[Libosinfo] [osinfo-db-tools 0/2] Start new osinfo-db-tools package

[Zeeshan Ali (Khattak)]

On Thu, May 19, 2016 at 9:36 AM, Daniel P. Berrange <berrange at redhat.com> wrote:
> On Fri, May 13, 2016 at 11:13:02PM +0100, Zeeshan Ali (Khattak) wrote:
>> On Fri, May 13, 2016 at 5:56 PM, Daniel P. Berrange <berrange at redhat.com> wrote:
>> > On Fri, May 13, 2016 at 05:35:15PM +0100, Zeeshan Ali (Khattak) wrote:
>> >> Hi Daniel,
>> >>
>> >> Thanks for the very long explanation (I especially appreciated the
>> >> ascii-art diagram) but I don't quite agree.
>> >>
>> >> While I don't see any harm in -db not depending on libosinfo, I really
>> >> don't see the benefit of going as far as to providing a separate
>> >> package for db tools. The -db does not necessarily be only data
>> >> either. I'd just keep the tools in -db or libosinfo and avoid the
>> >> maintenance and packaging headache.
>> >
>> > Actually I think it would make maintenance easier because it avoids
>> > tieing together things that would naturally evolve seprately. For
>> > example, there could be cases where a distro would want to update
>> > the DB management tools, but not pull in a new library version,
>> > or vica-verca. Letting these things evolve separately with their
>> > own release schedules will make life simpler for people consuming
>> > them, as they won't be faced with mutually exclusive decisions about
>> > which updates to pull in if they want to rebase some parts but not
>> > others.
>> >
>> > This may not sound like a big deal when all osinfo-db-tools contains
>> > is a couple of command line tools for validate & unpacking archives,
>> > but I've mentioned ideas before about providing tools to periodically
>> > query & download updated database version, so the osinfo-db-tools
>> > package is likely to gain more functionality & complexity over time.
>> > So I think over the long term the split will be beneficial to all
>> > involved.
>>
>> But why are we being religious about "no code in -db" package?
>
> It makes life much easier to get updates into distros. For example, if
> the package is 100% pure data, then for RHEL we can likely get an exception
> to allow updating of the data for every single release, as is done with
> updating things like pci.ids & usb.ids package. If the rebase included
> any amount of code / tools too, then there is increased risk and would
> need to go through the normal update approval workflow.

But distro packages do not need to have a 1-1 correspondence with
upstream. Downstreams already create multiple packages out of
libosinfo (-devel, -docs etc). I"m all of making lives easy for
packagers but I'd draw the line somewhere. Anyway, up to you.

-- 
Regards,

Zeeshan Ali (Khattak)