Skip to contentSkip to conversation

device_for_each_child(): kill pointless warning noise

PATCH In review Jeff Garzik 3 participants 4 messages Sep 28, 2006 Driver core

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 device_for_each_child(): kill pointless warning noise

As the last patch demonstrated, it is quite valid for a caller to ignore the return value of device_for_each_child(), given that the return value is wholly dependent on the actor -- which in practice often has a hardcoded return value.

Please apply, to reduce a portion of the warning explosion seen in current linux-2.6.git.

include/linux/device.h +1 -1
index 662e6a1..9d4f6a9 100644
@@ -393,7 +393,7 @@ extern void device_unregister(struct dev
 extern void device_initialize(struct device * dev);
 extern int __must_check device_add(struct device * dev);
 extern void device_del(struct device * dev);
-extern int __must_check device_for_each_child(struct device *, void *,
+extern int device_for_each_child(struct device *, void *,
 		     int (*fn)(struct device *, void *));
 extern int device_rename(struct device *dev, char *new_name);
 
#1 lore raw
Quoted 4 lines · On Wed, 27 Sep 2006 21:05:18 -0400 Jeff Garzik <[email protected]> wrote:

As the last patch demonstrated, it is quite valid for a caller to ignore the return value of device_for_each_child(), given that the return value is wholly dependent on the actor -- which in practice often has a hardcoded return value.

Yes, but almost all of the instances which you found are flat-out *wrong*. They're returning 0 or 1 at random places in the callchain because they're calling intermediate void-returning functions which are themselves dropping error codes on the floor instead of returning them.

Let no error go unhandled!

#2 lore raw
Quoted 11 lines · Andrew Morton
Quoted 4 lines · On Wed, 27 Sep 2006 21:05:18 -0400 Jeff Garzik <[email protected]> wrote:

As the last patch demonstrated, it is quite valid for a caller to ignore the return value of device_for_each_child(), given that the return value is wholly dependent on the actor -- which in practice often has a hardcoded return value.

Yes, but almost all of the instances which you found are flat-out *wrong*. They're returning 0 or 1 at random places in the callchain because they're calling intermediate void-returning functions which are themselves dropping
error codes on the floor instead of returning them.

"almost all" Thus it is wrong to _force_ the usage model on the caller.

It should be obvious that a simple search need not _require_ a dummy return value, that is promptly ignored.

See previous email for examples.

Jeff

#3 lore raw

Ar Mer, 2006-09-27 am 21:50 -0400, ysgrifennodd Jeff Garzik:

Quoted 4 lines

"almost all" Thus it is wrong to _force_ the usage model on the caller.

It should be obvious that a simple search need not _require_ a dummy return value, that is promptly ignored.

How about just adding an explicit device_for_each_child_noret() or similar function so that the safety check stays as well for the usual case.

Alan