lack of sigusr1 on windows

no-reply@allegro.cc (Mike Farrell) — Thu, 10 Dec 2009 08:44:10 +0000

Okay so I built this nice little IPC mechanism that works okay on mac. But on windows, I'm stuck without SIGUSR1. So now what the heck can I do? I don't want to use windows' sendmessage api because I'm using GTK for my GUI.

Can I just use SIGTERM or something and catch it in the same regard?

code:
[code]
static void ipc_recv(int i)
{
load(db_path.c_str(), 0, false);
refresh_gl_viewport();
}

void ipc_init()
{
string cdir = config_dir();
db_path = cdir + "/dropbox.xdf";
sig_path = cdir + "/sig";

signal(SIGUSR1, ipc_recv);

ofstream out(sig_path.c_str());
out << getpid() << endl;
out.close();
}

void ipc_send()
{
save(db_path);
if(kill(accu_pid, SIGUSR1) == -1)
perror("");
}
[/code]

no-reply@allegro.cc (GullRaDriel) — Thu, 10 Dec 2009 15:52:35 +0000

If you are using C++, maybe you can do the trick with try/catch
If you are using C, you can emulate that with setjmp and longjmp

Or I misunderstood the question.

no-reply@allegro.cc (Thomas Fjellstrom) — Thu, 10 Dec 2009 16:53:26 +0000

personally I don't know why you aren't using sockets or pipes.

no-reply@allegro.cc (GullRaDriel) — Thu, 10 Dec 2009 19:20:56 +0000

~~AFAIK, Windows = nopipes~~

I was wrong:
Pipes on Windows

I'd say go with pipes.

no-reply@allegro.cc (Mike Farrell) — Fri, 11 Dec 2009 01:20:38 +0000

Sockets cause an annoying popup regarding windows defender, the annoying block/unblock dialog. afaik this is right after calling WSAinit.

Pipes block until the both sides open them.

I need a single unidirectional signal to go from one process to the other.

no-reply@allegro.cc (X-G) — Fri, 11 Dec 2009 01:47:19 +0000

Mike Farrell said:

Pipes block until the both sides open them.

No they don't. That's the whole point of the overlapped I/O stuff.

no-reply@allegro.cc (Mike Farrell) — Fri, 11 Dec 2009 03:15:22 +0000

Wow. this really turned out to be a giant pain in the ass.

Unix solution: less than 10 lines
Windows: not so lucky

This is the only way I could do it without a loop requiring me to continuously poll the status of something whether it be a socket or pipe (I hate doing that).

This is quite ridiculous, but at least I can tuck it away and know that my app will do no processing on behalf of IPC unless the message window gets an event.

#SelectExpand
  1#ifdef WIN32
  2LRESULT CALLBACK ipc_window_proc(HWND hWnd, UINT message, WPARAM wParam, LPARAM lParam)
  3{
  4  switch (message)
  5  {
  6    case WM_USER:
  7      ipc_recv(0);
  8    break;
  9    default:
 10      return DefWindowProc(hWnd, message, wParam, lParam);
 11    break;
 12  }
 13
 14  return 0;
 15}
 16#endif
 17
 18void ipc_init()
 19{
 20  string cdir = config_dir();
 21  db_path = cdir + "/dropbox.xdf";
 22  sig_path = cdir + "/sig";
 23  
 24#ifndef WIN32
 25  signal(SIGUSR1, ipc_recv);
 26#else
 27  //welcome to hell
 28  HINSTANCE inst = (HINSTANCE)GetWindowLong((HWND)GDK_WINDOW_HWND(get_mainwin()->window), GWL_HINSTANCE);
 29  WNDCLASSEX wcex;
 30
 31  wcex.cbSize = sizeof(WNDCLASSEX);
 32  wcex.style          = 0;
 33  wcex.lpfnWndProc    = ipc_window_proc;
 34  wcex.cbClsExtra     = 0;
 35  wcex.cbWndExtra     = 0;
 36  wcex.hCursor        = 0;
 37  wcex.hbrBackground  = 0;
 38  wcex.hIcon          = 0;
 39  wcex.hIconSm        = 0;
 40  wcex.hInstance      = inst;
 41  wcex.lpszMenuName   = NULL;
 42  wcex.lpszClassName  = "MyApp Messages";
 43
 44  if(!RegisterClassEx(&wcex))
 45    cerr << "Cannot register IPC window class!" << endl;
 46  ipc_window = CreateWindow(wcex.lpszClassName, "MyApp IPC", WS_OVERLAPPEDWINDOW, CW_USEDEFAULT, CW_USEDEFAULT,
 47                            500, 100, HWND_MESSAGE, NULL, inst/*hInstance*/, NULL);
 48  if(!ipc_window)
 49    cerr << "Cannot create IPC window (" << (int)GetLastError() << ")" << endl;
 50
 51#endif
 52
 53  ofstream out(sig_path.c_str());
 54  out << getpid() << endl;
 55  out.close();  
 56}