fmt: `bad pointer in frame fmt.(*pp).printValue`

[gucio321]

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

$ go version
go version go1.21.4 linux/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=''
GOARCH='amd64'
GOBIN=''
GOCACHE='/home/my-username/.cache/go-build'
GOENV='/home/my-username/.config/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFLAGS=''
GOHOSTARCH='amd64'
GOHOSTOS='linux'
GOINSECURE=''
GOMODCACHE='/home/my-username/go/pkg/mod'
GONOPROXY=''
GONOSUMDB=''
GOOS='linux'
GOPATH='/home/my-username/go'
GOPRIVATE=''
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/usr/local/go'
GOSUMDB='sum.golang.org'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/usr/local/go/pkg/tool/linux_amd64'
GOVCS=''
GOVERSION='go1.21.4'
GCCGO='gccgo'
GOAMD64='v1'
AR='ar'
CC='gcc'
CXX='g++'
CGO_ENABLED='1'
GOMOD='/dev/null'
GOWORK=''
CGO_CFLAGS='-O2 -g'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-O2 -g'
CGO_FFLAGS='-O2 -g'
CGO_LDFLAGS='-O2 -g'
PKG_CONFIG='pkg-config'
GOGCCFLAGS='-fPIC -m64 -pthread -Wl,--no-gc-sections -fmessage-length=0 -ffile-prefix-map=/tmp/go-build497266722=/tmp/go-build -gno-record-gcc-switches'

What did you do?

I'm writting a wrapper of C void* pointer via CGO. I'm getting the following panic:

runtime: bad pointer in frame fmt.(*pp).printValue at 0xc00022d758: 0x1
fatal error: invalid pointer found on stack

Code causing this looks as follows:

        v := texture.ID()
        fmt.Println("%T\n", v)

BUT when I add this single line:

        fmt.Println(texture)
        v := texture.ID()
        fmt.Println("%T\n", v)

This works perfectly.

Full project (as I'm unable to extract a single reproduction code) is here:
https://github.com/gucio321/cimgui-go/tree/callbacks

What did you expect to see?

No panic should happen

What did you see instead?

Panic

[ericlagergren]

Any sort of repro would be helpful.

Out of curiosity, what happens when you add runtime.KeepAlive(texture) after the call to fmt.Println in your first code snippet?

Also, fyi: fmt.Println doesn't accept format specifiers.

[gucio321]

Also, fyi: fmt.Println doesn't accept format specifiers.

oh, sorry, its a typo ofc. I meant Printf, but it doesn't matter in fact. The problem is with the value returned by texture.ID() which returns unsafe.Pointer representing C void* pointer

Out of curiosity, what happens when you add runtime.KeepAlive(texture) after the call to fmt.Println in your first code snippet?

nothing changes

Any sort of repro would be helpful.

I know, I tried to write some code but with no result.

Also, I tried to use go-delve to debug this but with no result - when using debugger it worked fine...

[gucio321]

in fact, reproducable code is here: https://github.com/gucio321/cimgui-go/tree/callbacks
I tried to write something myself but it didn't work as expected
Generally speaking the idea looks as follows:

• C creates a void* pointer
• go takes it as an unsafe.pointer
  here are some code pieces:

type Texture struct {
        id     TextureID
        Width  int
        Height int
}

func NewTextureFromRgba(rgba *image.RGBA) *Texture {
        texID := textureManager.CreateTextureRgba(rgba, rgba.Bounds().Dx(), rgba.Bounds().Dy())

        if texID.Data == nil {
                return nil
        }

        texture := Texture{
                id:     texID,
                Width:  rgba.Bounds().Dx(),
                Height: rgba.Bounds().Dy(),
        }

        // Set finalizer
        runtime.SetFinalizer(&texture, (*Texture).release)

        return &texture
}

func (t *Texture) release() {
        textureManager.DeleteTexture(t.id)
}

func (t *Texture) ID() TextureID {
        return t.id
}

type TextureID struct {
        Data unsafe.Pointer
}

func (self *TextureID) handle() (*C.ImTextureID, func()) {
        result, fn := self.c()
        return &result, fn
}

func (selfStruct *TextureID) c() (result C.ImTextureID, fin func()) {
        self := selfStruct.Data
        selfArg, selfFin := WrapVoidPtr(self)
        return (C.ImTextureID)(selfArg), func() { selfFin() }
}

func newTextureIDFromC(cvalue *C.ImTextureID) *TextureID {
        v := (unsafe.Pointer)(*cvalue)
        return &TextureID{Data: unsafe.Pointer(v)}
}

[bcmills]

0x1 is almost certainly not a valid pointer. Per the C11 spec, “the result [of converting an integer to a pointer type] is implementation-defined, might not be correctly aligned, might not point to an entity of the referenced type, and might be a trap representation.” It just so happens that 0x1 acts as a trap representation for the Go runtime.

If you have a C type that may represent a mix of pointers and integers, you should generally pass it to Go as a uintptr rather than a void*.

[mknyszek]

@gucio321 Is there anything to follow-up with here? Closing optimistically, but please comment if you disagree. Thanks!

[gucio321]

@mknyszek After reading @bcmills's statement I still think that fmt shouldn't care what exactly is the pointer as it (in theory) could be whatever the programmer wants - even 0x1 (because why not? - its the idea of the unsafe package). So in my opinion using uintptr is a good workaround here but I think that this panic is a bug in go that should be fixed, shouldn't it?

[gucio321]

CC @mknyszek

[gucio321]

Also, using uintptr is in fact not a good solution at all since it violates race detector rules.