Re: [PATCH 1/3] btrfs: rst: remove encoding field from stripe_extent

[Johannes Thumshirn]

On 13.06.24 23:23, David Sterba wrote:
> On Tue, Jun 11, 2024 at 04:33:19PM +0000, Johannes Thumshirn wrote:
>> On 11.06.24 16:37, David Sterba wrote:
>>> On Mon, Jun 10, 2024 at 10:40:25AM +0200, Johannes Thumshirn wrote:
>>>> -#define BTRFS_STRIPE_RAID5	5
>>>> -#define BTRFS_STRIPE_RAID6	6
>>>> -#define BTRFS_STRIPE_RAID1C3	7
>>>> -#define BTRFS_STRIPE_RAID1C4	8
>>>> -
>>>>    struct btrfs_stripe_extent {
>>>> -	__u8 encoding;
>>>> -	__u8 reserved[7];
>>>>    	/* An array of raid strides this stripe is composed of. */
>>>> -	struct btrfs_raid_stride strides[];
>>>> +	__DECLARE_FLEX_ARRAY(struct btrfs_raid_stride, strides);
>>>
>>> Is there a reason to use the __ underscore macro? I see no difference
>>> between that and DECLARE_FLEX_ARRAY and underscore usually means that
>>> it's special in some way.
>>>
>>
>> Yes, the __ version is for UAPI, like __u8 or __le32 and so on.
> 
> I see, though I'd rather keep the on-disk definitions free of wrappers
> that hide the types. We use the __ int types but that's all and quite
> clear what it means.
> 
> There already are flexible members (btrfs_leaf, btrfs_node,
> btrfs_inode_extref), using the empty[] syntax. The macro wraps the
> distinction that c++ needs but so far the existing declarations have't
> been problematic.  So I'd rather keep the declarations consistent.
> 

Yes but all these examples have other members as well. After this patch, 
btrfs_stripe_extent is a container for btrfs_raid_stride, and C doesn't 
allow a flexmember only struct:

In file included from fs/btrfs/ctree.h:18,
                  from fs/btrfs/delayed-inode.h:19,
                  from fs/btrfs/super.c:32:
./include/uapi/linux/btrfs_tree.h:753:34: error: flexible array member 
in a struct with no named members
   753 |         struct btrfs_raid_stride strides[];
       |                                  ^~~~~~~