[dev.typeparams] cmd/compile: refactor export writing

This CL reorganizes export writing in preparation for unified IR:

1. It moves dumpexport into noder as noder.WriteExports so that it can
be extended to include unified IR's export data.

2. Adds an "extensions" flag to typecheck.WriteExports to control
whether the compiler-only extension data (e.g., function bodies and
linker symbol info) is included in the exports.

3. It moves the gc.exporter type into typecheck and renames it to
"crawler". The type originated as the implementation of
the (pre-iexport) binary exporter, but since the removal of bexport
it's been relegated to simply crawling the exported functions/bodies
graph to identify which inline bodies need to be included.

4. It changes inline.Inline_Flood into the method crawler.markInlBody.
Inline_Flood doesn't actually have anything to do with the rest of
inlining; its current name and location are just historical quirks.

Passes toolstash -cmp.

Change-Id: I6445e2de9d3ce500a3aded5a8e20b09f46d23dbc
Reviewed-on: https://go-review.googlesource.com/c/go/+/325212
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>

6 files changed, 224 insertions, 183 deletions

diff --git a/src/cmd/compile/internal/gc/export.go b/src/cmd/compile/internal/gc/export.go
index a11e5fdd30..9bf3c7240a 100644
--- a/src/cmd/compile/internal/gc/export.go
+++ b/src/cmd/compile/internal/gc/export.go
@@ -5,41 +5,16 @@
 package gc
 
 import (
+	"fmt"
+	"go/constant"
+
 	"cmd/compile/internal/base"
-	"cmd/compile/internal/inline"
 	"cmd/compile/internal/ir"
 	"cmd/compile/internal/typecheck"
 	"cmd/compile/internal/types"
 	"cmd/internal/bio"
-	"fmt"
-	"go/constant"
 )
 
-func exportf(bout *bio.Writer, format string, args ...interface{}) {
-	fmt.Fprintf(bout, format, args...)
-	if base.Debug.Export != 0 {
-		fmt.Printf(format, args...)
-	}
-}
-
-func dumpexport(bout *bio.Writer) {
-	p := &exporter{marked: make(map[*types.Type]bool)}
-	for _, n := range typecheck.Target.Exports {
-		p.markObject(n)
-	}
-
-	// The linker also looks for the $$ marker - use char after $$ to distinguish format.
-	exportf(bout, "\n$$B\n") // indicate binary export format
-	off := bout.Offset()
-	typecheck.WriteExports(bout.Writer)
-	size := bout.Offset() - off
-	exportf(bout, "\n$$\n")
-
-	if base.Debug.Export != 0 {
-		fmt.Printf("BenchmarkExportSize:%s 1 %d bytes\n", base.Ctxt.Pkgpath, size)
-	}
-}
-
 func dumpasmhdr() {
 	b, err := bio.Create(base.Flag.AsmHdr)
 	if err != nil {
@@ -74,92 +49,3 @@ func dumpasmhdr() {
 
 	b.Close()
 }
-
-type exporter struct {
-	marked map[*types.Type]bool // types already seen by markType
-}
-
-// markObject visits a reachable object.
-func (p *exporter) markObject(n ir.Node) {
-	if n.Op() == ir.ONAME {
-		n := n.(*ir.Name)
-		if n.Class == ir.PFUNC {
-			inline.Inline_Flood(n, typecheck.Export)
-		}
-	}
-
-	p.markType(n.Type())
-}
-
-// markType recursively visits types reachable from t to identify
-// functions whose inline bodies may be needed.
-func (p *exporter) markType(t *types.Type) {
-	if t.IsInstantiatedGeneric() {
-		// Re-instantiated types don't add anything new, so don't follow them.
-		return
-	}
-	if p.marked[t] {
-		return
-	}
-	p.marked[t] = true
-
-	// If this is a named type, mark all of its associated
-	// methods. Skip interface types because t.Methods contains
-	// only their unexpanded method set (i.e., exclusive of
-	// interface embeddings), and the switch statement below
-	// handles their full method set.
-	if t.Sym() != nil && t.Kind() != types.TINTER {
-		for _, m := range t.Methods().Slice() {
-			if types.IsExported(m.Sym.Name) {
-				p.markObject(ir.AsNode(m.Nname))
-			}
-		}
-	}
-
-	// Recursively mark any types that can be produced given a
-	// value of type t: dereferencing a pointer; indexing or
-	// iterating over an array, slice, or map; receiving from a
-	// channel; accessing a struct field or interface method; or
-	// calling a function.
-	//
-	// Notably, we don't mark function parameter types, because
-	// the user already needs some way to construct values of
-	// those types.
-	switch t.Kind() {
-	case types.TPTR, types.TARRAY, types.TSLICE:
-		p.markType(t.Elem())
-
-	case types.TCHAN:
-		if t.ChanDir().CanRecv() {
-			p.markType(t.Elem())
-		}
-
-	case types.TMAP:
-		p.markType(t.Key())
-		p.markType(t.Elem())
-
-	case types.TSTRUCT:
-		for _, f := range t.FieldSlice() {
-			if types.IsExported(f.Sym.Name) || f.Embedded != 0 {
-				p.markType(f.Type)
-			}
-		}
-
-	case types.TFUNC:
-		for _, f := range t.Results().FieldSlice() {
-			p.markType(f.Type)
-		}
-
-	case types.TINTER:
-		// TODO(danscales) - will have to deal with the types in interface
-		// elements here when implemented in types2 and represented in types1.
-		for _, f := range t.AllMethods().Slice() {
-			if types.IsExported(f.Sym.Name) {
-				p.markType(f.Type)
-			}
-		}
-
-	case types.TTYPEPARAM:
-		// No other type that needs to be followed.
-	}
-}
diff --git a/src/cmd/compile/internal/gc/obj.go b/src/cmd/compile/internal/gc/obj.go
index 0b10cb8a9e..a52696fbb6 100644
--- a/src/cmd/compile/internal/gc/obj.go
+++ b/src/cmd/compile/internal/gc/obj.go
@@ -7,6 +7,7 @@ package gc
 import (
 	"cmd/compile/internal/base"
 	"cmd/compile/internal/ir"
+	"cmd/compile/internal/noder"
 	"cmd/compile/internal/objw"
 	"cmd/compile/internal/reflectdata"
 	"cmd/compile/internal/staticdata"
@@ -103,7 +104,7 @@ func finishArchiveEntry(bout *bio.Writer, start int64, name string) {
 
 func dumpCompilerObj(bout *bio.Writer) {
 	printObjHeader(bout)
-	dumpexport(bout)
+	noder.WriteExports(bout)
 }
 
 func dumpdata() {
diff --git a/src/cmd/compile/internal/inline/inl.go b/src/cmd/compile/internal/inline/inl.go
index 594f280f03..e12a30f936 100644
--- a/src/cmd/compile/internal/inline/inl.go
+++ b/src/cmd/compile/internal/inline/inl.go
@@ -225,62 +225,6 @@ func canDelayResults(fn *ir.Func) bool {
 	return true
 }
 
-// Inline_Flood marks n's inline body for export and recursively ensures
-// all called functions are marked too.
-func Inline_Flood(n *ir.Name, exportsym func(*ir.Name)) {
-	if n == nil {
-		return
-	}
-	if n.Op() != ir.ONAME || n.Class != ir.PFUNC {
-		base.Fatalf("Inline_Flood: unexpected %v, %v, %v", n, n.Op(), n.Class)
-	}
-	fn := n.Func
-	if fn == nil {
-		base.Fatalf("Inline_Flood: missing Func on %v", n)
-	}
-	if fn.Inl == nil {
-		return
-	}
-
-	if fn.ExportInline() {
-		return
-	}
-	fn.SetExportInline(true)
-
-	typecheck.ImportedBody(fn)
-
-	var doFlood func(n ir.Node)
-	doFlood = func(n ir.Node) {
-		switch n.Op() {
-		case ir.OMETHEXPR, ir.ODOTMETH:
-			Inline_Flood(ir.MethodExprName(n), exportsym)
-
-		case ir.ONAME:
-			n := n.(*ir.Name)
-			switch n.Class {
-			case ir.PFUNC:
-				Inline_Flood(n, exportsym)
-				exportsym(n)
-			case ir.PEXTERN:
-				exportsym(n)
-			}
-
-		case ir.OCALLPART:
-			// Okay, because we don't yet inline indirect
-			// calls to method values.
-		case ir.OCLOSURE:
-			// VisitList doesn't visit closure bodies, so force a
-			// recursive call to VisitList on the body of the closure.
-			ir.VisitList(n.(*ir.ClosureExpr).Func.Body, doFlood)
-		}
-	}
-
-	// Recursively identify all referenced functions for
-	// reexport. We want to include even non-called functions,
-	// because after inlining they might be callable.
-	ir.VisitList(ir.Nodes(fn.Inl.Body), doFlood)
-}
-
 // hairyVisitor visits a function body to determine its inlining
 // hairiness and whether or not it can be inlined.
 type hairyVisitor struct {
diff --git a/src/cmd/compile/internal/noder/export.go b/src/cmd/compile/internal/noder/export.go
new file mode 100644
index 0000000000..9fb3b4da10
--- /dev/null
+++ b/src/cmd/compile/internal/noder/export.go
@@ -0,0 +1,26 @@
+// Copyright 2021 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 noder
+
+import (
+	"fmt"
+
+	"cmd/compile/internal/base"
+	"cmd/compile/internal/typecheck"
+	"cmd/internal/bio"
+)
+
+func WriteExports(out *bio.Writer) {
+	// The linker also looks for the $$ marker - use char after $$ to distinguish format.
+	out.WriteString("\n$$B\n") // indicate binary export format
+	off := out.Offset()
+	typecheck.WriteExports(out, true)
+	size := out.Offset() - off
+	out.WriteString("\n$$\n")
+
+	if base.Debug.Export != 0 {
+		fmt.Printf("BenchmarkExportSize:%s 1 %d bytes\n", base.Ctxt.Pkgpath, size)
+	}
+}
diff --git a/src/cmd/compile/internal/typecheck/crawler.go b/src/cmd/compile/internal/typecheck/crawler.go
new file mode 100644
index 0000000000..48fc61dbfd
--- /dev/null
+++ b/src/cmd/compile/internal/typecheck/crawler.go
@@ -0,0 +1,164 @@
+// Copyright 2021 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 typecheck
+
+import (
+	"cmd/compile/internal/base"
+	"cmd/compile/internal/ir"
+	"cmd/compile/internal/types"
+)
+
+// crawlExports crawls the type/object graph rooted at the given list
+// of exported objects. Any functions that are found to be potentially
+// callable by importers are marked with ExportInline so that
+// iexport.go knows to re-export their inline body.
+func crawlExports(exports []*ir.Name) {
+	p := crawler{marked: make(map[*types.Type]bool)}
+	for _, n := range exports {
+		p.markObject(n)
+	}
+}
+
+type crawler struct {
+	marked map[*types.Type]bool // types already seen by markType
+}
+
+// markObject visits a reachable object.
+func (p *crawler) markObject(n *ir.Name) {
+	if n.Op() == ir.ONAME && n.Class == ir.PFUNC {
+		p.markInlBody(n)
+	}
+
+	p.markType(n.Type())
+}
+
+// markType recursively visits types reachable from t to identify
+// functions whose inline bodies may be needed.
+func (p *crawler) markType(t *types.Type) {
+	if t.IsInstantiatedGeneric() {
+		// Re-instantiated types don't add anything new, so don't follow them.
+		return
+	}
+	if p.marked[t] {
+		return
+	}
+	p.marked[t] = true
+
+	// If this is a named type, mark all of its associated
+	// methods. Skip interface types because t.Methods contains
+	// only their unexpanded method set (i.e., exclusive of
+	// interface embeddings), and the switch statement below
+	// handles their full method set.
+	if t.Sym() != nil && t.Kind() != types.TINTER {
+		for _, m := range t.Methods().Slice() {
+			if types.IsExported(m.Sym.Name) {
+				p.markObject(m.Nname.(*ir.Name))
+			}
+		}
+	}
+
+	// Recursively mark any types that can be produced given a
+	// value of type t: dereferencing a pointer; indexing or
+	// iterating over an array, slice, or map; receiving from a
+	// channel; accessing a struct field or interface method; or
+	// calling a function.
+	//
+	// Notably, we don't mark function parameter types, because
+	// the user already needs some way to construct values of
+	// those types.
+	switch t.Kind() {
+	case types.TPTR, types.TARRAY, types.TSLICE:
+		p.markType(t.Elem())
+
+	case types.TCHAN:
+		if t.ChanDir().CanRecv() {
+			p.markType(t.Elem())
+		}
+
+	case types.TMAP:
+		p.markType(t.Key())
+		p.markType(t.Elem())
+
+	case types.TSTRUCT:
+		for _, f := range t.FieldSlice() {
+			if types.IsExported(f.Sym.Name) || f.Embedded != 0 {
+				p.markType(f.Type)
+			}
+		}
+
+	case types.TFUNC:
+		for _, f := range t.Results().FieldSlice() {
+			p.markType(f.Type)
+		}
+
+	case types.TINTER:
+		// TODO(danscales) - will have to deal with the types in interface
+		// elements here when implemented in types2 and represented in types1.
+		for _, f := range t.AllMethods().Slice() {
+			if types.IsExported(f.Sym.Name) {
+				p.markType(f.Type)
+			}
+		}
+
+	case types.TTYPEPARAM:
+		// No other type that needs to be followed.
+	}
+}
+
+// markInlBody marks n's inline body for export and recursively
+// ensures all called functions are marked too.
+func (p *crawler) markInlBody(n *ir.Name) {
+	if n == nil {
+		return
+	}
+	if n.Op() != ir.ONAME || n.Class != ir.PFUNC {
+		base.Fatalf("markInlBody: unexpected %v, %v, %v", n, n.Op(), n.Class)
+	}
+	fn := n.Func
+	if fn == nil {
+		base.Fatalf("markInlBody: missing Func on %v", n)
+	}
+	if fn.Inl == nil {
+		return
+	}
+
+	if fn.ExportInline() {
+		return
+	}
+	fn.SetExportInline(true)
+
+	ImportedBody(fn)
+
+	var doFlood func(n ir.Node)
+	doFlood = func(n ir.Node) {
+		switch n.Op() {
+		case ir.OMETHEXPR, ir.ODOTMETH:
+			p.markInlBody(ir.MethodExprName(n))
+
+		case ir.ONAME:
+			n := n.(*ir.Name)
+			switch n.Class {
+			case ir.PFUNC:
+				p.markInlBody(n)
+				Export(n)
+			case ir.PEXTERN:
+				Export(n)
+			}
+
+		case ir.OCALLPART:
+			// Okay, because we don't yet inline indirect
+			// calls to method values.
+		case ir.OCLOSURE:
+			// VisitList doesn't visit closure bodies, so force a
+			// recursive call to VisitList on the body of the closure.
+			ir.VisitList(n.(*ir.ClosureExpr).Func.Body, doFlood)
+		}
+	}
+
+	// Recursively identify all referenced functions for
+	// reexport. We want to include even non-called functions,
+	// because after inlining they might be callable.
+	ir.VisitList(fn.Inl.Body, doFlood)
+}
diff --git a/src/cmd/compile/internal/typecheck/iexport.go b/src/cmd/compile/internal/typecheck/iexport.go
index 3bfbea11c0..6987bc9918 100644
--- a/src/cmd/compile/internal/typecheck/iexport.go
+++ b/src/cmd/compile/internal/typecheck/iexport.go
@@ -204,7 +204,6 @@
 package typecheck
 
 import (
-	"bufio"
 	"bytes"
 	"crypto/md5"
 	"encoding/binary"
@@ -264,13 +263,22 @@ const (
 	magic = 0x6742937dc293105
 )
 
-func WriteExports(out *bufio.Writer) {
+// WriteExports writes the indexed export format to out. If extensions
+// is true, then the compiler-only extensions are included.
+func WriteExports(out io.Writer, extensions bool) {
+	if extensions {
+		// If we're exporting inline bodies, invoke the crawler to mark
+		// which bodies to include.
+		crawlExports(Target.Exports)
+	}
+
 	p := iexporter{
 		allPkgs:     map[*types.Pkg]bool{},
 		stringIndex: map[string]uint64{},
 		declIndex:   map[*types.Sym]uint64{},
 		inlineIndex: map[*types.Sym]uint64{},
 		typIndex:    map[*types.Type]uint64{},
+		extensions:  extensions,
 	}
 
 	for i, pt := range predeclared() {
@@ -397,6 +405,8 @@ type iexporter struct {
 	declIndex   map[*types.Sym]uint64
 	inlineIndex map[*types.Sym]uint64
 	typIndex    map[*types.Type]uint64
+
+	extensions bool
 }
 
 // stringOff returns the offset of s within the string section.
@@ -467,7 +477,9 @@ func (p *iexporter) doDecl(n *ir.Name) {
 			w.tag('V')
 			w.pos(n.Pos())
 			w.typ(n.Type())
-			w.varExt(n)
+			if w.p.extensions {
+				w.varExt(n)
+			}
 
 		case ir.PFUNC:
 			if ir.IsMethod(n) {
@@ -487,7 +499,9 @@ func (p *iexporter) doDecl(n *ir.Name) {
 				w.tparamList(n.Type().TParams().FieldSlice())
 			}
 			w.signature(n.Type())
-			w.funcExt(n)
+			if w.p.extensions {
+				w.funcExt(n)
+			}
 
 		default:
 			base.Fatalf("unexpected class: %v, %v", n, n.Class)
@@ -503,7 +517,9 @@ func (p *iexporter) doDecl(n *ir.Name) {
 		w.tag('C')
 		w.pos(n.Pos())
 		w.value(n.Type(), n.Val())
-		w.constExt(n)
+		if w.p.extensions {
+			w.constExt(n)
+		}
 
 	case ir.OTYPE:
 		if n.Type().Kind() == types.TTYPEPARAM && n.Type().Underlying() == n.Type() {
@@ -551,7 +567,9 @@ func (p *iexporter) doDecl(n *ir.Name) {
 
 		t := n.Type()
 		if t.IsInterface() {
-			w.typeExt(t)
+			if w.p.extensions {
+				w.typeExt(t)
+			}
 			break
 		}
 
@@ -567,9 +585,11 @@ func (p *iexporter) doDecl(n *ir.Name) {
 			w.signature(m.Type)
 		}
 
-		w.typeExt(t)
-		for _, m := range methods {
-			w.methExt(m)
+		if w.p.extensions {
+			w.typeExt(t)
+			for _, m := range methods {
+				w.methExt(m)
+			}
 		}
 
 	default: