* properly do absolute scoring now, the previous scoring where
preference could get reduced could have caused incorrect early
pruning of search tree
* backtracking is now separated from package state, and first
branching point is the decision if a name is left unassigned
or if something _has_ to be assigned. this allows multiple future
search tree optimizations like handling of common dependencies
early.
* merge common dependency names early to provide deeper forward
checking.
The array struct aligned size to 64 bit on x86_64 which caused bad things
to happen.
We use size_t to make sure the size element is correct regardless arch.
Solution found by Timo.
Instead of having a null pointer, use a dummy array which just
says the array is empty. This helps in multiple places of the code
which would otherwise need explicitly need to check first if the
array exists. This has been cause of multiple seg.faults in the
past as the array check is easily omitted.
This also removes (or fixes) all existing checks accordingly.
to never ever overwrite a while in the filesystem the user knows
about. it gives the impression of extraction succeeding even though
nothing was done. this is inteded to be used only for bootstrapping
with overlay.
this way we never change cwd, and relative filenames are always
parsed consistently. this also helps filename construction in many
places. this patch also changes '--root' to override location of
all configuration to be in the new root. previously it depended
on the file which one was used.
That will make the upgrade prefer packages available in repositories.
This is good if one want's to downgrade packages by removing an experimental
repository. Or to force re-install of locally built vs. repository version
when the package version are same, but checksum is different. Fixes#51.
Calculate changesets directly by stabilizating the package graph instead of
recalculating the whole graph and then diffing (similar approach as seen
in 'smart' package manager). The algorithm is not complete: defferred
search space forking is missing. So you don't always get a solution on
complex graphs.
Benefits:
- usually the search state tree is smaller (less memory used)
- speed relational to changeset size, not database size (usually faster)
- touch only packages related to users request (can work on partitially
broken state; upgrades only necessary packages, fixes#7)
Also implemented:
- command prompt to confirm operation if packages are deleted or downgraded
- requesting deletion of package suggests removal of all packages depending
on the package being removed (you'll get list of packages that also get
removed if you want package X removed)
- option --simulate to see what would have been done (mainly for testing)
- an untested implementation of versioned dependencies and conflicts
A lot has changed, so expect new bugs too.
Checksum of installed is computed on the fly when extracting them
and it'll be saved to fdb. When installing config files those are
diverted with suffix .apk-new if earlier version of same file with
local changes exist.