.gitignore added
[dotfiles/.git] / .config / coc / extensions / coc-go-data / tools / pkg / mod / honnef.co / go / tools@v0.1.1 / go / ir / func.go
diff --git a/.config/coc/extensions/coc-go-data/tools/pkg/mod/honnef.co/go/tools@v0.1.1/go/ir/func.go b/.config/coc/extensions/coc-go-data/tools/pkg/mod/honnef.co/go/tools@v0.1.1/go/ir/func.go
new file mode 100644 (file)
index 0000000..dca883d
--- /dev/null
@@ -0,0 +1,962 @@
+// 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 ir
+
+// This file implements the Function and BasicBlock types.
+
+import (
+       "bytes"
+       "fmt"
+       "go/ast"
+       "go/constant"
+       "go/format"
+       "go/token"
+       "go/types"
+       "io"
+       "os"
+       "strings"
+)
+
+// addEdge adds a control-flow graph edge from from to to.
+func addEdge(from, to *BasicBlock) {
+       from.Succs = append(from.Succs, to)
+       to.Preds = append(to.Preds, from)
+}
+
+// Control returns the last instruction in the block.
+func (b *BasicBlock) Control() Instruction {
+       if len(b.Instrs) == 0 {
+               return nil
+       }
+       return b.Instrs[len(b.Instrs)-1]
+}
+
+// SigmaFor returns the sigma node for v coming from pred.
+func (b *BasicBlock) SigmaFor(v Value, pred *BasicBlock) *Sigma {
+       for _, instr := range b.Instrs {
+               sigma, ok := instr.(*Sigma)
+               if !ok {
+                       // no more sigmas
+                       return nil
+               }
+               if sigma.From == pred && sigma.X == v {
+                       return sigma
+               }
+       }
+       return nil
+}
+
+// Parent returns the function that contains block b.
+func (b *BasicBlock) Parent() *Function { return b.parent }
+
+// String returns a human-readable label of this block.
+// It is not guaranteed unique within the function.
+//
+func (b *BasicBlock) String() string {
+       return fmt.Sprintf("%d", b.Index)
+}
+
+// emit appends an instruction to the current basic block.
+// If the instruction defines a Value, it is returned.
+//
+func (b *BasicBlock) emit(i Instruction, source ast.Node) Value {
+       i.setSource(source)
+       i.setBlock(b)
+       b.Instrs = append(b.Instrs, i)
+       v, _ := i.(Value)
+       return v
+}
+
+// predIndex returns the i such that b.Preds[i] == c or panics if
+// there is none.
+func (b *BasicBlock) predIndex(c *BasicBlock) int {
+       for i, pred := range b.Preds {
+               if pred == c {
+                       return i
+               }
+       }
+       panic(fmt.Sprintf("no edge %s -> %s", c, b))
+}
+
+// succIndex returns the i such that b.Succs[i] == c or -1 if there is none.
+func (b *BasicBlock) succIndex(c *BasicBlock) int {
+       for i, succ := range b.Succs {
+               if succ == c {
+                       return i
+               }
+       }
+       return -1
+}
+
+// hasPhi returns true if b.Instrs contains φ-nodes.
+func (b *BasicBlock) hasPhi() bool {
+       _, ok := b.Instrs[0].(*Phi)
+       return ok
+}
+
+func (b *BasicBlock) Phis() []Instruction {
+       return b.phis()
+}
+
+// phis returns the prefix of b.Instrs containing all the block's φ-nodes.
+func (b *BasicBlock) phis() []Instruction {
+       for i, instr := range b.Instrs {
+               if _, ok := instr.(*Phi); !ok {
+                       return b.Instrs[:i]
+               }
+       }
+       return nil // unreachable in well-formed blocks
+}
+
+// replacePred replaces all occurrences of p in b's predecessor list with q.
+// Ordinarily there should be at most one.
+//
+func (b *BasicBlock) replacePred(p, q *BasicBlock) {
+       for i, pred := range b.Preds {
+               if pred == p {
+                       b.Preds[i] = q
+               }
+       }
+}
+
+// replaceSucc replaces all occurrences of p in b's successor list with q.
+// Ordinarily there should be at most one.
+//
+func (b *BasicBlock) replaceSucc(p, q *BasicBlock) {
+       for i, succ := range b.Succs {
+               if succ == p {
+                       b.Succs[i] = q
+               }
+       }
+}
+
+// removePred removes all occurrences of p in b's
+// predecessor list and φ-nodes.
+// Ordinarily there should be at most one.
+//
+func (b *BasicBlock) removePred(p *BasicBlock) {
+       phis := b.phis()
+
+       // We must preserve edge order for φ-nodes.
+       j := 0
+       for i, pred := range b.Preds {
+               if pred != p {
+                       b.Preds[j] = b.Preds[i]
+                       // Strike out φ-edge too.
+                       for _, instr := range phis {
+                               phi := instr.(*Phi)
+                               phi.Edges[j] = phi.Edges[i]
+                       }
+                       j++
+               }
+       }
+       // Nil out b.Preds[j:] and φ-edges[j:] to aid GC.
+       for i := j; i < len(b.Preds); i++ {
+               b.Preds[i] = nil
+               for _, instr := range phis {
+                       instr.(*Phi).Edges[i] = nil
+               }
+       }
+       b.Preds = b.Preds[:j]
+       for _, instr := range phis {
+               phi := instr.(*Phi)
+               phi.Edges = phi.Edges[:j]
+       }
+}
+
+// Destinations associated with unlabelled for/switch/select stmts.
+// We push/pop one of these as we enter/leave each construct and for
+// each BranchStmt we scan for the innermost target of the right type.
+//
+type targets struct {
+       tail         *targets // rest of stack
+       _break       *BasicBlock
+       _continue    *BasicBlock
+       _fallthrough *BasicBlock
+}
+
+// Destinations associated with a labelled block.
+// We populate these as labels are encountered in forward gotos or
+// labelled statements.
+//
+type lblock struct {
+       _goto     *BasicBlock
+       _break    *BasicBlock
+       _continue *BasicBlock
+}
+
+// labelledBlock returns the branch target associated with the
+// specified label, creating it if needed.
+//
+func (f *Function) labelledBlock(label *ast.Ident) *lblock {
+       lb := f.lblocks[label.Obj]
+       if lb == nil {
+               lb = &lblock{_goto: f.newBasicBlock(label.Name)}
+               if f.lblocks == nil {
+                       f.lblocks = make(map[*ast.Object]*lblock)
+               }
+               f.lblocks[label.Obj] = lb
+       }
+       return lb
+}
+
+// addParam adds a (non-escaping) parameter to f.Params of the
+// specified name, type and source position.
+//
+func (f *Function) addParam(name string, typ types.Type, source ast.Node) *Parameter {
+       var b *BasicBlock
+       if len(f.Blocks) > 0 {
+               b = f.Blocks[0]
+       }
+       v := &Parameter{
+               name: name,
+       }
+       v.setBlock(b)
+       v.setType(typ)
+       v.setSource(source)
+       f.Params = append(f.Params, v)
+       if b != nil {
+               // There may be no blocks if this function has no body. We
+               // still create params, but aren't interested in the
+               // instruction.
+               f.Blocks[0].Instrs = append(f.Blocks[0].Instrs, v)
+       }
+       return v
+}
+
+func (f *Function) addParamObj(obj types.Object, source ast.Node) *Parameter {
+       name := obj.Name()
+       if name == "" {
+               name = fmt.Sprintf("arg%d", len(f.Params))
+       }
+       param := f.addParam(name, obj.Type(), source)
+       param.object = obj
+       return param
+}
+
+// addSpilledParam declares a parameter that is pre-spilled to the
+// stack; the function body will load/store the spilled location.
+// Subsequent lifting will eliminate spills where possible.
+//
+func (f *Function) addSpilledParam(obj types.Object, source ast.Node) {
+       param := f.addParamObj(obj, source)
+       spill := &Alloc{}
+       spill.setType(types.NewPointer(obj.Type()))
+       spill.source = source
+       f.objects[obj] = spill
+       f.Locals = append(f.Locals, spill)
+       f.emit(spill, source)
+       emitStore(f, spill, param, source)
+       // f.emit(&Store{Addr: spill, Val: param})
+}
+
+// startBody initializes the function prior to generating IR code for its body.
+// Precondition: f.Type() already set.
+//
+func (f *Function) startBody() {
+       entry := f.newBasicBlock("entry")
+       f.currentBlock = entry
+       f.objects = make(map[types.Object]Value) // needed for some synthetics, e.g. init
+}
+
+func (f *Function) blockset(i int) *BlockSet {
+       bs := &f.blocksets[i]
+       if len(bs.values) != len(f.Blocks) {
+               if cap(bs.values) >= len(f.Blocks) {
+                       bs.values = bs.values[:len(f.Blocks)]
+                       bs.Clear()
+               } else {
+                       bs.values = make([]bool, len(f.Blocks))
+               }
+       } else {
+               bs.Clear()
+       }
+       return bs
+}
+
+func (f *Function) exitBlock() {
+       old := f.currentBlock
+
+       f.Exit = f.newBasicBlock("exit")
+       f.currentBlock = f.Exit
+
+       ret := f.results()
+       results := make([]Value, len(ret))
+       // Run function calls deferred in this
+       // function when explicitly returning from it.
+       f.emit(new(RunDefers), nil)
+       for i, r := range ret {
+               results[i] = emitLoad(f, r, nil)
+       }
+
+       f.emit(&Return{Results: results}, nil)
+       f.currentBlock = old
+}
+
+// createSyntacticParams populates f.Params and generates code (spills
+// and named result locals) for all the parameters declared in the
+// syntax.  In addition it populates the f.objects mapping.
+//
+// Preconditions:
+// f.startBody() was called.
+// Postcondition:
+// len(f.Params) == len(f.Signature.Params) + (f.Signature.Recv() ? 1 : 0)
+//
+func (f *Function) createSyntacticParams(recv *ast.FieldList, functype *ast.FuncType) {
+       // Receiver (at most one inner iteration).
+       if recv != nil {
+               for _, field := range recv.List {
+                       for _, n := range field.Names {
+                               f.addSpilledParam(f.Pkg.info.Defs[n], n)
+                       }
+                       // Anonymous receiver?  No need to spill.
+                       if field.Names == nil {
+                               f.addParamObj(f.Signature.Recv(), field)
+                       }
+               }
+       }
+
+       // Parameters.
+       if functype.Params != nil {
+               n := len(f.Params) // 1 if has recv, 0 otherwise
+               for _, field := range functype.Params.List {
+                       for _, n := range field.Names {
+                               f.addSpilledParam(f.Pkg.info.Defs[n], n)
+                       }
+                       // Anonymous parameter?  No need to spill.
+                       if field.Names == nil {
+                               f.addParamObj(f.Signature.Params().At(len(f.Params)-n), field)
+                       }
+               }
+       }
+
+       // Named results.
+       if functype.Results != nil {
+               for _, field := range functype.Results.List {
+                       // Implicit "var" decl of locals for named results.
+                       for _, n := range field.Names {
+                               f.namedResults = append(f.namedResults, f.addLocalForIdent(n))
+                       }
+               }
+
+               if len(f.namedResults) == 0 {
+                       sig := f.Signature.Results()
+                       for i := 0; i < sig.Len(); i++ {
+                               // XXX position information
+                               v := f.addLocal(sig.At(i).Type(), nil)
+                               f.implicitResults = append(f.implicitResults, v)
+                       }
+               }
+       }
+}
+
+func numberNodes(f *Function) {
+       var base ID
+       for _, b := range f.Blocks {
+               for _, instr := range b.Instrs {
+                       if instr == nil {
+                               continue
+                       }
+                       base++
+                       instr.setID(base)
+               }
+       }
+}
+
+// buildReferrers populates the def/use information in all non-nil
+// Value.Referrers slice.
+// Precondition: all such slices are initially empty.
+func buildReferrers(f *Function) {
+       var rands []*Value
+       for _, b := range f.Blocks {
+               for _, instr := range b.Instrs {
+                       rands = instr.Operands(rands[:0]) // recycle storage
+                       for _, rand := range rands {
+                               if r := *rand; r != nil {
+                                       if ref := r.Referrers(); ref != nil {
+                                               *ref = append(*ref, instr)
+                                       }
+                               }
+                       }
+               }
+       }
+}
+
+func (f *Function) emitConsts() {
+       if len(f.Blocks) == 0 {
+               f.consts = nil
+               return
+       }
+
+       // TODO(dh): our deduplication only works on booleans and
+       // integers. other constants are represented as pointers to
+       // things.
+       if len(f.consts) == 0 {
+               return
+       } else if len(f.consts) <= 32 {
+               f.emitConstsFew()
+       } else {
+               f.emitConstsMany()
+       }
+}
+
+func (f *Function) emitConstsFew() {
+       dedup := make([]*Const, 0, 32)
+       for _, c := range f.consts {
+               if len(*c.Referrers()) == 0 {
+                       continue
+               }
+               found := false
+               for _, d := range dedup {
+                       if c.typ == d.typ && c.Value == d.Value {
+                               replaceAll(c, d)
+                               found = true
+                               break
+                       }
+               }
+               if !found {
+                       dedup = append(dedup, c)
+               }
+       }
+
+       instrs := make([]Instruction, len(f.Blocks[0].Instrs)+len(dedup))
+       for i, c := range dedup {
+               instrs[i] = c
+               c.setBlock(f.Blocks[0])
+       }
+       copy(instrs[len(dedup):], f.Blocks[0].Instrs)
+       f.Blocks[0].Instrs = instrs
+       f.consts = nil
+}
+
+func (f *Function) emitConstsMany() {
+       type constKey struct {
+               typ   types.Type
+               value constant.Value
+       }
+
+       m := make(map[constKey]Value, len(f.consts))
+       areNil := 0
+       for i, c := range f.consts {
+               if len(*c.Referrers()) == 0 {
+                       f.consts[i] = nil
+                       areNil++
+                       continue
+               }
+
+               k := constKey{
+                       typ:   c.typ,
+                       value: c.Value,
+               }
+               if dup, ok := m[k]; !ok {
+                       m[k] = c
+               } else {
+                       f.consts[i] = nil
+                       areNil++
+                       replaceAll(c, dup)
+               }
+       }
+
+       instrs := make([]Instruction, len(f.Blocks[0].Instrs)+len(f.consts)-areNil)
+       i := 0
+       for _, c := range f.consts {
+               if c != nil {
+                       instrs[i] = c
+                       c.setBlock(f.Blocks[0])
+                       i++
+               }
+       }
+       copy(instrs[i:], f.Blocks[0].Instrs)
+       f.Blocks[0].Instrs = instrs
+       f.consts = nil
+}
+
+// buildFakeExits ensures that every block in the function is
+// reachable in reverse from the Exit block. This is required to build
+// a full post-dominator tree, and to ensure the exit block's
+// inclusion in the dominator tree.
+func buildFakeExits(fn *Function) {
+       // Find back-edges via forward DFS
+       fn.fakeExits = BlockSet{values: make([]bool, len(fn.Blocks))}
+       seen := fn.blockset(0)
+       backEdges := fn.blockset(1)
+
+       var dfs func(b *BasicBlock)
+       dfs = func(b *BasicBlock) {
+               if !seen.Add(b) {
+                       backEdges.Add(b)
+                       return
+               }
+               for _, pred := range b.Succs {
+                       dfs(pred)
+               }
+       }
+       dfs(fn.Blocks[0])
+buildLoop:
+       for {
+               seen := fn.blockset(2)
+               var dfs func(b *BasicBlock)
+               dfs = func(b *BasicBlock) {
+                       if !seen.Add(b) {
+                               return
+                       }
+                       for _, pred := range b.Preds {
+                               dfs(pred)
+                       }
+                       if b == fn.Exit {
+                               for _, b := range fn.Blocks {
+                                       if fn.fakeExits.Has(b) {
+                                               dfs(b)
+                                       }
+                               }
+                       }
+               }
+               dfs(fn.Exit)
+
+               for _, b := range fn.Blocks {
+                       if !seen.Has(b) && backEdges.Has(b) {
+                               // Block b is not reachable from the exit block. Add a
+                               // fake jump from b to exit, then try again. Note that we
+                               // only add one fake edge at a time, as it may make
+                               // multiple blocks reachable.
+                               //
+                               // We only consider those blocks that have back edges.
+                               // Any unreachable block that doesn't have a back edge
+                               // must flow into a loop, which by definition has a
+                               // back edge. Thus, by looking for loops, we should
+                               // need fewer fake edges overall.
+                               fn.fakeExits.Add(b)
+                               continue buildLoop
+                       }
+               }
+
+               break
+       }
+}
+
+// finishBody() finalizes the function after IR code generation of its body.
+func (f *Function) finishBody() {
+       f.objects = nil
+       f.currentBlock = nil
+       f.lblocks = nil
+
+       // Remove from f.Locals any Allocs that escape to the heap.
+       j := 0
+       for _, l := range f.Locals {
+               if !l.Heap {
+                       f.Locals[j] = l
+                       j++
+               }
+       }
+       // Nil out f.Locals[j:] to aid GC.
+       for i := j; i < len(f.Locals); i++ {
+               f.Locals[i] = nil
+       }
+       f.Locals = f.Locals[:j]
+
+       optimizeBlocks(f)
+       buildFakeExits(f)
+       buildReferrers(f)
+       buildDomTree(f)
+       buildPostDomTree(f)
+
+       if f.Prog.mode&NaiveForm == 0 {
+               lift(f)
+       }
+
+       // emit constants after lifting, because lifting may produce new constants.
+       f.emitConsts()
+
+       f.namedResults = nil // (used by lifting)
+       f.implicitResults = nil
+
+       numberNodes(f)
+
+       defer f.wr.Close()
+       f.wr.WriteFunc("start", "start", f)
+
+       if f.Prog.mode&PrintFunctions != 0 {
+               printMu.Lock()
+               f.WriteTo(os.Stdout)
+               printMu.Unlock()
+       }
+
+       if f.Prog.mode&SanityCheckFunctions != 0 {
+               mustSanityCheck(f, nil)
+       }
+}
+
+func isUselessPhi(phi *Phi) (Value, bool) {
+       var v0 Value
+       for _, e := range phi.Edges {
+               if e == phi {
+                       continue
+               }
+               if v0 == nil {
+                       v0 = e
+               }
+               if v0 != e {
+                       if v0, ok := v0.(*Const); ok {
+                               if e, ok := e.(*Const); ok {
+                                       if v0.typ == e.typ && v0.Value == e.Value {
+                                               continue
+                                       }
+                               }
+                       }
+                       return nil, false
+               }
+       }
+       return v0, true
+}
+
+func (f *Function) RemoveNilBlocks() {
+       f.removeNilBlocks()
+}
+
+// removeNilBlocks eliminates nils from f.Blocks and updates each
+// BasicBlock.Index.  Use this after any pass that may delete blocks.
+//
+func (f *Function) removeNilBlocks() {
+       j := 0
+       for _, b := range f.Blocks {
+               if b != nil {
+                       b.Index = j
+                       f.Blocks[j] = b
+                       j++
+               }
+       }
+       // Nil out f.Blocks[j:] to aid GC.
+       for i := j; i < len(f.Blocks); i++ {
+               f.Blocks[i] = nil
+       }
+       f.Blocks = f.Blocks[:j]
+}
+
+// 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 (pkg *Package) SetDebugMode(debug bool) {
+       // TODO(adonovan): do we want ast.File granularity?
+       pkg.debug = debug
+}
+
+// debugInfo reports whether debug info is wanted for this function.
+func (f *Function) debugInfo() bool {
+       return f.Pkg != nil && f.Pkg.debug
+}
+
+// addNamedLocal creates a local variable, adds it to function f and
+// returns it.  Its name and type are taken from obj.  Subsequent
+// calls to f.lookup(obj) will return the same local.
+//
+func (f *Function) addNamedLocal(obj types.Object, source ast.Node) *Alloc {
+       l := f.addLocal(obj.Type(), source)
+       f.objects[obj] = l
+       return l
+}
+
+func (f *Function) addLocalForIdent(id *ast.Ident) *Alloc {
+       return f.addNamedLocal(f.Pkg.info.Defs[id], id)
+}
+
+// addLocal creates an anonymous local variable of type typ, adds it
+// to function f and returns it.  pos is the optional source location.
+//
+func (f *Function) addLocal(typ types.Type, source ast.Node) *Alloc {
+       v := &Alloc{}
+       v.setType(types.NewPointer(typ))
+       f.Locals = append(f.Locals, v)
+       f.emit(v, source)
+       return v
+}
+
+// lookup returns the address of the named variable identified by obj
+// that is local to function f or one of its enclosing functions.
+// If escaping, the reference comes from a potentially escaping pointer
+// expression and the referent must be heap-allocated.
+//
+func (f *Function) lookup(obj types.Object, escaping bool) Value {
+       if v, ok := f.objects[obj]; ok {
+               if alloc, ok := v.(*Alloc); ok && escaping {
+                       alloc.Heap = true
+               }
+               return v // function-local var (address)
+       }
+
+       // Definition must be in an enclosing function;
+       // plumb it through intervening closures.
+       if f.parent == nil {
+               panic("no ir.Value for " + obj.String())
+       }
+       outer := f.parent.lookup(obj, true) // escaping
+       v := &FreeVar{
+               name:   obj.Name(),
+               typ:    outer.Type(),
+               outer:  outer,
+               parent: f,
+       }
+       f.objects[obj] = v
+       f.FreeVars = append(f.FreeVars, v)
+       return v
+}
+
+// emit emits the specified instruction to function f.
+func (f *Function) emit(instr Instruction, source ast.Node) Value {
+       return f.currentBlock.emit(instr, source)
+}
+
+// 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 (f *Function) RelString(from *types.Package) string {
+       // Anonymous?
+       if f.parent != nil {
+               // An anonymous function's Name() looks like "parentName$1",
+               // but its String() should include the type/package/etc.
+               parent := f.parent.RelString(from)
+               for i, anon := range f.parent.AnonFuncs {
+                       if anon == f {
+                               return fmt.Sprintf("%s$%d", parent, 1+i)
+                       }
+               }
+
+               return f.name // should never happen
+       }
+
+       // Method (declared or wrapper)?
+       if recv := f.Signature.Recv(); recv != nil {
+               return f.relMethod(from, recv.Type())
+       }
+
+       // Thunk?
+       if f.method != nil {
+               return f.relMethod(from, f.method.Recv())
+       }
+
+       // Bound?
+       if len(f.FreeVars) == 1 && strings.HasSuffix(f.name, "$bound") {
+               return f.relMethod(from, f.FreeVars[0].Type())
+       }
+
+       // Package-level function?
+       // Prefix with package name for cross-package references only.
+       if p := f.pkg(); p != nil && p != from {
+               return fmt.Sprintf("%s.%s", p.Path(), f.name)
+       }
+
+       // Unknown.
+       return f.name
+}
+
+func (f *Function) relMethod(from *types.Package, recv types.Type) string {
+       return fmt.Sprintf("(%s).%s", relType(recv, from), f.name)
+}
+
+// writeSignature writes to buf the signature sig in declaration syntax.
+func writeSignature(buf *bytes.Buffer, from *types.Package, name string, sig *types.Signature, params []*Parameter) {
+       buf.WriteString("func ")
+       if recv := sig.Recv(); recv != nil {
+               buf.WriteString("(")
+               if n := params[0].Name(); n != "" {
+                       buf.WriteString(n)
+                       buf.WriteString(" ")
+               }
+               types.WriteType(buf, params[0].Type(), types.RelativeTo(from))
+               buf.WriteString(") ")
+       }
+       buf.WriteString(name)
+       types.WriteSignature(buf, sig, types.RelativeTo(from))
+}
+
+func (f *Function) pkg() *types.Package {
+       if f.Pkg != nil {
+               return f.Pkg.Pkg
+       }
+       return nil
+}
+
+var _ io.WriterTo = (*Function)(nil) // *Function implements io.Writer
+
+func (f *Function) WriteTo(w io.Writer) (int64, error) {
+       var buf bytes.Buffer
+       WriteFunction(&buf, f)
+       n, err := w.Write(buf.Bytes())
+       return int64(n), err
+}
+
+// WriteFunction writes to buf a human-readable "disassembly" of f.
+func WriteFunction(buf *bytes.Buffer, f *Function) {
+       fmt.Fprintf(buf, "# Name: %s\n", f.String())
+       if f.Pkg != nil {
+               fmt.Fprintf(buf, "# Package: %s\n", f.Pkg.Pkg.Path())
+       }
+       if syn := f.Synthetic; syn != 0 {
+               fmt.Fprintln(buf, "# Synthetic:", syn)
+       }
+       if pos := f.Pos(); pos.IsValid() {
+               fmt.Fprintf(buf, "# Location: %s\n", f.Prog.Fset.Position(pos))
+       }
+
+       if f.parent != nil {
+               fmt.Fprintf(buf, "# Parent: %s\n", f.parent.Name())
+       }
+
+       from := f.pkg()
+
+       if f.FreeVars != nil {
+               buf.WriteString("# Free variables:\n")
+               for i, fv := range f.FreeVars {
+                       fmt.Fprintf(buf, "# % 3d:\t%s %s\n", i, fv.Name(), relType(fv.Type(), from))
+               }
+       }
+
+       if len(f.Locals) > 0 {
+               buf.WriteString("# Locals:\n")
+               for i, l := range f.Locals {
+                       fmt.Fprintf(buf, "# % 3d:\t%s %s\n", i, l.Name(), relType(deref(l.Type()), from))
+               }
+       }
+       writeSignature(buf, from, f.Name(), f.Signature, f.Params)
+       buf.WriteString(":\n")
+
+       if f.Blocks == nil {
+               buf.WriteString("\t(external)\n")
+       }
+
+       for _, b := range f.Blocks {
+               if b == nil {
+                       // Corrupt CFG.
+                       fmt.Fprintf(buf, ".nil:\n")
+                       continue
+               }
+               fmt.Fprintf(buf, "b%d:", b.Index)
+               if len(b.Preds) > 0 {
+                       fmt.Fprint(buf, " ←")
+                       for _, pred := range b.Preds {
+                               fmt.Fprintf(buf, " b%d", pred.Index)
+                       }
+               }
+               if b.Comment != "" {
+                       fmt.Fprintf(buf, " # %s", b.Comment)
+               }
+               buf.WriteByte('\n')
+
+               if false { // CFG debugging
+                       fmt.Fprintf(buf, "\t# CFG: %s --> %s --> %s\n", b.Preds, b, b.Succs)
+               }
+
+               buf2 := &bytes.Buffer{}
+               for _, instr := range b.Instrs {
+                       buf.WriteString("\t")
+                       switch v := instr.(type) {
+                       case Value:
+                               // Left-align the instruction.
+                               if name := v.Name(); name != "" {
+                                       fmt.Fprintf(buf, "%s = ", name)
+                               }
+                               buf.WriteString(instr.String())
+                       case nil:
+                               // Be robust against bad transforms.
+                               buf.WriteString("<deleted>")
+                       default:
+                               buf.WriteString(instr.String())
+                       }
+                       buf.WriteString("\n")
+
+                       if f.Prog.mode&PrintSource != 0 {
+                               if s := instr.Source(); s != nil {
+                                       buf2.Reset()
+                                       format.Node(buf2, f.Prog.Fset, s)
+                                       for {
+                                               line, err := buf2.ReadString('\n')
+                                               if len(line) == 0 {
+                                                       break
+                                               }
+                                               buf.WriteString("\t\t> ")
+                                               buf.WriteString(line)
+                                               if line[len(line)-1] != '\n' {
+                                                       buf.WriteString("\n")
+                                               }
+                                               if err != nil {
+                                                       break
+                                               }
+                                       }
+                               }
+                       }
+               }
+               buf.WriteString("\n")
+       }
+}
+
+// newBasicBlock adds to f a new basic block and returns it.  It does
+// not automatically become the current block for subsequent calls to emit.
+// comment is an optional string for more readable debugging output.
+//
+func (f *Function) newBasicBlock(comment string) *BasicBlock {
+       b := &BasicBlock{
+               Index:   len(f.Blocks),
+               Comment: comment,
+               parent:  f,
+       }
+       b.Succs = b.succs2[:0]
+       f.Blocks = append(f.Blocks, b)
+       return b
+}
+
+// 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 (prog *Program) NewFunction(name string, sig *types.Signature, provenance Synthetic) *Function {
+       return &Function{Prog: prog, name: name, Signature: sig, Synthetic: provenance}
+}
+
+//lint:ignore U1000 we may make use of this for functions loaded from export data
+type extentNode [2]token.Pos
+
+func (n extentNode) Pos() token.Pos { return n[0] }
+func (n extentNode) End() token.Pos { return n[1] }
+
+func (f *Function) initHTML(name string) {
+       if name == "" {
+               return
+       }
+       if rel := f.RelString(nil); rel == name {
+               f.wr = NewHTMLWriter("ir.html", rel, "")
+       }
+}