[bug-diffutils] bug#33467: bug#33467: bug#33467: bug#33467: bug#33467: "diff" can not scan directories with same dev/inode

[Paul Eggert]

On 11/30/18 5:09 AM, Delian Krustev wrote:
> This can be easily cross-reported to LKML. If "diff" developers blame the
> kernel/fs both parties should discuss it as its unacceptable a core utility to
> be unusable. ( I'm pretty sure though, that the kernel developers will be
> convinced that it does not need change. )

In that case I suppose we are at an impasse, and it's the poor users who 
will have to suffer. :-)

Seriously, POSIX is clear that dev+ino uniquely identifies a file, and 
many applications rely on this property. If a filesystem chooses to 
depart from standard behavior it hardly seems realistic to expect 
developers to change all their programs to work around the 
incompatibility, particularly if workarounds lead to slow behavior or 
incorrect behavior or both.

It shouldn't be that hard for the filesystem in question to invent inode 
numbers specific to the particular instance of the file. This sort of 
game has been played for years for Microsoft filesystems that lack 
inodes, and although the game can't be played perfectly in practice it 
seems to work well enough to keep users happy. Better, of course, would 
be proper support for inode numbers.