GitLab

Remove the Ubuntu 18.04 box, add macOS 12.
Make macOS11 and 12 boxes available only for vmware_desktop, and make
that provider also the default.
Remove the S3 bucket mirror for boxes downloads, they've never been
there.

Those environment setup scripts are not maintained and have never been
particularly useful ever since running of jobs is done from the host
system.

Support simplified yaml syntax, and don't set "role" and "role_path"
values until we call a script provisioner. Simplify provisioner
extensions accordingly.

The problem is gone now that we configure things properly, and we don't
want instances to termiante when there's a shutdown happening from
inside.

Also, hide the public key from the log.

Using x11vnc since it allows attaching to the existing X11 session,
including an Xvfb one.

Now we have a GUI for an AWS instance running Linux, and can do visual
testing.

If the GUI desktop cannot start because there is no screen, then install
and run Xvfb instead. Revert to the multi-user systemd target and inject
the starting of the desktop environment into the run envelope.

This way we can run UI tests on Linux cloud instances.

The current boxes have a working X11 driver.

With the Azure and AWS providers, the cleanup method is never called
when terminating an instance. So we have to use an action hook. Since
the hook gets called a lot, so use a hash to do the work only once.

Optionally, clean up the disk before packaging. Not well tested yet.

Use default values for now, should make this configurable.
This could in theory be done via a regular local_command provisioner.

Subclass the providers we support, and add helper methods there.
Move code into provider specific file, and clean up global states
a bit.

As the first sub-command, get the ID of a created instance.

Even if neither "packages" attribute is an array, merge them. Otherwise
we end up with two script provisioners called "install" and vagrant will
only run one of them.

Add unit test.

We need to sort the tags correctly, already done on Unix via
fa115f9d.

On Windows, we can't reliably find vmrun, so always install virtualbox
until we find a way and time to wrestle with the batch script.

Those boxes can be published on the vagrant cloud, so we don't need to
maintain our own metadata files. The 'minicoin' organization on the
vagrant cloud is the namespace for the boxes when published, so adjust
workaround accordingly.

Add an info.json file with some more meta data.

Consolidate the box packaging script and add a script for publishing
everything to vagrant.

That file can then be loaded from the generic AWS code, so we don't need
to package a new box whenever a modification is needed. Instead, move the
instance and EBS configuration back into each box file so that we don't
need platform-specific logic in the generic code.

With this change, we no longer need to pass data around through global
variables.

It's not needed for anything else.

Vagrant changes (?) broke the test runner, so changing the code to what
the status command prints and adjust the runner accordingly.

It's not needed. Without the group, the basic provisioning works with
Ubuntu, OpenSUSE, and RedHat AMIs equally well.

The macOS root file system is read only, which we could work around on a
local VM by adding a link to /etc/synthetic.conf, but it's both hacky
and not feasibly for macOS machines living in the cloud. And adding stuff
to / is also awkward on Linux.

So move things to /opt instead, which on Windows because c:\opt\minicoin
which isn't any more weird that c:\minicoin. Adjust all the job scripts
and the internal explicit handling of the default shared folder.

Add ingress rules that allow only the host through from which the rule
was created. Test what's missing, and add needed rules, including those
for VNC/Apple screen sharing.

By some definition of "finish"; as long as there is a dedicated host
with auto-allocation and capacity in the account, it works.

Otherwise the group won't exist and the chown'ing of .ssh will fail.

https://github.com/mitchellh/vagrant-aws/issues/538 would block the
automatic download of AWS vagrant boxes unless we explicitly specify
an AMI, which in turn would then always overwrite the default from
the box.

Workaround as per the comment.

Due to https://github.com/mitchellh/vagrant-aws/issues/538 they won't be
downloaded automatically unless the minicoin.yml specifies an AMI, which
then needs to be the correct AMI.

So unfortunately the boxes are not quite as useful as they could be to
get started.

Store the public key we use in the global $settings and read it from
there in the box's Vagrantfile.

A minicoin machine could simply override the AMI to use, there should
be no need to have a different box for different Linux distros, or for
different Windows versions.

Move common code to general provider configuration so that boxes only
have to provide the AMI and the user data.

Passing the admin password for the box through to the box-specific
override is messy, but since all boxes of the same provider will have
the same password, we can use an environment variable.
To pass the admin password through the cloud_prepare plugin we can use
folder options.

Read the public key, and quote it directly in the user data script so
that it's written to the authorization file.