Previous implementation didn't work as intended because there was still
override in init. Current implementation instead of playing with
guessing separates default and instance setttings. Also update test
cases to handle this scenario correctly
it has been found that on big (>100) repos it starts lagging on reload.
This commit adds guard to avoid updating rows whose package statuses
were not changed
* store built packages in archive tree instead of repository
* write tests to support new changes
* implement atomic_move method, move files only with lock
* use generic packages tree for all repos
* lookup through archive packages before build
* add archive trigger
* add archive trigger
* regenerate docs
* gpg loader fix
* support requires repostory flag
* drop excess REQUIRES_REPOSITORY
* simplify symlionk creation
* remove generators
* fix sttyle
* add separate function for symlinks creation
* fix rebase
* add note about slicing
* smol refactoring of archive_tree class
* remove duplicate code
* fix typos
* few review fixes
* monor fixes and typos
* clean empty directories
* remove side effect from getter
* drop recursive remove
* ensure_exists now accepts only argument
* add package like guard to symlinks fix
* speedup archive_lookup processing by iterrupting cycle
* remove custom filelock
* fix naming
* remove remove flag from repo
* review fixes
* restore wrapper around filelock
* extract repository explorer to separate class
* docs update
* fix ide findings
Previous implementation was somewhat working in the most) scenarios, but
was super slow to handle permissions. However, it is actually very
limited operations in which the application can do anything, so it is
much easier to just drop privileged user to normal one
It has been found that during cold start (e.g. in docker container),
some permissions are invalid. In order to handle that, some operations
are not guarded with RepositoryPaths.preserve_root guard
In addition, it has been also found that in some cases (e.g. web server
start) migrations are performed on empty repository identifier which may
lead to wrong data (see also 435375721d),
as well as some unexpected results during database operations. In order
to handle that, now all watcher instances have their own databases (and
configurations)
