1
0
mirror of https://github.com/golang/go synced 2024-11-19 02:04:42 -07:00
go/ssa/emit.go

455 lines
12 KiB
Go
Raw Normal View History

// Copyright 2013 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package ssa
// Helpers for emitting SSA instructions.
import (
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
"go/ast"
"go/token"
"code.google.com/p/go.tools/go/types"
)
// emitNew emits to f a new (heap Alloc) instruction allocating an
// object of type typ. pos is the optional source location.
//
func emitNew(f *Function, typ types.Type, pos token.Pos) *Alloc {
v := &Alloc{Heap: true}
v.setType(types.NewPointer(typ))
v.setPos(pos)
f.emit(v)
return v
}
// emitLoad emits to f an instruction to load the address addr into a
// new temporary, and returns the value so defined.
//
func emitLoad(f *Function, addr Value) *UnOp {
v := &UnOp{Op: token.MUL, X: addr}
v.setType(deref(addr.Type()))
f.emit(v)
return v
}
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
// emitDebugRef emits to f a DebugRef pseudo-instruction associating
// expression e with value v.
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
//
func emitDebugRef(f *Function, e ast.Expr, v Value, isAddr bool) {
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
if !f.debugInfo() {
return // debugging not enabled
}
if v == nil || e == nil {
panic("nil")
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
}
var obj types.Object
e = unparen(e)
if id, ok := e.(*ast.Ident); ok {
if isBlankIdent(id) {
return
}
obj = f.Pkg.objectOf(id)
switch obj.(type) {
case *types.Nil, *types.Const, *types.Builtin:
return
}
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
}
f.emit(&DebugRef{
X: v,
Expr: e,
IsAddr: isAddr,
go.tools/ssa: add debug information for all ast.Idents. This CL adds three new functions to determine the SSA Value for a given syntactic var, func or const object: Program.{Const,Func,Var}Value. Since constants and functions are immutable, the first two only need a types.Object; but each distinct reference to a var may return a distinct Value, so the third requires an ast.Ident parameter too. Debug information for local vars is encoded in the instruction stream in the form of DebugRef instructions, which are a no-op but relate their operand to a particular ident in the AST. The beauty of this approach is that it naturally stays consistent during optimisation passes (e.g. lifting) without additional bookkeeping. DebugRef instructions are only generated if the DebugMode builder flag is set; I plan to make the policy more fine- grained (per function). DebugRef instructions are inserted for: - expr(Ident) for rvalue idents - address.store() for idents that update an lvalue - address.address() for idents that take address of lvalue (this new method replaces all uses of lval.(address).addr) - expr() for all constant expressions - local ValueSpecs with implicit zero initialization (no RHS) (this case doesn't call store() or address()) To ensure we don't forget to emit debug info for uses of Idents, we must use the lvalue mechanism consistently. (Previously, many simple cases had effectively inlined these functions.) Similarly setCallFunc no longer inlines expr(Ident). Also: - Program.Value() has been inlined & specialized. - Program.Package() has moved nearer the new lookup functions. - refactoring: funcSyntax has lost paramFields, resultFields; gained funcType, which provides access to both. - add package-level constants to Package.values map. - opt: don't call localValueSpec for constants. (The resulting code is always optimised away.) There are a number of comments asking whether Literals should have positions. Will address in a follow-up. Added tests of all interesting cases. R=gri CC=golang-dev https://golang.org/cl/11259044
2013-07-15 11:56:46 -06:00
object: obj,
})
}
// emitArith emits to f code to compute the binary operation op(x, y)
// where op is an eager shift, logical or arithmetic operation.
// (Use emitCompare() for comparisons and Builder.logicalBinop() for
// non-eager operations.)
//
func emitArith(f *Function, op token.Token, x, y Value, t types.Type, pos token.Pos) Value {
switch op {
case token.SHL, token.SHR:
x = emitConv(f, x, t)
// y may be signed or an 'untyped' constant.
// TODO(adonovan): whence signed values?
if b, ok := y.Type().Underlying().(*types.Basic); ok && b.Info()&types.IsUnsigned == 0 {
y = emitConv(f, y, types.Typ[types.Uint64])
}
case token.ADD, token.SUB, token.MUL, token.QUO, token.REM, token.AND, token.OR, token.XOR, token.AND_NOT:
x = emitConv(f, x, t)
y = emitConv(f, y, t)
default:
panic("illegal op in emitArith: " + op.String())
}
v := &BinOp{
Op: op,
X: x,
Y: y,
}
v.setPos(pos)
v.setType(t)
return f.emit(v)
}
// emitCompare emits to f code compute the boolean result of
// comparison comparison 'x op y'.
//
func emitCompare(f *Function, op token.Token, x, y Value, pos token.Pos) Value {
xt := x.Type().Underlying()
yt := y.Type().Underlying()
// Special case to optimise a tagless SwitchStmt so that
// these are equivalent
// switch { case e: ...}
// switch true { case e: ... }
// if e==true { ... }
// even in the case when e's type is an interface.
// TODO(adonovan): opt: generalise to x==true, false!=y, etc.
if x == vTrue && op == token.EQL {
if yt, ok := yt.(*types.Basic); ok && yt.Info()&types.IsBoolean != 0 {
return y
}
}
if types.IsIdentical(xt, yt) {
// no conversion necessary
} else if _, ok := xt.(*types.Interface); ok {
y = emitConv(f, y, x.Type())
} else if _, ok := yt.(*types.Interface); ok {
x = emitConv(f, x, y.Type())
} else if _, ok := x.(*Const); ok {
x = emitConv(f, x, y.Type())
} else if _, ok := y.(*Const); ok {
y = emitConv(f, y, x.Type())
} else {
// other cases, e.g. channels. No-op.
}
v := &BinOp{
Op: op,
X: x,
Y: y,
}
v.setPos(pos)
v.setType(tBool)
return f.emit(v)
}
// isValuePreserving returns true if a conversion from ut_src to
// ut_dst is value-preserving, i.e. just a change of type.
// Precondition: neither argument is a named type.
//
func isValuePreserving(ut_src, ut_dst types.Type) bool {
// Identical underlying types?
if types.IsIdentical(ut_dst, ut_src) {
return true
}
switch ut_dst.(type) {
case *types.Chan:
// Conversion between channel types?
_, ok := ut_src.(*types.Chan)
return ok
case *types.Pointer:
// Conversion between pointers with identical base types?
_, ok := ut_src.(*types.Pointer)
return ok
case *types.Signature:
// Conversion from (T) func f() method to f(T) function?
_, ok := ut_src.(*types.Signature)
return ok
}
return false
}
// emitConv emits to f code to convert Value val to exactly type typ,
// and returns the converted value. Implicit conversions are required
// by language assignability rules in assignments, parameter passing,
// etc. Conversions cannot fail dynamically.
//
func emitConv(f *Function, val Value, typ types.Type) Value {
t_src := val.Type()
// Identical types? Conversion is a no-op.
if types.IsIdentical(t_src, typ) {
return val
}
ut_dst := typ.Underlying()
ut_src := t_src.Underlying()
// Just a change of type, but not value or representation?
if isValuePreserving(ut_src, ut_dst) {
c := &ChangeType{X: val}
c.setType(typ)
return f.emit(c)
}
// Conversion to, or construction of a value of, an interface type?
if _, ok := ut_dst.(*types.Interface); ok {
// Assignment from one interface type to another?
if _, ok := ut_src.(*types.Interface); ok {
c := &ChangeInterface{X: val}
c.setType(typ)
return f.emit(c)
}
// Untyped nil constant? Return interface-typed nil constant.
if ut_src == tUntypedNil {
return nilConst(typ)
}
// Convert (non-nil) "untyped" literals to their default type.
if t, ok := ut_src.(*types.Basic); ok && t.Info()&types.IsUntyped != 0 {
val = emitConv(f, val, DefaultType(ut_src))
}
go.tools/ssa: fix computation of set of types requiring method sets. Motivation: Previously, we assumed that the set of types for which a complete method set (containing all synthesized wrapper functions) is required at runtime was the set of types used as operands to some *ssa.MakeInterface instruction. In fact, this is an underapproximation because types can be derived from other ones via reflection, and some of these may need methods. The reflect.Type API allows *T to be derived from T, and these may have different method sets. Reflection also allows almost any subcomponent of a type to be accessed (with one exception: given T, defined 'type T struct{S}', you can reach S but not struct{S}). As a result, the pointer analysis was unable to generate all necessary constraints before running the solver, causing a crash when reflection derives types whose methods are unavailable. (A similar problem would afflict an ahead-of-time compiler based on ssa. The ssa/interp interpreter was immune only because it does not require all wrapper methods to be created before execution begins.) Description: This change causes the SSA builder to record, for each package, the set of all types with non-empty method sets that are referenced within that package. This set is accessed via Packages.TypesWithMethodSets(). Program.TypesWithMethodSets() returns its union across all packages. The set of references that matter are: - types of operands to some MakeInterface instruction (as before) - types of all exported package members - all subcomponents of the above, recursively. This is a conservative approximation to the set of types whose methods may be called dynamically. We define the owning package of a type as follows: - the owner of a named type is the package in which it is defined; - the owner of a pointer-to-named type is the owner of that named type; - the owner of all other types is nil. A package must include the method sets for all types that it owns, and all subcomponents of that type that are not owned by another package, recursively. Types with an owner appear in exactly one package; types with no owner (such as struct{T}) may appear within multiple packages. (A typical Go compiler would emit multiple copies of these methods as weak symbols; a typical linker would eliminate duplicates.) Also: - go/types/typemap: implement hash function for *Tuple. - pointer: generate nodes/constraints for all of ssa.Program.TypesWithMethodSets(). Add rtti.go regression test. - Add API test of Package.TypesWithMethodSets(). - Set Function.Pkg to nil (again) for wrapper functions, since these may be shared by many packages. - Remove a redundant logging statement. - Document that ssa CREATE phase is in fact sequential. Fixes golang/go#6605 R=gri CC=golang-dev https://golang.org/cl/14920056
2013-10-23 15:07:52 -06:00
f.Pkg.needMethodsOf(val.Type())
mi := &MakeInterface{X: val}
mi.setType(typ)
return f.emit(mi)
}
// Conversion of a constant to a non-interface type results in
// a new constant of the destination type and (initially) the
// same abstract value. We don't compute the representation
// change yet; this defers the point at which the number of
// possible representations explodes.
if c, ok := val.(*Const); ok {
return NewConst(c.Value, typ)
}
// A representation-changing conversion.
c := &Convert{X: val}
c.setType(typ)
return f.emit(c)
}
// emitStore emits to f an instruction to store value val at location
// addr, applying implicit conversions as required by assignabilty rules.
//
func emitStore(f *Function, addr, val Value) *Store {
s := &Store{
Addr: addr,
Val: emitConv(f, val, deref(addr.Type())),
}
f.emit(s)
return s
}
// emitJump emits to f a jump to target, and updates the control-flow graph.
// Postcondition: f.currentBlock is nil.
//
func emitJump(f *Function, target *BasicBlock) {
b := f.currentBlock
b.emit(new(Jump))
addEdge(b, target)
f.currentBlock = nil
}
// emitIf emits to f a conditional jump to tblock or fblock based on
// cond, and updates the control-flow graph.
// Postcondition: f.currentBlock is nil.
//
func emitIf(f *Function, cond Value, tblock, fblock *BasicBlock) {
b := f.currentBlock
b.emit(&If{Cond: cond})
addEdge(b, tblock)
addEdge(b, fblock)
f.currentBlock = nil
}
// emitExtract emits to f an instruction to extract the index'th
// component of tuple, ascribing it type typ. It returns the
// extracted value.
//
func emitExtract(f *Function, tuple Value, index int, typ types.Type) Value {
e := &Extract{Tuple: tuple, Index: index}
// In all cases but one (tSelect's recv), typ is redundant w.r.t.
// tuple.Type().(*types.Tuple).Values[index].Type.
e.setType(typ)
return f.emit(e)
}
// emitTypeAssert emits to f a type assertion value := x.(t) and
// returns the value. x.Type() must be an interface.
//
func emitTypeAssert(f *Function, x Value, t types.Type, pos token.Pos) Value {
a := &TypeAssert{X: x, AssertedType: t}
a.setPos(pos)
a.setType(t)
return f.emit(a)
}
// emitTypeTest emits to f a type test value,ok := x.(t) and returns
// a (value, ok) tuple. x.Type() must be an interface.
//
func emitTypeTest(f *Function, x Value, t types.Type, pos token.Pos) Value {
a := &TypeAssert{
X: x,
AssertedType: t,
CommaOk: true,
}
a.setPos(pos)
a.setType(types.NewTuple(
types.NewVar(token.NoPos, nil, "value", t),
varOk,
))
return f.emit(a)
}
// emitTailCall emits to f a function call in tail position. The
// caller is responsible for all fields of 'call' except its type.
// Intended for wrapper methods.
// Precondition: f does/will not use deferred procedure calls.
// Postcondition: f.currentBlock is nil.
//
func emitTailCall(f *Function, call *Call) {
tresults := f.Signature.Results()
nr := tresults.Len()
if nr == 1 {
call.typ = tresults.At(0).Type()
} else {
call.typ = tresults
}
tuple := f.emit(call)
var ret Return
switch nr {
case 0:
// no-op
case 1:
ret.Results = []Value{tuple}
default:
for i := 0; i < nr; i++ {
v := emitExtract(f, tuple, i, tresults.At(i).Type())
// TODO(adonovan): in principle, this is required:
// v = emitConv(f, o.Type, f.Signature.Results[i].Type)
// but in practice emitTailCall is only used when
// the types exactly match.
ret.Results = append(ret.Results, v)
}
}
f.emit(&ret)
f.currentBlock = nil
}
// emitImplicitSelections emits to f code to apply the sequence of
// implicit field selections specified by indices to base value v, and
// returns the selected value.
//
go.tools/ssa: (another) major refactoring of method-set logic. We now use LookupFieldOrMethod for all SelectorExprs, and simplify the logic to discriminate the various cases. We inline static calls to promoted/indirected functions, dramatically reducing the number of functions created. More tests are needed, but I'd like to submit this as-is. In this CL, we: - rely less on Id strings. Internally we now use *types.Method (and its components) almost everywhere. - stop thinking of types.Methods as objects. They don't have stable identities. (Hopefully they will become plain-old structs soon.) - eliminate receiver indirection wrappers: indirection and promotion are handled together by makeWrapper. - Handle the interactions of promotion, indirection and abstract methods much more cleanly. - support receiver-bound interface method closures. - break up builder.selectField so we can re-use parts (emitFieldSelection). - add importer.PackageInfo.classifySelector utility. - delete interfaceMethodIndex() - delete namedTypeMethodIndex() - delete isSuperInterface() (replaced by types.IsAssignable) - call memberFromObject on each declared concrete method's *types.Func, not on every Method frem each method set, in the CREATE phase for packages loaded by gcimporter. go/types: - document Func, Signature.Recv() better. - use fmt in {Package,Label}.String - reimplement Func.String to be prettier and to include method receivers. API changes: - Function.method now holds the types.Method (soon to be not-an-object) for synthetic wrappers. - CallCommon.Method now contains an abstract (interface) method object; was an abstract method index. - CallCommon.MethodId() gone. - Program.LookupMethod now takes a *Method not an Id string. R=gri CC=golang-dev https://golang.org/cl/11674043
2013-07-26 09:22:34 -06:00
// If v is the address of a struct, the result will be the address of
// a field; if it is the value of a struct, the result will be the
// value of a field.
//
func emitImplicitSelections(f *Function, v Value, indices []int) Value {
for _, index := range indices {
fld := deref(v.Type()).Underlying().(*types.Struct).Field(index)
if isPointer(v.Type()) {
instr := &FieldAddr{
X: v,
Field: index,
}
instr.setType(types.NewPointer(fld.Type()))
v = f.emit(instr)
// Load the field's value iff indirectly embedded.
if isPointer(fld.Type()) {
v = emitLoad(f, v)
}
} else {
instr := &Field{
X: v,
Field: index,
}
instr.setType(fld.Type())
v = f.emit(instr)
}
}
return v
}
go.tools/ssa: (another) major refactoring of method-set logic. We now use LookupFieldOrMethod for all SelectorExprs, and simplify the logic to discriminate the various cases. We inline static calls to promoted/indirected functions, dramatically reducing the number of functions created. More tests are needed, but I'd like to submit this as-is. In this CL, we: - rely less on Id strings. Internally we now use *types.Method (and its components) almost everywhere. - stop thinking of types.Methods as objects. They don't have stable identities. (Hopefully they will become plain-old structs soon.) - eliminate receiver indirection wrappers: indirection and promotion are handled together by makeWrapper. - Handle the interactions of promotion, indirection and abstract methods much more cleanly. - support receiver-bound interface method closures. - break up builder.selectField so we can re-use parts (emitFieldSelection). - add importer.PackageInfo.classifySelector utility. - delete interfaceMethodIndex() - delete namedTypeMethodIndex() - delete isSuperInterface() (replaced by types.IsAssignable) - call memberFromObject on each declared concrete method's *types.Func, not on every Method frem each method set, in the CREATE phase for packages loaded by gcimporter. go/types: - document Func, Signature.Recv() better. - use fmt in {Package,Label}.String - reimplement Func.String to be prettier and to include method receivers. API changes: - Function.method now holds the types.Method (soon to be not-an-object) for synthetic wrappers. - CallCommon.Method now contains an abstract (interface) method object; was an abstract method index. - CallCommon.MethodId() gone. - Program.LookupMethod now takes a *Method not an Id string. R=gri CC=golang-dev https://golang.org/cl/11674043
2013-07-26 09:22:34 -06:00
// emitFieldSelection emits to f code to select the index'th field of v.
//
// If wantAddr, the input must be a pointer-to-struct and the result
// will be the field's address; otherwise the result will be the
// field's value.
//
func emitFieldSelection(f *Function, v Value, index int, wantAddr bool, pos token.Pos) Value {
fld := deref(v.Type()).Underlying().(*types.Struct).Field(index)
if isPointer(v.Type()) {
instr := &FieldAddr{
X: v,
Field: index,
}
instr.setPos(pos)
instr.setType(types.NewPointer(fld.Type()))
v = f.emit(instr)
// Load the field's value iff we don't want its address.
if !wantAddr {
v = emitLoad(f, v)
}
} else {
instr := &Field{
X: v,
Field: index,
}
instr.setPos(pos)
instr.setType(fld.Type())
v = f.emit(instr)
}
return v
}
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
// createRecoverBlock emits to f a block of code to return after a
// recovered panic, and sets f.Recover to it.
//
// If f's result parameters are named, the code loads and returns
// their current values, otherwise it returns the zero values of their
// type.
//
// Idempotent.
//
func createRecoverBlock(f *Function) {
if f.Recover != nil {
return // already created
}
saved := f.currentBlock
f.Recover = f.newBasicBlock("recover")
f.currentBlock = f.Recover
var results []Value
if f.namedResults != nil {
// Reload NRPs to form value tuple.
for _, r := range f.namedResults {
results = append(results, emitLoad(f, r))
}
} else {
R := f.Signature.Results()
for i, n := 0, R.Len(); i < n; i++ {
T := R.At(i).Type()
var v Value
// Return zero value of each result type.
switch T.Underlying().(type) {
case *types.Struct, *types.Array:
v = emitLoad(f, f.addLocal(T, token.NoPos))
default:
v = zeroConst(T)
}
results = append(results, v)
}
}
f.emit(&Return{Results: results})
f.currentBlock = saved
}