On Sun, Feb 18, 2018 at 04:06:31PM +0000, Zbigniew Jędrzejewski-Szmek wrote:
On Sun, Feb 18, 2018 at 12:47:24PM +0000, Richard W.M. Jones wrote:
> On Sat, Feb 17, 2018 at 04:08:09PM +0000, Zbigniew Jędrzejewski-Szmek wrote:
> > On Fri, Feb 16, 2018 at 09:53:35PM +0000, Richard W.M. Jones wrote:
> > > sln (staticly linked ‘ln’) was removed from glibc recently:
> > >
> > >
https://bugzilla.redhat.com/show_bug.cgi?id=1531546
> > >
> > > The explanation for this was:
> > >
> > > "The sln program is no longer useful, so we will not ship it
anymore."
> > >
> > > and it was removed from Rawhide 3 days later.
> > >
> > > There are some %post scripts which still use this, notably
> > > ca-certificates, so that's now broken. But more to the point what do
> > > you do if you are in a situation where you need to make a symlink to
> > > save some core shared library on the currently running system?
> > >
> > > I also don't think rather fundamental, useful and old tools like this
> > > should be removed without discussion.
> >
> > Why is normal ln not enough? It should be installed and runnable on
> > pretty much any system, even during upgrades of glibc and other basic
> > libraries.
>
> The idea behind sln is that it works even if you've broken the libc
> symlink.
Right. But does this really happen? In my experience, 10 years ago
this kind of stuff happened, either because the tools were less
reliable, or rather more likely because the human operating them
didn't yet know how to operate them properly. I think we have all moved
far in the direction of systems-as-cattle, and when we get to the point
where libc is hosed, well, that's it, we provision a new installation.
The greater point was that our toolchain shouldn't make loads of
changes with minimal discussion. This is just one example of several
which have happened recently:
- removal of crypt
- removal of xdr_* functions (breaking ABIs!) and rpcgen
- adding flags which break CLANG
- changes which break very long-standing assumptions around linking
These things need to be signalled _long_ (years) in advance and
coordinated with users.
Rich.
--
Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
Read my programming and virtualization blog:
http://rwmj.wordpress.com
virt-top is 'top' for virtual machines. Tiny program with many
powerful monitoring features, net stats, disk stats, logging, etc.
http://people.redhat.com/~rjones/virt-top