On Thu, Mar 25, 2021 at 11:38 AM Marco Ronchese <marcoronch...@gmail.com> wrote: > > I am calling certain Windows API from Go code (without CGO), everything works > flawlessly, but now I bumped into this issue. > > I want to register a callback that accepts (also) a float as argument > (https://docs.microsoft.com/en-us/windows/win32/api/audiopolicy/nf-audiopolicy-iaudiosessionevents-onsimplevolumechanged) > and I get a runtime error: > > panic: compileCallback: float arguments not supported > > This panic is thrown at https://golang.org/src/runtime/syscall_windows.go#L125 > > I tried to receive an uint32 and convert it with math.Float64frombits (well, > basically just with some pointer arithmetic) but no luck. > > This issue on Github (https://github.com/golang/go/issues/6510), which got > fixed, is related, but there the syscall itself was returning floats, here is > a callback to register with a syscall. > > The questions are: > 1) Can I work around this with some clever pointer conversion? From what I > understand this is not the case since basically Go runtime is ignoring some > CPU registers where that value is stored, am I right? > 2) What is the philosophy behind the choice of not supporting this kind of > stuff? Is something like: "Go runtime needs to support the basic syscall the > language, its standard library and most users need, and for the rest C is the > way" > > A while back I though using CGO for these kind of stuff was the only way, few > weeks ago I discover that this was not necessarily the case. I was thrilled I > could write everything in Go, but maybe this is not true after all. Well, > quite a journey. Of course I hope I am wrong
The problem is the calling convention. syscall.NewCallback has to take the C values, which arrive using the C calling convention, and pass them to Go using the Go calling convention. On amd64 the main step here is callbackasm1 in runtime/sys_windows_amd64.s That function takes the C argument registers and stores them in the right place for the Go code. Unfortunately, on x86 floats are passed in floating point registers, not the ordinary argument registers. So to handle floating point arguments the code would need to get them from the right register. Not supporting this case is not a philosophical issue. It's that nobody really knows how to implement it. Ian -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/golang-nuts/CAOyqgcVwgQ9QyBuYzBxU9SxhsDMr%2Buwr%2Bp5BYGV8rA6-dn4%3DHw%40mail.gmail.com.
