On Feb 10 10:34, Eric Blake wrote: > On 02/10/2015 09:46 AM, Helmut Karlowski wrote: > >> The same test from the command line: > >> > >> $ stat -c %i /dev/tty - 0>/dev/tty > >> 327680 > >> 8912896 > >> > >> use any other (already-existing) file to see that the two numbers should > >> normally be the same. > > > > Don't know why, but cygwin is not the only OS that does that. I'd > > suggest to check for terminal or something. > > Umm, what other OS changes the inode from what you called stat() on to > what you open()d? Okay, there are some special device files like > /dev/ptmx that intentionally create a new device (and therefore new > inode) every time it is opened, but is /dev/tty in that boat? POSIX > requires /dev/tty to resolve to the process's controlling terminal. > > If there _are_ other OS that change the inode of the fd returned by > opening /dev/tty, that argues that any change in bash to special-case > /dev/tty should be made upstream, so I'm reluctant to write a > cygwin-specific bash patch. If there aren't other OS that change the > inode of /dev/tty, then it's better to figure out how to patch > cygwin1.dll to be more like Linux.
That's that. The other is that the way /dev/tty works on Linux and Cygwin is different. On Linux you open /dev/tty and the kernel redirects all calls to the descriptor to your current controlling tty. On Cygwin you open /dev/tty and the open call itself is redirected to an open call of your current ctty. That doesn't naturally mean that the stat/fstat inode numbers must be different, but they are right now due to the underlying behaviour. Whatever POSIX has to say about /dev/tty, I don't think this really qualifies as a bug. There are apparently just a couple of ingrained expectations floating around... Corinna -- Corinna Vinschen Please, send mails regarding Cygwin to Cygwin Maintainer cygwin AT cygwin DOT com Red Hat
pgpw6DZ6GCabD.pgp
Description: PGP signature
