With the adjustments provided in #11455 the location of injecting
within the call stack has been changed slightly. With the entire
stack now being generated before execution instead of dynamic
hooks being wrapped around actions at run time, this update
ensures that the method is called correctly after the entirety of
the synced folders action has completed.
Windows commands that run over SSH are wrapped in a script that writes a
special marker to the two output streams (stdout and stderr). This
allows Vagrant to consume the output streams.
Unfortunately, this leads to a sort of chicken-and-egg problem where no
commands can be run before a wrapper script exists. For example, you
can't make a destination directory to upload the wrapper script without
first creating a wrapper script to make the directory. :)
This commit changes the behavior of the WinSSH communicator to assume
that the destination directory already exists for provisioning scripts.
It also moves the default `upload_path` from the shell provisioner
config so we can have OS-specific defaults.
Finally, it introduces a Windows-specific #upload method which will
properly use a Windows path separator on a non-Windows host.
This prevents any unexpected connection related error from breaking
the wait loop while windows reboots. If an underlying problem unrelated
to the guest is causing exceptions, the exception will still be raised
to the user, simply after the loop has exceeded the defined maximum
wait time.
Fixes#11238
Since the root file system is marked as read-only, attempting to
link the shared directory to `/vagrant` will fail. If the guest
path is on the root file system and APFS is used, create the
link as a firmlink instead.
This ensures that rsync can be installed on an Alpine Linux machine where the
apk cache may not be current.
Display a warning if the vagrant-alpine plugin is installed, since Alpine guest
support has been merged into Vagrant core.
CentOS 8+ and Fedora 30+ no longer have the alias "nfs" for "nfs-server"
systemd service.
This shouldn't break backward compatibility, since "nfs-server" service
is available on all supported redhat systems that have systemctl binary.
Fixes#10838
getent queries the system resolver for the hostname - but it's not
the resolver we're interested in. In fact, the hostname-to-be-set
may already exist in DNS (becuase DNS really is a nifty thing and
can do a lot of things which are not that possible with /etc/hosts
alone), in which case getent will "not fail" and vagrant will believe
the hostname had already been set.
Instead, query hostnamectl for the "static" hostname - that's the
one we will be setting, so we're ok IFF hostnamectl returns exactly
what we would be setting.
- nfs.service got recently removed in openSUSE Tumbleweed and calling service
restart nfs errors out on Tumbleweed. nfs.service has been an alias to
nfs-client.target for a very long time and can thus be safely substituted.
- all actively supported versions of openSUSE & SLE are using systemd now
=> no reason not to use systemctl
The workaround for the broken repository should be safe to be removed,
since the last affected Alpine version (<=3.3) EOL'd in November of 2017.
The remaining important commands can be split out into seperate calls
of sudo(), which removes the need for manual exit-code checking
(since it aborts by itself when a command fails) and makes the code
easier to handle in general.
