[PATCH] os: added the -upstart option for Upstart's signaling method.

[Aaron Plattner]

On Tue, Feb 23, 2010 at 10:15:31PM -0800, Oliver McFadden wrote:
> This is very similar to the RunFromSmartParent option, except we do not
> send the signal to our parent process, but our own process instead; also
> the signal is SIGSTOP, not SIGUSR1.
> 
> Upstart will detect that our process has stopped, and interprets this as
> "process ready" (to accept clients), send us SIGCONT, and move our job
> from the SPAWNED to RUNNING status.
> 
> Signed-off-by: Oliver McFadden <oliver.mcfadden at nokia.com>
> ---
> Unfortunately we're stuck with using Upstart, and in the tradition of trying to
> keep as close to upstream as possible, I'm sending this for review...
> 
> According to http://upstart.ubuntu.com/ Upstart is used by a number of
> distributions (Ubuntu, Fedora, Debian, and of course Maemo) so this is probably
> a worthwhile patch.
> 
>  include/opaque.h |    1 +
>  os/connection.c  |    5 +++++
>  os/utils.c       |    5 +++++
>  3 files changed, 11 insertions(+), 0 deletions(-)
> 
> diff --git a/include/opaque.h b/include/opaque.h
> index b3c7c70..f8c0000 100644
> --- a/include/opaque.h
> +++ b/include/opaque.h
> @@ -55,6 +55,7 @@ extern _X_EXPORT int defaultBackingStore;
>  extern _X_EXPORT Bool disableBackingStore;
>  extern _X_EXPORT Bool enableBackingStore;
>  extern _X_EXPORT Bool PartialNetwork;
> +extern _X_EXPORT Bool RunFromUpstart;
>  #ifndef NOLOGOHACK
>  extern _X_EXPORT int logoScreenSaver;
>  #endif
> diff --git a/os/connection.c b/os/connection.c
> index 3ff93bb..92aec38 100644
> --- a/os/connection.c
> +++ b/os/connection.c
> @@ -146,6 +146,8 @@ Bool NewOutputPending;		/* not yet attempted to write some new output */
>  Bool AnyClientsWriteBlocked;	/* true if some client blocked on write */
>  
>  static Bool RunFromSmartParent;	/* send SIGUSR1 to parent process */
> +Bool RunFromUpstart;		/* send SIGSTOP to our own process;
> +				   Upstart will send SIGCONT back. */
>  Bool PartialNetwork;	/* continue even if unable to bind all addrs */
>  static Pid_t ParentProcess;
>  
> @@ -357,6 +359,9 @@ NotifyParentProcess(void)
>  	    kill (ParentProcess, SIGUSR1);
>  	}
>      }
> +    if (RunFromUpstart) {
> +	kill (getpid (), SIGSTOP);

raise(SIGSTOP) is simpler.

Is there any way to tell for sure that Upstart is running and is going to
wake us up, or do we just deadlock if something goes wrong with it?

> +    }
>  #endif
>  }
>  
> diff --git a/os/utils.c b/os/utils.c
> index 21e25e0..d19ffb0 100644
> --- a/os/utils.c
> +++ b/os/utils.c
> @@ -538,6 +538,7 @@ void UseMsg(void)
>  #endif
>      ErrorF("-dumbSched             Disable smart scheduling, enable old behavior\n");
>      ErrorF("-schedInterval int     Set scheduler interval in msec\n");
> +    ErrorF("-upstart               Enable support for Upstart's signaling method\n");

This almost implies that -upstart is something somebody might actually want
to pass to the X server.  In cases like this, less documentation is almost
better; "Do not use this option" conveys more useful information to the
user.

>      ErrorF("+extension name        Enable extension\n");
>      ErrorF("-extension name        Disable extension\n");
>  #ifdef XDMCP
> @@ -936,6 +937,10 @@ ProcessCommandLine(int argc, char *argv[])
>  		UseMsg ();
>  	}
>  #endif
> +	else if ( strcmp( argv[i], "-upstart") == 0)
> +	{
> +	    RunFromUpstart = TRUE;

Running away from Upstart sounds like a good idea.  ;)

> +	}
>  	else if ( strcmp( argv[i], "+extension") == 0)
>  	{
>  	    if (++i < argc)
> -- 
> 1.6.1
>