// translatePackage translates an *ssa.Package into an LLVM module, and returns
// the translation unit information.
func (u *unit) translatePackage(pkg *ssa.Package) {
ms := make([]ssa.Member, len(pkg.Members))
i := 0
for _, m := range pkg.Members {
ms[i] = m
i++
}
sort.Sort(byMemberName(ms))
// Initialize global storage and type descriptors for this package.
// We must create globals regardless of whether they're referenced,
// hence the duplication in frame.value.
for _, m := range ms {
switch v := m.(type) {
case *ssa.Global:
elemtyp := deref(v.Type())
llelemtyp := u.llvmtypes.ToLLVM(elemtyp)
vname := u.types.mc.mangleGlobalName(v)
global := llvm.AddGlobal(u.module.Module, llelemtyp, vname)
if !v.Object().Exported() {
global.SetLinkage(llvm.InternalLinkage)
}
u.addGlobal(global, elemtyp)
global = llvm.ConstBitCast(global, u.llvmtypes.ToLLVM(v.Type()))
u.globals[v] = global
case *ssa.Type:
u.types.getTypeDescriptorPointer(v.Type())
}
}
// Define functions.
u.defineFunctionsInOrder(ssautil.AllFunctions(pkg.Prog))
// Emit initializers for type descriptors, which may trigger
// the resolution of additional functions.
u.types.emitTypeDescInitializers()
// Define remaining functions that were resolved during
// runtime type mapping, but not defined.
u.defineFunctionsInOrder(u.undefinedFuncs)
// Set initializers for globals.
for global, init := range u.globalInits {
initval := init.build(global.Type().ElementType())
global.SetInitializer(initval)
}
}
// CallGraph computes the call graph of the specified program
// considering only static calls.
//
func CallGraph(prog *ssa.Program) *callgraph.Graph {
cg := callgraph.New(nil) // TODO(adonovan) eliminate concept of rooted callgraph
// TODO(adonovan): opt: use only a single pass over the ssa.Program.
for f := range ssautil.AllFunctions(prog) {
fnode := cg.CreateNode(f)
for _, b := range f.Blocks {
for _, instr := range b.Instrs {
if site, ok := instr.(ssa.CallInstruction); ok {
if g := site.Common().StaticCallee(); g != nil {
gnode := cg.CreateNode(g)
callgraph.AddEdge(fnode, site, gnode)
}
}
}
}
}
return cg
}
func doOneInput(input, filename string) bool {
var conf loader.Config
// Parsing.
f, err := conf.ParseFile(filename, input)
if err != nil {
fmt.Println(err)
return false
}
// Create single-file main package and import its dependencies.
conf.CreateFromFiles("main", f)
iprog, err := conf.Load()
if err != nil {
fmt.Println(err)
return false
}
mainPkgInfo := iprog.Created[0].Pkg
// SSA creation + building.
prog := ssa.Create(iprog, ssa.SanityCheckFunctions)
prog.BuildAll()
mainpkg := prog.Package(mainPkgInfo)
ptrmain := mainpkg // main package for the pointer analysis
if mainpkg.Func("main") == nil {
// No main function; assume it's a test.
ptrmain = prog.CreateTestMainPackage(mainpkg)
}
// Find all calls to the built-in print(x). Analytically,
// print is a no-op, but it's a convenient hook for testing
// the PTS of an expression, so our tests use it.
probes := make(map[*ssa.CallCommon]bool)
for fn := range ssautil.AllFunctions(prog) {
if fn.Pkg == mainpkg {
for _, b := range fn.Blocks {
for _, instr := range b.Instrs {
if instr, ok := instr.(ssa.CallInstruction); ok {
call := instr.Common()
if b, ok := call.Value.(*ssa.Builtin); ok && b.Name() == "print" && len(call.Args) == 1 {
probes[instr.Common()] = true
}
}
}
}
}
}
ok := true
lineMapping := make(map[string]string) // maps "file:line" to @line tag
// Parse expectations in this input.
var exps []*expectation
re := regexp.MustCompile("// *@([a-z]*) *(.*)$")
lines := strings.Split(input, "\n")
for linenum, line := range lines {
linenum++ // make it 1-based
if matches := re.FindAllStringSubmatch(line, -1); matches != nil {
match := matches[0]
kind, rest := match[1], match[2]
e := &expectation{kind: kind, filename: filename, linenum: linenum}
if kind == "line" {
if rest == "" {
ok = false
e.errorf("@%s expectation requires identifier", kind)
} else {
lineMapping[fmt.Sprintf("%s:%d", filename, linenum)] = rest
}
continue
}
if e.needsProbe() && !strings.Contains(line, "print(") {
ok = false
e.errorf("@%s expectation must follow call to print(x)", kind)
continue
}
switch kind {
case "pointsto":
e.args = split(rest, "|")
case "types":
for _, typstr := range split(rest, "|") {
var t types.Type = types.Typ[types.Invalid] // means "..."
if typstr != "..." {
texpr, err := parser.ParseExpr(typstr)
if err != nil {
ok = false
// Don't print err since its location is bad.
e.errorf("'%s' is not a valid type", typstr)
continue
}
mainFileScope := mainpkg.Object.Scope().Child(0)
tv, err := types.EvalNode(prog.Fset, texpr, mainpkg.Object, mainFileScope)
if err != nil {
ok = false
// Don't print err since its location is bad.
e.errorf("'%s' is not a valid type: %s", typstr, err)
continue
}
t = tv.Type
}
e.types = append(e.types, t)
}
case "calls":
e.args = split(rest, "->")
// TODO(adonovan): eagerly reject the
// expectation if fn doesn't denote
// existing function, rather than fail
// the expectation after analysis.
if len(e.args) != 2 {
ok = false
e.errorf("@calls expectation wants 'caller -> callee' arguments")
continue
}
case "warning":
lit, err := strconv.Unquote(strings.TrimSpace(rest))
if err != nil {
ok = false
e.errorf("couldn't parse @warning operand: %s", err.Error())
continue
}
e.args = append(e.args, lit)
default:
ok = false
e.errorf("unknown expectation kind: %s", e)
continue
}
exps = append(exps, e)
}
}
var log bytes.Buffer
fmt.Fprintf(&log, "Input: %s\n", filename)
// Run the analysis.
config := &pointer.Config{
Reflection: true,
BuildCallGraph: true,
Mains: []*ssa.Package{ptrmain},
Log: &log,
}
for probe := range probes {
v := probe.Args[0]
if pointer.CanPoint(v.Type()) {
config.AddQuery(v)
}
}
// Print the log is there was an error or a panic.
complete := false
defer func() {
if !complete || !ok {
log.WriteTo(os.Stderr)
}
}()
result, err := pointer.Analyze(config)
if err != nil {
panic(err) // internal error in pointer analysis
}
// Check the expectations.
for _, e := range exps {
var call *ssa.CallCommon
var pts pointer.PointsToSet
var tProbe types.Type
if e.needsProbe() {
if call, pts = findProbe(prog, probes, result.Queries, e); call == nil {
ok = false
e.errorf("unreachable print() statement has expectation %s", e)
continue
}
tProbe = call.Args[0].Type()
if !pointer.CanPoint(tProbe) {
ok = false
e.errorf("expectation on non-pointerlike operand: %s", tProbe)
continue
}
}
switch e.kind {
case "pointsto":
if !checkPointsToExpectation(e, pts, lineMapping, prog) {
ok = false
}
case "types":
if !checkTypesExpectation(e, pts, tProbe) {
ok = false
}
case "calls":
if !checkCallsExpectation(prog, e, result.CallGraph) {
ok = false
}
case "warning":
if !checkWarningExpectation(prog, e, result.Warnings) {
ok = false
}
}
}
complete = true
// ok = false // debugging: uncomment to always see log
return ok
}
// CallGraph computes the call graph of the specified program using the
// Class Hierarchy Analysis algorithm.
//
func CallGraph(prog *ssa.Program) *callgraph.Graph {
cg := callgraph.New(nil) // TODO(adonovan) eliminate concept of rooted callgraph
allFuncs := ssautil.AllFunctions(prog)
// funcsBySig contains all functions, keyed by signature. It is
// the effective set of address-taken functions used to resolve
// a dynamic call of a particular signature.
var funcsBySig typeutil.Map // value is []*ssa.Function
// methodsByName contains all methods,
// grouped by name for efficient lookup.
methodsByName := make(map[string][]*ssa.Function)
// methodsMemo records, for every abstract method call call I.f on
// interface type I, the set of concrete methods C.f of all
// types C that satisfy interface I.
methodsMemo := make(map[*types.Func][]*ssa.Function)
lookupMethods := func(m *types.Func) []*ssa.Function {
methods, ok := methodsMemo[m]
if !ok {
I := m.Type().(*types.Signature).Recv().Type().Underlying().(*types.Interface)
for _, f := range methodsByName[m.Name()] {
C := f.Signature.Recv().Type() // named or *named
if types.Implements(C, I) {
methods = append(methods, f)
}
}
methodsMemo[m] = methods
}
return methods
}
for f := range allFuncs {
if f.Signature.Recv() == nil {
// Package initializers can never be address-taken.
if f.Name() == "init" && f.Synthetic == "package initializer" {
continue
}
funcs, _ := funcsBySig.At(f.Signature).([]*ssa.Function)
funcs = append(funcs, f)
funcsBySig.Set(f.Signature, funcs)
} else {
methodsByName[f.Name()] = append(methodsByName[f.Name()], f)
}
}
addEdge := func(fnode *callgraph.Node, site ssa.CallInstruction, g *ssa.Function) {
gnode := cg.CreateNode(g)
callgraph.AddEdge(fnode, site, gnode)
}
addEdges := func(fnode *callgraph.Node, site ssa.CallInstruction, callees []*ssa.Function) {
// Because every call to a highly polymorphic and
// frequently used abstract method such as
// (io.Writer).Write is assumed to call every concrete
// Write method in the program, the call graph can
// contain a lot of duplication.
//
// TODO(adonovan): opt: consider factoring the callgraph
// API so that the Callers component of each edge is a
// slice of nodes, not a singleton.
for _, g := range callees {
addEdge(fnode, site, g)
}
}
for f := range allFuncs {
fnode := cg.CreateNode(f)
for _, b := range f.Blocks {
for _, instr := range b.Instrs {
if site, ok := instr.(ssa.CallInstruction); ok {
call := site.Common()
if call.IsInvoke() {
addEdges(fnode, site, lookupMethods(call.Method))
} else if g := call.StaticCallee(); g != nil {
addEdge(fnode, site, g)
} else if _, ok := call.Value.(*ssa.Builtin); !ok {
callees, _ := funcsBySig.At(call.Signature()).([]*ssa.Function)
addEdges(fnode, site, callees)
}
}
}
}
}
return cg
}
func TestStdlib(t *testing.T) {
// Load, parse and type-check the program.
t0 := time.Now()
alloc0 := bytesAllocated()
// Load, parse and type-check the program.
ctxt := build.Default // copy
ctxt.GOPATH = "" // disable GOPATH
conf := loader.Config{Build: &ctxt}
if _, err := conf.FromArgs(buildutil.AllPackages(conf.Build), true); err != nil {
t.Errorf("FromArgs failed: %v", err)
return
}
iprog, err := conf.Load()
if err != nil {
t.Fatalf("Load failed: %v", err)
}
t1 := time.Now()
alloc1 := bytesAllocated()
// Create SSA packages.
var mode ssa.BuilderMode
// Comment out these lines during benchmarking. Approx SSA build costs are noted.
mode |= ssa.SanityCheckFunctions // + 2% space, + 4% time
mode |= ssa.GlobalDebug // +30% space, +18% time
prog := ssa.Create(iprog, mode)
t2 := time.Now()
// Build SSA.
prog.BuildAll()
t3 := time.Now()
alloc3 := bytesAllocated()
numPkgs := len(prog.AllPackages())
if want := 140; numPkgs < want {
t.Errorf("Loaded only %d packages, want at least %d", numPkgs, want)
}
// Keep iprog reachable until after we've measured memory usage.
if len(iprog.AllPackages) == 0 {
print() // unreachable
}
allFuncs := ssautil.AllFunctions(prog)
// Check that all non-synthetic functions have distinct names.
// Synthetic wrappers for exported methods should be distinct too,
// except for unexported ones (explained at (*Function).RelString).
byName := make(map[string]*ssa.Function)
for fn := range allFuncs {
if fn.Synthetic == "" || ast.IsExported(fn.Name()) {
str := fn.String()
prev := byName[str]
byName[str] = fn
if prev != nil {
t.Errorf("%s: duplicate function named %s",
prog.Fset.Position(fn.Pos()), str)
t.Errorf("%s: (previously defined here)",
prog.Fset.Position(prev.Pos()))
}
}
}
// Dump some statistics.
var numInstrs int
for fn := range allFuncs {
for _, b := range fn.Blocks {
numInstrs += len(b.Instrs)
}
}
// determine line count
var lineCount int
prog.Fset.Iterate(func(f *token.File) bool {
lineCount += f.LineCount()
return true
})
// NB: when benchmarking, don't forget to clear the debug +
// sanity builder flags for better performance.
t.Log("GOMAXPROCS: ", runtime.GOMAXPROCS(0))
t.Log("#Source lines: ", lineCount)
t.Log("Load/parse/typecheck: ", t1.Sub(t0))
t.Log("SSA create: ", t2.Sub(t1))
t.Log("SSA build: ", t3.Sub(t2))
// SSA stats:
t.Log("#Packages: ", numPkgs)
t.Log("#Functions: ", len(allFuncs))
t.Log("#Instructions: ", numInstrs)
t.Log("#MB AST+types: ", int64(alloc1-alloc0)/1e6)
t.Log("#MB SSA: ", int64(alloc3-alloc1)/1e6)
}
// TestSyntheticFuncs checks that the expected synthetic functions are
// created, reachable, and not duplicated.
func TestSyntheticFuncs(t *testing.T) {
const input = `package P
type T int
func (T) f() int
func (*T) g() int
var (
// thunks
a = T.f
b = T.f
c = (struct{T}).f
d = (struct{T}).f
e = (*T).g
f = (*T).g
g = (struct{*T}).g
h = (struct{*T}).g
// bounds
i = T(0).f
j = T(0).f
k = new(T).g
l = new(T).g
// wrappers
m interface{} = struct{T}{}
n interface{} = struct{T}{}
o interface{} = struct{*T}{}
p interface{} = struct{*T}{}
q interface{} = new(struct{T})
r interface{} = new(struct{T})
s interface{} = new(struct{*T})
t interface{} = new(struct{*T})
)
`
// Parse
var conf loader.Config
f, err := conf.ParseFile("<input>", input)
if err != nil {
t.Fatalf("parse: %v", err)
}
conf.CreateFromFiles(f.Name.Name, f)
// Load
iprog, err := conf.Load()
if err != nil {
t.Fatalf("Load: %v", err)
}
// Create and build SSA
prog := ssa.Create(iprog, 0)
prog.BuildAll()
// Enumerate reachable synthetic functions
want := map[string]string{
"(*P.T).g$bound": "bound method wrapper for func (*P.T).g() int",
"(P.T).f$bound": "bound method wrapper for func (P.T).f() int",
"(*P.T).g$thunk": "thunk for func (*P.T).g() int",
"(P.T).f$thunk": "thunk for func (P.T).f() int",
"(struct{*P.T}).g$thunk": "thunk for func (*P.T).g() int",
"(struct{P.T}).f$thunk": "thunk for func (P.T).f() int",
"(*P.T).f": "wrapper for func (P.T).f() int",
"(*struct{*P.T}).f": "wrapper for func (P.T).f() int",
"(*struct{*P.T}).g": "wrapper for func (*P.T).g() int",
"(*struct{P.T}).f": "wrapper for func (P.T).f() int",
"(*struct{P.T}).g": "wrapper for func (*P.T).g() int",
"(struct{*P.T}).f": "wrapper for func (P.T).f() int",
"(struct{*P.T}).g": "wrapper for func (*P.T).g() int",
"(struct{P.T}).f": "wrapper for func (P.T).f() int",
"P.init": "package initializer",
}
for fn := range ssautil.AllFunctions(prog) {
if fn.Synthetic == "" {
continue
}
name := fn.String()
wantDescr, ok := want[name]
if !ok {
t.Errorf("got unexpected/duplicate func: %q: %q", name, fn.Synthetic)
continue
}
delete(want, name)
if wantDescr != fn.Synthetic {
t.Errorf("(%s).Synthetic = %q, want %q", name, fn.Synthetic, wantDescr)
}
}
for fn, descr := range want {
t.Errorf("want func: %q: %q", fn, descr)
}
}
func TestStdlib(t *testing.T) {
if !*runStdlibTest {
t.Skip("skipping (slow) stdlib test (use --stdlib)")
}
// Load, parse and type-check the program.
ctxt := build.Default // copy
ctxt.GOPATH = "" // disable GOPATH
conf := loader.Config{Build: &ctxt}
if _, err := conf.FromArgs(buildutil.AllPackages(conf.Build), true); err != nil {
t.Errorf("FromArgs failed: %v", err)
return
}
iprog, err := conf.Load()
if err != nil {
t.Fatalf("Load failed: %v", err)
}
// Create SSA packages.
prog := ssa.Create(iprog, 0)
prog.BuildAll()
numPkgs := len(prog.AllPackages())
if want := 240; numPkgs < want {
t.Errorf("Loaded only %d packages, want at least %d", numPkgs, want)
}
// Determine the set of packages/tests to analyze.
var testPkgs []*ssa.Package
for _, info := range iprog.InitialPackages() {
testPkgs = append(testPkgs, prog.Package(info.Pkg))
}
testmain := prog.CreateTestMainPackage(testPkgs...)
if testmain == nil {
t.Fatal("analysis scope has tests")
}
// Run the analysis.
config := &Config{
Reflection: false, // TODO(adonovan): fix remaining bug in rVCallConstraint, then enable.
BuildCallGraph: true,
Mains: []*ssa.Package{testmain},
}
// TODO(adonovan): add some query values (affects track bits).
t0 := time.Now()
result, err := Analyze(config)
if err != nil {
t.Fatal(err) // internal error in pointer analysis
}
_ = result // TODO(adonovan): measure something
t1 := time.Now()
// Dump some statistics.
allFuncs := ssautil.AllFunctions(prog)
var numInstrs int
for fn := range allFuncs {
for _, b := range fn.Blocks {
numInstrs += len(b.Instrs)
}
}
// determine line count
var lineCount int
prog.Fset.Iterate(func(f *token.File) bool {
lineCount += f.LineCount()
return true
})
t.Log("#Source lines: ", lineCount)
t.Log("#Instructions: ", numInstrs)
t.Log("Pointer analysis: ", t1.Sub(t0))
}