Subject: Re: bin/3011: ftp could be smarter with host:/path and URL's
To: None <Chris_G_Demetriou@ux2.sp.cs.cmu.edu>
From: Gordon W. Ross <email@example.com>
Date: 12/11/1996 09:49:14
> Date: Tue, 10 Dec 1996 19:19:51 -0500
> From: Chris G Demetriou <Chris_G_Demetriou@ux2.sp.cs.cmu.edu>
> > Actually, that is exactly the reason I have been lobying for an
> > implementation of the nsswitch code that removes all the direct
> > references to the YP stuff, etc. from libc (and using rtld to load
> > those when needed).
> Unfortunately, any way you look at it, that's quite likely a step
> backward on systems which don't support shared libraries, because if
> they don't support shared libraries they're not very likely to support
> rtld and dynamic loading of those modules.
Right, this would not make sense until rtld works everywhere.
> Sure, it helps the size of a "few" static executables on systems which
> already support shared libraries, but those aren't really _that_ big a
> problem, are they?
These are certainly not a "show-stopper" problems.
The two biggest problems I see are:
(1) A minimal program that calls gethostbyname is much bigger
than it needs to be. I deal with this for the RAMDISK root
by creating stubs for the YP stuff, etc. (Yuck!)
(2) When someone decides to add support for a new type of lookup
in nsswitch, then a new version of libc.so is required. It
would be nice if we could avoid having libc.so know what all
the lookup methods are.
Note: even if rtld is used for most lookup methods, I would keep the
simple file-based lookup code in libc.so (a special case) to make sure
"file" lookup works (i.e. /etc/hosts) even if rtld lookups are broken.