With BuildWorld() we get more results back (now we return the whole
model, including the false assertions).
Besides, now solving with BuildWorld() detects an invalid case:
when we supply a provided, the definitionDB shouldn't explictly supply
also the package that has to be provided. This would cause to 'shadow'
packages between repositories.
The test was invalid before, and shouldn't have contained A1. Moved the
test to Pending to inspect it further in subsequent dev iterations
Instead of using the DefinitionDB which supposedly contains only the
relations present in the online repositories. In this way the solver its
more consistent and tries to solve with only the internal definitions.
This also fixes quirks with luet upgrade --universe
In this way when we order, we always return the same solution order in
case there are weak deps.
The following is optional - it doesn't change the "correctness" of the
solver results: We add an extra edge between deps that
share common dependendencies. This makes the link more stronger and
balances the graph so it doesn't show different results for the same query, as they
could be shuffled as don't have a direct connection.
With this change the solver during install now considers only the part
of the tree which is required to calculate the solution, it doesn't
consider anymore World() as the search space.
The search space now is narrowed down to the packages that related to
the one which we are considering.
In this subset of changes we are also optimizing the Parallel solver
avoiding an useless loop.
This change boost overall performance on large datasets which don't
necessarly have relations touching the whole tree.
- Adds upgrade --universe and upgrade --universe --clean. It will
attempt to bring the system as much close as the content available in
the repositories. It differs from a standard upgrade which checks
directly that what is pulled in doesn't conflict with the system. In
this new way, we just query the SAT solver to decide that on our
behalf.
- Add uninstall --full-clean. It uses only the SAT solver to uninstall
the package and it will drop as many packages as required (including
revdeps of packages too.
- Don't sign installed packages during finalizer execution
- Enforce solver constraints: build ALO and AMO rules taking into account
that the current package might not be selected at all.
- Force uninstalls on upgrade
- Enable option to tell uninstall to ignore conflict with the analized system state,
as we don't want any conflict with the installed to raise during the upgrade.
In this way we both force uninstalls and we avoid to check with conflicts
against the current system state which is pending to deletion.
This is due to the fact that now the solver enforces the constraints
and explictly denies two packages of the same version installed.
- Adapt test as now we generate more constraints, which makes the solver more
noisy on the package that are explictly selected or not
It creates cycle and we don't want to output anything from the computation process.
We should handle output in different stages
Also create constructor for solver to be able to consume resolvers.