Re: [PATCH v3] tst-ttyname: skip the test if failed to become root

[Luke Shumaker <lukeshu at lukeshu dot com>]

On Sat, 30 Dec 2017 19:11:52 -0500,
Dmitry V. Levin wrote:
> On Sat, Dec 30, 2017 at 06:09:22PM -0500, Luke Shumaker wrote:
> > Florian Weimer wrote:
> > > I don't think the glibc test suite is supposed to pass in such an
> > > environment [without ptmx].  If you don't provide /dev/null, /sys,
> > > or /proc to the tests, some of them will fail as well.  I still
> > > think that the current test accurately reflects the inadequacy of
> > > your test environment.
>
> Also, one cannot declare these environments unsupported in master branch
> without reaching a consensus.

On Fri, 29 Dec 2017 08:55:17 -0500,
Florian Weimer wrote:
> * Dmitry V. Levin:
> > It used to work perfectly for at least 15 years, and it still works
> > when tst-ttyname is fixed.
> 
> This just reflects that we did not have sufficient test coverage for
> the PTY code.

Obviously, I'm the outsider here; and can't authoritatively comment on
whether glibc supports environments without ptmx / UNIX 98 PTYs, but
from looking over the codebase:

 1. There is insufficient test coverate of posix_openpt and friends to
    conclude that Linux environments without ptmx were previously
    supported.

    The only other tests that call posix_openpt or other PTY-opening
    functions are

     - debug/tst-chk1.c
     - login/tst-grantpt.c
     - login/tst-ptsname.c
    
    I don't have a good idea of what tst-chk1 does, or why it doesn't
    error if posix_openpt fails; but the other 2 ignore an error
    because "maybe the system does not have SUS pseudo terminals", and
    the posix_openpt bit is just one subtest case.

    Which is to say that posix_openpt is previously untested by the
    test suite.

    But, Linux does support SUS / UNIX 98 pseudo terminals; and
    tst-ttyname lives in sysdeps/unix/sysv/linux.  Linux has no config
    option to disable ptmx.  The only variable there is whatever sets
    up /dev.

 2. The Linux implementation of getpt (a GNU extension) first tries
    posix_openpt to get a UNIX 98 PTY, but if that fails, it falls
    back to trying to get a BSD PTY.  Which suggests that glibc
    supports Linux environments without ptmx.

I find #2 fairly compelling.  That is to say, I've changed my mind,
and do support applying and backporting a patch to EXIT_UNSUPPORTED if
posix_openpt returns ENOENT.

-- 
Happy hacking,
~ Luke Shumaker