weak: weak.Pointer[T] can be misused with type conversions

[mateusz834]

This is the second time something like this happens in the std, see #56603.

func main() {
        val := (weak.Pointer[[1024 * 1024]int])(weak.Make(new(int))).Value()
        val[len(val)-1] = 999
        fmt.Println(val[len(val)-1])
}

$ go run .
unexpected fault address 0xc000810108
fatal error: fault
[signal SIGSEGV: segmentation violation code=0x2 addr=0xc000810108 pc=0x49203d]

goroutine 1 gp=0xc000002380 m=0 mp=0x560480 [running]:
runtime.throw({0x4b4e20?, 0x7f54c773b108?})
        /home/mateusz/code/go/src/runtime/panic.go:1096 +0x48 fp=0xc000084ea8 sp=0xc000084e78 pc=0x463328
runtime.sigpanic()
        /home/mateusz/code/go/src/runtime/signal_unix.go:939 +0x26c fp=0xc000084f08 sp=0xc000084ea8 pc=0x4646cc
main.main()
        /tmp/testing/main.go:10 +0x3d fp=0xc000084f50 sp=0xc000084f08 pc=0x49203d
runtime.main()
        /home/mateusz/code/go/src/runtime/proc.go:283 +0x28b fp=0xc000084fe0 sp=0xc000084f50 pc=0x43434b
runtime.goexit({})
        /home/mateusz/code/go/src/runtime/asm_amd64.s:1700 +0x1 fp=0xc000084fe8 sp=0xc

(........)

Previously reported at: #67552 (comment)

CC @mknyszek

[mateusz834]

Marking this as a release-blocker since doing the same thing like in #56603 is technically a backwards incompatible change.

[gopherbot]

Change https://go.dev/cl/647195 mentions this issue: weak: prevent unsafe conversions using weak pointers

[gopherbot]

Change https://go.dev/cl/647215 mentions this issue: weak: disallow type conversions of Pointer[T]

[thepudds]

Hi @mknyszek, I sent https://go.dev/cl/647195, which seems to pass the trybots. @qiulaidongfeng also sent https://go.dev/cl/647215.

Do those seem like reasonable approaches?

[mknyszek]

Shoot. Maybe we should have a more generic (ha) test for this on the standard library, or perhaps it should be part of the per-release API checker.

Thank you both @qiulaidongfeng and @thepudds for sending CLs! I'll probably go with @thepudds' CL just because it adds a couple of tests, for the next person who trips over the same issue.

[mateusz834]

Just want to note that this needs also to be fixed on the release-branch.go1.24 branch, as this is a new API.

[mknyszek]

Yep, we'll have to cherry-pick the change.

[dmitshur]

CC @cherrymui.