Comment 25 for bug 1650688

This is really messy. As Pat pointed out, one option is to teach systemd to understand that there might be /etc/writable/localtime. I started with such a patch but its more involved (we need to tweak context_write_data_timezone() and get_timezone() to make it understand the different layout) and also very unlikely be taken by upstream (and a high likelyhood for conflicts).

A cheap(er) alternative might be to simply makee /etc/localtime writable and stop it being a symlink to writable/localtime. Then we patch systemd to retry the symlinking of that file in a non-atomic way if the atomic one failed. This patch is much more isolated and easier to carry. *If* all apps are using systemd to write this symlink we should be fine. It is still non-atomic but its just the timezone that would be wrong if in the middle of unlink/symlink() things go bad.