Date: Fri, 6 Sep 2019 13:56:28 +0100
>> I agree with this. This is why I support a handle approach rather than
>> a byte based approach.
>
> I was very clear in the subject of the email: filenames for interchange. I
> knew the discussion of handles would come about.
>
> You can't store a handle in a file, not in any OS I am familiar with.
Actually, you can, every inode is uniquely identified on a system by
st_dev + st_ino. What you do is convert those into hex, and hardlink
your inode into some common directory e.g. /store/inodes/HEX. Your JSON
file then just lists the unique number identifying the inode, no paths.
P1031 LLFIO makes that sort of stuff trivially easy to implement. But as
WG21 hasn't chosen a direction on modern i/o yet, it should be omitted
from the discussion of P1689.
Niall
>> a byte based approach.
>
> I was very clear in the subject of the email: filenames for interchange. I
> knew the discussion of handles would come about.
>
> You can't store a handle in a file, not in any OS I am familiar with.
Actually, you can, every inode is uniquely identified on a system by
st_dev + st_ino. What you do is convert those into hex, and hardlink
your inode into some common directory e.g. /store/inodes/HEX. Your JSON
file then just lists the unique number identifying the inode, no paths.
P1031 LLFIO makes that sort of stuff trivially easy to implement. But as
WG21 hasn't chosen a direction on modern i/o yet, it should be omitted
from the discussion of P1689.
Niall
Received on 2019-09-06 14:56:31