ir

package
v0.4.1 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Aug 2, 2022 License: MIT, BSD-3-Clause Imports: 24 Imported by: 0

Documentation

Overview

Package ir defines a representation of the elements of Go programs (packages, types, functions, variables and constants) using a static single-information (SSI) form intermediate representation (IR) for the bodies of functions.

THIS INTERFACE IS EXPERIMENTAL AND IS LIKELY TO CHANGE.

For an introduction to SSA form, upon which SSI builds, see http://en.wikipedia.org/wiki/Static_single_assignment_form. This page provides a broader reading list: http://www.dcs.gla.ac.uk/~jsinger/ssa.html.

For an introduction to SSI form, see The static single information form by C. Scott Ananian.

The level of abstraction of the IR form is intentionally close to the source language to facilitate construction of source analysis tools. It is not intended for machine code generation.

The simplest way to create the IR of a package is to load typed syntax trees using golang.org/x/tools/go/packages, then invoke the irutil.Packages helper function. See ExampleLoadPackages and ExampleWholeProgram for examples. The resulting ir.Program contains all the packages and their members, but IR code is not created for function bodies until a subsequent call to (*Package).Build or (*Program).Build.

The builder initially builds a naive IR form in which all local variables are addresses of stack locations with explicit loads and stores. Registerization of eligible locals and φ-node insertion using dominance and dataflow are then performed as a second pass called "lifting" to improve the accuracy and performance of subsequent analyses; this pass can be skipped by setting the NaiveForm builder flag.

The primary interfaces of this package are:

  • Member: a named member of a Go package.
  • Value: an expression that yields a value.
  • Instruction: a statement that consumes values and performs computation.
  • Node: a Value or Instruction (emphasizing its membership in the IR value graph)

A computation that yields a result implements both the Value and Instruction interfaces. The following table shows for each concrete type which of these interfaces it implements.

                   Value?          Instruction?    Member?
*Alloc                ✔               ✔
*BinOp                ✔               ✔
*BlankStore                           ✔
*Builtin              ✔
*Call                 ✔               ✔
*ChangeInterface      ✔               ✔
*ChangeType           ✔               ✔
*Const                ✔               ✔
*Convert              ✔               ✔
*DebugRef                             ✔
*Defer                ✔               ✔
*Extract              ✔               ✔
*Field                ✔               ✔
*FieldAddr            ✔               ✔
*FreeVar              ✔
*Function             ✔                               ✔ (func)
*Global               ✔                               ✔ (var)
*Go                   ✔               ✔
*If                                   ✔
*Index                ✔               ✔
*IndexAddr            ✔               ✔
*Jump                                 ✔
*Load                 ✔               ✔
*MakeChan             ✔               ✔
*MakeClosure          ✔               ✔
*MakeInterface        ✔               ✔
*MakeMap              ✔               ✔
*MakeSlice            ✔               ✔
*MapLookup            ✔               ✔
*MapUpdate            ✔               ✔
*NamedConst                                           ✔ (const)
*Next                 ✔               ✔
*Panic                                ✔
*Parameter            ✔               ✔
*Phi                  ✔               ✔
*Range                ✔               ✔
*Recv                 ✔               ✔
*Return                               ✔
*RunDefers                            ✔
*Select               ✔               ✔
*Send                 ✔               ✔
*Sigma                ✔               ✔
*Slice                ✔               ✔
*SliceToArrayPointer  ✔               ✔
*Store                ✔               ✔
*StringLookup         ✔               ✔
*Type                                                 ✔ (type)
*TypeAssert           ✔               ✔
*UnOp                 ✔               ✔
*Unreachable                          ✔

Other key types in this package include: Program, Package, Function and BasicBlock.

The program representation constructed by this package is fully resolved internally, i.e. it does not rely on the names of Values, Packages, Functions, Types or BasicBlocks for the correct interpretation of the program. Only the identities of objects and the topology of the IR and type graphs are semantically significant. (There is one exception: Ids, used to identify field and method names, contain strings.) Avoidance of name-based operations simplifies the implementation of subsequent passes and can make them very efficient. Many objects are nonetheless named to aid in debugging, but it is not essential that the names be either accurate or unambiguous. The public API exposes a number of name-based maps for client convenience.

The ir/irutil package provides various utilities that depend only on the public API of this package.

TODO(adonovan): Consider the exceptional control-flow implications of defer and recover().

TODO(adonovan): write a how-to document for all the various cases of trying to determine corresponding elements across the four domains of source locations, ast.Nodes, types.Objects, ir.Values/Instructions.

Example (BuildPackage)

This program demonstrates how to run the IR builder on a single package of one or more already-parsed files. Its dependencies are loaded from compiler export data. This is what you'd typically use for a compiler; it does not depend on golang.org/x/tools/go/loader.

It shows the printed representation of packages, functions, and instructions. Within the function listing, the name of each BasicBlock such as ".0.entry" is printed left-aligned, followed by the block's Instructions.

For each instruction that defines an IR virtual register (i.e. implements Value), the type of that value is shown in the right column.

Build and run the irdump.go program if you want a standalone tool with similar functionality. It is located at github.com/Abirdcfly/go-tools/internal/cmd/irdump.

package main

import (
	"bytes"
	"fmt"
	"go/ast"
	"go/importer"
	"go/parser"
	"go/token"
	"go/types"
	"os"
	"strings"

	"github.com/Abirdcfly/go-tools/go/ir"
	"github.com/Abirdcfly/go-tools/go/ir/irutil"
)

const hello = `
package main

import "fmt"

const message = "Hello, World!"

func main() {
	fmt.Println(message)
}
`

func main() {
	// Parse the source files.
	fset := token.NewFileSet()
	f, err := parser.ParseFile(fset, "hello.go", hello, parser.ParseComments|parser.SkipObjectResolution)
	if err != nil {
		fmt.Print(err) // parse error
		return
	}
	files := []*ast.File{f}

	// Create the type-checker's package.
	pkg := types.NewPackage("hello", "")

	// Type-check the package, load dependencies.
	// Create and build the IR program.
	hello, _, err := irutil.BuildPackage(
		&types.Config{Importer: importer.Default()}, fset, pkg, files, ir.SanityCheckFunctions)
	if err != nil {
		fmt.Print(err) // type error in some package
		return
	}

	// Print out the package.
	hello.WriteTo(os.Stdout)

	// Print out the package-level functions.
	// Replace interface{} with any so the tests work for Go 1.17 and Go 1.18.
	{
		var buf bytes.Buffer
		ir.WriteFunction(&buf, hello.Func("init"))
		fmt.Print(strings.ReplaceAll(buf.String(), "interface{}", "any"))
	}
	{
		var buf bytes.Buffer
		ir.WriteFunction(&buf, hello.Func("main"))
		fmt.Print(strings.ReplaceAll(buf.String(), "interface{}", "any"))
	}

}
Output:

package hello:
  func  init       func()
  var   init$guard bool
  func  main       func()
  const message    message = Const <untyped string> {"Hello, World!"}

# Name: hello.init
# Package: hello
# Synthetic: package initializer
func init():
b0: # entry
	t1 = Const <bool> {true}
	t2 = Load <bool> init$guard
	If t2 → b1 b2

b1: ← b0 b2 # exit
	Return

b2: ← b0 # init.start
	Store {bool} init$guard t1
	t6 = Call <()> fmt.init
	Jump → b1

# Name: hello.main
# Package: hello
# Location: hello.go:8:1
func main():
b0: # entry
	t1 = Const <string> {"Hello, World!"}
	t2 = Const <int> {0}
	t3 = HeapAlloc <*[1]any>
	t4 = IndexAddr <*any> t3 t2
	t5 = MakeInterface <any> t1
	Store {any} t4 t5
	t7 = Slice <[]any> t3 <nil> <nil> <nil>
	t8 = Call <(n int, err error)> fmt.Println t7
	Jump → b1

b1: ← b0 # exit
	Return
Example (LoadPackages)

This example builds IR code for a set of packages using the x/tools/go/packages API. This is what you would typically use for a analysis capable of operating on a single package.

package main

import (
	"log"

	"github.com/Abirdcfly/go-tools/go/ir"
	"github.com/Abirdcfly/go-tools/go/ir/irutil"

	"golang.org/x/tools/go/packages"
)

func main() {
	// Load, parse, and type-check the initial packages.
	cfg := &packages.Config{Mode: packages.NeedName | packages.NeedFiles | packages.NeedCompiledGoFiles | packages.NeedImports | packages.NeedTypes | packages.NeedTypesSizes | packages.NeedSyntax | packages.NeedTypesInfo}
	initial, err := packages.Load(cfg, "fmt", "net/http")
	if err != nil {
		log.Fatal(err)
	}

	// Stop if any package had errors.
	// This step is optional; without it, the next step
	// will create IR for only a subset of packages.
	if packages.PrintErrors(initial) > 0 {
		log.Fatalf("packages contain errors")
	}

	// Create IR packages for all well-typed packages.
	prog, pkgs := irutil.Packages(initial, ir.PrintPackages, nil)
	_ = prog

	// Build IR code for the well-typed initial packages.
	for _, p := range pkgs {
		if p != nil {
			p.Build()
		}
	}
}
Output:

Example (LoadWholeProgram)

This example builds IR code for a set of packages plus all their dependencies, using the x/tools/go/packages API. This is what you'd typically use for a whole-program analysis.

package main

import (
	"log"

	"github.com/Abirdcfly/go-tools/go/ir"
	"github.com/Abirdcfly/go-tools/go/ir/irutil"

	"golang.org/x/tools/go/packages"
)

func main() {
	// Load, parse, and type-check the whole program.
	cfg := packages.Config{Mode: packages.NeedName | packages.NeedFiles | packages.NeedCompiledGoFiles | packages.NeedImports | packages.NeedTypes | packages.NeedTypesSizes | packages.NeedSyntax | packages.NeedTypesInfo | packages.NeedDeps}
	initial, err := packages.Load(&cfg, "fmt", "net/http")
	if err != nil {
		log.Fatal(err)
	}

	// Create IR packages for well-typed packages and their dependencies.
	prog, pkgs := irutil.AllPackages(initial, ir.PrintPackages, nil)
	_ = pkgs

	// Build IR code for the whole program.
	prog.Build()
}
Output:

Index

Examples

Constants

View Source
const BuilderModeDoc = `` /* 480-byte string literal not displayed */

Variables

This section is empty.

Functions

func HasEnclosingFunction

func HasEnclosingFunction(pkg *Package, path []ast.Node) bool

HasEnclosingFunction returns true if the AST node denoted by path is contained within the declaration of some function or package-level variable.

Unlike EnclosingFunction, the behaviour of this function does not depend on whether IR code for pkg has been built, so it can be used to quickly reject check inputs that will cause EnclosingFunction to fail, prior to IR building.

func WriteFunction

func WriteFunction(buf *bytes.Buffer, f *Function)

WriteFunction writes to buf a human-readable "disassembly" of f.

func WritePackage

func WritePackage(buf *bytes.Buffer, p *Package)

WritePackage writes to buf a human-readable summary of p.

Types

type AggregateConst

type AggregateConst struct {
	Values []Value
	// contains filtered or unexported fields
}

func (*AggregateConst) Name

func (v *AggregateConst) Name() string

func (*AggregateConst) Operands

func (v *AggregateConst) Operands(rands []*Value) []*Value

func (*AggregateConst) Referrers

func (v *AggregateConst) Referrers() *[]Instruction

func (*AggregateConst) RelString

func (v *AggregateConst) RelString(pkg *types.Package) string

func (*AggregateConst) String

func (v *AggregateConst) String() string

func (*AggregateConst) Type

func (v *AggregateConst) Type() types.Type

type Alloc

type Alloc struct {
	Heap bool
	// contains filtered or unexported fields
}

The Alloc instruction reserves space for a variable of the given type, zero-initializes it, and yields its address.

Alloc values are always addresses, and have pointer types, so the type of the allocated variable is actually Type().Underlying().(*types.Pointer).Elem().

If Heap is false, Alloc allocates space in the function's activation record (frame); we refer to an Alloc(Heap=false) as a "stack" alloc. Each stack Alloc returns the same address each time it is executed within the same activation; the space is re-initialized to zero.

If Heap is true, Alloc allocates space in the heap; we refer to an Alloc(Heap=true) as a "heap" alloc. Each heap Alloc returns a different address each time it is executed.

When Alloc is applied to a channel, map or slice type, it returns the address of an uninitialized (nil) reference of that kind; store the result of MakeSlice, MakeMap or MakeChan in that location to instantiate these types.

Pos() returns the ast.CompositeLit.Lbrace for a composite literal, or the ast.CallExpr.Rparen for a call to new() or for a call that allocates a varargs slice.

Example printed form:

t1 = StackAlloc <*int>
t2 = HeapAlloc <*int> (new)

func (*Alloc) Name

func (v *Alloc) Name() string

func (*Alloc) Operands

func (v *Alloc) Operands(rands []*Value) []*Value

func (*Alloc) Referrers

func (v *Alloc) Referrers() *[]Instruction

func (*Alloc) String

func (v *Alloc) String() string

func (*Alloc) Type

func (v *Alloc) Type() types.Type

type ArrayConst

type ArrayConst struct {
	// contains filtered or unexported fields
}

TODO add the element's zero constant to ArrayConst

func (*ArrayConst) Name

func (v *ArrayConst) Name() string

func (*ArrayConst) Operands

func (v *ArrayConst) Operands(rands []*Value) []*Value

func (*ArrayConst) Referrers

func (v *ArrayConst) Referrers() *[]Instruction

func (*ArrayConst) RelString

func (v *ArrayConst) RelString(pkg *types.Package) string

func (*ArrayConst) String

func (v *ArrayConst) String() string

func (*ArrayConst) Type

func (v *ArrayConst) Type() types.Type

type BasicBlock

type BasicBlock struct {
	Index   int    // index of this block within Parent().Blocks
	Comment string // optional label; no semantic significance

	Instrs       []Instruction // instructions in order
	Preds, Succs []*BasicBlock // predecessors and successors
	// contains filtered or unexported fields
}

BasicBlock represents an IR basic block.

The final element of Instrs is always an explicit transfer of control (If, Jump, Return, Panic, or Unreachable).

A block may contain no Instructions only if it is unreachable, i.e., Preds is nil. Empty blocks are typically pruned.

BasicBlocks and their Preds/Succs relation form a (possibly cyclic) graph independent of the IR Value graph: the control-flow graph or CFG. It is illegal for multiple edges to exist between the same pair of blocks.

Each BasicBlock is also a node in the dominator tree of the CFG. The tree may be navigated using Idom()/Dominees() and queried using Dominates().

The order of Preds and Succs is significant (to Phi and If instructions, respectively).

func (*BasicBlock) Control

func (b *BasicBlock) Control() Instruction

Control returns the last instruction in the block.

func (*BasicBlock) Dominates

func (b *BasicBlock) Dominates(c *BasicBlock) bool

Dominates reports whether b dominates c.

func (*BasicBlock) Dominees

func (b *BasicBlock) Dominees() []*BasicBlock

Dominees returns the list of blocks that b immediately dominates: its children in the dominator tree.

func (*BasicBlock) Idom

func (b *BasicBlock) Idom() *BasicBlock

Idom returns the block that immediately dominates b: its parent in the dominator tree, if any. The entry node (b.Index==0) does not have a parent.

func (*BasicBlock) Parent

func (b *BasicBlock) Parent() *Function

Parent returns the function that contains block b.

func (*BasicBlock) Phis

func (b *BasicBlock) Phis() []Instruction

func (*BasicBlock) SigmaFor

func (b *BasicBlock) SigmaFor(v Value, pred *BasicBlock) *Sigma

SigmaFor returns the sigma node for v coming from pred.

func (*BasicBlock) String

func (b *BasicBlock) String() string

String returns a human-readable label of this block. It is not guaranteed unique within the function.

type BinOp

type BinOp struct {

	// One of:
	// ADD SUB MUL QUO REM          + - * / %
	// AND OR XOR SHL SHR AND_NOT   & | ^ << >> &^
	// EQL NEQ LSS LEQ GTR GEQ      == != < <= < >=
	Op   token.Token
	X, Y Value
	// contains filtered or unexported fields
}

The BinOp instruction yields the result of binary operation X Op Y.

Pos() returns the ast.BinaryExpr.OpPos, if explicit in the source.

Example printed form:

t3 = BinOp <int> {+} t2 t1

func (*BinOp) Name

func (v *BinOp) Name() string

func (*BinOp) Operands

func (v *BinOp) Operands(rands []*Value) []*Value

func (*BinOp) Referrers

func (v *BinOp) Referrers() *[]Instruction

func (*BinOp) String

func (v *BinOp) String() string

func (*BinOp) Type

func (v *BinOp) Type() types.Type

type BlankStore

type BlankStore struct {
	Val Value
	// contains filtered or unexported fields
}

The BlankStore instruction is emitted for assignments to the blank identifier.

BlankStore is a pseudo-instruction: it has no dynamic effect.

Pos() returns NoPos.

Example printed form:

BlankStore t1

func (*BlankStore) Block

func (v *BlankStore) Block() *BasicBlock

func (BlankStore) Comment

func (instr BlankStore) Comment() string

func (*BlankStore) Operands

func (s *BlankStore) Operands(rands []*Value) []*Value

func (*BlankStore) Parent

func (v *BlankStore) Parent() *Function

func (*BlankStore) Referrers

func (v *BlankStore) Referrers() *[]Instruction

func (*BlankStore) String

func (s *BlankStore) String() string

type BlockMap

type BlockMap[T any] []T

BlockMap is a mapping from basic blocks (identified by their indices) to values.

type BlockSet

type BlockSet struct {
	// contains filtered or unexported fields
}

func NewBlockSet

func NewBlockSet(size int) *BlockSet

func (*BlockSet) Add

func (s *BlockSet) Add(b *BasicBlock) bool

add adds b to the set and returns true if the set changed.

func (*BlockSet) Clear

func (s *BlockSet) Clear()

func (*BlockSet) Has

func (s *BlockSet) Has(b *BasicBlock) bool

func (*BlockSet) Num

func (s *BlockSet) Num() int

func (*BlockSet) Set

func (s *BlockSet) Set(s2 *BlockSet)

func (*BlockSet) Take

func (s *BlockSet) Take() int

take removes an arbitrary element from a set s and returns its index, or returns -1 if empty.

type BuilderMode

type BuilderMode uint

BuilderMode is a bitmask of options for diagnostics and checking.

*BuilderMode satisfies the flag.Value interface. Example:

var mode = ir.BuilderMode(0)
func init() { flag.Var(&mode, "build", ir.BuilderModeDoc) }
const (
	PrintPackages            BuilderMode = 1 << iota // Print package inventory to stdout
	PrintFunctions                                   // Print function IR code to stdout
	PrintSource                                      // Print source code when printing function IR
	LogSource                                        // Log source locations as IR builder progresses
	SanityCheckFunctions                             // Perform sanity checking of function bodies
	NaiveForm                                        // Build naïve IR form: don't replace local loads/stores with registers
	GlobalDebug                                      // Enable debug info for all packages
	SplitAfterNewInformation                         // Split live range after we learn something new about a value
)

func (BuilderMode) Get

func (m BuilderMode) Get() interface{}

Get returns m.

func (*BuilderMode) Set

func (m *BuilderMode) Set(s string) error

Set parses the flag characters in s and updates *m.

func (BuilderMode) String

func (m BuilderMode) String() string

type Builtin

type Builtin struct {
	// contains filtered or unexported fields
}

A Builtin represents a specific use of a built-in function, e.g. len.

Builtins are immutable values. Builtins do not have addresses. Builtins can only appear in CallCommon.Func.

Name() indicates the function: one of the built-in functions from the Go spec (excluding "make" and "new") or one of these ir-defined intrinsics:

// wrapnilchk returns ptr if non-nil, panics otherwise.
// (For use in indirection wrappers.)
func ir:wrapnilchk(ptr *T, recvType, methodName string) *T

// noreturnWasPanic returns true if the previously called
// function panicked, false if it exited the process.
func ir:noreturnWasPanic() bool

Object() returns a *types.Builtin for built-ins defined by the spec, nil for others.

Type() returns a *types.Signature representing the effective signature of the built-in for this call.

func (Builtin) ID

func (n Builtin) ID() ID

func (*Builtin) Name

func (v *Builtin) Name() string

func (*Builtin) Object

func (v *Builtin) Object() types.Object

func (*Builtin) Operands

func (v *Builtin) Operands(rands []*Value) []*Value

Non-Instruction Values:

func (*Builtin) Parent

func (v *Builtin) Parent() *Function

func (*Builtin) Pos

func (v *Builtin) Pos() token.Pos

func (*Builtin) Referrers

func (*Builtin) Referrers() *[]Instruction

func (*Builtin) Source

func (n *Builtin) Source() ast.Node

func (*Builtin) String

func (v *Builtin) String() string

func (*Builtin) Type

func (v *Builtin) Type() types.Type

type Call

type Call struct {
	Call CallCommon
	// contains filtered or unexported fields
}

The Call instruction represents a function or method call.

The Call instruction yields the function result if there is exactly one. Otherwise it returns a tuple, the components of which are accessed via Extract.

See CallCommon for generic function call documentation.

Pos() returns the ast.CallExpr.Lparen, if explicit in the source.

Example printed form:

t3 = Call <()> println t1 t2
t4 = Call <()> foo$1
t6 = Invoke <string> t5.String

func (*Call) Common

func (s *Call) Common() *CallCommon

func (*Call) Name

func (v *Call) Name() string

func (*Call) Operands

func (s *Call) Operands(rands []*Value) []*Value

func (*Call) Referrers

func (v *Call) Referrers() *[]Instruction

func (*Call) String

func (v *Call) String() string

func (*Call) Type

func (v *Call) Type() types.Type

func (*Call) Value

func (s *Call) Value() *Call

type CallCommon

type CallCommon struct {
	Value    Value       // receiver (invoke mode) or func value (call mode)
	Method   *types.Func // abstract method (invoke mode)
	Args     []Value     // actual parameters (in static method call, includes receiver)
	TypeArgs []types.Type
	Results  Value
}

CallCommon is contained by Go, Defer and Call to hold the common parts of a function or method call.

Each CallCommon exists in one of two modes, function call and interface method invocation, or "call" and "invoke" for short.

1. "call" mode: when Method is nil (!IsInvoke), a CallCommon represents an ordinary function call of the value in Value, which may be a *Builtin, a *Function or any other value of kind 'func'.

Value may be one of:

(a) a *Function, indicating a statically dispatched call
    to a package-level function, an anonymous function, or
    a method of a named type.
(b) a *MakeClosure, indicating an immediately applied
    function literal with free variables.
(c) a *Builtin, indicating a statically dispatched call
    to a built-in function.
(d) any other value, indicating a dynamically dispatched
    function call.

StaticCallee returns the identity of the callee in cases (a) and (b), nil otherwise.

Args contains the arguments to the call. If Value is a method, Args[0] contains the receiver parameter.

Example printed form:

t3 = Call <()> println t1 t2
Go t3
Defer t3

2. "invoke" mode: when Method is non-nil (IsInvoke), a CallCommon represents a dynamically dispatched call to an interface method. In this mode, Value is the interface value and Method is the interface's abstract method. Note: an abstract method may be shared by multiple interfaces due to embedding; Value.Type() provides the specific interface used for this call.

Value is implicitly supplied to the concrete method implementation as the receiver parameter; in other words, Args[0] holds not the receiver but the first true argument.

Example printed form:

t6 = Invoke <string> t5.String
GoInvoke t4.Bar t2
DeferInvoke t4.Bar t2

For all calls to variadic functions (Signature().Variadic()), the last element of Args is a slice.

func (*CallCommon) Description

func (c *CallCommon) Description() string

Description returns a description of the mode of this call suitable for a user interface, e.g., "static method call".

func (*CallCommon) IsInvoke

func (c *CallCommon) IsInvoke() bool

IsInvoke returns true if this call has "invoke" (not "call") mode.

func (*CallCommon) Operands

func (c *CallCommon) Operands(rands []*Value) []*Value

func (*CallCommon) Signature

func (c *CallCommon) Signature() *types.Signature

Signature returns the signature of the called function.

For an "invoke"-mode call, the signature of the interface method is returned.

In either "call" or "invoke" mode, if the callee is a method, its receiver is represented by sig.Recv, not sig.Params().At(0).

func (*CallCommon) StaticCallee

func (c *CallCommon) StaticCallee() *Function

StaticCallee returns the callee if this is a trivially static "call"-mode call to a function.

func (*CallCommon) String

func (c *CallCommon) String() string

type CallInstruction

type CallInstruction interface {
	Instruction
	Common() *CallCommon // returns the common parts of the call
	Value() *Call
}

The CallInstruction interface, implemented by *Go, *Defer and *Call, exposes the common parts of function-calling instructions, yet provides a way back to the Value defined by *Call alone.

type ChangeInterface

type ChangeInterface struct {
	X Value
	// contains filtered or unexported fields
}

ChangeInterface constructs a value of one interface type from a value of another interface type known to be assignable to it. This operation cannot fail.

Pos() returns the ast.CallExpr.Lparen if the instruction arose from an explicit T(e) conversion; the ast.TypeAssertExpr.Lparen if the instruction arose from an explicit e.(T) operation; or token.NoPos otherwise.

Example printed form:

t2 = ChangeInterface <I1> t1

func (*ChangeInterface) Name

func (v *ChangeInterface) Name() string

func (*ChangeInterface) Operands

func (v *ChangeInterface) Operands(rands []*Value) []*Value

func (*ChangeInterface) Referrers

func (v *ChangeInterface) Referrers() *[]Instruction

func (*ChangeInterface) String

func (v *ChangeInterface) String() string

func (*ChangeInterface) Type

func (v *ChangeInterface) Type() types.Type

type ChangeType

type ChangeType struct {
	X Value
	// contains filtered or unexported fields
}

The ChangeType instruction applies to X a value-preserving type change to Type().

Type changes are permitted:

  • between a named type and its underlying type.
  • between two named types of the same underlying type.
  • between (possibly named) pointers to identical base types.
  • from a bidirectional channel to a read- or write-channel, optionally adding/removing a name.

This operation cannot fail dynamically.

Pos() returns the ast.CallExpr.Lparen, if the instruction arose from an explicit conversion in the source.

Example printed form:

t2 = ChangeType <*T> t1

func (*ChangeType) Name

func (v *ChangeType) Name() string

func (*ChangeType) Operands

func (v *ChangeType) Operands(rands []*Value) []*Value

func (*ChangeType) Referrers

func (v *ChangeType) Referrers() *[]Instruction

func (*ChangeType) String

func (v *ChangeType) String() string

func (*ChangeType) Type

func (v *ChangeType) Type() types.Type

type CompositeValue

type CompositeValue struct {

	// Bitmap records which elements were explicitly provided. For example, [4]byte{2: x} would have a bitmap of 0010.
	Bitmap big.Int
	// The number of bits set in Bitmap
	NumSet int
	// Dense list of values in the composite literal. Omitted elements are filled in with zero values.
	Values []Value
	// contains filtered or unexported fields
}

func (*CompositeValue) Name

func (v *CompositeValue) Name() string

func (*CompositeValue) Operands

func (v *CompositeValue) Operands(rands []*Value) []*Value

func (*CompositeValue) Referrers

func (v *CompositeValue) Referrers() *[]Instruction

func (*CompositeValue) String

func (v *CompositeValue) String() string

func (*CompositeValue) Type

func (v *CompositeValue) Type() types.Type

type Const

type Const struct {
	Value constant.Value
	// contains filtered or unexported fields
}

A Const represents the value of a constant expression.

The underlying type of a constant may be any boolean, numeric, or string type. In addition, a Const may represent the nil value of any reference type---interface, map, channel, pointer, slice, or function---but not "untyped nil".

All source-level constant expressions are represented by a Const of the same type and value.

Value holds the exact value of the constant, independent of its Type(), using the same representation as package go/constant uses for constants, or nil for a typed nil value.

Pos() returns token.NoPos.

Example printed form:

Const <int> {42}
Const <untyped string> {"test"}
Const <MyComplex> {(3 + 4i)}

func NewConst

func NewConst(val constant.Value, typ types.Type) *Const

NewConst returns a new constant of the specified value and type. val must be valid according to the specification of Const.Value.

func (*Const) Complex128

func (c *Const) Complex128() complex128

Complex128 returns the complex value of this constant truncated to fit a complex128.

func (*Const) Float64

func (c *Const) Float64() float64

Float64 returns the numeric value of this constant truncated to fit a float64.

func (*Const) Int64

func (c *Const) Int64() int64

Int64 returns the numeric value of this constant truncated to fit a signed 64-bit integer.

func (*Const) IsNil

func (c *Const) IsNil() bool

IsNil returns true if this constant represents a typed or untyped nil value.

func (*Const) Name

func (v *Const) Name() string

func (*Const) Operands

func (v *Const) Operands(rands []*Value) []*Value

func (*Const) Referrers

func (v *Const) Referrers() *[]Instruction

func (*Const) RelString

func (c *Const) RelString(from *types.Package) string

func (*Const) String

func (c *Const) String() string

func (*Const) Type

func (v *Const) Type() types.Type

func (*Const) Uint64

func (c *Const) Uint64() uint64

Uint64 returns the numeric value of this constant truncated to fit an unsigned 64-bit integer.

type Constant

type Constant interface {
	Instruction
	Value

	RelString(*types.Package) string
	// contains filtered or unexported methods
}

type ConstantSwitch

type ConstantSwitch struct {
	Tag Value
	// Constant branch conditions. A nil Value denotes the (implicit
	// or explicit) default branch.
	Conds []Value
	// contains filtered or unexported fields
}

func (*ConstantSwitch) Block

func (v *ConstantSwitch) Block() *BasicBlock

func (ConstantSwitch) Comment

func (instr ConstantSwitch) Comment() string

func (*ConstantSwitch) Operands

func (s *ConstantSwitch) Operands(rands []*Value) []*Value

func (*ConstantSwitch) Parent

func (v *ConstantSwitch) Parent() *Function

func (*ConstantSwitch) Referrers

func (v *ConstantSwitch) Referrers() *[]Instruction

func (*ConstantSwitch) String

func (s *ConstantSwitch) String() string

type Convert

type Convert struct {
	X Value
	// contains filtered or unexported fields
}

The Convert instruction yields the conversion of value X to type Type(). One or both of those types is basic (but possibly named).

A conversion may change the value and representation of its operand. Conversions are permitted:

  • between real numeric types.
  • between complex numeric types.
  • between string and []byte or []rune.
  • between pointers and unsafe.Pointer.
  • between unsafe.Pointer and uintptr.
  • from (Unicode) integer to (UTF-8) string.

A conversion may imply a type name change also.

This operation cannot fail dynamically.

Conversions of untyped string/number/bool constants to a specific representation are eliminated during IR construction.

Pos() returns the ast.CallExpr.Lparen, if the instruction arose from an explicit conversion in the source.

Example printed form:

t2 = Convert <[]byte> t1

func (*Convert) Name

func (v *Convert) Name() string

func (*Convert) Operands

func (v *Convert) Operands(rands []*Value) []*Value

func (*Convert) Referrers

func (v *Convert) Referrers() *[]Instruction

func (*Convert) String

func (v *Convert) String() string

func (*Convert) Type

func (v *Convert) Type() types.Type

type Copy

type Copy struct {
	X    Value
	Why  Instruction
	Info CopyInfo
	// contains filtered or unexported fields
}

func (*Copy) Name

func (v *Copy) Name() string

func (*Copy) Operands

func (s *Copy) Operands(rands []*Value) []*Value

func (*Copy) Referrers

func (v *Copy) Referrers() *[]Instruction

func (*Copy) String

func (v *Copy) String() string

func (*Copy) Type

func (v *Copy) Type() types.Type

type CopyInfo

type CopyInfo uint64
const (
	CopyInfoUnspecified CopyInfo = 0
	CopyInfoNotNil      CopyInfo = 1 << iota
	CopyInfoNotZeroLength
	CopyInfoNotNegative
	CopyInfoSingleConcreteType
	CopyInfoClosed
)

type DebugRef

type DebugRef struct {
	Expr ast.Expr // the referring expression (never *ast.ParenExpr)

	IsAddr bool  // Expr is addressable and X is the address it denotes
	X      Value // the value or address of Expr
	// contains filtered or unexported fields
}

A DebugRef instruction maps a source-level expression Expr to the IR value X that represents the value (!IsAddr) or address (IsAddr) of that expression.

DebugRef is a pseudo-instruction: it has no dynamic effect.

Pos() returns Expr.Pos(), the start position of the source-level expression. This is not the same as the "designated" token as documented at Value.Pos(). e.g. CallExpr.Pos() does not return the position of the ("designated") Lparen token.

DebugRefs are generated only for functions built with debugging enabled; see Package.SetDebugMode() and the GlobalDebug builder mode flag.

DebugRefs are not emitted for ast.Idents referring to constants or predeclared identifiers, since they are trivial and numerous. Nor are they emitted for ast.ParenExprs.

(By representing these as instructions, rather than out-of-band, consistency is maintained during transformation passes by the ordinary SSA renaming machinery.)

Example printed form:

; *ast.CallExpr @ 102:9 is t5
; var x float64 @ 109:72 is x
; address of *ast.CompositeLit @ 216:10 is t0

func (*DebugRef) Block

func (v *DebugRef) Block() *BasicBlock

func (DebugRef) Comment

func (instr DebugRef) Comment() string

func (*DebugRef) Operands

func (s *DebugRef) Operands(rands []*Value) []*Value

func (*DebugRef) Parent

func (v *DebugRef) Parent() *Function

func (*DebugRef) Pos

func (s *DebugRef) Pos() token.Pos

func (*DebugRef) Referrers

func (v *DebugRef) Referrers() *[]Instruction

func (*DebugRef) String

func (s *DebugRef) String() string

type Defer

type Defer struct {
	Call CallCommon
	// contains filtered or unexported fields
}

The Defer instruction pushes the specified call onto a stack of functions to be called by a RunDefers instruction or by a panic.

See CallCommon for generic function call documentation.

Pos() returns the ast.DeferStmt.Defer.

Example printed form:

Defer println t1
Defer t3
DeferInvoke t4.Bar t2

func (*Defer) Block

func (v *Defer) Block() *BasicBlock

func (Defer) Comment

func (instr Defer) Comment() string

func (*Defer) Common

func (s *Defer) Common() *CallCommon

func (*Defer) Operands

func (s *Defer) Operands(rands []*Value) []*Value

func (*Defer) Parent

func (v *Defer) Parent() *Function

func (*Defer) Referrers

func (v *Defer) Referrers() *[]Instruction

func (*Defer) String

func (s *Defer) String() string

func (*Defer) Value

func (s *Defer) Value() *Call

type Extract

type Extract struct {
	Tuple Value
	Index int
	// contains filtered or unexported fields
}

The Extract instruction yields component Index of Tuple.

This is used to access the results of instructions with multiple return values, such as Call, TypeAssert, Next, Recv, MapLookup and others.

Example printed form:

t7 = Extract <bool> [1] (ok) t4

func (*Extract) Name

func (v *Extract) Name() string

func (*Extract) Operands

func (v *Extract) Operands(rands []*Value) []*Value

func (*Extract) Referrers

func (v *Extract) Referrers() *[]Instruction

func (*Extract) String

func (v *Extract) String() string

func (*Extract) Type

func (v *Extract) Type() types.Type

type Field

type Field struct {
	X     Value // struct
	Field int   // index into X.Type().(*types.Struct).Fields
	// contains filtered or unexported fields
}

The Field instruction yields the Field of struct X.

The field is identified by its index within the field list of the struct type of X; by using numeric indices we avoid ambiguity of package-local identifiers and permit compact representations.

Pos() returns the position of the ast.SelectorExpr.Sel for the field, if explicit in the source.

Example printed form:

t2 = FieldAddr <int> [0] (X) t1

func (*Field) Name

func (v *Field) Name() string

func (*Field) Operands

func (v *Field) Operands(rands []*Value) []*Value

func (*Field) Referrers

func (v *Field) Referrers() *[]Instruction

func (*Field) String

func (v *Field) String() string

func (*Field) Type

func (v *Field) Type() types.Type

type FieldAddr

type FieldAddr struct {
	X     Value // *struct
	Field int   // field is X.Type().Underlying().(*types.Pointer).Elem().Underlying().(*types.Struct).Field(Field)
	// contains filtered or unexported fields
}

The FieldAddr instruction yields the address of Field of *struct X.

The field is identified by its index within the field list of the struct type of X.

Dynamically, this instruction panics if X evaluates to a nil pointer.

Type() returns a (possibly named) *types.Pointer.

Pos() returns the position of the ast.SelectorExpr.Sel for the field, if explicit in the source.

Example printed form:

t2 = FieldAddr <*int> [0] (X) t1

func (*FieldAddr) Name

func (v *FieldAddr) Name() string

func (*FieldAddr) Operands

func (v *FieldAddr) Operands(rands []*Value) []*Value

func (*FieldAddr) Referrers

func (v *FieldAddr) Referrers() *[]Instruction

func (*FieldAddr) String

func (v *FieldAddr) String() string

func (*FieldAddr) Type

func (v *FieldAddr) Type() types.Type

type FreeVar

type FreeVar struct {
	// contains filtered or unexported fields
}

A FreeVar represents a free variable of the function to which it belongs.

FreeVars are used to implement anonymous functions, whose free variables are lexically captured in a closure formed by MakeClosure. The value of such a free var is an Alloc or another FreeVar and is considered a potentially escaping heap address, with pointer type.

FreeVars are also used to implement bound method closures. Such a free var represents the receiver value and may be of any type that has concrete methods.

Pos() returns the position of the value that was captured, which belongs to an enclosing function.

func (FreeVar) ID

func (n FreeVar) ID() ID

func (*FreeVar) Name

func (v *FreeVar) Name() string

func (*FreeVar) Operands

func (v *FreeVar) Operands(rands []*Value) []*Value

func (*FreeVar) Parent

func (v *FreeVar) Parent() *Function

func (*FreeVar) Pos

func (n *FreeVar) Pos() token.Pos

func (*FreeVar) Referrers

func (v *FreeVar) Referrers() *[]Instruction

func (*FreeVar) Source

func (n *FreeVar) Source() ast.Node

func (*FreeVar) String

func (v *FreeVar) String() string

func (*FreeVar) Type

func (v *FreeVar) Type() types.Type

type Function

type Function struct {
	Signature *types.Signature

	Synthetic Synthetic

	Pkg       *Package      // enclosing package; nil for shared funcs (wrappers and error.Error)
	Prog      *Program      // enclosing program
	Params    []*Parameter  // function parameters; for methods, includes receiver
	FreeVars  []*FreeVar    // free variables whose values must be supplied by closure
	Locals    []*Alloc      // local variables of this function
	Blocks    []*BasicBlock // basic blocks of the function; nil => external
	Exit      *BasicBlock   // The function's exit block
	AnonFuncs []*Function   // anonymous functions directly beneath this one

	NoReturn NoReturn // Calling this function will always terminate control flow.
	// contains filtered or unexported fields
}

Function represents the parameters, results, and code of a function or method.

If Blocks is nil, this indicates an external function for which no Go source code is available. In this case, FreeVars and Locals are nil too. Clients performing whole-program analysis must handle external functions specially.

Blocks contains the function's control-flow graph (CFG). Blocks[0] is the function entry point; block order is not otherwise semantically significant, though it may affect the readability of the disassembly. To iterate over the blocks in dominance order, use DomPreorder().

A nested function (Parent()!=nil) that refers to one or more lexically enclosing local variables ("free variables") has FreeVars. Such functions cannot be called directly but require a value created by MakeClosure which, via its Bindings, supplies values for these parameters.

If the function is a method (Signature.Recv() != nil) then the first element of Params is the receiver parameter.

A Go package may declare many functions called "init". For each one, Object().Name() returns "init" but Name() returns "init#1", etc, in declaration order.

Pos() returns the declaring ast.FuncLit.Type.Func or the position of the ast.FuncDecl.Name, if the function was explicit in the source. Synthetic wrappers, for which Synthetic != "", may share the same position as the function they wrap. Syntax.Pos() always returns the position of the declaring "func" token.

Type() returns the function's Signature.

func EnclosingFunction

func EnclosingFunction(pkg *Package, path []ast.Node) *Function

EnclosingFunction returns the function that contains the syntax node denoted by path.

Syntax associated with package-level variable specifications is enclosed by the package's init() function.

Returns nil if not found; reasons might include:

  • the node is not enclosed by any function.
  • the node is within an anonymous function (FuncLit) and its IR function has not been created yet (pkg.Build() has not yet been called).

func (*Function) DomPreorder

func (f *Function) DomPreorder() []*BasicBlock

DomPreorder returns a new slice containing the blocks of f in dominator tree preorder.

func (Function) ID

func (n Function) ID() ID

func (*Function) Name

func (v *Function) Name() string

func (*Function) Object

func (v *Function) Object() types.Object

func (*Function) Operands

func (v *Function) Operands(rands []*Value) []*Value

func (*Function) Package

func (v *Function) Package() *Package

func (*Function) Parent

func (v *Function) Parent() *Function

func (*Function) Pos

func (n *Function) Pos() token.Pos

func (*Function) Referrers

func (v *Function) Referrers() *[]Instruction

func (*Function) RelString

func (f *Function) RelString(from *types.Package) string

RelString returns the full name of this function, qualified by package name, receiver type, etc.

The specific formatting rules are not guaranteed and may change.

Examples:

"math.IsNaN"                  // a package-level function
"(*bytes.Buffer).Bytes"       // a declared method or a wrapper
"(*bytes.Buffer).Bytes$thunk" // thunk (func wrapping method; receiver is param 0)
"(*bytes.Buffer).Bytes$bound" // bound (func wrapping method; receiver supplied by closure)
"main.main$1"                 // an anonymous function in main
"main.init#1"                 // a declared init function
"main.init"                   // the synthesized package initializer

When these functions are referred to from within the same package (i.e. from == f.Pkg.Object), they are rendered without the package path. For example: "IsNaN", "(*Buffer).Bytes", etc.

All non-synthetic functions have distinct package-qualified names. (But two methods may have the same name "(T).f" if one is a synthetic wrapper promoting a non-exported method "f" from another package; in that case, the strings are equal but the identifiers "f" are distinct.)

func (*Function) RemoveNilBlocks

func (f *Function) RemoveNilBlocks()

func (*Function) Source

func (n *Function) Source() ast.Node

func (*Function) String

func (v *Function) String() string

func (*Function) Token

func (v *Function) Token() token.Token

func (*Function) Type

func (v *Function) Type() types.Type

func (*Function) ValueForExpr

func (f *Function) ValueForExpr(e ast.Expr) (value Value, isAddr bool)

ValueForExpr returns the IR Value that corresponds to non-constant expression e.

It returns nil if no value was found, e.g.

  • the expression is not lexically contained within f;
  • f was not built with debug information; or
  • e is a constant expression. (For efficiency, no debug information is stored for constants. Use go/types.Info.Types[e].Value instead.)
  • e is a reference to nil or a built-in function.
  • the value was optimised away.

If e is an addressable expression used in an lvalue context, value is the address denoted by e, and isAddr is true.

The types of e (or &e, if isAddr) and the result are equal (modulo "untyped" bools resulting from comparisons).

(Tip: to find the ir.Value given a source position, use astutil.PathEnclosingInterval to locate the ast.Node, then EnclosingFunction to locate the Function, then ValueForExpr to find the ir.Value.)

func (*Function) WriteTo

func (f *Function) WriteTo(w io.Writer) (int64, error)

type GenericConst

type GenericConst struct {
	// contains filtered or unexported fields
}

func (*GenericConst) Name

func (v *GenericConst) Name() string

func (*GenericConst) Operands

func (v *GenericConst) Operands(rands []*Value) []*Value

func (*GenericConst) Referrers

func (v *GenericConst) Referrers() *[]Instruction

func (*GenericConst) RelString

func (v *GenericConst) RelString(pkg *types.Package) string

func (*GenericConst) String

func (v *GenericConst) String() string

func (*GenericConst) Type

func (v *GenericConst) Type() types.Type

type Global

type Global struct {
	Pkg *Package
	// contains filtered or unexported fields
}

A Global is a named Value holding the address of a package-level variable.

Pos() returns the position of the ast.ValueSpec.Names[*] identifier.

func (Global) ID

func (n Global) ID() ID

func (*Global) Name

func (v *Global) Name() string

func (*Global) Object

func (v *Global) Object() types.Object

func (*Global) Operands

func (v *Global) Operands(rands []*Value) []*Value

func (*Global) Package

func (v *Global) Package() *Package

func (*Global) Parent

func (v *Global) Parent() *Function

func (*Global) Pos

func (n *Global) Pos() token.Pos

func (*Global) Referrers

func (v *Global) Referrers() *[]Instruction

func (*Global) RelString

func (v *Global) RelString(from *types.Package) string

func (*Global) Source

func (n *Global) Source() ast.Node

func (*Global) String

func (v *Global) String() string

func (*Global) Token

func (v *Global) Token() token.Token

func (*Global) Type

func (v *Global) Type() types.Type

type Go

type Go struct {
	Call CallCommon
	// contains filtered or unexported fields
}

The Go instruction creates a new goroutine and calls the specified function within it.

See CallCommon for generic function call documentation.

Pos() returns the ast.GoStmt.Go.

Example printed form:

Go println t1
Go t3
GoInvoke t4.Bar t2

func (*Go) Block

func (v *Go) Block() *BasicBlock

func (Go) Comment

func (instr Go) Comment() string

func (*Go) Common

func (s *Go) Common() *CallCommon

func (*Go) Operands

func (s *Go) Operands(rands []*Value) []*Value

func (*Go) Parent

func (v *Go) Parent() *Function

func (*Go) Referrers

func (v *Go) Referrers() *[]Instruction

func (*Go) String

func (s *Go) String() string

func (*Go) Value

func (s *Go) Value() *Call

type HTMLWriter

type HTMLWriter struct {
	// contains filtered or unexported fields
}

func NewHTMLWriter

func NewHTMLWriter(path string, funcname, cfgMask string) *HTMLWriter

func (*HTMLWriter) Close

func (w *HTMLWriter) Close()

func (*HTMLWriter) Printf

func (w *HTMLWriter) Printf(msg string, v ...interface{})

func (*HTMLWriter) WriteColumn

func (w *HTMLWriter) WriteColumn(phase, title, class, html string)

WriteColumn writes raw HTML in a column headed by title. It is intended for pre- and post-compilation log output.

func (*HTMLWriter) WriteFunc

func (w *HTMLWriter) WriteFunc(phase, title string, f *Function)

WriteFunc writes f in a column headed by title. phase is used for collapsing columns and should be unique across the table.

func (*HTMLWriter) WriteString

func (w *HTMLWriter) WriteString(s string)

type ID

type ID int

type If

type If struct {
	Cond Value
	// contains filtered or unexported fields
}

The If instruction transfers control to one of the two successors of its owning block, depending on the boolean Cond: the first if true, the second if false.

An If instruction must be the last instruction of its containing BasicBlock.

Pos() returns the *ast.IfStmt, if explicit in the source.

Example printed form:

If t2 → b1 b2

func (*If) Block

func (v *If) Block() *BasicBlock

func (If) Comment

func (instr If) Comment() string

func (*If) Operands

func (s *If) Operands(rands []*Value) []*Value

func (*If) Parent

func (v *If) Parent() *Function

func (*If) Referrers

func (v *If) Referrers() *[]Instruction

func (*If) String

func (s *If) String() string

type Index

type Index struct {
	X     Value // array
	Index Value // integer index
	// contains filtered or unexported fields
}

The Index instruction yields element Index of array X.

Pos() returns the ast.IndexExpr.Lbrack for the index operation, if explicit in the source.

Example printed form:

t3 = Index <int> t2 t1

func (*Index) Name

func (v *Index) Name() string

func (*Index) Operands

func (v *Index) Operands(rands []*Value) []*Value

func (*Index) Referrers

func (v *Index) Referrers() *[]Instruction

func (*Index) String

func (v *Index) String() string

func (*Index) Type

func (v *Index) Type() types.Type

type IndexAddr

type IndexAddr struct {
	X     Value // slice or *array,
	Index Value // numeric index
	// contains filtered or unexported fields
}

The IndexAddr instruction yields the address of the element at index Index of collection X. Index is an integer expression.

The elements of maps and strings are not addressable; use StringLookup, MapLookup or MapUpdate instead.

Dynamically, this instruction panics if X evaluates to a nil *array pointer.

Type() returns a (possibly named) *types.Pointer.

Pos() returns the ast.IndexExpr.Lbrack for the index operation, if explicit in the source.

Example printed form:

t3 = IndexAddr <*int> t2 t1

func (*IndexAddr) Name

func (v *IndexAddr) Name() string

func (*IndexAddr) Operands

func (v *IndexAddr) Operands(rands []*Value) []*Value

func (*IndexAddr) Referrers

func (v *IndexAddr) Referrers() *[]Instruction

func (*IndexAddr) String

func (v *IndexAddr) String() string

func (*IndexAddr) Type

func (v *IndexAddr) Type() types.Type

type Instruction

type Instruction interface {
	Comment() string

	// String returns the disassembled form of this value.
	//
	// Examples of Instructions that are Values:
	//       "BinOp <int> {+} t1 t2"  (BinOp)
	//       "Call <int> len t1"      (Call)
	// Note that the name of the Value is not printed.
	//
	// Examples of Instructions that are not Values:
	//       "Return t1"              (Return)
	//       "Store {int} t2 t1"      (Store)
	//
	// (The separation of Value.Name() from Value.String() is useful
	// for some analyses which distinguish the operation from the
	// value it defines, e.g., 'y = local int' is both an allocation
	// of memory 'local int' and a definition of a pointer y.)
	String() string

	// ID returns the ID of this instruction. IDs are unique within a single
	// function and are densely numbered, but may contain gaps.
	// Globally, instructions are identified by their addresses. However,
	// IDs exist to facilitate efficient storage of mappings between
	// instructions and data when analysing functions.
	//
	// NB: IDs are allocated late in the IR construction process and
	// are not available to early stages of said process.
	ID() ID

	// Parent returns the function to which this instruction
	// belongs.
	Parent() *Function

	// Block returns the basic block to which this instruction
	// belongs.
	Block() *BasicBlock

	// Operands returns the operands of this instruction: the
	// set of Values it references.
	//
	// Specifically, it appends their addresses to rands, a
	// user-provided slice, and returns the resulting slice,
	// permitting avoidance of memory allocation.
	//
	// The operands are appended in undefined order, but the order
	// is consistent for a given Instruction; the addresses are
	// always non-nil but may point to a nil Value.  Clients may
	// store through the pointers, e.g. to effect a value
	// renaming.
	//
	// Value.Referrers is a subset of the inverse of this
	// relation.  (Referrers are not tracked for all types of
	// Values.)
	Operands(rands []*Value) []*Value

	Referrers() *[]Instruction // nil for non-Values

	// Source returns the AST node responsible for creating this
	// instruction. A single AST node may be responsible for more than
	// one instruction, and not all instructions have an associated
	// AST node.
	Source() ast.Node

	// Pos returns Source().Pos() if Source is not nil, else it
	// returns token.NoPos.
	Pos() token.Pos
	// contains filtered or unexported methods
}

An Instruction is an IR instruction that computes a new Value or has some effect.

An Instruction that defines a value (e.g. BinOp) also implements the Value interface; an Instruction that only has an effect (e.g. Store) does not.

type Jump

type Jump struct {
	// contains filtered or unexported fields
}

The Jump instruction transfers control to the sole successor of its owning block.

A Jump must be the last instruction of its containing BasicBlock.

Pos() returns NoPos.

Example printed form:

Jump → b1

func NewJump

func NewJump(parent *BasicBlock) *Jump

func (*Jump) Block

func (v *Jump) Block() *BasicBlock

func (Jump) Comment

func (instr Jump) Comment() string

func (*Jump) Operands

func (*Jump) Operands(rands []*Value) []*Value

func (*Jump) Parent

func (v *Jump) Parent() *Function

func (*Jump) Referrers

func (v *Jump) Referrers() *[]Instruction

func (*Jump) String

func (s *Jump) String() string

type Load

type Load struct {
	X Value
	// contains filtered or unexported fields
}

The Load instruction loads a value from a memory address.

For implicit memory loads, Pos() returns the position of the most closely associated source-level construct; the details are not specified.

Example printed form:

t2 = Load <int> t1

func (*Load) Name

func (v *Load) Name() string

func (*Load) Operands

func (v *Load) Operands(rands []*Value) []*Value

func (*Load) Referrers

func (v *Load) Referrers() *[]Instruction

func (*Load) String

func (v *Load) String() string

func (*Load) Type

func (v *Load) Type() types.Type

type MakeChan

type MakeChan struct {
	Size Value // int; size of buffer; zero => synchronous.
	// contains filtered or unexported fields
}

The MakeChan instruction creates a new channel object and yields a value of kind chan.

Type() returns a (possibly named) *types.Chan.

Pos() returns the ast.CallExpr.Lparen for the make(chan) that created it.

Example printed form:

t3 = MakeChan <chan int> t1
t4 = MakeChan <chan IntChan> t2

func (*MakeChan) Name

func (v *MakeChan) Name() string

func (*MakeChan) Operands

func (v *MakeChan) Operands(rands []*Value) []*Value

func (*MakeChan) Referrers

func (v *MakeChan) Referrers() *[]Instruction

func (*MakeChan) String

func (v *MakeChan) String() string

func (*MakeChan) Type

func (v *MakeChan) Type() types.Type

type MakeClosure

type MakeClosure struct {
	Fn       Value   // always a *Function
	Bindings []Value // values for each free variable in Fn.FreeVars
	// contains filtered or unexported fields
}

The MakeClosure instruction yields a closure value whose code is Fn and whose free variables' values are supplied by Bindings.

Type() returns a (possibly named) *types.Signature.

Pos() returns the ast.FuncLit.Type.Func for a function literal closure or the ast.SelectorExpr.Sel for a bound method closure.

Example printed form:

t1 = MakeClosure <func()> foo$1 t1 t2
t5 = MakeClosure <func(int)> (T).foo$bound t4

func (*MakeClosure) Name

func (v *MakeClosure) Name() string

func (*MakeClosure) Operands

func (v *MakeClosure) Operands(rands []*Value) []*Value

func (*MakeClosure) Referrers

func (v *MakeClosure) Referrers() *[]Instruction

func (*MakeClosure) String

func (v *MakeClosure) String() string

func (*MakeClosure) Type

func (v *MakeClosure) Type() types.Type

type MakeInterface

type MakeInterface struct {
	X Value
	// contains filtered or unexported fields
}

MakeInterface constructs an instance of an interface type from a value of a concrete type.

Use Program.MethodSets.MethodSet(X.Type()) to find the method-set of X, and Program.MethodValue(m) to find the implementation of a method.

To construct the zero value of an interface type T, use:

NewConst(constant.MakeNil(), T, pos)

Pos() returns the ast.CallExpr.Lparen, if the instruction arose from an explicit conversion in the source.

Example printed form:

t2 = MakeInterface <interface{}> t1

func (*MakeInterface) Name

func (v *MakeInterface) Name() string

func (*MakeInterface) Operands

func (v *MakeInterface) Operands(rands []*Value) []*Value

func (*MakeInterface) Referrers

func (v *MakeInterface) Referrers() *[]Instruction

func (*MakeInterface) String

func (v *MakeInterface) String() string

func (*MakeInterface) Type

func (v *MakeInterface) Type() types.Type

type MakeMap

type MakeMap struct {
	Reserve Value // initial space reservation; nil => default
	// contains filtered or unexported fields
}

The MakeMap instruction creates a new hash-table-based map object and yields a value of kind map.

Type() returns a (possibly named) *types.Map.

Pos() returns the ast.CallExpr.Lparen, if created by make(map), or the ast.CompositeLit.Lbrack if created by a literal.

Example printed form:

t1 = MakeMap <map[string]int>
t2 = MakeMap <StringIntMap> t1

func (*MakeMap) Name

func (v *MakeMap) Name() string

func (*MakeMap) Operands

func (v *MakeMap) Operands(rands []*Value) []*Value

func (*MakeMap) Referrers

func (v *MakeMap) Referrers() *[]Instruction

func (*MakeMap) String

func (v *MakeMap) String() string

func (*MakeMap) Type

func (v *MakeMap) Type() types.Type

type MakeSlice

type MakeSlice struct {
	Len Value
	Cap Value
	// contains filtered or unexported fields
}

The MakeSlice instruction yields a slice of length Len backed by a newly allocated array of length Cap.

Both Len and Cap must be non-nil Values of integer type.

(Alloc(types.Array) followed by Slice will not suffice because Alloc can only create arrays of constant length.)

Type() returns a (possibly named) *types.Slice.

Pos() returns the ast.CallExpr.Lparen for the make([]T) that created it.

Example printed form:

t3 = MakeSlice <[]string> t1 t2
t4 = MakeSlice <StringSlice> t1 t2

func (*MakeSlice) Name

func (v *MakeSlice) Name() string

func (*MakeSlice) Operands

func (v *MakeSlice) Operands(rands []*Value) []*Value

func (*MakeSlice) Referrers

func (v *MakeSlice) Referrers() *[]Instruction

func (*MakeSlice) String

func (v *MakeSlice) String() string

func (*MakeSlice) Type

func (v *MakeSlice) Type() types.Type

type MapLookup

type MapLookup struct {
	X       Value // map
	Index   Value // key-typed index
	CommaOk bool  // return a value,ok pair
	// contains filtered or unexported fields
}

The MapLookup instruction yields element Index of collection X, a map.

If CommaOk, the result is a 2-tuple of the value above and a boolean indicating the result of a map membership test for the key. The components of the tuple are accessed using Extract.

Pos() returns the ast.IndexExpr.Lbrack, if explicit in the source.

Example printed form:

t4 = MapLookup <string> t3 t1
t6 = MapLookup <(string, bool)> t3 t2

func (*MapLookup) Name

func (v *MapLookup) Name() string

func (*MapLookup) Operands

func (v *MapLookup) Operands(rands []*Value) []*Value

func (*MapLookup) Referrers

func (v *MapLookup) Referrers() *[]Instruction

func (*MapLookup) String

func (v *MapLookup) String() string

func (*MapLookup) Type

func (v *MapLookup) Type() types.Type

type MapUpdate

type MapUpdate struct {
	Map   Value
	Key   Value
	Value Value
	// contains filtered or unexported fields
}

The MapUpdate instruction updates the association of Map[Key] to Value.

Pos() returns the ast.KeyValueExpr.Colon or ast.IndexExpr.Lbrack, if explicit in the source.

Example printed form:

MapUpdate t3 t1 t2

func (*MapUpdate) Block

func (v *MapUpdate) Block() *BasicBlock

func (MapUpdate) Comment

func (instr MapUpdate) Comment() string

func (*MapUpdate) Operands

func (v *MapUpdate) Operands(rands []*Value) []*Value

func (*MapUpdate) Parent

func (v *MapUpdate) Parent() *Function

func (*MapUpdate) Referrers

func (v *MapUpdate) Referrers() *[]Instruction

func (*MapUpdate) String

func (s *MapUpdate) String() string

type Member

type Member interface {
	Name() string                    // declared name of the package member
	String() string                  // package-qualified name of the package member
	RelString(*types.Package) string // like String, but relative refs are unqualified
	Object() types.Object            // typechecker's object for this member, if any
	Type() types.Type                // type of the package member
	Token() token.Token              // token.{VAR,FUNC,CONST,TYPE}
	Package() *Package               // the containing package
}

A Member is a member of a Go package, implemented by *NamedConst, *Global, *Function, or *Type; they are created by package-level const, var, func and type declarations respectively.

type NamedConst

type NamedConst struct {
	Value *Const
	// contains filtered or unexported fields
}

A NamedConst is a Member of a Package representing a package-level named constant.

Pos() returns the position of the declaring ast.ValueSpec.Names[*] identifier.

NB: a NamedConst is not a Value; it contains a constant Value, which it augments with the name and position of its 'const' declaration.

func (*NamedConst) Name

func (c *NamedConst) Name() string

func (*NamedConst) Object

func (c *NamedConst) Object() types.Object

func (*NamedConst) Package

func (c *NamedConst) Package() *Package

func (*NamedConst) Pos

func (c *NamedConst) Pos() token.Pos

func (*NamedConst) RelString

func (c *NamedConst) RelString(from *types.Package) string

func (*NamedConst) String

func (c *NamedConst) String() string

func (*NamedConst) Token

func (c *NamedConst) Token() token.Token

func (*NamedConst) Type

func (c *NamedConst) Type() types.Type

type Next

type Next struct {
	Iter     Value
	IsString bool // true => string iterator; false => map iterator.
	// contains filtered or unexported fields
}

The Next instruction reads and advances the (map or string) iterator Iter and returns a 3-tuple value (ok, k, v). If the iterator is not exhausted, ok is true and k and v are the next elements of the domain and range, respectively. Otherwise ok is false and k and v are undefined.

Components of the tuple are accessed using Extract.

The IsString field distinguishes iterators over strings from those over maps, as the Type() alone is insufficient: consider map[int]rune.

Type() returns a *types.Tuple for the triple (ok, k, v). The types of k and/or v may be types.Invalid.

Example printed form:

t5 = Next <(ok bool, k int, v rune)> t2
t5 = Next <(ok bool, k invalid type, v invalid type)> t2

func (*Next) Name

func (v *Next) Name() string

func (*Next) Operands

func (v *Next) Operands(rands []*Value) []*Value

func (*Next) Referrers

func (v *Next) Referrers() *[]Instruction

func (*Next) String

func (v *Next) String() string

func (*Next) Type

func (v *Next) Type() types.Type

type NoReturn

type NoReturn uint8
const (
	Returns NoReturn = iota
	AlwaysExits
	AlwaysUnwinds
	NeverReturns
)

type Node

type Node interface {

	// Common methods:
	ID() ID
	String() string
	Source() ast.Node
	Pos() token.Pos
	Parent() *Function

	// Partial methods:
	Operands(rands []*Value) []*Value // nil for non-Instructions
	Referrers() *[]Instruction        // nil for non-Values
	// contains filtered or unexported methods
}

A Node is a node in the IR value graph. Every concrete type that implements Node is also either a Value, an Instruction, or both.

Node contains the methods common to Value and Instruction, plus the Operands and Referrers methods generalized to return nil for non-Instructions and non-Values, respectively.

Node is provided to simplify IR graph algorithms. Clients should use the more specific and informative Value or Instruction interfaces where appropriate.

type Package

type Package struct {
	Prog      *Program          // the owning program
	Pkg       *types.Package    // the corresponding go/types.Package
	Members   map[string]Member // all package members keyed by name (incl. init and init#%d)
	Functions []*Function       // all functions, excluding anonymous ones
	// contains filtered or unexported fields
}

A Package is a single analyzed Go package containing Members for all package-level functions, variables, constants and types it declares. These may be accessed directly via Members, or via the type-specific accessor methods Func, Type, Var and Const.

Members also contains entries for "init" (the synthetic package initializer) and "init#%d", the nth declared init function, and unspecified other things too.

func (*Package) Build

func (p *Package) Build()

Build builds IR code for all functions and vars in package p.

Precondition: CreatePackage must have been called for all of p's direct imports (and hence its direct imports must have been error-free).

Build is idempotent and thread-safe.

func (*Package) Const

func (p *Package) Const(name string) (c *NamedConst)

Const returns the package-level constant of the specified name, or nil if not found.

func (*Package) Func

func (p *Package) Func(name string) (f *Function)

Func returns the package-level function of the specified name, or nil if not found.

func (*Package) SetDebugMode

func (pkg *Package) SetDebugMode(debug bool)

SetDebugMode sets the debug mode for package pkg. If true, all its functions will include full debug info. This greatly increases the size of the instruction stream, and causes Functions to depend upon the ASTs, potentially keeping them live in memory for longer.

func (*Package) String

func (p *Package) String() string

func (*Package) Type

func (p *Package) Type(name string) (t *Type)

Type returns the package-level type of the specified name, or nil if not found.

func (*Package) Var

func (p *Package) Var(name string) (g *Global)

Var returns the package-level variable of the specified name, or nil if not found.

func (*Package) WriteTo

func (p *Package) WriteTo(w io.Writer) (int64, error)

type Panic

type Panic struct {
	X Value // an interface{}
	// contains filtered or unexported fields
}

The Panic instruction initiates a panic with value X.

A Panic instruction must be the last instruction of its containing BasicBlock, which must have one successor, the exit block.

NB: 'go panic(x)' and 'defer panic(x)' do not use this instruction; they are treated as calls to a built-in function.

Pos() returns the ast.CallExpr.Lparen if this panic was explicit in the source.

Example printed form:

Panic t1

func (*Panic) Block

func (v *Panic) Block() *BasicBlock

func (Panic) Comment

func (instr Panic) Comment() string

func (*Panic) Operands

func (s *Panic) Operands(rands []*Value) []*Value

func (*Panic) Parent

func (v *Panic) Parent() *Function

func (*Panic) Referrers

func (v *Panic) Referrers() *[]Instruction

func (*Panic) String

func (s *Panic) String() string

type Parameter

type Parameter struct {
	// contains filtered or unexported fields
}

A Parameter represents an input parameter of a function.

func (*Parameter) Name

func (v *Parameter) Name() string

func (*Parameter) Object

func (v *Parameter) Object() types.Object

func (*Parameter) Operands

func (v *Parameter) Operands(rands []*Value) []*Value

func (*Parameter) Referrers

func (v *Parameter) Referrers() *[]Instruction

func (*Parameter) String

func (v *Parameter) String() string

func (*Parameter) Type

func (v *Parameter) Type() types.Type

type Phi

type Phi struct {
	Edges []Value // Edges[i] is value for Block().Preds[i]
	// contains filtered or unexported fields
}

The Phi instruction represents an SSA φ-node, which combines values that differ across incoming control-flow edges and yields a new value. Within a block, all φ-nodes must appear before all non-φ, non-σ nodes.

Pos() returns the position of the && or || for short-circuit control-flow joins, or that of the *Alloc for φ-nodes inserted during SSA renaming.

Example printed form:

t3 = Phi <int> 2:t1 4:t2 (x)

func (*Phi) Name

func (v *Phi) Name() string

func (*Phi) Operands

func (v *Phi) Operands(rands []*Value) []*Value

func (*Phi) Referrers

func (v *Phi) Referrers() *[]Instruction

func (*Phi) String

func (v *Phi) String() string

func (*Phi) Type

func (v *Phi) Type() types.Type

type Program

type Program struct {
	Fset      *token.FileSet // position information for the files of this Program
	PrintFunc string         // create ir.html for function specified in PrintFunc

	MethodSets typeutil.MethodSetCache // cache of type-checker's method-sets
	// contains filtered or unexported fields
}

A Program is a partial or complete Go program converted to IR form.

func NewProgram

func NewProgram(fset *token.FileSet, mode BuilderMode) *Program

NewProgram returns a new IR Program.

mode controls diagnostics and checking during IR construction.

func (*Program) AllPackages

func (prog *Program) AllPackages() []*Package

AllPackages returns a new slice containing all packages in the program prog in unspecified order.

func (*Program) Build

func (prog *Program) Build()

Build calls Package.Build for each package in prog.

Build is intended for whole-program analysis; a typical compiler need only build a single package.

Build is idempotent and thread-safe.

func (*Program) ConstValue

func (prog *Program) ConstValue(obj *types.Const) *Const

ConstValue returns the IR Value denoted by the source-level named constant obj.

func (*Program) CreatePackage

func (prog *Program) CreatePackage(pkg *types.Package, files []*ast.File, info *types.Info, importable bool) *Package

CreatePackage constructs and returns an IR Package from the specified type-checked, error-free file ASTs, and populates its Members mapping.

importable determines whether this package should be returned by a subsequent call to ImportedPackage(pkg.Path()).

The real work of building IR form for each function is not done until a subsequent call to Package.Build().

func (*Program) FuncValue

func (prog *Program) FuncValue(obj *types.Func) *Function

FuncValue returns the concrete Function denoted by the source-level named function obj, or nil if obj denotes an interface method.

TODO(adonovan): check the invariant that obj.Type() matches the result's Signature, both in the params/results and in the receiver.

func (*Program) ImportedPackage

func (prog *Program) ImportedPackage(path string) *Package

ImportedPackage returns the importable Package whose PkgPath is path, or nil if no such Package has been created.

A parameter to CreatePackage determines whether a package should be considered importable. For example, no import declaration can resolve to the ad-hoc main package created by 'go build foo.go'.

TODO(adonovan): rethink this function and the "importable" concept; most packages are importable. This function assumes that all types.Package.Path values are unique within the ir.Program, which is false---yet this function remains very convenient. Clients should use (*Program).Package instead where possible. IR doesn't really need a string-keyed map of packages.

func (*Program) LookupMethod

func (prog *Program) LookupMethod(T types.Type, pkg *types.Package, name string) *Function

LookupMethod returns the implementation of the method of type T identified by (pkg, name). It returns nil if the method exists but is abstract, and panics if T has no such method.

func (*Program) MethodValue

func (prog *Program) MethodValue(sel *types.Selection) *Function

MethodValue returns the Function implementing method sel, building wrapper methods on demand. It returns nil if sel denotes an abstract (interface) method.

Precondition: sel.Kind() == MethodVal.

Thread-safe.

EXCLUSIVE_LOCKS_ACQUIRED(prog.methodsMu)

func (*Program) NewFunction

func (prog *Program) NewFunction(name string, sig *types.Signature, provenance Synthetic) *Function

NewFunction returns a new synthetic Function instance belonging to prog, with its name and signature fields set as specified.

The caller is responsible for initializing the remaining fields of the function object, e.g. Pkg, Params, Blocks.

It is practically impossible for clients to construct well-formed IR functions/packages/programs directly, so we assume this is the job of the Builder alone. NewFunction exists to provide clients a little flexibility. For example, analysis tools may wish to construct fake Functions for the root of the callgraph, a fake "reflect" package, etc.

TODO(adonovan): think harder about the API here.

func (*Program) Package

func (prog *Program) Package(obj *types.Package) *Package

Package returns the IR Package corresponding to the specified type-checker package object. It returns nil if no such IR package has been created.

func (*Program) RuntimeTypes

func (prog *Program) RuntimeTypes() []types.Type

RuntimeTypes returns a new unordered slice containing all concrete types in the program for which a complete (non-empty) method set is required at run-time.

Thread-safe.

EXCLUSIVE_LOCKS_ACQUIRED(prog.methodsMu)

func (*Program) VarValue

func (prog *Program) VarValue(obj *types.Var, pkg *Package, ref []ast.Node) (value Value, isAddr bool)

VarValue returns the IR Value that corresponds to a specific identifier denoting the source-level named variable obj.

VarValue returns nil if a local variable was not found, perhaps because its package was not built, the debug information was not requested during IR construction, or the value was optimized away.

ref is the path to an ast.Ident (e.g. from PathEnclosingInterval), and that ident must resolve to obj.

pkg is the package enclosing the reference. (A reference to a var always occurs within a function, so we need to know where to find it.)

If the identifier is a field selector and its base expression is non-addressable, then VarValue returns the value of that field. For example:

func f() struct {x int}
f().x  // VarValue(x) returns a *Field instruction of type int

All other identifiers denote addressable locations (variables). For them, VarValue may return either the variable's address or its value, even when the expression is evaluated only for its value; the situation is reported by isAddr, the second component of the result.

If !isAddr, the returned value is the one associated with the specific identifier. For example,

var x int    // VarValue(x) returns Const 0 here
x = 1        // VarValue(x) returns Const 1 here

It is not specified whether the value or the address is returned in any particular case, as it may depend upon optimizations performed during IR code generation, such as registerization, constant folding, avoidance of materialization of subexpressions, etc.

type Range

type Range struct {
	X Value // string or map
	// contains filtered or unexported fields
}

The Range instruction yields an iterator over the domain and range of X, which must be a string or map.

Elements are accessed via Next.

Type() returns an opaque and degenerate "rangeIter" type.

Pos() returns the ast.RangeStmt.For.

Example printed form:

t2 = Range <iter> t1

func (*Range) Name

func (v *Range) Name() string

func (*Range) Operands

func (v *Range) Operands(rands []*Value) []*Value

func (*Range) Referrers

func (v *Range) Referrers() *[]Instruction

func (*Range) String

func (v *Range) String() string

func (*Range) Type

func (v *Range) Type() types.Type

type Recv

type Recv struct {
	Chan    Value
	CommaOk bool
	// contains filtered or unexported fields
}

The Recv instruction receives from channel Chan.

If CommaOk, the result is a 2-tuple of the value above and a boolean indicating the success of the receive. The components of the tuple are accessed using Extract.

Pos() returns the ast.UnaryExpr.OpPos, if explicit in the source. For receive operations implicit in ranging over a channel, Pos() returns the ast.RangeStmt.For.

Example printed form:

t2 = Recv <int> t1
t3 = Recv <(int, bool)> t1

func (*Recv) Name

func (v *Recv) Name() string

func (*Recv) Operands

func (recv *Recv) Operands(rands []*Value) []*Value

func (*Recv) Referrers

func (v *Recv) Referrers() *[]Instruction

func (*Recv) String

func (recv *Recv) String() string

func (*Recv) Type

func (v *Recv) Type() types.Type

type Return

type Return struct {
	Results []Value
	// contains filtered or unexported fields
}

The Return instruction returns values and control back to the calling function.

len(Results) is always equal to the number of results in the function's signature.

If len(Results) > 1, Return returns a tuple value with the specified components which the caller must access using Extract instructions.

There is no instruction to return a ready-made tuple like those returned by a "value,ok"-mode TypeAssert, MapLookup or Recv or a tail-call to a function with multiple result parameters.

Return must be the last instruction of its containing BasicBlock. Such a block has no successors.

Pos() returns the ast.ReturnStmt.Return, if explicit in the source.

Example printed form:

Return
Return t1 t2

func (*Return) Block

func (v *Return) Block() *BasicBlock

func (Return) Comment

func (instr Return) Comment() string

func (*Return) Operands

func (s *Return) Operands(rands []*Value) []*Value

func (*Return) Parent

func (v *Return) Parent() *Function

func (*Return) Referrers

func (v *Return) Referrers() *[]Instruction

func (*Return) String

func (s *Return) String() string

type RunDefers

type RunDefers struct {
	// contains filtered or unexported fields
}

The RunDefers instruction pops and invokes the entire stack of procedure calls pushed by Defer instructions in this function.

It is legal to encounter multiple 'rundefers' instructions in a single control-flow path through a function; this is useful in the combined init() function, for example.

Pos() returns NoPos.

Example printed form:

RunDefers

func (*RunDefers) Block

func (v *RunDefers) Block() *BasicBlock

func (RunDefers) Comment

func (instr RunDefers) Comment() string

func (*RunDefers) Operands

func (*RunDefers) Operands(rands []*Value) []*Value

func (*RunDefers) Parent

func (v *RunDefers) Parent() *Function

func (*RunDefers) Referrers

func (v *RunDefers) Referrers() *[]Instruction

func (*RunDefers) String

func (*RunDefers) String() string

type Select

type Select struct {
	States   []*SelectState
	Blocking bool
	// contains filtered or unexported fields
}

The Select instruction tests whether (or blocks until) one of the specified sent or received states is entered.

Let n be the number of States for which Dir==RECV and Tᵢ (0 ≤ i < n) be the element type of each such state's Chan. Select returns an n+2-tuple

(index int, recvOk bool, r₀ T₀, ... rₙ-1 Tₙ-1)

The tuple's components, described below, must be accessed via the Extract instruction.

If Blocking, select waits until exactly one state holds, i.e. a channel becomes ready for the designated operation of sending or receiving; select chooses one among the ready states pseudorandomly, performs the send or receive operation, and sets 'index' to the index of the chosen channel.

If !Blocking, select doesn't block if no states hold; instead it returns immediately with index equal to -1.

If the chosen channel was used for a receive, the rᵢ component is set to the received value, where i is the index of that state among all n receive states; otherwise rᵢ has the zero value of type Tᵢ. Note that the receive index i is not the same as the state index index.

The second component of the triple, recvOk, is a boolean whose value is true iff the selected operation was a receive and the receive successfully yielded a value.

Pos() returns the ast.SelectStmt.Select.

Example printed form:

t6 = SelectNonBlocking <(index int, ok bool, int)> [<-t4, t5<-t1]
t11 = SelectBlocking <(index int, ok bool)> []

func (*Select) Name

func (v *Select) Name() string

func (*Select) Operands

func (v *Select) Operands(rands []*Value) []*Value

func (*Select) Referrers

func (v *Select) Referrers() *[]Instruction

func (*Select) String

func (s *Select) String() string

func (*Select) Type

func (v *Select) Type() types.Type

type SelectState

type SelectState struct {
	Dir       types.ChanDir // direction of case (SendOnly or RecvOnly)
	Chan      Value         // channel to use (for send or receive)
	Send      Value         // value to send (for send)
	Pos       token.Pos     // position of token.ARROW
	DebugNode ast.Node      // ast.SendStmt or ast.UnaryExpr(<-) [debug mode]
}

SelectState is a helper for Select. It represents one goal state and its corresponding communication.

type Send

type Send struct {
	Chan, X Value
	// contains filtered or unexported fields
}

The Send instruction sends X on channel Chan.

Pos() returns the ast.SendStmt.Arrow, if explicit in the source.

Example printed form:

Send t2 t1

func (*Send) Block

func (v *Send) Block() *BasicBlock

func (Send) Comment

func (instr Send) Comment() string

func (*Send) Operands

func (s *Send) Operands(rands []*Value) []*Value

func (*Send) Parent

func (v *Send) Parent() *Function

func (*Send) Referrers

func (v *Send) Referrers() *[]Instruction

func (*Send) String

func (s *Send) String() string

type Sigma

type Sigma struct {
	From *BasicBlock
	X    Value
	// contains filtered or unexported fields
}

The Sigma instruction represents an SSI σ-node, which splits values at branches in the control flow.

Conceptually, σ-nodes exist at the end of blocks that branch and constitute parallel assignments to one value per destination block. However, such a representation would be awkward to work with, so instead we place σ-nodes at the beginning of branch targets. The From field denotes to which incoming edge the node applies.

Within a block, all σ-nodes must appear before all non-σ nodes.

Example printed form:

t2 = Sigma <int> [#0] t1 (x)

func (*Sigma) Name

func (v *Sigma) Name() string

func (*Sigma) Operands

func (v *Sigma) Operands(rands []*Value) []*Value

func (*Sigma) Referrers

func (v *Sigma) Referrers() *[]Instruction

func (*Sigma) String

func (v *Sigma) String() string

func (*Sigma) Type

func (v *Sigma) Type() types.Type

type Slice

type Slice struct {
	X              Value // slice, string, or *array
	Low, High, Max Value // each may be nil
	// contains filtered or unexported fields
}

The Slice instruction yields a slice of an existing string, slice or *array X between optional integer bounds Low and High.

Dynamically, this instruction panics if X evaluates to a nil *array pointer.

Type() returns string if the type of X was string, otherwise a *types.Slice with the same element type as X.

Pos() returns the ast.SliceExpr.Lbrack if created by a x[:] slice operation, the ast.CompositeLit.Lbrace if created by a literal, or NoPos if not explicit in the source (e.g. a variadic argument slice).

Example printed form:

t4 = Slice <[]int> t3 t2 t1 <nil>

func (*Slice) Name

func (v *Slice) Name() string

func (*Slice) Operands

func (v *Slice) Operands(rands []*Value) []*Value

func (*Slice) Referrers

func (v *Slice) Referrers() *[]Instruction

func (*Slice) String

func (v *Slice) String() string

func (*Slice) Type

func (v *Slice) Type() types.Type

type SliceToArrayPointer

type SliceToArrayPointer struct {
	X Value
	// contains filtered or unexported fields
}

The SliceToArrayPointer instruction yields the conversion of slice X to array pointer.

Pos() returns the ast.CallExpr.Lparen, if the instruction arose from an explicit conversion in the source.

Example printed form:

t1 = SliceToArrayPointer <*[4]byte> t1

func (*SliceToArrayPointer) Name

func (v *SliceToArrayPointer) Name() string

func (*SliceToArrayPointer) Operands

func (v *SliceToArrayPointer) Operands(rands []*Value) []*Value

func (*SliceToArrayPointer) Referrers

func (v *SliceToArrayPointer) Referrers() *[]Instruction

func (*SliceToArrayPointer) String

func (v *SliceToArrayPointer) String() string

func (*SliceToArrayPointer) Type

func (v *SliceToArrayPointer) Type() types.Type

type StackMap

type StackMap struct {
	// contains filtered or unexported fields
}

func (*StackMap) Get

func (m *StackMap) Get(key Value) (Value, bool)

func (*StackMap) Pop

func (m *StackMap) Pop()

func (*StackMap) Push

func (m *StackMap) Push()

func (*StackMap) Set

func (m *StackMap) Set(k Value, v Value)

type Store

type Store struct {
	Addr Value
	Val  Value
	// contains filtered or unexported fields
}

The Store instruction stores Val at address Addr. Stores can be of arbitrary types.

Pos() returns the position of the source-level construct most closely associated with the memory store operation. Since implicit memory stores are numerous and varied and depend upon implementation choices, the details are not specified.

Example printed form:

Store {int} t2 t1

func (*Store) Block

func (v *Store) Block() *BasicBlock

func (Store) Comment

func (instr Store) Comment() string

func (*Store) Operands

func (s *Store) Operands(rands []*Value) []*Value

func (*Store) Parent

func (v *Store) Parent() *Function

func (*Store) Referrers

func (v *Store) Referrers() *[]Instruction

func (*Store) String

func (s *Store) String() string

type StringLookup

type StringLookup struct {
	X     Value // string
	Index Value // numeric index
	// contains filtered or unexported fields
}

The StringLookup instruction yields element Index of collection X, a string. Index is an integer expression.

Pos() returns the ast.IndexExpr.Lbrack, if explicit in the source.

Example printed form:

t3 = StringLookup <uint8> t2 t1

func (*StringLookup) Name

func (v *StringLookup) Name() string

func (*StringLookup) Operands

func (v *StringLookup) Operands(rands []*Value) []*Value

func (*StringLookup) Referrers

func (v *StringLookup) Referrers() *[]Instruction

func (*StringLookup) String

func (v *StringLookup) String() string

func (*StringLookup) Type

func (v *StringLookup) Type() types.Type

type Synthetic

type Synthetic int
const (
	SyntheticLoadedFromExportData Synthetic = iota + 1
	SyntheticPackageInitializer
	SyntheticThunk
	SyntheticWrapper
	SyntheticBound
	SyntheticGeneric
)

func (Synthetic) String

func (syn Synthetic) String() string

type Type

type Type struct {
	// contains filtered or unexported fields
}

A Type is a Member of a Package representing a package-level named type.

func (*Type) Name

func (t *Type) Name() string

func (*Type) Object

func (t *Type) Object() types.Object

func (*Type) Package

func (t *Type) Package() *Package

func (*Type) Pos

func (t *Type) Pos() token.Pos

func (*Type) RelString

func (t *Type) RelString(from *types.Package) string

func (*Type) String

func (t *Type) String() string

func (*Type) Token

func (t *Type) Token() token.Token

func (*Type) Type

func (t *Type) Type() types.Type

type TypeAssert

type TypeAssert struct {
	X            Value
	AssertedType types.Type
	CommaOk      bool
	// contains filtered or unexported fields
}

The TypeAssert instruction tests whether interface value X has type AssertedType.

If !CommaOk, on success it returns v, the result of the conversion (defined below); on failure it panics.

If CommaOk: on success it returns a pair (v, true) where v is the result of the conversion; on failure it returns (z, false) where z is AssertedType's zero value. The components of the pair must be accessed using the Extract instruction.

If AssertedType is a concrete type, TypeAssert checks whether the dynamic type in interface X is equal to it, and if so, the result of the conversion is a copy of the value in the interface.

If AssertedType is an interface, TypeAssert checks whether the dynamic type of the interface is assignable to it, and if so, the result of the conversion is a copy of the interface value X. If AssertedType is a superinterface of X.Type(), the operation will fail iff the operand is nil. (Contrast with ChangeInterface, which performs no nil-check.)

Type() reflects the actual type of the result, possibly a 2-types.Tuple; AssertedType is the asserted type.

Pos() returns the ast.CallExpr.Lparen if the instruction arose from an explicit T(e) conversion; the ast.TypeAssertExpr.Lparen if the instruction arose from an explicit e.(T) operation; or the ast.CaseClause.Case if the instruction arose from a case of a type-switch statement.

Example printed form:

t2 = TypeAssert <int> t1
t4 = TypeAssert <(value fmt.Stringer, ok bool)> t1

func (*TypeAssert) Name

func (v *TypeAssert) Name() string

func (*TypeAssert) Operands

func (v *TypeAssert) Operands(rands []*Value) []*Value

func (*TypeAssert) Referrers

func (v *TypeAssert) Referrers() *[]Instruction

func (*TypeAssert) String

func (v *TypeAssert) String() string

func (*TypeAssert) Type

func (v *TypeAssert) Type() types.Type

type TypeSwitch

type TypeSwitch struct {
	Tag   Value
	Conds []types.Type
	// contains filtered or unexported fields
}

func (*TypeSwitch) Name

func (v *TypeSwitch) Name() string

func (*TypeSwitch) Operands

func (s *TypeSwitch) Operands(rands []*Value) []*Value

func (*TypeSwitch) Referrers

func (v *TypeSwitch) Referrers() *[]Instruction

func (*TypeSwitch) String

func (s *TypeSwitch) String() string

func (*TypeSwitch) Type

func (v *TypeSwitch) Type() types.Type

type UnOp

type UnOp struct {
	Op token.Token // One of: NOT SUB XOR ! - ^
	X  Value
	// contains filtered or unexported fields
}

The UnOp instruction yields the result of Op X. XOR is bitwise complement. SUB is negation. NOT is logical negation.

Example printed form:

t2 = UnOp <int> {^} t1

func (*UnOp) Name

func (v *UnOp) Name() string

func (*UnOp) Operands

func (v *UnOp) Operands(rands []*Value) []*Value

func (*UnOp) Referrers

func (v *UnOp) Referrers() *[]Instruction

func (*UnOp) String

func (v *UnOp) String() string

func (*UnOp) Type

func (v *UnOp) Type() types.Type

type Unreachable

type Unreachable struct {
	// contains filtered or unexported fields
}

The Unreachable pseudo-instruction signals that execution cannot continue after the preceding function call because it terminates the process.

The instruction acts as a control instruction, jumping to the exit block. However, this jump will never execute.

An Unreachable instruction must be the last instruction of its containing BasicBlock.

Example printed form:

Unreachable → b1

func (*Unreachable) Block

func (v *Unreachable) Block() *BasicBlock

func (Unreachable) Comment

func (instr Unreachable) Comment() string

func (*Unreachable) Operands

func (*Unreachable) Operands(rands []*Value) []*Value

func (*Unreachable) Parent

func (v *Unreachable) Parent() *Function

func (*Unreachable) Referrers

func (v *Unreachable) Referrers() *[]Instruction

func (*Unreachable) String

func (s *Unreachable) String() string

type Value

type Value interface {

	// Name returns the name of this value, and determines how
	// this Value appears when used as an operand of an
	// Instruction.
	//
	// This is the same as the source name for Parameters,
	// Builtins, Functions, FreeVars, Globals.
	// For constants, it is a representation of the constant's value
	// and type.  For all other Values this is the name of the
	// virtual register defined by the instruction.
	//
	// The name of an IR Value is not semantically significant,
	// and may not even be unique within a function.
	Name() string

	// ID returns the ID of this value. IDs are unique within a single
	// function and are densely numbered, but may contain gaps.
	// Values and other Instructions share the same ID space.
	// Globally, values are identified by their addresses. However,
	// IDs exist to facilitate efficient storage of mappings between
	// values and data when analysing functions.
	//
	// NB: IDs are allocated late in the IR construction process and
	// are not available to early stages of said process.
	ID() ID

	// If this value is an Instruction, String returns its
	// disassembled form; otherwise it returns unspecified
	// human-readable information about the Value, such as its
	// kind, name and type.
	String() string

	// Type returns the type of this value.  Many instructions
	// (e.g. IndexAddr) change their behaviour depending on the
	// types of their operands.
	Type() types.Type

	// Parent returns the function to which this Value belongs.
	// It returns nil for named Functions, Builtin and Global.
	Parent() *Function

	// Referrers returns the list of instructions that have this
	// value as one of their operands; it may contain duplicates
	// if an instruction has a repeated operand.
	//
	// Referrers actually returns a pointer through which the
	// caller may perform mutations to the object's state.
	//
	// Referrers is currently only defined if Parent()!=nil,
	// i.e. for the function-local values FreeVar, Parameter,
	// Functions (iff anonymous) and all value-defining instructions.
	// It returns nil for named Functions, Builtin and Global.
	//
	// Instruction.Operands contains the inverse of this relation.
	Referrers() *[]Instruction

	Operands(rands []*Value) []*Value // nil for non-Instructions

	// Source returns the AST node responsible for creating this
	// value. A single AST node may be responsible for more than one
	// value, and not all values have an associated AST node.
	//
	// Do not use this method to find a Value given an ast.Expr; use
	// ValueForExpr instead.
	Source() ast.Node

	// Pos returns Source().Pos() if Source is not nil, else it
	// returns token.NoPos.
	Pos() token.Pos
	// contains filtered or unexported methods
}

A Value is an IR value that can be referenced by an instruction.

func Unwrap

func Unwrap(v Value) Value

Unwrap recursively unwraps Sigma and Copy nodes.

Directories

Path Synopsis

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL