fix a refleak

TODO

@@ -1,11 +1,38 @@
-- try:
+- do we need vips_object_unref_outputs() in more places? check all calls to
+  _build()
+
+
+
+
+- how about
+
+	vips max add[babe.jpg,babe2.jpg]
+
+  does that make any sense?
+
+	vips copy add[babe.jpg,add[babe2.jpg,babe3.jpg]] sum.v
+
+
 
-	$ vips copy z.tif[page=12] babe.v
 
-  get a couple of leaked objects from the error
 
 - make im_tiff2vips.c into a stub
 
+- what should we do?
+
+	vips_image_new_from_file( "x.tif[page=12]" );
+
+  or maybe:
+
+	vips_image_new_from_file( "x.tif", "page", 12, NULL );
+
+  or maybe neither? we'd like this to work though:
+
+  	header x.tif[page=12]
+
+  object_new_from_string()?
+
+
 
 
 
@@ -20,8 +47,6 @@
 - "header fred.png" does not work, since header uses im_open() which uses
   VipsForeign
 
-- "header x.tif[page=7]" does not work
-
 
 
 

libvips/iofuncs/object.c

@@ -1280,16 +1280,20 @@ vips_object_set_argument_from_string( VipsObject *object,
 
 		if( vips_cache_operation_build( 
 			(VipsOperation **) &new_object ) ) {
+			/* The build may have made some output objects before
+			 * failing.
+			 */
+			vips_object_unref_outputs( new_object );
 			g_object_unref( new_object );
 			return( -1 );
 		}
 
 		g_object_get( new_object, "out", &out, NULL );
 
-		/* Getting @out will have upped it's count and we want to
-		 * hold the only ref to it.
+		/* Getting @out will have upped its count so it'll be safe.
+		 * We can junk all other outputs,
 		 */
-		g_object_unref( out );
+		vips_object_unref_outputs( new_object );
 
 		/* @out holds a ref to new_object, we can drop ours.
 		 */