Paolo Bonzini wrote: > On 07/23/2011 11:12 PM, Jim Meyering wrote: >> >> In my experience, dev/ino is sufficient, as long as you're not using one of >> a few POSIX-violating fringe file systems (clearcase's MVS comes to mind). > > What about a few POSIX-violating fringe operating systems (Windows and > DJGPP come to mind)? :) For Windows we can write our own stat > function in cygwin, but for DJGPP I think we're in a bad situation...
AFAIK, DJGPP is not relevant these days. What Windows environment are you worried about? As far as I know, the vast majority of sys/fs pairs in use today have usable dev/ino. There used to be support in GNU rm for systems that lack useful stat.st_ino (it was 0 for every file), but when that and st_dev became important enough from a security/reliability standpoint, I ended up removing the work-around code. No one has mentioned it for at least 3 or 4 years. IMHO, any system without usable dev/ino is not a reasonable portability target. > The attribute tests fix it in practice, but since this is not an > optimization, perhaps gnulib's SAME_INODE module should provide an > indicator of the reliability of the macro.
