| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
You can now specify relationships directly in the language:
File[/foo] -> Service[bar]
Specifies a normal dependency while:
File[/foo] ~> Service[bar]
Specifies a subscription.
You can also do relationship chaining, specifying multiple
relationships on a single line:
File[/foo] -> Package[baz] -> Service[bar]
Note that while it's confusing, you don't have to have all
of the arrows be the same direction:
File[/foo] -> Service[bar] <~ Package[baz]
This can provide some succinctness at the cost of readability.
You can also specify full resources, rather than just
resource refs:
file { "/foo": ensure => present } -> package { bar: ensure => installed }
But wait! There's more! You can also specify a subscription on either side
of the relationship marker:
yumrepo { foo: .... }
package { bar: provider => yum, ... }
Yumrepo <| |> -> Package <| provider == yum |>
This, finally, provides easy many to many relationships in Puppet, but it also opens
the door to massive dependency cycles. This last feature is a very powerful stick,
and you can considerably hurt yourself with it.
Signed-off-by: Luke Kanies <luke@puppetlabs.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit is hopefully less messy than it
first appears, but it's certainly cross-cutting.
The reason for all of this is that we previously only
looked up builtin resource types from outside the parser,
but now that the defined resource types are available globally
via environments, we can push that lookup code to Resource.
Once we do that, however, we have to have environment and
namespace information in every resource.
Here I remove the Resource::Reference classes (except
the AST class), and use Resource instances instead. I
did this because the shared code between the two classes
got incredibly complicated, such that they should have had
a hierarchical relationship disallowed by their constants.
This complexity convinced me just to get rid of References
entirely.
I also make Puppet::Parser::Resource a subclass
of Puppet::Resource.
There are still broken tests in test/, but this was a big
enough commit I wanted to get it in.
Signed-off-by: Luke Kanies <luke@reductivelabs.com>
|
| |
|
|
|
|
|
|
|
|
| |
Basically, these classes (ResourceType and ResourceTypeCollection)
don't really belong in Parser, so I'm moving them to the
Resource namespace. This will be where anything RAL-related goes
from now on, and as we migrate functionality out of Puppet::Type,
it should go here.
Signed-off-by: Luke Kanies <luke@reductivelabs.com>
|
| |
|
|
|
|
|
| |
It's no longer necessary, given the new ResourceTypeCollection
class.
Signed-off-by: Luke Kanies <luke@reductivelabs.com>
|
| |
|
|
|
|
|
|
|
| |
We now use references to the ResourceTypeCollection
instances through the environment, which is much cleaner.
The next step is to remove the Interpreter class.
Signed-off-by: Luke Kanies <luke@reductivelabs.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This refactor fixes about a quarter of the test failures on master and (I
hope) will simplify some of the integration issues on the testing branch.
It is my best guess at The Right Thing To Do (or at least a step in that
direction) but I could be persuaded otherwise.
The basic idea is to take responsibility for maintaining scope hierarchy and
class_name -> class_scope mapping out of the compiler class and put it in the
scope class where it arguably belongs. To maintain the semantics, class
scopes are all tracked by the "top level" scope, though this could be relaxed
if the nesting semantics were ever needed.
If this winds up being the right thing to do, related routines (e.g. newscope)
should be sorted out as well.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit extracts these three classes into a single
ResourceType class in the Parser heirarchy, now completely
independent of the AST heirarchy.
Most of the other changes are just changing the interface
to the new class, which is greatly simplified over the previous
classes.
This opens up the possibility of drastically simplifying a lot
of this other code, too -- in particular, replacing the reference
to the parser with a reference to the (soon to be renamed)
LoadedCode class.
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| |
|
|
|
|
|
|
|
| |
This commit adds any external node classes to the classlist
at compiler initialization, so that at least those classes
will be isolated from any ordering issues encountered when
testing the contents of the class list during compilation.
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| |
|
|
|
|
|
| |
We need to be able to do compatibility testing, and this
allows us to do so.
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We were getting strange dependency cycles because our class structure
mirrored our scope structure. We can't change the scope structure
without switching from a dynamically scoped language to a lexically scoped
language, which is too big of a change to make right now. Instead,
I'm changing the resource graph so that all classes default to just
having an edge to the 'main' graph.
This will be a behaviour change for many, in that you were getting
automatic dependencies from this old behaviour, but it will bring
consistency.
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| |
|
|
|
|
|
|
| |
I also took the opportunity to clean up and simplify
the interface to the parts of the parser that interact
with this. Mostly it was method renames.
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| |
|
|
|
|
|
|
|
|
|
| |
This is mostly a configuration change, with some
code getting removed.
Also adding an extra require in Format;
Puppet::Provider requires Puppet::Provider::Confiner,
so the constant lookup is weird.
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
While evaluating the AST, catalog vertices are not always ordered
the same way on different run, leading to some tags (which should
have been applied in evaluation order) to not be associated with
some underlying resources.
This changeset change all accesses to resources inside the compiler
to always use an ordered (in evaluation order) list of added
resources.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
| |
|
|
| |
Signed-off-by: Luke Kanies <luke@madstop.com>
|
| | |
|
| |
|
|
|
|
| |
stacked metaparameter values do not result in all resources
that receive a given default also getting those stacked
values.
|
| |
|
|
|
|
| |
look resources up, which means there's no concern about
not finding the resource, which is where the nil was coming
from. We now just iterate over the vertices.
|
| | |
|
| |
|
|
|
|
|
| |
*sworn* I did this weekend). In the process, I fixed
a couple of bugs related to differentiating between
nodes and classes, and then cleaned up quite a few
error messages.
|
| |
|
|
| |
removing the bangs from 'add_vertex!' and 'add_edge!'.
|
|
|
since it's stupid to have a class named after
a verb.
|
