--- projects/pjob/pjob.c	2006/01/24 20:15:25	1.3
+++ projects/pjob/pjob.c	2006/01/30 21:46:59	1.6
@@ -25,6 +25,9 @@
 #include <glib.h>
 #include <glib/gprintf.h>
 #include <poll.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <sys/resource.h>
 
 #define DEFNUMTHREAD 10
 #define DEFTIMEOUT 60
@@ -35,6 +38,7 @@ struct _process_t{
 	gchar *jobname;
 	GTimer *timer;
 	GPid pid;
+	gint stat_loc;
 	gchar *file_stdout;
 	gchar *file_stderr;
 
@@ -124,12 +128,12 @@ void process_child(gpointer data, gpointer user_data){
 
 	/* Exec the job */
 	if (infile == NULL){
-		if( ! g_spawn_async_with_pipes(NULL, execargv, NULL, 0, NULL, NULL, &(proc->pid), NULL, &(outpipes[0]), &(outpipes[1]), &err)){
+		if( ! g_spawn_async_with_pipes(NULL, execargv, NULL, G_SPAWN_DO_NOT_REAP_CHILD, NULL, NULL, &(proc->pid), NULL, &(outpipes[0]), &(outpipes[1]), &err)){
 			g_printerr("Failed to execute job %s: %s\n", proc->jobname, err->message);
 			return;
 		}
 	}else{
-		if( ! g_spawn_async_with_pipes(NULL, execargv, NULL, 0, NULL, NULL, &(proc->pid), &(inpipes[1]), &(outpipes[0]), &(outpipes[1]), &err)){
+		if( ! g_spawn_async_with_pipes(NULL, execargv, NULL, G_SPAWN_DO_NOT_REAP_CHILD, NULL, NULL, &(proc->pid), &(inpipes[1]), &(outpipes[0]), &(outpipes[1]), &err)){
 			g_printerr("Failed to execute job %s: %s\n", proc->jobname, err->message);
 			return;
 		}
@@ -165,7 +169,7 @@ void process_child(gpointer data, gpointer user_data){
 		poll(fds, fdssize, -1);
 		/* For stdout and stderr see if there is any data, and read it */
 		for(x=0; x<2; x++){
-			if((fds[x].revents & POLLIN) == 0){
+			if((fds[x].revents & POLLIN) != 0){
 				/* We have data to read */
 				g_io_channel_read_line(sout[x], &readbuf, &rdatasize, NULL, NULL);
 				if(rdatasize > 0){
@@ -208,8 +212,24 @@ void process_child(gpointer data, gpointer user_data){
 		if(((fds[0].revents & POLLHUP) != 0) && ((fds[1].revents & POLLHUP) != 0)) break;
 	}	
 
+	while((waitpid(proc->pid, &(proc->stat_loc), 0)) != proc->pid);
+
 	g_timer_stop(proc->timer);
 
+	/* If process exited cleanly */
+	if (WIFEXITED(proc->stat_loc)){
+		/* Get the exit code */
+		if (verbose) g_fprintf(stderr, "Job '%s' exited with code %d. Exec time: %.2f\n", proc->jobname, WEXITSTATUS(proc->stat_loc), g_timer_elapsed(proc->timer, 0));
+	}else{
+		/* Otherwise - find out what it died with */
+		/* TODO - this doesn't work quite right.. Mainly because its looking at the shell process, so the
+		 * child of the /bin/sh which does get a signal, this isn't passed up. Although, it handly tells
+		 * us if the /bin/sh gets a SEGV etc ;)
+		 */
+		g_fprintf(stderr, "Job %s exited with signal %d. Exec time: %.2f\n", proc->jobname, (WTERMSIG(proc->stat_loc)), g_timer_elapsed(proc->timer, 0));
+	}
+
+
 	g_io_channel_shutdown(sout[0], TRUE, NULL);
 	g_io_channel_shutdown(sout[1], TRUE, NULL);
 
@@ -223,7 +243,6 @@ void process_child(gpointer data, gpointer user_data){
 
 	g_spawn_close_pid(proc->pid);
 
-	if (verbose) g_fprintf(stderr, "Ending job '%s'\n", proc->jobname);
 
 }
 
@@ -313,6 +332,8 @@ int main(int argc, char **argv){
 	GError *pp_err = NULL, *err = NULL;
 	gint x;
 
+	struct rlimit rlp;
+
 	GOptionContext *optcontext;
 
 	optcontext = g_option_context_new(" - parallel job executer");
@@ -342,6 +363,14 @@ int main(int argc, char **argv){
 	}else{
 		g_printerr("Threading not supported\n");
 	}
+
+	/* Up the number of FD's to the "hard" limit.
+	 * This is mainly to get around the very small default
+	 * solaris has, or 256
+	 */
+	getrlimit(RLIMIT_NOFILE, &rlp);
+	rlp.rlim_cur = rlp.rlim_max;
+	setrlimit(RLIMIT_NOFILE, &rlp);
 	
 	if(verbose) g_printerr("Creating a threadpool %d in size\n", numthreads);
 	procpool = g_thread_pool_new(process_child, NULL, numthreads, FALSE, &pp_err);