On Mon, Apr 21, 2014 at 7:49 AM, Adam Thompson <athom...@athompso.net> wrote: > On 14-04-21 09:36 AM, Philip Guenther wrote: >> >> Hmm, I haven't seen anything like that in many many years. $ENV used to be >> processed by non-interactive scripts, but that was a bug fixed in 2007. I >> would be interested in hearing details of specific cases where .profile is >> parsed by non-terminal software. > > I have developers using Ant deploy scripts that SSH into the target host > repeatedly, once for every build step. Ant does a reasonably good job of > emulating a terminal and handling strangeness in the output, but it's still > clearly "non-terminal software".
I don't understand: if it's it's invoking ssh in a way that gives it a pty on the remote side, then the stty's aren't a problem. If it's not, then it must be explicitly going out of its way to get a login shell without pty. Lacking the "why we did this" of doing, I don't see what problem it would be the correct solution for. > There's no reasonable way (that I'm aware of) for Ant to *not* cause a login > shell to occur when operating in this mode. > Many non-terminal things can process .profile: GDM & KDM, I believe, still > do so. Not sure about XDM. > > I believe that's not exactly what you meant... I agree that the standard use > case of: > .profile - login shells only > $ENV - interactive shells only > is still appropriate. > > This generally means, IIRC, that environment variables (ironically, given > the name "$ENV") should go into .profile, but things that screw with the > terminal should go into $ENV instead. > > Funny - I know intuitively where I want to put different things, but > actually writing it down and coming up with a clear set of guidelines is > surprisingly hard. That probably means my knowledge is incomplete. > > -- > -Adam Thompson > athom...@athompso.net
