RepoSack: New API to reload system repo

[m-blaha]

Introduce a new API to clear all solvables from the system repo and
reload fresh data from the RPM database.

Re-creating the Base object can be resource-intensive, and there is a
use case in dnfdaemon where the current session (which wraps around the
Base instance) needs to be reused even after executing an RPM
transaction - whether for queries or running another transaction.

For: #1653

[m-blaha]

@jrohel @kontura do you see any issues with this?
It can also be used by dnfdaemon to handle messages from rpm-plugin-dbus-announce rpm plugin.

[kontura]

@jrohel @kontura do you see any issues with this? It can also be used by dnfdaemon to handle messages from rpm-plugin-dbus-announce rpm plugin.

It does make me feel uneasy. It basically invalidates the pool right? Libsolv ids might have changed. So any Package object, Goal or PackageQuery in progress should be invalidated as well.

If I understand correctly for example existing Package could simply transform into a different one after the reload is used. I could see this causing quite hard to find bugs.

[m-blaha]

Yes, that's right. Any place that keeps an id of an installed package is invalid after this call.

[jrohel]

Code from libsolv:

  if (reuseids && repo->end == pool->nsolvables)
    {
      /* it's ok to reuse the ids. As this is the last repo, we can
         just shrink the solvable array */

It looks like reuseids works only when the repo is last. DNF places the system repo first. Behind it are available repos. The system repo is last only if it is the only one (we don't have available repos loaded).

Note:
OK. Maybe after the first reload the system repository will be the last one.

[jrohel]

You are introducing a new API method void reload_system_repo(). DNF saves the system repository to the pool first, followed by the others. Thus, installed solvables have a lower id than available. What do you expect to happen after calling the libsolv function repo_empty(repo, true) when there are other repos after the repo? Will space and solvables ids be able to be reused? I'm not sure about that. And what if the new system repo is larger (a package has been installed)? Surely the repo won't fit in the original space. Will the entire repo be saved to the end of the pool? Or will it be scattered in the pool - part in the original location and part at the end of the pool?

When we sort packages and they have the same nevra, the solvable id comes into play. Thus, moving the system repo to the end affects the sorting of the packages. The question is how much does it matter.

In general we are getting into the untested area of using libsolv.