Re: [bug-gnulib] using dev_t and ino_t

[Paul Eggert]

address@hidden (Karl Berry) writes:

> Are there portability problems with using dev_t and ino_t?

They were in 7th Edition Unix <sys/types.h> so they're universal in
Unix-like hosts.  POSIX requires them.  However, I vaguely recall that
there are some non-Unix non-POSIX hosts that lack ino_t (because they
don't have inodes).  From the diffutils changelog:

 Fri Aug 27 06:59:03 1993  Paul Eggert  (address@hidden)

         * cmp.c (main): ...
         Avoid mentioning `dev_t', `ino_t' for portability to nonstandard hosts.

Also, coreutils has an AC_CHECK_TYPE(ino_t, unsigned long int).  But
interestingly enough gnulib does not check for ino_t -- perhaps a
portability glitch in gnulib?

Anyway, I wouldn't lose a lot of sleep over assuming them.

PS.  While we're on the portability subject, POSIX says that dev_t
might be a floating-point or even complex type, but ino_t must be an
unsigned integer type.  A little weird, but that's what it says.