Architectures are now a list and pyalpm's pycman was not able to handle
this. For backwards compatibility convert a string to a list.

As syncdb's are not signed in Arch Linux providing SIG_DATABASE returns
no search results.

Fixes: FS#71146

In 67596333 we removed the total download callback as it no
longer exists in pacman but forgot to clean up enum for it leading to
the a segfault and callbacks being wrong.

Minor update to resolve compatibility with libalpm callback api.

For frontends to manage their callback state libalpm added a void
*context argument as this is ABI breakage.

This release adds compatibility with libalpm v13 and is not backwards
compatible.

Pacman now supports multiple 'Architecture' values and change the
setter/getter library functions which now accepts a list of
architectures.

The "total download" callback has been removed from libalpm in
1e60a5f00674a9a9b0

alpm_db_update now supports updating multiple databases at once the so
called parallel download feature. This requires an adjustment of the
alpm_db_update function call.

alpm_db_search now returns an int and set's pm_errno when the result is
-1 add a todo as this needs to be handled later

Allows us to test our next branch periodically with a Gitlab CI weekly
schedule.

It is easier to read if you don't need to try to figure out the flow of
a list of strings, and the non-issueful setuptools.Extension()
invocation, as they relate to __main__.

It's declarative and easier to read, and comes with convenient helper
functions for reading files.

The source code headers all state "or later", though some specify 2+ and
some specify 3+.

Evgeniy Alekseev authored 3 years ago and Jelle van der Waa committed 3 years ago

Signed-off-by: Jelle van der Waa <jelle@archlinux.org>

The code in find_file never matched the files itself but a tuple of
file, 0, 0).

Fixes: FS#69414

This function can't be tested and it would be testing CPython.

The current pyalpm implementation has no proper refcounting for Package
objects. Add refcount handling for the Database Object in the Package
model and Decref when the package object goes out of scope.

The current pyalpm implementation does not handle proper refcounting
which means the alpm_handle was never freed otherwise accessing an
object when the Handle goes out of scope leads to segfaults.

This means everything needs to keep a reference to the pyalpm Handle
object for every object created from it. The database object now
receives a Handle and increases the refcount for it to stop it from
being removed by the Python garbage collector.

Signed-off-by: Jelle van der Waa <jelle@archlinux.org>

Python's tarfile can actually generate tarfile's in memory which makes
it possible to remove the terrible chdir/cwd code and do less I/O.