg *types.Package) {
for _, decl := range f.Decls {
switch n := decl.(type) {
case *ast.FuncDecl:
2021-02-04 12:41:00 +00:00
fs := c.newFunc(n)
fs.file = f
}
}
}
func (c *codegen) writeJumps(b []byte) ([]byte, error) {
ctx := vm.NewContext(b)
var offsets []int
for op, param, err := ctx.Next(); err == nil && ctx.IP() < len(b); op, param, err = ctx.Next() {
switch op {
case opcode.JMP, opcode.JMPIFNOT, opcode.JMPIF, opcode.CALL,
opcode.JMPEQ, opcode.JMPNE,
opcode.JMPGT, opcode.JMPGE, opcode.JMPLE, opcode.JMPLT:
case opcode.TRYL:
_, err := c.replaceLabelWithOffset(ctx.IP(), param)
2020-08-21 12:37:46 +00:00
if err != nil {
return nil, err
}
_, err = c.replaceLabelWithOffset(ctx.IP(), param[4:])
if err != nil {
return nil, err
}
case opcode.JMPL, opcode.JMPIFL, opcode.JMPIFNOTL,
opcode.JMPEQL, opcode.JMPNEL,
opcode.JMPGTL, opcode.JMPGEL, opcode.JMPLEL, opcode.JMPLTL,
opcode.CALLL, opcode.PUSHA, opcode.ENDTRYL:
offset, err := c.replaceLabelWithOffset(ctx.IP(), param)
if err != nil {
return nil, err
}
if op != opcode.PUSHA && math.MinInt8 <= offset && offset <= math.MaxInt8 {
offsets = append(offsets, ctx.IP())
}
case opcode.INITSLOT:
nextIP := ctx.NextIP()
info := c.reverseOffsetMap[ctx.IP()]
if argCount := b[nextIP-1]; info.count == 0 && argCount == 0 {
offsets = append(offsets, ctx.IP())
continue
}
if info.count > 255 {
return nil, fmt.Errorf("func '%s' has %d local variables (maximum is 255)", info.name, info.count)
}
b[nextIP-2] = byte(info.count)
}
}
if c.deployEndOffset >= 0 {
_, end := correctRange(uint16(c.initEndOffset+1), uint16(c.deployEndOffset), offsets)
c.deployEndOffset = int(end)
}
if c.initEndOffset > 0 {
_, end := correctRange(0, uint16(c.initEndOffset), offsets)
c.initEndOffset = int(end)
}
// Correct function ip range.
// Note: indices are sorted in increasing order.
for _, f := range c.funcs {
f.rng.Start, f.rng.End = correctRange(f.rng.Start, f.rng.End, offsets)
}
return shortenJumps(b, offsets), nil
}
func correctRange(start, end uint16, offsets []int) (uint16, uint16) {
newStart, newEnd := start, end
loop:
for _, ind := range offsets {
switch {
case ind > int(end):
break loop
case ind < int(start):
newStart -= longToShortRemoveCount
newEnd -= longToShortRemoveCount
case ind >= int(start):
newEnd -= longToShortRemoveCount
}
}
return newStart, newEnd
}
func (c *codegen) replaceLabelWithOffset(ip int, arg []byte) (int, error) {
index := binary.LittleEndian.Uint16(arg)
if int(index) > len(c.l) {
return 0, fmt.Errorf("unexpected label number: %d (max %d)", index, len(c.l))
}
if c.l[index] < 0 {
return 0, fmt.Errorf("invalid label target: %d at %d", c.l[index], ip)
}
offset := c.l[index] - ip
if offset > math.MaxInt32 || offset < math.MinInt32 {
return 0, fmt.Errorf("label offset is too big at the instruction %d: %d (max %d, min %d)",
ip, offset, math.MaxInt32, math.MinInt32)
}
binary.LittleEndian.PutUint32(arg, uint32(offset))
return offset, nil
}
// longToShortRemoveCount is a difference between short and long instruction sizes in bytes.
// By pure coincidence, this is also the size of `INITSLOT` instruction.
const longToShortRemoveCount = 3
// shortenJumps returns converts b to a program where all long JMP*/CALL* specified by absolute offsets,
// are replaced with their corresponding short counterparts. It panics if either b or offsets are invalid.
// This is done in 2 passes:
// 1. Alter jump offsets taking into account parts to be removed.
// 2. Perform actual removal of jump targets.
// Note: after jump offsets altering, there can appear new candidates for conversion.
// These are ignored for now.
func shortenJumps(b []byte, offsets []int) []byte {
if len(offsets) == 0 {
return b
}
// 1. Alter existing jump offsets.
ctx := vm.NewContext(b)
for op, _, err := ctx.Next(); err == nil && ctx.IP() < len(b); op, _, err = ctx.Next() {
// we can't use arg returned by ctx.Next() because it is copied
nextIP := ctx.NextIP()
ip := ctx.IP()
switch op {
case opcode.JMP, opcode.JMPIFNOT, opcode.JMPIF, opcode.CALL,
opcode.JMPEQ, opcode.JMPNE,
opcode.JMPGT, opcode.JMPGE, opcode.JMPLE, opcode.JMPLT, opcode.ENDTRY:
offset := int(int8(b[nextIP-1]))
offset += calcOffsetCorrection(ip, ip+offset, offsets)
b[nextIP-1] = byte(offset)
case opcode.TRY:
2020-08-21 12:37:46 +00:00
catchOffset := int(int8(b[nextIP-2]))
catchOffset += calcOffsetCorrection(ip, ip+catchOffset, offsets)
b[nextIP-1] = byte(catchOffset)
finallyOffset := int(int8(b[nextIP-1]))
finallyOffset += calcOffsetCorrection(ip, ip+finallyOffset, offsets)
b[nextIP-1] = byte(finallyOffset)
case opcode.JMPL, opcode.JMPIFL, opcode.JMPIFNOTL,
opcode.JMPEQL, opcode.JMPNEL,
opcode.JMPGTL, opcode.JMPGEL, opcode.JMPLEL, opcode.JMPLTL,
opcode.CALLL, opcode.PUSHA, opcode.ENDTRYL:
arg := b[nextIP-4:]
offset := int(int32(binary.LittleEndian.Uint32(arg)))
offset += calcOffsetCorrection(ip, ip+offset, offsets)
binary.LittleEndian.PutUint32(arg, uint32(offset))
case opcode.TRYL:
2020-08-21 12:37:46 +00:00
arg := b[nextIP-8:]
catchOffset := int(int32(binary.LittleEndian.Uint32(arg)))
catchOffset += calcOffsetCorrection(ip, ip+catchOffset, offsets)
binary.LittleEndian.PutUint32(arg, uint32(catchOffset))
arg = b[nextIP-4:]
finallyOffset := int(int32(binary.LittleEndian.Uint32(arg)))
finallyOffset += calcOffsetCorrection(ip, ip+finallyOffset, offsets)
binary.LittleEndian.PutUint32(arg, uint32(finallyOffset))
}
}
// 2. Convert instructions.
copyOffset := 0
l := len(offsets)
if op := opcode.Opcode(b[offsets[0]]); op != opcode.INITSLOT {
b[offsets[0]] = byte(toShortForm(op))
}
for i := 0; i < l; i++ {
start := offsets[i] + 2
if b[offsets[i]] == byte(opcode.INITSLOT) {
start = offsets[i]
}
end := len(b)
if i != l-1 {
end = offsets[i+1]
if op := opcode.Opcode(b[offsets[i+1]]); op != opcode.INITSLOT {
end += 2
b[offsets[i+1]] = byte(toShortForm(op))
}
}
copy(b[start-copyOffset:], b[start+3:end])
copyOffset += longToShortRemoveCount
}
return b[:len(b)-copyOffset]
}
func calcOffsetCorrection(ip, target int, offsets []int) int {
cnt := 0
start := sort.Search(len(offsets), func(i int) bool {
return offsets[i] >= ip || offsets[i] >= target
})
for i := start; i < len(offsets) && (offsets[i] < target || offsets[i] <= ip); i++ {
ind := offsets[i]
if ip <= ind && ind < target ||
ind != ip && target <= ind && ind <= ip {
cnt += longToShortRemoveCount
}
}
if ip < target {
return -cnt
}
return cnt
}
func negateJmp(op opcode.Opcode) opcode.Opcode {
switch op {
case opcode.JMPIFL:
return opcode.JMPIFNOTL
case opcode.JMPIFNOTL:
return opcode.JMPIFL
case opcode.JMPEQL:
return opcode.JMPNEL
case opcode.JMPNEL:
return opcode.JMPEQL
case opcode.JMPGTL:
return opcode.JMPLEL
case opcode.JMPGEL:
return opcode.JMPLTL
case opcode.JMPLEL:
return opcode.JMPGTL
case opcode.JMPLTL:
return opcode.JMPGEL
default:
panic(fmt.Errorf("invalid opcode in negateJmp: %s", op))
}
}
func toShortForm(op opcode.Opcode) opcode.Opcode {
switch op {
case opcode.JMPL:
return opcode.JMP
case opcode.JMPIFL:
return opcode.JMPIF
case opcode.JMPIFNOTL:
return opcode.JMPIFNOT
case opcode.JMPEQL:
return opcode.JMPEQ
case opcode.JMPNEL:
return opcode.JMPNE
case opcode.JMPGTL:
return opcode.JMPGT
case opcode.JMPGEL:
return opcode.JMPGE
case opcode.JMPLEL:
return opcode.JMPLE
case opcode.JMPLTL:
return opcode.JMPLT
case opcode.CALLL:
return opcode.CALL
case opcode.ENDTRYL:
return opcode.ENDTRY
default:
panic(fmt.Errorf("invalid opcode: %s", op))
}
}