Re: Status of revivesa

["Steven M. Bellovin" <smb%cs.columbia.edu@localhost>]

On Fri, 26 Sep 2008 14:56:05 +0100
Mindaugas Rasiukevicius <rmind%NetBSD.org@localhost> wrote:

> Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost> wrote:
> > christos%astron.com@localhost wrote:
> > 
> > > >I wonder if the binary compatibility for pthread is
> > > >a serious showstopper, i.e. it's really more important
> > > >than other new features like MPsafe syscalls or
> > > >WAPBL etc. for most ordinary users.
> > > 
> > > >Can't we defer revivesa to 5.1, as fixsa on 4.1?
> > > >Users who actually need working SA support could use
> > > >(forthcomming) 4.1 with fixsa and I guess there are
> > > >quite few such users.
> > > 
> > > Once you release 5.0 without revive-sa you have screwed the users
> > > who want to upgrade from 4.x to 5.0. Adding it to 5.1 or
> > > announcing that you are going to add it in 5.1 wil just hurt the
> > > adoption of 5.0 or just hurt the users who will not wait.
> > 
> > In the perfect world?
> > 
> > How many users will actually be screwed seriously with SA,
> > rather than pleased by advantages on other features?
> > 
> > Which actual application is expected to be run on 5.0 with revivesa
> > in the real world?  Is it really expected to work without any
> > benchmark?
> 
> We had accidentally broken libpthread compatibility (after move of
> some functionality to librt) in -current. It was noticed 6 months
> later, by one developer. So extent of the problem is probably not
> that large.
> 
I had a serious issue that I believe is related to threads; I was about
to post about it.

Briefly, I upgraded a -current system across some magic boundary that
broke Perl.  Fine; I expected that.  But that left things in a state
where I couldn't cleanly pkg_delete perl -- it deleted some stuff
from /var/db/pkg, but not all.  So -- 'make install' wouldn't work,
because perl was there.  'make replace' and 'make update' wouldn't
work, because it wasn't there.  And pkg_info simply dropped core.

This happened on two different machines...



                --Steve Bellovin, http://www.cs.columbia.edu/~smb