[1067] in Kerberos_V5_Development
Re: [Doug Engert ] Krlogind and ss-962301
daemon@ATHENA.MIT.EDU (Marc Horowitz)
Mon Apr 15 17:44:23 1996
To: "Richard Basch" <basch@lehman.com>
Cc: Sam Hartman <hartmans@MIT.EDU>, krbdev@MIT.EDU
In-Reply-To: Your message of "Mon, 15 Apr 1996 15:36:54 EDT."
<199604151936.PAA00525@badger.lehman.com>
Date: Mon, 15 Apr 1996 17:43:54 EDT
From: Marc Horowitz <marc@MIT.EDU>
I think you misunderstand the issue, Richard. I'm aware of the
difference between capability testing and os conditionalizing, and we
agree on this point. I believe the right thing to do is to have two
major paths through the code: one for bsd-style pty mangling, and
another which uses streams. libpty should be able to encapsulate any
pty dependencies, or the library isn't doing its job. Then, using as
automated a method we can, select one or the other. However, this
doesn't always work: Some OS's have the hooks for streams, but they
don't actually work.
The issue here is hpux 10.x. /dev/ptmx and all the other stream hooks
seem to be present, but the vendor utilities don't use streams. I
don't think that it is wise for us to be using streams when the vendor
doesn't, because I'm afraid of what might break.
If everything else is done right, the only hpux 10.x specific place in
the system should be a platform name test in configure.in in libpty.
I do not see how this makes the code ugly, or has any risk of breaking
a new OS. Platform name tests are required for IRIX anyway, since
streams on that platform really is broken; this just adds one more
line to a case statement.
Marc
