On Wed, 1 Aug 2007, Shawn Walker wrote:
> Thousands of FreeBSD servers pale in comparison to the number of
> GNU/Linux, Windows, Apple, and Solaris systems. Those systems
> prove that the binary model is a good one.
Not exactly, those numbers only illustrate that Linux has more
device drivers, and Solaris better large system support and more
enterprise-oriented software.
> The point is that binary configurations are the only ones that
> are truly supportable from a vendor perspective, since they have
> a known, dependable, signable, verifiable configuration.
How do you figure? Plenty of software vendors, mysql for one,
support both models. Both have known, signed, and verifiable
configurations. This is a red herring.
> While I agree that having an easy way to customise and rebuild
> binaries for a given distribution would be nice, I do not
> believe that it should be the primary method of package
> distribution.
When I run Synaptic, and select an application to install, it is
not "nice" that the list of dependencies is not optional. It is
not nice when I cannot de-install a package because of unneeded
dependencies. It is not nice when I have to write a specfile and
run rpmbuild for any difference from the binary package. And it is
really not nice to have dozens of unnecessary packages installed
on a server simply because the developer compiled-in what
dependencies he thought most end-users would want.
> Again, I support the idea of making what you want to do easy, I
> just don't support that as being a replacement for a
> binary-based system, because it is not.
How so? Given the Synaptic example. Both offer the exact same
front-end. The only difference is that dependencies would also
have check-boxes. If you left alone it could install the binary
version like you advocate. If, OTOH, you selected or deselected
additional features it would install the source version. The same
database of checksums, inter-dependencies, and other information
is kept nuder /var/{db/pkg,sadm,lib/apt,/var/lib/rpm,...} either
way. This is a wholly viable, proven, and superior model to
binary-based systems. But I am interested in specific examples of
where and why it might not be. The vendor perspective you describe
above is not a correct example though, if only because vendors
could easily decide to support only binary versions.
Any BSD-admins care to weigh in on this? So far the only
detractors appear to be sysadmins or users with no experience
managing ports-based systems.
--
Roger Marquis
Roble Systems Consulting
http://www.roble.com/
_______________________________________________
indiana-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/indiana-discuss
