On Fri, 23 Mar 2007, Jon Peatfield wrote:

> I've been (fairly) quietly listening to the various threads about 
> repositories in SL5, and have a couple of questions about things. 
> Apologies in advance if what I'm asking about is something I should be 
> able to find in the notes etc, and I certainly don't want to distract 
> people from the SL5 beta work...
> 
> I've seen mention that the new anaconda can be told to look at several 
> repos during install, and that these will use the same info generating as 
> is needed for yum to also be able to use them later.
> 
> How will one tell anaconda the repos to use?  I see the kickstart 'repo' 
> option to add extra repos, but is there also a similar mechanism for 
> manual installs?

There was a option in the installer to add these extra repos in Beta2(very 
alpha for SL5) so I expect it will be in SL5.  This option was not on in 
the first "very alpha" , not sure if we turned it on in the 2nd "very 
alpha" but it should be on in the real SL5 releases.  The Upstream vendor 
does not show this option in their Beta of 5.

> 
> Will the anaconda stuff cope with repos which provide different versions 
> of packages (e.g. updates), or packages which clash/obsolete ones in the 
> 'main' repo?  (basically does it have a complete dependency solver?)

It is a complete dep solver,  but sometimes you did not get what you 
expected because of newer versions where you do not expect them.

> 
> As someone who hasn't used yum much before, can someone point me at a 
> gentle introduction to constructing the repo data that it and anaconda 
> will need?

Put your rpms in a directory.  Go to that directory and type

createrepo .

> 
> Previously (in SL3,4 and even back using RH) we constructed the install 
> trees by taking the latest version of each package from various sources, 
> and using genhdlist/pkgorder/genhdlist but I'm assuming that with the new 
> anaconda features that won't be needed any more.  Is that right?

You will have to do a "createrepo" instead.  Same idea different utility.
I suspect I will still do pkgorder as if the data does exist then the 
installer will use it during a non iso install.
> 
> In 5rolling/i386/SL/build/scripts/ I see code in build.hdlist.sh (say) 
> which does essentially pkgorder/createrepo so I guess that the pkgorder 

pkgorder is mostly needed for the iso's images where order matters.  In 
one of the Updates to 4 the Upstream Vendor broke the pkgorder code in a 
way that required it for a non iso image install.  So I assume they have 
fixed that bug and in reality it is not needed for a non iso install 
anymore but will probably still do it.  The createrepo will still have to 
be done.

> stuff still needs to be done like before, but that worries me about 
> whether anaconda still needs it in which case the extra repos may be 2nd 
> class...
> 

pkgorder was only needed to determine order of package installs.

No 2nd class here.

> 

-Connie Sieh