Bundle registry file monitoring clashes with its own operations

[dsd]

There are several races within bundleregistry.py because the gio file monitor does not collaborate with the install/remove/upgrade operations in the file.

For example, on an upgrade, the code first uninstalls the old version (by deleting the whole activity dir tree) and then calling self.remove_bundle() to remove it from self._bundles. However, the file monitor will *also* notice the file deletion (uninstallation) and call self.remove_bundle. One of the two will fail.

Then, continuing with the upgrade, install() is called to install the new bundle, first unpacking the bundle and then calling self.add_bundle(). However, during the unpack, the file monitor will also note the presence of the new activity and call self.add_bundle(). One of the two calls will fail as the bundle is already present.

I suspect this is the cause of ​http://dev.laptop.org/ticket/11373

This behaviour can be seen by sprinkling a few debug log messages in bundleregistry.py then upgrading an activity via the journal or the activity updater.

Reproduced with sugar-0.96.3.

[JerryV]

related ticket with patch ​http://bugs.sugarlabs.org/ticket/3342

[dnarvaez]

dsd fixed this recently.