.. and back to read-only after finishing with modifications.

fixes #512

if package arch is not set, do not append anything to repository.
if arch is set, it is appended to repository.

Use statfs() filesystem type, instead of the device IDs.

Packages without architecture should get the platforms default
architecture (so we are backwards compatible). Only the virtual
packages should get 'noarch' by default. Also print full path
to the index file which failed.

The atomization change broke virtual packages because they don't
have license or arch set.

Parse install_if from package metadata and include it in the
indexes. Also setup the reverse install_if dependencies when
loading a database. ref #443.

Actual install_if functionality is not yet implemented.

Just disable installation of packages using the new stuff. Also
flag lower case package info fields as non-critical and allow
installation even if that features is not supported.

 - implement a hash table for commonly shared fields such as
   license, version and architecture
 - use macroes to print blobs or pkgname-pkgver strings
 - fix some old cruft

Architecture is now:
 - parsed from .PKGINFO
 - written to index and installed db
 - appended to repository URL when fetching files

Unsigned indexes should not be needed anymore anyway.

.. instead of the pkgname-pkgver string.

id cache needs to be initialized early. and group database filename
is surprisingly 'group'.

Otherwise we end up using wrong uid/gid mappings when doing install
to alternate system root. Fixes #434.

We want remove dirs when they are unreferenced so we remove all dirs on
apk del, but we don't want remove dirs when closing database. So we make
removing dir optional when unreferencing it.

This partially reverts commit c7ffc96a.

fixes #406

Seems like recent eglibc requires that you include sys/stat.h

The trigger list can be reused after it's cleared, make sure
the list node is initialized all the time.

So open the db in read-only mode instead. This allows --simulate to
be run as non-root user.

The triggers are read during apk_db_unpack_pkg(). If we delete the
triggers list after then unpack we delete the triggers which is not
what we want.

This fixes bug introduced in ce3cf8bf

got broken to the previous array cleanup and grepping struct
definition only from headers.

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.

clean up the triggers properly, in proper order.

we do not create mirror repositories from other valid repositories,
not from what was installed locally.

add the missing variable.

otherwise regeneration of non-repository index might go wrong or we
might delete too many files from cache when doing 'cache clean'.

got broke few commits ago when apk_pkg_installed() call was
moved to happen after the package name has been read.

this is now mandatory after the overlay fixes. otherwise the
package will not get listed as installed.

turns out the logic does not work with overlays as expected due
to busybox symlinks being unmanaged. remove the useless option.

appears this is possible, so support it.

lookup proper directory instance as the tar output might not be
linear.

get list of overlay files from stdin, so those do not get overwritten.

less syscalls.

do not overwrite target if we have --never-overwrite and target exists.

ref #197

the old code treated a symlink to directory as file; it tried
to calculate regular has of it. fix this by: 1) using no follow
on migration and pruning stats, and 2) the helper function to
check if it's point to directory and not calculate hash in that
case. fixes #188.

ability embed description information to repository indexes
(e.g. repository name and version) and show it via "apk version -I".