[packagekit] thinking about moving forward with libhif

[Colin Walters]

On Wed, May 7, 2014 at 2:53 PM, Richard Hughes <hughsient at gmail.com> 
wrote:
> On 6 May 2014 14:20, Colin Walters <walters at verbum.org> wrote:
>>  Would that make sense?
> 
> I think making ostree use libhif for installation too is a good idea.

I'm interested in that, but I'm not sure how practical it's going to be 
in the short term.  The thing is that rpm-ostree is going to need to 
poke deeply into the RPM structure.  See:

http://lists.rpm.org/pipermail/rpm-maint/2014-April/003682.html

I'll need to be able to unpack the RPMs into a scratch area for 
inspection before attempting to install them.  Furthermore, I want to 
keep the RPMs inside the local ostree repository itself, instead of in 
a lookaside cache.

For example, with your normal package manager flow (yum/apt-get), they 
do depsolving, and download packages into a scratch area.  Then once 
all downloads are complete, they are all unpacked in serial.

This leads to potentially large peak space usage.  The way I have 
planned for rpm-ostree is quite different - we unpack the packages as 
they're downloaded and stage the content into the de-duplicating ostree 
repository.  Each object is carefully fsync()'d to disk at this time, 
the same way "ostree pull" works.

Unpacking as they're downloaded will also help me analyze them per 
above.

It seems difficult to capture this flow in libhif versus the 
traditional flow.  Maybe you could scatter the API with callbacks or 
something, but I think what I want in the short term is a convenient 
GObject wrapper around hawkey/librepo, which libhif is a good starting 
point.

> I'll have some more example code in a few days time.

Cool!  Looking forward to it.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/packagekit/attachments/20140507/45141ffb/attachment.html>