This is a subtle issue that was found that caused a panic. This should
solve things for now, but it would be wise to build embedded or
composite resources sparingly until we we're certain this would work the
way we wanted for all scenarios.
This improves some of the closing in the svc resource. This still needs
lots of improvements, and it's sort of terrible because it was some very
early code written.
This patch adds a util function, SessionBusUsable, that makes and returns
a new usable dbus session bus. If the svc bool session is true, the resource
will use a bus created with that function.
This adds the edgeable trait to the group resource and adds an
AutoEdges method which returns nil, nil. These changes are necessary
to allow UserRes to make autoedges to GroupRes.
This patch corrects the destination path in CopyFs to use the source's
base filepath, instead of the entire source path. Now copying /foo/bar
to /baz results in /baz/bar instead of /baz/foo/bar. This commit also adds
a test to verify this behaviour.
This requires breaking changes in gofmt. It is hilarious that this was
changed. Oh well. This also moves to the latest stable etcd. Lastly,
this changes the `go vet` testing to test by package, since the new go
vet changed how it works and now fails without this change.
This patch fixes a previously undiscovered bug which prevented
the use of the source field in the file resource. CheckApply was
returning early if obj.Content was nil. It is also necessary to
check that obj.Source is empty before returning, otherwise
syncCheckApply never runs.
This patch adds a shell test for net, which creates a dummy interface
and runs mgmt to bring it up and assign it with an address. It then
checks if the state was applied correctly. Finally, it runs mgmt again
to bring the interface down, and tests that it comes down and stays
down.
In some scenarios it is desirable to set the addrs and gateway
independently, i.e. if a default gateway is already set on
the machine. This patch removes the requirement to set them
together.
Per default, the Ruby gems renerate documentation in two distinct formats
during installation. By passing --no-ri and --no-rdoc, gem is instructed
to skip this step for both formats.
If the user needs documentation for any of the gems after all, they can
manually generate the docs themselves.
On Ubuntu, the apt-get install call to ruby, ruby-devel, and rubygems will
fail because there is no "rubygems" package in Ubuntu.
In Debian, this package is virtual only. In both cases, the ruby package
is sufficient. (See also https://packages.debian.org/jessie/rubygems)
W: GPG error: https://packagecloud.io/rabbitmq/rabbitmq-server/ubuntu trusty InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY F6609E60DC62814E
E: The repository 'https://packagecloud.io/rabbitmq/rabbitmq-server/ubuntu trusty InRelease' is not signed.
N: Updating from such a repository can't be done securely, and is therefore disabled by default.
N: See apt-secure(8) manpage for repository creation and user configuration details.
Hopefully this makes releases a little better for users.
In particular, this avoids listing old build artifacts in the SHA256SUMS
files when we make new releases, and users can now download them
directly.
Now to make a release you run: `make tag && make release`.
After the first make session ends, you'll have a new tag released
publicly, and then during the second make session, the release target
will notice this new tag, build some assets, and upload them!
This commit adds new make targets for rpm, deb, and pacman packages.
It also adds a phony target that uploads tarballs of the packages,
along with their signed (and unsigned) checksums to the github release
page. Once the current commit is tagged as a release, run `make release`
to build the packages and upload them to github.
The docker project absurdly *copies* all of the dependencies into the
vendor/ directory instead of using git submodules or avoiding
unnecessary vendoring entirely. We manually remove these changes until
they learn to use tools how they're intended.
As an aside, we recommend using a more intelligent, modern tool like
systemd-nspawn instead.
It turns out that some planned additions to the parser make it so that
the map type definition can be ambiguous. As a result, this patch
updates the definition so that the map definition is not confused with
an open curly bracket anywhere.
Thanks to pestle and stbenjamin for their help understanding yacc!
This adds the additional bits onto the class/include statements to
support or detect class recursion. It's not currently supported, but
I figured I'd commit the detection code as a variant of the recursion
implementation, since I think this is correct, and it was a bit tricky
for me to get it right.
This adds support for the class definition statement and the include
statement which produces the output from the corresponding class.
The classes in this language support optional input parameters.
In contrast with other tools, the class is *not* a singleton, although
it can be used as one. Using include with equivalent input parameters
will cause the class to act as a singleton, although it can also be used
to produce distinct output.
The output produced by including a class is actually a list of
statements (a prog) which is ultimately a list of resources and edges.
This is different from functions which produces values.
This means that it's legal to produce two compatible (usually identical)
resources without a compile error and without causing two of them to get
run. It's too bad puppet never got this right.
It's probably worth checking if this could be done for edges too, and if
the logic can be contained in the engine and not in the frontend.
