Re: Exit hooks not run at logout on w32

[Lennart Borgman]

Jason Rumney wrote:
> Lennart Borgman <address@hidden> writes:
>
>   
> > Where is the code that discovers that WM_CLOSE has been sent to the
> > last frame and that Emacs should be killed?
> >     
>
> WM_CLOSE is received in w32fns.c and passed on to lisp in w32term.c,
> the same as most other windows messages. But I think you will find
> that the system does not send a WM_CLOSE in these circumstances,
> because that would trigger the exit hooks if it was the case.
>   
Yes, you are right. There was no WM_CLOSE and indeed also no 
WM_QUERYENDSESSION. At least not in the case that I have been looking 
at: starting emacs.exe from cmd.exe. I believe that this might be due to 
that MS Windows believes that emacs.exe is a console app.

Actually emacs recieve a control event instead, CTRL_LOGOFF_EVENT during 
my testing. (It could also be CTRL_SHUTDOWN_EVENT and probably 
CTRL_CLOSE_EVENT.) This event can be catched in ctrl_c_handler in 
w32console.c. I have tried that and then used PostThreadMessage 
(dwWindowsThreadId, WM_EMACS_KILL, 0, 0) to be able to call 
save-buffers-kill-emacs later on.

WM_EMACS_KILL seems ok to use here, it is not used anywhere else as far 
as I can see. It is passed on and shows up in w32_msg_pump. There I am 
trying to use my_post_msg to pass it to the lisp thread. However it does 
not show up in w32_read_socket so I guess I am doing something wrong. My 
call to my_post_msg looks like this:

static void
w32_msg_pump (deferred_msg * msg_buf)
{
 MSG msg;
 int result;
 HWND focus_window;
 BOOL bRet;

 msh_mousewheel = RegisterWindowMessage (MSH_MOUSEWHEEL);

 while ( (bRet = GetMessage (&msg, NULL, 0, 0)) != 0)
   {
     if (-1 == bRet) {
       // error, handle?
     } else {
     if (msg.hwnd == NULL)
   {
     switch (msg.message)
       {
           case WM_EMACS_KILL:
             {
               W32Msg wmsg;
               printf("w32_msg_pump: WM_EMACS_KILL\n"); fflush(stdout);
               my_post_msg (&wmsg, 0, msg, 0, 0);
               break;
             }

Should it show up in w32_read_socket after this or am I misunderstanding 
something? In w32_read_socket I do

       case WM_EMACS_KILL:
         printf("w32_read_socket WM_EMACS_KILL\n"); fflush(stdout);

I would appreciate some help here.