[issue176] cd channel

[Jeff Johnson]

On Jun 18, 2006, at 11:38 AM, Christoph Thiel wrote:

> On Sun, 18 Jun 2006, Jeff Johnson wrote:
>
>>>> I just compared
>>>> ftp://opensuse.mirrors.skynet.be/pub/ftp.opensuse.org/opensuse/ 
>>>> distribution/SL-10.1/inst-source/content
>>>> to that on the SUSE cd. They are exactly the same. Also Fedora uses
>>>> repodata (rpm-md).
>>>
>>> rpm-md isn't used on the SUSE CDs or DVDs. It's only used in the ftp
>>> trees.
>>
>> Any particularly reason other than no round 'tuits? I'm patiently
>> collecting rpm-md wartlets, there are any number of deficiencies  
>> in the
>> representation.
>
> It was done this way, because when we started playing around with  
> rpm-md,
> there wasn't any "distributed media support" in rpm-md and we were  
> facing
> significant performance (mainly because of memory usage) issues with
> rpm-md. The YaST metadata format's main advantage is, that you can  
> easily
> seek through the metadata, which saves a lot of memory (which might be
> very limited at installation time, were you don't have swap yet).
>
> If you are interested in a list of problems that we experienced with
> rpm-md, we should probably take this discussion to the rpm-metadata
> mailinglist and try to agree on some fixes & extensions. (We at Novell
> have already added quite some extensions to rpm-md, which we use  
> for our
> online update repositories. Unfortunately it's still lacking some  
> decent
> documentation.)
>

I'm interested in problems, but I believe the politics of rpm-md have  
doomed the format forevermore,
better markup is needed instead. I've currently implemented --yaml in  
rpm-4.4.7 which is already
60% of the size of the equivalent (and rather terser than what is in  
rpm-md) dump in --xml. Even
the compressed yaml metadata if 10% smaller than the equivalent xml.

But by all means, suggest your extensions on rpm-metadata. I love to  
see vendor flame wars! ;-)

>
>>> The CD change problem is an known issue with the yast2 channel  
>>> support.
>>>
>>
>> Hmmm, I can describe how the multi-cd swapping problem is solved in
>> anaconda if you wish.
>
> That would be interesting -- I don't know if there is any logic in  
> smart
> to minimize the number of CD changes, and stuff like that.
>

For hysterical Conectiva related reasons smart overloads PreReq:  
dependency flags with
some CD switching semantics, I've forgot the details.

Since PreReq: has (finally!) been ripped out of rpm, a better  
implementation in smart will eventually be needed.

Otherwise smart has probably the best known (by me) solution for  
multiple cd swapping.

Not that anaconda gets it right, but at least the CD's don't (or  
didn't) swap.

73 de Jeff