[Cluster-devel] [PATCH 0/6] list: introduce list_last_entry_or_null()
Hi Folks, This patch is trying to introduce a new list helper to retrieve the last entry or return NULL if the list is empty corresponding to it, which is inspired by Jiri Pirko's list_first_entry_or_null(). With this trivial helper, we could get a little benefit to simplify the code logic of xfs_ail_max() and make the code looks a bit more consistent with xfs_ail_min() which has already been simplified via list_first_entry_or_null(). Moreover, as a per simple grep against fs as well as driver's tree, I think there are some other places could make use of it for similar purpose, here I spread this helper over some file systems as well. Any comments are welcome! Thanks, -Jeff
Re: [Cluster-devel] [PATCH 0/6] list: introduce list_last_entry_or_null()
On Fri, Nov 15, 2013 at 12:29:48PM +0800, Jeff Liu wrote: Hi Folks, This patch is trying to introduce a new list helper to retrieve the last entry or return NULL if the list is empty corresponding to it, which is inspired by Jiri Pirko's list_first_entry_or_null(). How did we end up with the stupid _or_null prefix there? I think the functionality is useful, but the naming is way to verbose, especially given that a list_first_entry or list_last_entry that optimizes away that one check doesn't seem useful. Instead of encoding detailed semantics in the name a good kerneldoc comment is way better.
Re: [Cluster-devel] [PATCH 0/6] list: introduce list_last_entry_or_null()
Hi, On Fri, 2013-11-15 at 06:11 -0800, Christoph Hellwig wrote: On Fri, Nov 15, 2013 at 12:29:48PM +0800, Jeff Liu wrote: Hi Folks, This patch is trying to introduce a new list helper to retrieve the last entry or return NULL if the list is empty corresponding to it, which is inspired by Jiri Pirko's list_first_entry_or_null(). How did we end up with the stupid _or_null prefix there? I think the functionality is useful, but the naming is way to verbose, especially given that a list_first_entry or list_last_entry that optimizes away that one check doesn't seem useful. Instead of encoding detailed semantics in the name a good kerneldoc comment is way better. Also, in two of the cases (patch 3 and patch 6) we don't want NULL in the case that the list is empty, and instead want another pointer which, after the patches, requires an extra if. Now I suspect that the optimizer will sort that out (I've not checked) so its probably not too big an issue. However, if instead of returning NULL for list empty, we had an extra argument to the function, then it could return whatever pointer was required for each case, whether NULL or otherwise, Steve.
Re: [Cluster-devel] [PATCH 0/6] list: introduce list_last_entry_or_null()
On 11/15 2013 22:11, Christoph Hellwig wrote: On Fri, Nov 15, 2013 at 12:29:48PM +0800, Jeff Liu wrote: Hi Folks, This patch is trying to introduce a new list helper to retrieve the last entry or return NULL if the list is empty corresponding to it, which is inspired by Jiri Pirko's list_first_entry_or_null(). How did we end up with the stupid _or_null prefix there? I think the functionality is useful, but the naming is way to verbose, especially given that a list_first_entry or list_last_entry that optimizes away that one check doesn't seem useful. Instead of encoding detailed semantics in the name a good kerneldoc comment is way better. When it comes to this point, looks the name is really a bit verbose. And yes, they have the same kernel doc to list_first_entry/list_last_entry. But at rculist.h, we have another similar routine with this suffix, i.e, list_fist_or_null_rcu(). Maybe we can figure out a better name and replace the existing code with it as list_first_entry_or_null() has not yet been widely used, it mainly being used in drivers code as per a grep against linux-next tree. $ grep list_first_entry_or_null -r linux-next ./drivers/acpi/acpi_ipmi.c: driver_data.selected_smi = list_first_entry_or_null( ./drivers/base/regmap/regmap.c: async = list_first_entry_or_null(map-async_free, ./drivers/base/regmap/regmap.c: async = list_first_entry_or_null(map-async_free, ./drivers/net/team/team.c: port = list_first_entry_or_null(team-port_list, ./drivers/net/ethernet/sfc/mcdi.c: async = list_first_entry_or_null( ./drivers/net/ethernet/intel/i40e/i40e_main.c: return list_first_entry_or_null(vsi-mac_filter_list, ./drivers/net/vxlan.c: rd = list_first_entry_or_null(f-remotes, struct vxlan_rdst, list); ./drivers/of/irq.c: desc = list_first_entry_or_null(intc_parent_list, ./fs/xfs/xfs_trans_priv.h: return list_first_entry_or_null(ailp-xa_ail, struct xfs_log_item, ./net/nfc/digital_core.c: cmd = list_first_entry_or_null(ddev-cmd_queue, struct digital_cmd, ./net/nfc/digital_core.c: cmd = list_first_entry_or_null(ddev-cmd_queue, struct digital_cmd, Thanks, -Jeff