Re: git: 824bbb9a4082 - main - diff: consider two files with same inodes as identical

From: Ian Lepore <ian_at_freebsd.org>
Date: Thu, 07 Oct 2021 15:40:54 UTC
On Thu, 2021-10-07 at 09:32 -0600, Alan Somers wrote:
> On Thu, Oct 7, 2021 at 9:09 AM Mariusz Zaborski <oshogbo@freebsd.org>
> wrote:
> > 
> > The branch main has been updated by oshogbo:
> > 
> > URL:
> > https://cgit.FreeBSD.org/src/commit/?id=824bbb9a40820fb62bde0a91c0f13e0b894da149
> > 
> > commit 824bbb9a40820fb62bde0a91c0f13e0b894da149
> > Author:     Mariusz Zaborski <oshogbo@FreeBSD.org>
> > AuthorDate: 2021-10-07 15:07:00 +0000
> > Commit:     Mariusz Zaborski <oshogbo@FreeBSD.org>
> > CommitDate: 2021-10-07 15:07:00 +0000
> > 
> >     diff: consider two files with same inodes as identical
> > 
> >     Obtained from:  OpenBSD
> >     MFC after:      1 week
> > ---
> >  usr.bin/diff/diffreg.c | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/usr.bin/diff/diffreg.c b/usr.bin/diff/diffreg.c
> > index fc3c3406a073..995843f9e539 100644
> > --- a/usr.bin/diff/diffreg.c
> > +++ b/usr.bin/diff/diffreg.c
> > @@ -440,6 +440,10 @@ files_differ(FILE *f1, FILE *f2, int flags)
> >         if ((flags & (D_EMPTY1|D_EMPTY2)) || stb1.st_size !=
> > stb2.st_size ||
> >             (stb1.st_mode & S_IFMT) != (stb2.st_mode & S_IFMT))
> >                 return (1);
> > +
> > +       if (stb1.st_dev == stb2.st_dev && stb1.st_ino ==
> > stb2.st_ino)
> > +               return (0);
> > +
> 
> Checking st_dev is not correct.  It does necessarily bear any
> relation
> to the specific mounted file system.  It might, but that's up to the
> file system driver.  fusefs, for example, allows the server to
> populate that field with whatever it damn well pleases.  diff should
> use statfs instead, and check the f_fsid field.  That would probably
> work.
> -Alan

That may be a fusefs bug, then.  Posix states

   The st_ino and st_dev fields taken together uniquely identify the
   file within the system.

And I've seen real-world code more than once that uses st_dev equality
to decide whether to rename versus copy-and-delete a file.

-- Ian