cmd/compile: compile failed with "Value live at entry"

[YueMengGu]

What version of Go are you using (go version)?

$ go version
go version go1.18.3 darwin/amd64

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GO111MODULE="on"
GOARCH="amd64"
GOBIN=""
GOCACHE="/Users/xxx/.cache/go-build"
GOENV="/Users/xxx/Library/Application Support/go/env"
GOEXE=""
GOEXPERIMENT=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOINSECURE=""
GOMODCACHE="/Users/xxx/Coding/go/pkg/mod"
GONOPROXY="*.everphoto.cn,git.smartisan.com"
GONOSUMDB="*.everphoto.cn,git.smartisan.com"
GOOS="darwin"
GOPATH="/Users/xxx/Coding/go"
GOPRIVATE="*.everphoto.cn,git.smartisan.com"
GOPROXY="https://goproxy.cn,https://proxy.golang.org,direct"
GOROOT="/Users/xxx/Coding/github/go"
GOSUMDB="sum.golang.google.cn"
GOTMPDIR=""
GOTOOLDIR="/Users/xxx/Coding/github/go/pkg/tool/darwin_amd64"
GOVCS=""
GOVERSION="go1.18.3"
GCCGO="gccgo"
GOAMD64="v1"
AR="ar"
CC="clang"
CXX="clang++"
CGO_ENABLED="1"
GOMOD="/dev/null"
GOWORK=""
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -arch x86_64 -m64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/77/qcdftwkd6jb6945_b2k6n60h0000gp/T/go-build1272012810=/tmp/go-build -gno-record-gcc-switches -fno-common"

What did you do?

package main

type Elem struct {}
func  (*Elem) Wait(callback func()) {}

type Base struct {
  elem [8]*Elem
}

var g_val = 1
func (s *Base) Do() *int {
  resp := &g_val
  for _, e := range s.elem {
    e.Wait(func() {
      *resp = 0
    })
  }
  return resp
}

type Sub struct {
  *Base
}

func main() {
  a := Sub{}
  a.Do()
}

What did you expect to see?

go build test.go

compile with no error

What did you see instead?

command-line-arguments

./diversity_wrap.go:15:7: internal compiler error: 'Sub.Do.func1': Value live at entry. It shouldn't be. func Sub.Do.func1, node resp, value nil

Please file a bug report including a short program that triggers the error.
https://go.dev/issue/new

it's ok with go1.16 and go1.17

I have tried patch which didn't work:

• https://go-review.googlesource.com/c/go/+/402854/
• https://go-review.googlesource.com/c/go/+/403995/

[ZekeLu]

This is the commit that introduced the bug: a3bb28e

[YueMengGu]

This is the commit that introduced the bug: a3bb28e

I tried to change the loop from range to by step，just like

  func (s *Base) Do() *int {
    resp := &g_val
    for i := 0; i < 8; i++ {
    //for _, e := range s.elem {
      s.elem[i].Wait(func() {
        *resp = 0
      })
    }
    return resp
  }

Compiled with go build -gcflags="-m=2" test.go

Without the change, compiled with go1.17 we got

./diversity_wrap.go:11:6: cannot inline (*Base).Do: unhandled op RANGE

And with the change, compiled with go1.17 we got

./diversity_wrap.go:11:6: can inline (*Base).Do with cost 46 as: method(*Base) func() *int { resp := &g_val; for loop; return resp }

The range is missing, and then compiled with go1.18.3 we got

./diversity_wrap.go:11:6: can inline (*Base).Do with cost 46 as: method(*Base) func() *int { resp := &g_val; for loop; return resp }
./diversity_wrap.go:16:7: internal compiler error: 'Sub.Do.func1': Value live at entry. It shouldn't be. func Sub.Do.func1, node resp, value nil

Please file a bug report including a short program that triggers the error.
https://go.dev/issue/new

So the bug maybe introduced not only a3bb28e

[cuonglm]

Writing Base.Do as:

func (s *Base) Do() *int {
	resp := &g_val
	for *resp != 0 {
		s.elem[0].Wait(func() { *resp = 0 })
	}
	return resp
}

is enough to trigger the bug.

[cuonglm]

git bisect points to https://go-review.googlesource.com/c/go/+/327871 cc @mdempsky

Unified IR does not have this bug, because it generates its own wrapper instead of relying on reflectdata.

[gopherbot]

Change https://go.dev/cl/423434 mentions this issue: test: add test case for issue 53702

[mdempsky]

git bisect points to https://go-review.googlesource.com/c/go/+/327871 cc @mdempsky

That's surprising. That commit should have only affect devirtualizationed, but the test case doesn't have any interface types.

If we revert just the typecheck.NeedITab stuff, does it fix the issue? (We shouldn't need it for unified IR, but maybe it was important to !unified mode.)

[cuonglm]

That's surprising. That commit should have only affect devirtualizationed, but the test case doesn't have any interface types.

I think the problem comes from generating method wrapper, not interface types. After CL 327871, we now always do inlining for method wrapper after global escape analysis. Note that disable inlining will make it compiled in non-unified.

If we revert just the typecheck.NeedITab stuff, does it fix the issue?

No.

[gopherbot]

Change https://go.dev/cl/423834 mentions this issue: test: fix issue53702.go for noopt builder

[cuonglm]

This is now fixed at tip, could we close? @mdempsky

[mdempsky]

@cuonglm The issue is reportedly still a regression from Go 1.17. It's also not specifically a generics issue, so I think we should try to backport fixes for Go 1.18/1.19.

[gopherbot]

Change https://go.dev/cl/426334 mentions this issue: cmd/compile: only inline method wrapper if method don't contain closures

[cuonglm]

@gopherbot please backport this issue to go1.18 and go1.19

[gopherbot]

Backport issue(s) opened: #54725 (for 1.18), #54726 (for 1.19).

Remember to create the cherry-pick CL(s) as soon as the patch is submitted to master, according to https://go.dev/wiki/MinorReleases.