startup: more reliable startup of multiple instances

Until now, when a new soffice instance (S2) started and tried to
connect to an existing soffice process (S1), S2 may have failed to
boostrap due to race condition in communication over the shared pipe.

S1 can be shutdown after S2 connected to it but _before_ S1 handled its
arguments (code run after 'accept' method in OfficeIPCThread).
This patch introduces a new message, sent by the main soffice after it
has called accept if and only if it's not shutting down (see mbDowning
member).
The other soffice waits for this message before enabling going in
PIPE_CONNECTED mode. If soffice fails to receive this message, pipe mode is
left unchanged and after a quick pause, it will try again.

Change-Id: I2e099a5804e1e8dd535cfd31ef454cffa44efa62
Signed-off-by: Stephan Bergmann <sbergman@redhat.com>

1 files changed, 13 insertions, 4 deletions

diff --git a/desktop/unx/source/start.c b/desktop/unx/source/start.c
index 2d12d89e89cc..0e32c590f9de 100644
--- a/desktop/unx/source/start.c
+++ b/desktop/unx/source/start.c
@@ -839,10 +839,19 @@ SAL_IMPLEMENT_MAIN_WITH_ARGS( argc, argv )
 
         if ( ( fd = connect_pipe( pPipePath ) ) >= 0 )
         {
-            rtl_uString *pCwdPath = NULL;
-            osl_getProcessWorkingDir( &pCwdPath );
-
-            bSentArgs = send_args( fd, pCwdPath );
+            // Wait for answer
+            char resp[ strlen( "InternalIPC::SendArguments" ) + 1];
+            ssize_t n = read( fd, resp, SAL_N_ELEMENTS( resp ) );
+            if (n == (ssize_t) SAL_N_ELEMENTS( resp )
+                && (memcmp(
+                resp, "InternalIPC::SendArguments",
+                SAL_N_ELEMENTS( resp ) - 1) == 0)) {
+                rtl_uString *pCwdPath = NULL;
+                osl_getProcessWorkingDir( &pCwdPath );
+
+                // Then send args
+                bSentArgs = send_args( fd, pCwdPath );
+           }
 
             close( fd );
         }