The spec was setting the order's state to "complete" but didn't save
that state to the database. The new locking mechanism is was reloading
the order which loaded the cart state again. And since the order.next
method was mocked to just return true, the controller was trying to do
that in an infinite loop.
When two people tried to buy the same item at the same time, it was
possible to oversell the item and end up with negative stock.
Parallel checkouts could also lead to other random failures. This spec
is testing that scenario by starting two threads which would run into a
race condition unless they use effective synchronisation. The added spec
fails if the synchronisation is removed from the CheckoutController.
OrderUpdater#shipping_address_from_distributor uses order.address_from_distributor to set order.ship_address when order is not delivery: this will clear the ship address as it was done previously without setting an empty address like Spree::Address.default
If the order is allowed to retain a shipping_method_id, then subsequent
saves of the order will cause a new shipment to be initialised. Seems to
only happen for delivery shipping methods. This is undesirable because
fees for the new shipment will appear in the checkout summary, which is
not smart enough to recognise existing shipment fees and adjust the order
total accordingly.
Spree does not call after_<order.state> methods any more as of
https://github.com/spree/spree/pull/2557, so our #after_complete method
is never triggered and thus the order never reset.
This makes the condition:
```ruby
if current_order.andand.distributor == @order.distributor
```
in app/views/spree/orders/form/_update_buttons.html.haml return false
and as a result the "Back To Cart" button is not shown.
This commit resets the order (emptying the session[:order_id] and
creating a new order, aka. cart) right from the
CheckoutController#update rather than relying on infernal callbacks (of
what the Spree core team itself was unhappy about since long ago
https://github.com/spree/spree/issues/2488). There is the first place
where we know the order has been successfully completed.
