[dev.typeparams] cmd/compile: record writer's stack at export data sync points

This CL extends the unified export data format's existing sync
mechanism to save writer stacks, controlled by the -d=syncframes debug
flag. This allows readers to provide more details when reporting
desync errors, which should simplify development of the data format
and the various reader/writer implementations.

For example, CL 328051 updated reader and writer, but missed making a
similar change to the linker (fix in CL 328054). Re-reviewing the CL
in isolation after the failure, it was not immediately obvious what
was going wrong. But the pair of stack traces below identifies exactly
what happened: it should have updated linker.relocFuncExt to write out
the new sync marker too.

```
data sync error: package "internal/abi", section 6, index 4, offset 536

found UseReloc, written at:
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/encoder.go:221: (*encoder).reloc +0x44
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/linker.go:214: (*linker).relocFuncExt +0x580
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/linker.go:233: (*linker).relocTypeExt +0x234
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/linker.go:161: (*linker).relocObj +0x2198
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/linker.go:64: (*linker).relocIdx +0x196

expected ImplicitTypes, reading at:
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/reader.go:796: (*reader).implicitTypes +0x36
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/reader.go:810: (*reader).addBody +0x81
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/reader.go:727: (*reader).funcExt +0x542
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/reader.go:651: (*reader).method +0x324
	/home/mdempsky/wd/go/src/cmd/compile/internal/noder/reader.go:557: (*pkgReader).objIdx +0x2704
```

Change-Id: I911193edd2a965f81b7459f15fb613a773584685
Reviewed-on: https://go-review.googlesource.com/c/go/+/328909
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
This commit is contained in:
Matthew Dempsky 2021-06-17 01:12:23 -07:00
parent 78aa251ace
commit 54fe57bc22
8 changed files with 219 additions and 46 deletions

View file

@ -44,6 +44,7 @@ type DebugFlags struct {
Panic int `help:"show all compiler panics"`
Slice int `help:"print information about slice compilation"`
SoftFloat int `help:"force compiler to emit soft-float code"`
SyncFrames int `help:"how many writer stack frames to include at sync points in unified export data"`
TypeAssert int `help:"print information about type assertion inlining"`
TypecheckInl int `help:"eager typechecking of inline function bodies"`
Unified int `help:"enable unified IR construction"`

View file

@ -13,6 +13,7 @@ import (
"go/token"
"math/big"
"os"
"runtime"
"strings"
"cmd/compile/internal/base"
@ -131,17 +132,82 @@ func (r *decoder) checkErr(err error) {
}
}
func (r *decoder) sync(m syncMarker) {
if debug {
pos, err0 := r.data.Seek(0, os.SEEK_CUR)
x, err := r.data.ReadByte()
r.checkErr(err)
if x != byte(m) {
// TODO(mdempsky): Revisit this error message, and make it more
// useful (e.g., include r.p.pkgPath).
base.Fatalf("data sync error: found %v at %v (%v) in (%v:%v), but expected %v", syncMarker(x), pos, err0, r.k, r.idx, m)
}
func (r *decoder) rawUvarint() uint64 {
x, err := binary.ReadUvarint(&r.data)
r.checkErr(err)
return x
}
func (r *decoder) rawVarint() int64 {
ux := r.rawUvarint()
// Zig-zag decode.
x := int64(ux >> 1)
if ux&1 != 0 {
x = ^x
}
return x
}
func (r *decoder) rawReloc(k reloc, idx int) int {
e := r.relocs[idx]
assert(e.kind == k)
return e.idx
}
func (r *decoder) sync(mWant syncMarker) {
if !enableSync {
return
}
pos, _ := r.data.Seek(0, os.SEEK_CUR) // TODO(mdempsky): io.SeekCurrent after #44505 is resolved
mHave := syncMarker(r.rawUvarint())
writerPCs := make([]int, r.rawUvarint())
for i := range writerPCs {
writerPCs[i] = int(r.rawUvarint())
}
if mHave == mWant {
return
}
// There's some tension here between printing:
//
// (1) full file paths that tools can recognize (e.g., so emacs
// hyperlinks the "file:line" text for easy navigation), or
//
// (2) short file paths that are easier for humans to read (e.g., by
// omitting redundant or irrelevant details, so it's easier to
// focus on the useful bits that remain).
//
// The current formatting favors the former, as it seems more
// helpful in practice. But perhaps the formatting could be improved
// to better address both concerns. For example, use relative file
// paths if they would be shorter, or rewrite file paths to contain
// "$GOROOT" (like objabi.AbsFile does) if tools can be taught how
// to reliably expand that again.
fmt.Printf("export data desync: package %q, section %v, index %v, offset %v\n", r.common.pkgPath, r.k, r.idx, pos)
fmt.Printf("\nfound %v, written at:\n", mHave)
if len(writerPCs) == 0 {
fmt.Printf("\t[stack trace unavailable; recompile package %q with -d=syncframes]\n", r.common.pkgPath)
}
for _, pc := range writerPCs {
fmt.Printf("\t%s\n", r.common.stringIdx(r.rawReloc(relocString, pc)))
}
fmt.Printf("\nexpected %v, reading at:\n", mWant)
var readerPCs [32]uintptr // TODO(mdempsky): Dynamically size?
n := runtime.Callers(2, readerPCs[:])
for _, pc := range fmtFrames(readerPCs[:n]...) {
fmt.Printf("\t%s\n", pc)
}
// We already printed a stack trace for the reader, so now we can
// simply exit. Printing a second one with panic or base.Fatalf
// would just be noise.
os.Exit(1)
}
func (r *decoder) bool() bool {
@ -154,16 +220,12 @@ func (r *decoder) bool() bool {
func (r *decoder) int64() int64 {
r.sync(syncInt64)
x, err := binary.ReadVarint(&r.data)
r.checkErr(err)
return x
return r.rawVarint()
}
func (r *decoder) uint64() uint64 {
r.sync(syncUint64)
x, err := binary.ReadUvarint(&r.data)
r.checkErr(err)
return x
return r.rawUvarint()
}
func (r *decoder) len() int { x := r.uint64(); v := int(x); assert(uint64(v) == x); return v }
@ -177,11 +239,7 @@ func (r *decoder) code(mark syncMarker) int {
func (r *decoder) reloc(k reloc) int {
r.sync(syncUseReloc)
idx := r.len()
e := r.relocs[idx]
assert(e.kind == k)
return e.idx
return r.rawReloc(k, r.len())
}
func (r *decoder) string() string {

View file

@ -13,6 +13,7 @@ import (
"go/constant"
"io"
"math/big"
"runtime"
"cmd/compile/internal/base"
)
@ -93,6 +94,8 @@ type encoder struct {
relocs []relocEnt
data bytes.Buffer
encodingRelocHeader bool
k reloc
idx int
}
@ -107,6 +110,10 @@ func (w *encoder) flush() int {
// TODO(mdempsky): Consider writing these out separately so they're
// easier to strip, along with function bodies, so that we can prune
// down to just the data that's relevant to go/types.
if w.encodingRelocHeader {
base.Fatalf("encodingRelocHeader already true; recursive flush?")
}
w.encodingRelocHeader = true
w.sync(syncRelocs)
w.len(len(w.relocs))
for _, rent := range w.relocs {
@ -128,10 +135,58 @@ func (w *encoder) checkErr(err error) {
}
}
func (w *encoder) rawUvarint(x uint64) {
var buf [binary.MaxVarintLen64]byte
n := binary.PutUvarint(buf[:], x)
_, err := w.data.Write(buf[:n])
w.checkErr(err)
}
func (w *encoder) rawVarint(x int64) {
// Zig-zag encode.
ux := uint64(x) << 1
if x < 0 {
ux = ^ux
}
w.rawUvarint(ux)
}
func (w *encoder) rawReloc(r reloc, idx int) int {
// TODO(mdempsky): Use map for lookup.
for i, rent := range w.relocs {
if rent.kind == r && rent.idx == idx {
return i
}
}
i := len(w.relocs)
w.relocs = append(w.relocs, relocEnt{r, idx})
return i
}
func (w *encoder) sync(m syncMarker) {
if debug {
err := w.data.WriteByte(byte(m))
w.checkErr(err)
if !enableSync {
return
}
// Writing out stack frame string references requires working
// relocations, but writing out the relocations themselves involves
// sync markers. To prevent infinite recursion, we simply trim the
// stack frame for sync markers within the relocation header.
var frames []string
if !w.encodingRelocHeader && base.Debug.SyncFrames > 0 {
pcs := make([]uintptr, base.Debug.SyncFrames)
n := runtime.Callers(2, pcs)
frames = fmtFrames(pcs[:n]...)
}
// TODO(mdempsky): Save space by writing out stack frames as a
// linked list so we can share common stack frames.
w.rawUvarint(uint64(m))
w.rawUvarint(uint64(len(frames)))
for _, frame := range frames {
w.rawUvarint(uint64(w.rawReloc(relocString, w.p.stringIdx(frame))))
}
}
@ -148,18 +203,12 @@ func (w *encoder) bool(b bool) bool {
func (w *encoder) int64(x int64) {
w.sync(syncInt64)
var buf [binary.MaxVarintLen64]byte
n := binary.PutVarint(buf[:], x)
_, err := w.data.Write(buf[:n])
w.checkErr(err)
w.rawVarint(x)
}
func (w *encoder) uint64(x uint64) {
w.sync(syncUint64)
var buf [binary.MaxVarintLen64]byte
n := binary.PutUvarint(buf[:], x)
_, err := w.data.Write(buf[:n])
w.checkErr(err)
w.rawUvarint(x)
}
func (w *encoder) len(x int) { assert(x >= 0); w.uint64(uint64(x)) }
@ -168,17 +217,7 @@ func (w *encoder) uint(x uint) { w.uint64(uint64(x)) }
func (w *encoder) reloc(r reloc, idx int) {
w.sync(syncUseReloc)
// TODO(mdempsky): Use map for lookup.
for i, rent := range w.relocs {
if rent.kind == r && rent.idx == idx {
w.len(i)
return
}
}
w.len(len(w.relocs))
w.relocs = append(w.relocs, relocEnt{r, idx})
w.len(w.rawReloc(r, idx))
}
func (w *encoder) code(c code) {

View file

@ -0,0 +1,20 @@
// 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.
// +build !go1.7
// TODO(mdempsky): Remove after #44505 is resolved
package noder
import "runtime"
func walkFrames(pcs []uintptr, visit frameVisitor) {
for _, pc := range pcs {
fn := runtime.FuncForPC(pc)
file, line := fn.FileLine(pc)
visit(file, line, fn.Name(), pc-fn.Entry())
}
}

View file

@ -0,0 +1,24 @@
// 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.
// +build go1.7
package noder
import "runtime"
func walkFrames(pcs []uintptr, visit frameVisitor) {
if len(pcs) == 0 {
return
}
frames := runtime.CallersFrames(pcs)
for {
frame, more := frames.Next()
visit(frame.File, frame.Line, frame.Function, frame.PC-frame.Entry)
if !more {
return
}
}
}

View file

@ -910,7 +910,7 @@ func (r *reader) addLocal(name *ir.Name, ctxt ir.Class) {
assert(ctxt == ir.PAUTO || ctxt == ir.PPARAM || ctxt == ir.PPARAMOUT)
r.sync(syncAddLocal)
if debug {
if enableSync {
want := r.int()
if have := len(r.locals); have != want {
base.FatalfAt(name.Pos(), "locals table has desynced")

View file

@ -6,8 +6,39 @@
package noder
const debug = true
import (
"fmt"
"strings"
)
// enableSync controls whether sync markers are written into unified
// IR's export data format and also whether they're expected when
// reading them back in. They're inessential to the correct
// functioning of unified IR, but are helpful during development to
// detect mistakes.
//
// When sync is enabled, writer stack frames will also be included in
// the export data. Currently, a fixed number of frames are included,
// controlled by -d=syncframes (default 0).
const enableSync = true
// fmtFrames formats a backtrace for reporting reader/writer desyncs.
func fmtFrames(pcs ...uintptr) []string {
res := make([]string, 0, len(pcs))
walkFrames(pcs, func(file string, line int, name string, offset uintptr) {
// Trim package from function name. It's just redundant noise.
name = strings.TrimPrefix(name, "cmd/compile/internal/noder.")
res = append(res, fmt.Sprintf("%s:%v: %s +0x%v", file, line, name, offset))
})
return res
}
type frameVisitor func(file string, line int, name string, offset uintptr)
// syncMarker is an enum type that represents markers that may be
// written to export data to ensure the reader and writer stay
// synchronized.
type syncMarker int
//go:generate stringer -type=syncMarker -trimprefix=sync

View file

@ -733,7 +733,7 @@ func (w *writer) funcarg(param *types2.Var, result bool) {
func (w *writer) addLocal(obj types2.Object) {
w.sync(syncAddLocal)
idx := len(w.localsIdx)
if debug {
if enableSync {
w.int(idx)
}
w.localsIdx[obj] = idx