cmd/compile: loopvar doesn't trigger nocopy detection

[zigo101]

Go version

go version go1.22.0 linux/amd64

Output of go env in your module/workspace:

.

What did you do?

package main

import (
	"fmt"
	"strings"
)

func foo() string {
	for b, i := (strings.Builder{}), byte('a'); ; i++ {
		b.WriteByte(i)
		if i == 'z' {
			return b.String()
		}
	}
}

func bar(callback func(*strings.Builder)) string {
	for b, i := (strings.Builder{}), byte('a'); ; i++ {
		b.WriteByte(i)
		callback(&b) // <-- difference here
		if i == 'z' {
			return b.String()
		}
	}
}

func main() {
	debugProcess := func(pb *strings.Builder) {
		// do nothing
	}
	fmt.Println("foo:", foo())
	fmt.Println("bar:", bar(debugProcess))
}

What did you see happen?

Consistent behavior between foo and bar with 1.22 compiler.

What did you expect to see?

Inconsistent behavior between foo and bar with 1.22 compiler.

[zigo101]

The problem should exist for every value which contains self-reference pointers when it is used as loop variables. (edit: looks not true)

[ianlancetaylor]

CC @dr2chase

[zigo101]

In the following program, bar function panics, but foo doesn't.
The fact reflects that, the compiler will check whether or not a declared
function will put the references of the argument somewhere after the
function exits, but not check this for local anonymous functions.

package main

import  "strings"

//go:noinline
func f(b *strings.Builder) {
	println(b.String())
}


func foo() {
	for b := (strings.Builder{}); b.Len() < 2; {
		b.WriteByte('!')
		f(&b)
	}
}

var p *strings.Builder

//go:noinline
func g(b *strings.Builder) {
	println(b.String())
	p = b
}

func bar() {
	for b := (strings.Builder{}); b.Len() < 2; {
		b.WriteByte('!')
		g(&b)
	}
}

func main() {
	println("------------ foo:")
	foo() // not panic
	println("------------ bar:")
	bar() // panic
}

[dr2chase]

The panic I get is

panic: strings: illegal use of non-zero Builder copied by value

goroutine 1 [running]:
strings.(*Builder).copyCheck(...)
	/Users/drchase/work/go/src/strings/builder.go:47
strings.(*Builder).WriteByte(...)
	/Users/drchase/work/go/src/strings/builder.go:102
main.bar()
	/Users/drchase/work/go/src/go101/main.go:28 +0xad
main.main()
	/Users/drchase/work/go/src/go101/main.go:37 +0x4f
exit status 2

Is that what you get? You could have provided this information in your bug report and it would have been helpful.

[zigo101]

Yes, we should get it for both functions (foo and bar), but only get it for the second one (bar).

[zigo101]

An example to extend one of my previous comment:

package main

import (
	"fmt"
	"strings"
)

//go:noinline
func globalDebugProcess(pb *strings.Builder) {
	// do nothing
}

func foo() string {
	for b, i := (strings.Builder{}), byte('a'); ; i++ {
		b.WriteByte(i)
		globalDebugProcess(&b)
		if i == 'z' {
			return b.String()
		}
	}
}

func bar(callback func(*strings.Builder)) string {
	for b, i := (strings.Builder{}), byte('a'); ; i++ {
		b.WriteByte(i)
		callback(&b) // <-- difference here
		if i == 'z' {
			return b.String()
		}
	}
}

func main() {
	localDebugProcess := func(pb *strings.Builder) {
		// do nothing
	}
	fmt.Println("foo:", foo())
	fmt.Println("    foo done.")
	fmt.Println("bar:", bar(localDebugProcess))
	fmt.Println("    bar done.")
}

[zigo101]

BTW, such cases can never happen in JS world. So this is a fundamental difference from JS.

[dr2chase]

Is this causing problems in a real program? Also, in your examples, it is helpful if you provide information about what happened when you ran it. Just because I observe a problem when I run it, does not mean I observe the same problem.

[zigo101]

? The problem is just there. I just expected the behavior doesn't change when adding/removing a no-op line.

The example is real.