On Thu, Mar 13, 2008 at 04:19:46PM -0700, Mike Shapiro wrote:
> On Fri, Mar 14, 2008 at 12:07:17AM +0100, Joerg Schilling wrote:
> > If you don't know, such a library already exists since August 1986, so why 
> > should there be another one.
> 
> Solaris is allowed to have whatever namespace of whatever pieces of software
> we all choose.  The ARC's job is to own that namespace and approve changes.
> There is no common library with this name in Solaris.  This case adds one.

Well...

While working on the SQLite3 integration into SFW Mike Sullivan
requested that a heads up be sent to the w-team just in case there are
library name conflicts elsewhere.  Mike (Sullivan) went so far as to
assert that we should do this when we add any libraries where other
software might pick it up.  It's not enough to install the WOS and see
that there's not conflict because auto-configured items in, say, SFW,
might incorrectly think that your new library is something they should
use.

I'm not sure how likely that is though.  For example, there are two
things in SFW (Apache2 modules) that can use SQLite3, but won't use it
unless you add --with-sqlite3=... to the ./configure command line.

But it's always possible that conflicts of this sort will arise.

If we had tighter control over the Solaris library namespace then this
might not be an issue, but such control would imply more process for
folks integrating new libraries.  There's a price to pay either.

Finally, if there is a FOSS libscsi that predates ours and their ABIs
are different (which, unless we're doing a re-write, they will be) then
it would probably be better to avoid the conflict.  The problem with
this is that having to search freshmeat.net, sourceforge, etc..., gauge
popularity/deployment of any conflicts found, and so on means that the
Solaris library namespace has exploded and there is no more control.

Which brings me to the frightening possibility that anyone integrating a
new library into /usr/lib in Solaris would have to build and test *all*
the Solaris consolidations.  Yes, Mike (Sullivan) almost went as far as
requiring that I do just that, but settled on requiring that I build
SFWNV and ONNV and send a heads up to the w-team and the JDS team.

I would rather we avoid such a namespace explosion, but I'm not sure we
can, in the long-term anyways.

Nico
-- 

Reply via email to