Re: packaging system (was: Re: GCC 3.3.2 kernel)

[Craig Dooley <craig@xxxxxxxxxx>]

On Sunday 02 November 2003 09:08, ichel Talon wrote:
> William Dean DeVries <look_in_message@xxxxxxxxxxx> wrote in
>
> news:pan.2003.11.02.07.31.31.787827@xxxxxxxxxxx:
> >     You know, you could put the needed headers in your source
> >     directory, and
> > if the compiler still can't find them I bet you could tell it
> > wheretolook. Just a thought.
>
> Yes i know that. This means however spending time editing makefiles and
> so on. What i call a pain in the ***

Not if the packaging system is designed for it.  Something like
find $pkgdir -name "*.so*" > plist
find $pkgdir -name "/usr/share/doc/*" > plist-doc
find $pkgdir > plist-dev

I personally think that the -doc is unneeded and could be put with -dev so 
that only runtimes are in the normal package.  How many gnome users need the 
gtk api reference or even know what it is?  If you know you need it, then you 
probably would know how to get it.

>
> >     As for splitting packages into 2 or even better 3 parts.
> > The port system is redesigned it could be made to have flags
> > which would tell it to install the other parts of packages by
> > default.  Thus making everyone happy.
> >     In regards to having three plist in a packages, having three
>
> I am still waiting good arguments why installing the whole stuff on hard
> disk causes problem. At present the cheapest disk is around 50 Gigs, in
> two or three years the disks will be so big as being practically
> infinite. Introducing the smallest complexity or inconvenience for
> solving a non problem is in my opinion a major form of stupidity. Even
> USB sticks are now up to 1 Gig. It is not the same as when disks were
> smaller than 1 Gig and one had to be careful. Being careful today and
> introducing a lot of problems for users is really really dumb.

Because not everyone went out and bought a brand new machine the day 
DragonflyBSD came out.  Granted, I sacrificed my primary work machine to it, 
and am happy I did, but not everyone wants to go buy something new to test 
out an OS.  In the server segment, yes, but not everyone.  I dont see ncurses 
wasting 50g, but it's just an options.  Also consider in debian the sizes of 
big packages.  libc6 is like 3.5 megs, I think libc6-dev is like 6.  There 
are still some people on 28.8, and I dont think they'd really like the extra 
half an hour of downloading if we did things like weekly binary snapshots of 
the tree.  Some people just dont have the need for dev packages and compilers 
in their systems.  Maybe im just talking about the linux desktop crowd here, 
but it's an option to the user.

> >     I am currently reading this thread and making a wish list of
> >     feature
> > for a new package management system.  I will probably post it later.
> > I also plan to look at the features of dpkg discussed here.  I hate
> > the system used in freebsd, but like the idea behind it.
>
> I would be very happy that you explain us why the ports system is so bad
> that you hate it. Personnally i have used it since FreeBSD 2.2.5 and i
> have been reasonably happy with it. On the other hand my own experience
> leads me to hate the packaging systems that one finds on Linux distros,
> even on Debian. It is not that i have anything against Linux itself,
> simply that the FreeBSD ports sytem has always worked infinitely better
> for me than what can be found on these distros. The better by far is the
> Debian one, and even this one causes more problems than it solves.
> In particular, that to make it reasonably safe, the Debian developers
> need to wait a very unreasonable time before updating their stable
> version. I prefer any day having recent programs, even with patches,
> than waiting three years to get them on a Debian system.

The port system is okay.  It's easier than slackware, and it's easy to install 
stuff, but not to keep the system up to day.  Debian stable is exactly that.  
Unless it's for security reasons, they pretty much dont update their software 
in stable.  Unstable IS up-to date, and is not unstable, but most people base 
their judgement of debian on the stable branch.  Portupgrade is such a hack.  
If anything, there should at least be a way to save your compiler flags when 
packages were built so if you REALLY dont want svgalib on your system, you 
dont get it when you upgrade something.  Binary updates are much easier on 
users, nicer on the wire, and more convenient.  In debian the openssl bugs 
took a 30k download.  In FreeBSD it required a cvsup thats about 400m for the 
whole tree, about 2 hours for a buildworld on my machine, and overwriting all 
your binaries since the last update, possibly introducing things you didnt 
know changed.  I love that the whole source tree is in one central 
repository, and it makes things so much easier to learn than hunting around 
for disparate source packages, but I think for distribution, binaries are 
much more convenient.  

-Craig
-- 
Craig Dooley										craig@xxxxxxxxxx

Attachment: pgp00000.pgp
Description: signature