gccgo: confusing closure names

[cherrymui]

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

gccgo2 (GCC) 9.0.0 20190108 (experimental)

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

Linux/AMD64

What did you do?

package main

type T struct {
    closure1, closure2 func()
}

func (t *T) M1()  { t.closure1 = func() { t.closure2() } }
func (t *T) M2()  { t.closure2 = func() { panic("XXX") } }

func main() {
    var t T
    t.M1()
    t.M2()
    t.closure1()
}

Built with gccgo, this program prints

panic: XXX

goroutine 1 [running]:
panic
    /tmp/src/gccgo/libgo/go/runtime/panic.go:588
main.func1
    /tmp/c.go:8
main.func1
    /tmp/c.go:7
main.main
    /tmp/c.go:14

The closures on line 7 and 8 are different functions, yet have the same name, which is confusing. It is ok that closures have somewhat artificial names, but it is super confusing if different closures have the name.

It appears to me (without checking the source code) that if the closures are defined in global scope or inside a function (NOT method), they are properly named as package_name.funcN or package_name.function_name.funcN with proper N. However, if they are defined in methods, they are all named package_name.funcN with N starting from 1 in each method. In this case, the two closures are defined in two different methods, and both named main.func1.

cc @ianlancetaylor @thanm

[ianlancetaylor]

The symbols do have different names in the object file: main..1main.T.M1..func1 and main..1main.T.M2..func1. (These names are set by Gogo::nested_function_name in names.cc). But the names in the debug info are indeed both main.func1. I think the confusion is happening in Named_object::get_id; I think that does an unfortunate munging of the full names.