cmd/compile/internal/gc: compact binary export format

The binary import/export format is significantly more
compact than the existing textual format. It should
also be faster to read and write (to be measured).

Use -newexport to enable, for instance:
export GO_GCFLAGS=-newexport; make.bash

The compiler can import packages using both the old
and the new format ("mixed mode").

Missing: export info for inlined functions bodies
(performance issue, does not affect correctness).

Disabled by default until we have inlined function
bodies and confirmation of no regression and equality
of binaries.

For golang#6110.
For golang#1909.

This change depends on:

   https://go-review.googlesource.com/16220
   https://go-review.googlesource.com/16222

(already submitted) for all.bash to work.

Some initial export data sizes for std lib packages. This data
is without exported functions with inlineable function bodies.

Package                                       old      new    new/old

archive/tar.................................13875.....3883    28%
archive/zip.................................19464.....5046    26%
bufio....................................... 7733.....2222    29%
bytes.......................................10342.....3347    32%
cmd/addr2line.................................242.......26    11%
cmd/api.....................................39305....10368    26%
cmd/asm/internal/arch.......................27732.....7939    29%
cmd/asm/internal/asm........................35264....10295    29%
cmd/asm/internal/flags........................629......178    28%
cmd/asm/internal/lex........................39248....11128    28%
cmd/asm.......................................306.......26     8%
cmd/cgo.....................................40197....10570    26%
cmd/compile/internal/amd64...................1106......214    19%
cmd/compile/internal/arm....................27891.....7710    28%
cmd/compile/internal/arm64....................891......153    17%
cmd/compile/internal/big....................21637.....8336    39%
cmd/compile/internal/gc....................109845....29727    27%
cmd/compile/internal/mips64...................972......168    17%
cmd/compile/internal/ppc64....................972......168    17%
cmd/compile/internal/x86.....................1104......195    18%
cmd/compile...................................329.......26     8%
cmd/cover...................................12986.....3749    29%
cmd/dist......................................477.......67    14%
cmd/doc.....................................23043.....6793    29%
cmd/expdump...................................167.......26    16%
cmd/fix......................................1190......208    17%
cmd/go......................................26399.....5629    21%
cmd/gofmt.....................................499.......26     5%
cmd/internal/gcprog..........................1342......490    37%
cmd/internal/goobj...........................2690......980    36%
cmd/internal/obj/arm........................32740....10057    31%
cmd/internal/obj/arm64......................46542....15364    33%
cmd/internal/obj/mips.......................42140....13731    33%
cmd/internal/obj/ppc64......................42140....13731    33%
cmd/internal/obj/x86........................52732....19015    36%
cmd/internal/obj............................36729....11690    32%
cmd/internal/objfile........................36365....10287    28%
cmd/link/internal/amd64.....................45893....12220    27%
cmd/link/internal/arm.........................307.......96    31%
cmd/link/internal/arm64.......................345.......98    28%
cmd/link/internal/ld.......................109300....46326    42%
cmd/link/internal/ppc64.......................344.......99    29%
cmd/link/internal/x86.........................334......107    32%
cmd/link......................................314.......26     8%
cmd/newlink..................................8110.....2544    31%
cmd/nm........................................210.......26    12%
cmd/objdump...................................244.......26    11%
cmd/pack....................................14248.....4066    29%
cmd/pprof/internal/commands..................5239.....1285    25%
cmd/pprof/internal/driver...................37967.....8860    23%
cmd/pprof/internal/fetch....................30962.....7337    24%
cmd/pprof/internal/plugin...................47734.....7719    16%
cmd/pprof/internal/profile..................22286.....6922    31%
cmd/pprof/internal/report...................31187.....7838    25%
cmd/pprof/internal/svg.......................4315......965    22%
cmd/pprof/internal/symbolizer...............30051.....7397    25%
cmd/pprof/internal/symbolz..................28545.....6949    24%
cmd/pprof/internal/tempfile.................12550.....3356    27%
cmd/pprof.....................................563.......26     5%
cmd/trace....................................1455......636    44%
cmd/vendor/golang.org/x/arch/arm/armasm....168035....64737    39%
cmd/vendor/golang.org/x/arch/x86/x86asm.....26871.....8578    32%
cmd/vet.....................................38980.....9913    25%
cmd/vet/whitelist.............................102.......49    48%
cmd/yacc.....................................2518......926    37%
compress/bzip2...............................6326......129     2%
compress/flate...............................7069.....2541    36%
compress/gzip...............................20143.....5069    25%
compress/lzw..................................828......295    36%
compress/zlib...............................10676.....2692    25%
container/heap................................523......181    35%
container/list...............................3517......740    21%
container/ring................................881......229    26%
crypto/aes....................................550......187    34%
crypto/cipher................................1966......825    42%
crypto.......................................1836......646    35%
crypto/des....................................632......235    37%
crypto/dsa..................................18718.....5035    27%
crypto/ecdsa................................23131.....6097    26%
crypto/elliptic.............................20790.....5740    28%
crypto/hmac...................................455......186    41%
crypto/md5...................................1375......171    12%
crypto/rand.................................18132.....4748    26%
crypto/rc4....................................561......240    43%
crypto/rsa..................................22094.....6380    29%
crypto/sha1..................................1416......172    12%
crypto/sha256.................................551......238    43%
crypto/sha512.................................839......378    45%
crypto/subtle................................1153......250    22%
crypto/tls..................................58203....17984    31%
crypto/x509/pkix............................29447.....8161    28%
database/sql/driver..........................3318.....1096    33%
database/sql................................11258.....3942    35%
debug/dwarf.................................18416.....7006    38%
debug/elf...................................57530....21014    37%
debug/gosym..................................4992.....2058    41%
debug/macho.................................23037.....6538    28%
debug/pe....................................21063.....6619    31%
debug/plan9obj...............................2467......802    33%
encoding/ascii85.............................1523......360    24%
encoding/asn1................................1718......527    31%
encoding/base32..............................2642......686    26%
encoding/base64..............................3077......800    26%
encoding/binary..............................4727.....1040    22%
encoding/csv................................12223.....2850    23%
encoding......................................383......217    57%
encoding/gob................................37563....10113    27%
encoding/hex.................................1327......390    29%
encoding/json...............................30897.....7804    25%
encoding/pem..................................595......200    34%
encoding/xml................................37798.....9336    25%
errors........................................274.......36    13%
expvar.......................................3155.....1021    32%
flag........................................19860.....2849    14%
fmt..........................................3137.....1263    40%
go/ast......................................44729....13422    30%
go/build....................................16336.....4657    29%
go/constant..................................3703......846    23%
go/doc.......................................9877.....2807    28%
go/format....................................5472.....1575    29%
go/importer..................................4980.....1301    26%
go/internal/gccgoimporter....................5587.....1525    27%
go/internal/gcimporter.......................8979.....2186    24%
go/parser...................................20692.....5304    26%
go/printer...................................7015.....2029    29%
go/scanner...................................9719.....2824    29%
go/token.....................................7933.....2465    31%
go/types....................................64569....19978    31%
hash/adler32.................................1176......176    15%
hash/crc32...................................1663......360    22%
hash/crc64...................................1587......306    19%
hash/fnv.....................................3964......260     7%
hash..........................................591......278    47%
html..........................................217.......74    34%
html/template...............................69623....12588    18%
image/color/palette...........................315.......98    31%
image/color..................................5565.....1036    19%
image/draw...................................6917.....1028    15%
image/gif....................................8894.....1654    19%
image/internal/imageutil.....................9112.....1476    16%
image/jpeg...................................6647.....1026    15%
image/png....................................6906.....1069    15%
image.......................................28992.....6139    21%
index/suffixarray...........................17106.....4773    28%
internal/singleflight........................1614......506    31%
internal/testenv............................12212.....3152    26%
internal/trace...............................2762.....1323    48%
io/ioutil...................................13502.....3682    27%
io...........................................6765.....2482    37%
log.........................................11620.....3317    29%
log/syslog..................................13516.....3821    28%
math/big....................................21819.....8320    38%
math/cmplx...................................2816......438    16%
math/rand....................................2317......929    40%
math.........................................7511.....2444    33%
mime/multipart..............................12679.....3360    27%
mime/quotedprintable.........................5458.....1235    23%
mime.........................................6076.....1628    27%
net/http/cgi................................59796....17173    29%
net/http/cookiejar..........................14781.....3739    25%
net/http/fcgi...............................57861....16426    28%
net/http/httptest...........................84100....24365    29%
net/http/httputil...........................67763....18869    28%
net/http/internal............................6907......637     9%
net/http/pprof..............................57945....16316    28%
net/http....................................95391....30210    32%
net/internal/socktest........................4555.....1453    32%
net/mail....................................14481.....3608    25%
net/rpc/jsonrpc.............................33335......988     3%
net/rpc.....................................79950....23106    29%
net/smtp....................................57790....16468    28%
net/textproto...............................11356.....3248    29%
net/url......................................3123.....1009    32%
os/exec.....................................20738.....5769    28%
os/signal.....................................437......167    38%
os..........................................24875.....6668    27%
path/filepath...............................11340.....2826    25%
path..........................................778......285    37%
reflect.....................................15469.....5198    34%
regexp......................................13627.....4661    34%
regexp/syntax................................5539.....2249    41%
runtime/debug................................9275.....2322    25%
runtime/pprof................................1355......477    35%
runtime/race...................................39.......17    44%
runtime/trace.................................228.......92    40%
runtime.....................................13498.....1821    13%
sort.........................................2848......842    30%
strconv......................................2947.....1252    42%
strings......................................7983.....2456    31%
sync/atomic..................................2666.....1149    43%
sync.........................................2568......845    33%
syscall.....................................81252....38398    47%
testing/iotest...............................2444......302    12%
testing/quick...............................18890.....5076    27%
testing.....................................16502.....4800    29%
text/scanner.................................6849.....2052    30%
text/tabwriter...............................6607.....1863    28%
text/template/parse.........................22978.....6183    27%
text/template...............................64153....11518    18%
time........................................12103.....3546    29%
unicode......................................9706.....3320    34%
unicode/utf16................................1055......148    14%
unicode/utf8.................................1118......513    46%
vendor/golang.org/x/net/http2/hpack..........8905.....2636    30%

All packages                              3518505  1017774    29%

Change-Id: Id657334f276383ff1e6fa91472d3d1db5a03349c
Reviewed-on: https://go-review.googlesource.com/13937
Run-TryBot: Robert Griesemer <[email protected]>
Reviewed-by: Chris Manghane <[email protected]>

Author: griesemer

src/cmd/compile/internal/gc/bexport.go

@@ -874,11 +874,18 @@ func checkdupfields(t *Type, what string) {
  * a type for struct/interface/arglist
  */
 func tostruct(l *NodeList) *Type {
-	var f *Type
 	t := typ(TSTRUCT)
+	tostruct0(t, l)
+	return t
+}
+
+func tostruct0(t *Type, l *NodeList) {
+	if t == nil || t.Etype != TSTRUCT {
+		Fatalf("struct expected")
+	}
 
 	for tp := &t.Type; l != nil; l = l.Next {
-		f = structfield(l.N)
+		f := structfield(l.N)
 
 		*tp = f
 		tp = &f.Down
@@ -896,8 +903,6 @@ func tostruct(l *NodeList) *Type {
 	if !t.Broke {
 		checkwidth(t)
 	}
-
-	return t
 }
 
 func tofunargs(l *NodeList) *Type {
@@ -996,18 +1001,23 @@ func interfacefield(n *Node) *Type {
 }
 
 func tointerface(l *NodeList) *Type {
-	var f *Type
-	var t1 *Type
-
 	t := typ(TINTER)
+	tointerface0(t, l)
+	return t
+}
+
+func tointerface0(t *Type, l *NodeList) *Type {
+	if t == nil || t.Etype != TINTER {
+		Fatalf("interface expected")
+	}
 
 	tp := &t.Type
 	for ; l != nil; l = l.Next {
-		f = interfacefield(l.N)
+		f := interfacefield(l.N)
 
 		if l.N.Left == nil && f.Type.Etype == TINTER {
 			// embedded interface, inline methods
-			for t1 = f.Type.Type; t1 != nil; t1 = t1.Down {
+			for t1 := f.Type.Type; t1 != nil; t1 = t1.Down {
 				f = typ(TFIELD)
 				f.Type = t1.Type
 				f.Broke = t1.Broke
@@ -1200,6 +1210,14 @@ func isifacemethod(f *Type) bool {
  */
 func functype(this *Node, in *NodeList, out *NodeList) *Type {
 	t := typ(TFUNC)
+	functype0(t, this, in, out)
+	return t
+}
+
+func functype0(t *Type, this *Node, in *NodeList, out *NodeList) {
+	if t == nil || t.Etype != TFUNC {
+		Fatalf("function type expected")
+	}
 
 	var rcvr *NodeList
 	if this != nil {
@@ -1230,8 +1248,6 @@ func functype(this *Node, in *NodeList, out *NodeList) *Type {
 			t.Outnamed = true
 		}
 	}
-
-	return t
 }
 
 var methodsym_toppkg *Pkg

src/cmd/compile/internal/gc/bimport.go

@@ -5,13 +5,28 @@
 package gc
 
 import (
+	"bytes"
 	"cmd/internal/obj"
 	"fmt"
 	"sort"
 	"unicode"
 	"unicode/utf8"
 )
 
+var (
+	newexport    int // if set, use new export format
+	Debug_export int // if set, print debugging information about export data
+	exportsize   int
+)
+
+func exportf(format string, args ...interface{}) {
+	n, _ := fmt.Fprintf(bout, format, args...)
+	exportsize += n
+	if Debug_export != 0 {
+		fmt.Printf(format, args...)
+	}
+}
+
 var asmlist *NodeList
 
 // Mark n's symbol as exported
@@ -35,8 +50,8 @@ func exportsym(n *Node) {
 }
 
 func exportname(s string) bool {
-	if s[0] < utf8.RuneSelf {
-		return 'A' <= s[0] && s[0] <= 'Z'
+	if r := s[0]; r < utf8.RuneSelf {
+		return 'A' <= r && r <= 'Z'
 	}
 	r, _ := utf8.DecodeRuneInString(s)
 	return unicode.IsUpper(r)
@@ -87,7 +102,7 @@ func dumppkg(p *Pkg) {
 	if !p.Direct {
 		suffix = " // indirect"
 	}
-	fmt.Fprintf(bout, "\timport %s %q%s\n", p.Name, p.Path, suffix)
+	exportf("\timport %s %q%s\n", p.Name, p.Path, suffix)
 }
 
 // Look for anything we need for the inline body
@@ -128,7 +143,7 @@ func reexportdep(n *Node) {
 			}
 		}
 
-		// Local variables in the bodies need their type.
+	// Local variables in the bodies need their type.
 	case ODCL:
 		t := n.Left.Type
 
@@ -167,7 +182,7 @@ func reexportdep(n *Node) {
 			exportlist = append(exportlist, n)
 		}
 
-		// for operations that need a type when rendered, put the type on the export list.
+	// for operations that need a type when rendered, put the type on the export list.
 	case OCONV,
 		OCONVIFACE,
 		OCONVNOP,
@@ -216,9 +231,9 @@ func dumpexportconst(s *Sym) {
 	dumpexporttype(t)
 
 	if t != nil && !isideal(t) {
-		fmt.Fprintf(bout, "\tconst %v %v = %v\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtSharp), Vconv(n.Val(), obj.FmtSharp))
+		exportf("\tconst %v %v = %v\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtSharp), Vconv(n.Val(), obj.FmtSharp))
 	} else {
-		fmt.Fprintf(bout, "\tconst %v = %v\n", Sconv(s, obj.FmtSharp), Vconv(n.Val(), obj.FmtSharp))
+		exportf("\tconst %v = %v\n", Sconv(s, obj.FmtSharp), Vconv(n.Val(), obj.FmtSharp))
 	}
 }
 
@@ -242,14 +257,14 @@ func dumpexportvar(s *Sym) {
 			}
 
 			// NOTE: The space after %#S here is necessary for ld's export data parser.
-			fmt.Fprintf(bout, "\tfunc %v %v { %v }\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtShort|obj.FmtSharp), Hconv(n.Func.Inl, obj.FmtSharp))
+			exportf("\tfunc %v %v { %v }\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtShort|obj.FmtSharp), Hconv(n.Func.Inl, obj.FmtSharp))
 
 			reexportdeplist(n.Func.Inl)
 		} else {
-			fmt.Fprintf(bout, "\tfunc %v %v\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtShort|obj.FmtSharp))
+			exportf("\tfunc %v %v\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtShort|obj.FmtSharp))
 		}
 	} else {
-		fmt.Fprintf(bout, "\tvar %v %v\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtSharp))
+		exportf("\tvar %v %v\n", Sconv(s, obj.FmtSharp), Tconv(t, obj.FmtSharp))
 	}
 }
 
@@ -287,10 +302,10 @@ func dumpexporttype(t *Type) {
 	}
 	sort.Sort(methodbyname(m))
 
-	fmt.Fprintf(bout, "\ttype %v %v\n", Sconv(t.Sym, obj.FmtSharp), Tconv(t, obj.FmtSharp|obj.FmtLong))
+	exportf("\ttype %v %v\n", Sconv(t.Sym, obj.FmtSharp), Tconv(t, obj.FmtSharp|obj.FmtLong))
 	for _, f := range m {
 		if f.Nointerface {
-			fmt.Fprintf(bout, "\t//go:nointerface\n")
+			exportf("\t//go:nointerface\n")
 		}
 		if f.Type.Nname != nil && f.Type.Nname.Func.Inl != nil { // nname was set by caninl
 
@@ -299,10 +314,10 @@ func dumpexporttype(t *Type) {
 			if Debug['l'] < 2 {
 				typecheckinl(f.Type.Nname)
 			}
-			fmt.Fprintf(bout, "\tfunc (%v) %v %v { %v }\n", Tconv(getthisx(f.Type).Type, obj.FmtSharp), Sconv(f.Sym, obj.FmtShort|obj.FmtByte|obj.FmtSharp), Tconv(f.Type, obj.FmtShort|obj.FmtSharp), Hconv(f.Type.Nname.Func.Inl, obj.FmtSharp))
+			exportf("\tfunc (%v) %v %v { %v }\n", Tconv(getthisx(f.Type).Type, obj.FmtSharp), Sconv(f.Sym, obj.FmtShort|obj.FmtByte|obj.FmtSharp), Tconv(f.Type, obj.FmtShort|obj.FmtSharp), Hconv(f.Type.Nname.Func.Inl, obj.FmtSharp))
 			reexportdeplist(f.Type.Nname.Func.Inl)
 		} else {
-			fmt.Fprintf(bout, "\tfunc (%v) %v %v\n", Tconv(getthisx(f.Type).Type, obj.FmtSharp), Sconv(f.Sym, obj.FmtShort|obj.FmtByte|obj.FmtSharp), Tconv(f.Type, obj.FmtShort|obj.FmtSharp))
+			exportf("\tfunc (%v) %v %v\n", Tconv(getthisx(f.Type).Type, obj.FmtSharp), Sconv(f.Sym, obj.FmtShort|obj.FmtByte|obj.FmtSharp), Tconv(f.Type, obj.FmtShort|obj.FmtSharp))
 		}
 	}
 }
@@ -341,33 +356,83 @@ func dumpsym(s *Sym) {
 }
 
 func dumpexport() {
-	lno := lineno
-
 	if buildid != "" {
-		fmt.Fprintf(bout, "build id %q\n", buildid)
-	}
-	fmt.Fprintf(bout, "\n$$\npackage %s", localpkg.Name)
-	if safemode != 0 {
-		fmt.Fprintf(bout, " safe")
-	}
-	fmt.Fprintf(bout, "\n")
+		exportf("build id %q\n", buildid)
+	}
+
+	size := 0 // size of export section without enclosing markers
+	if forceNewExport || newexport != 0 {
+		// binary export
+		// The linker also looks for the $$ marker - use char after $$ to distinguish format.
+		exportf("\n$$B\n")        // indicate binary format
+		const verifyExport = true // enable to check format changes
+		if verifyExport {
+			// save a copy of the export data
+			var copy bytes.Buffer
+			bcopy := obj.Binitw(&copy)
+			size = Export(bcopy, Debug_export != 0)
+			bcopy.Flush() // flushing to bytes.Buffer cannot fail
+			if n, err := bout.Write(copy.Bytes()); n != size || err != nil {
+				Fatalf("error writing export data: got %d bytes, want %d bytes, err = %v", n, size, err)
+			}
+
+			// verify there's no "\n$$\n" inside the export data
+			// TODO(gri) fragile - the end marker needs to be fixed
+			// TODO(gri) investigate if exporting a string containing "\n$$\n"
+			//           causes problems (old and new format)
+			if bytes.Index(copy.Bytes(), []byte("\n$$\n")) >= 0 {
+				Fatalf("export data contains end marker in its midst")
+			}
 
-	for _, p := range pkgs {
-		if p.Direct {
-			dumppkg(p)
+			// verify that we can read the copied export data back in
+			// (use empty package map to avoid collisions)
+			savedPkgMap := pkgMap
+			savedPkgs := pkgs
+			pkgMap = make(map[string]*Pkg)
+			pkgs = nil
+			importpkg = mkpkg("")
+			Import(obj.Binitr(&copy)) // must not die
+			importpkg = nil
+			pkgs = savedPkgs
+			pkgMap = savedPkgMap
+		} else {
+			size = Export(bout, Debug_export != 0)
+		}
+		exportf("\n$$\n")
+	} else {
+		// textual export
+		lno := lineno
+
+		exportf("\n$$\n") // indicate textual format
+		exportsize = 0
+		exportf("package %s", localpkg.Name)
+		if safemode != 0 {
+			exportf(" safe")
+		}
+		exportf("\n")
+
+		for _, p := range pkgs {
+			if p.Direct {
+				dumppkg(p)
+			}
 		}
-	}
 
-	// exportlist grows during iteration - cannot use range
-	for len(exportlist) > 0 {
-		n := exportlist[0]
-		exportlist = exportlist[1:]
-		lineno = n.Lineno
-		dumpsym(n.Sym)
+		// exportlist grows during iteration - cannot use range
+		for len(exportlist) > 0 {
+			n := exportlist[0]
+			exportlist = exportlist[1:]
+			lineno = n.Lineno
+			dumpsym(n.Sym)
+		}
+
+		size = exportsize
+		exportf("\n$$\n")
+		lineno = lno
 	}
 
-	fmt.Fprintf(bout, "\n$$\n")
-	lineno = lno
+	if Debug_export != 0 {
+		fmt.Printf("export data size = %d bytes\n", size)
+	}
 }
 
 /*

src/cmd/compile/internal/gc/dcl.go

@@ -288,7 +288,7 @@ const (
 
 	TFUNC
 	TARRAY
-	T_old_DARRAY
+	T_old_DARRAY // Doesn't seem to be used in existing code. Used now for Isddd export (see bexport.go). TODO(gri) rename.
 	TSTRUCT
 	TCHAN
 	TMAP

src/cmd/compile/internal/gc/export.go

@@ -254,6 +254,7 @@ import_stmt:
 			break;
 		}
 		if my.Name == "init" {
+			lineno = int32($1)
 			Yyerror("cannot import package as init - init must be a func");
 			break;
 		}