This patch is the same approach as the one that want into 2.3.x. It
covers the functions in master that do not exist in 2.3.x.
Without this patch all of the spec tests for parser functions in stdlib
would instantiate their own scope instances. This is a problem because
the standard library is tightly coupled with the internal behavior of
Puppet. Tight coupling like this creates failures when we change the
internal behavior of Puppet. This is exactly what happened recently
when we changed the method signature for the initializer of
Puppet::Parser::Scope instances.
This patch fixes the problem by creating scope instances using the
puppet labs spec helper. The specific method that provides scope
instances in Puppet-version-independent way is something like this:
let(:scope) { PuppetlabsSpec::PuppetInternals.scope }
This patch simply implements this across the board.
Paired-with: Andrew Parker <andy@puppetlabs.com>
In Puppet 2.6.x there is a bug where a function may be incorrectly detected as
an rvalue when it is not, or not detected when it is. This means that in tests
the correct syntax for calling a function will be rejected. This disables
those tests on 2.6.x, as there is no straightforward way to write them to be
compatible with both 2.6.x and newer versions of Puppet.
* 2.3.x:
Make sure functions are loaded for each test
Use rvalue functions correctly
(Maint) Don't mock with mocha
(Maint) Fix up the get_module_path parser function
(Maint) use PuppetlabsSpec::PuppetSeams.parser_scope (2.3.x)
(Maint) Rename PuppetlabsSpec::Puppet{Seams,Internals}
(Maint) use PuppetlabsSpec::PuppetSeams.parser_scope
(Maint) Fix interpreter lines
Update CHANGELOG, Modulefile for 2.3.3
fix regression in #11017 properly
Fix spec tests using the new spec_helper
Update CHANGELOG for 2.3.2 release
Make file_line default to ensure => present
Memoize file_line spec instance variables
Fix spec tests using the new spec_helper
Revert "Merge remote-tracking branch 'eshamow/tickets/bug/13595_restrict_initialize_everything_for_tests' into 2.2.x"
(#13595) initialize_everything_for_tests couples modules Puppet ver
The test_helper code in Puppet now resets function state between each test.
This patch fixes two spec files where the function was not actually loaded in
the tests, causing them to fail.
A bug fix in Puppet exposed that several tests were using rvalue functions
incorrectly (this was not properly checked by puppet before). This fixes those
tests.
Without this patch applied the stdlib module has load-order issues with mocha
and rspec-puppet. The root cause has yet to be determined, but we've narrowed
it down to this description:
"If any rspec-puppet example groups run before parser function example groups
and the parser function example groups use mock() then you'll get this error:"
You can exercise this explicitly with:
rspec -fd spec/unit/puppet/{provider,type,parser}
This will ensure rspec runs all of the provider and type spec tests, which are
rspec-puppet ones, before the parser function specs are run. I should also
note we empted out the test in the file_line provider to be nothing except an
empty describe block and this was still sufficient to trigger the load order
error described here.
Failures:
1) function_get_module_path when locating a module should be able to find module paths from the modulepath setting
Failure/Error: mod = mock("Puppet::Module")
NoMethodError:
undefined method `mock' for #<RSpec::Core::ExampleGroup::Nested_14::Nested_1:0x107b946c0>
# ./spec/unit/puppet/parser/functions/get_module_path_spec.rb:21
# ./spec/unit/puppet/parser/functions/get_module_path_spec.rb:29
2) function_get_module_path when locating a module should be able to find module paths when the modulepath is a list
Failure/Error: mod = mock("Puppet::Module")
NoMethodError:
undefined method `mock' for #<RSpec::Core::ExampleGroup::Nested_14::Nested_1:0x107b81ea8>
# ./spec/unit/puppet/parser/functions/get_module_path_spec.rb:21
# ./spec/unit/puppet/parser/functions/get_module_path_spec.rb:34
3) function_get_module_path when locating a module should respect the environment
Failure/Error: mod = mock("Puppet::Module")
NoMethodError:
undefined method `mock' for #<RSpec::Core::ExampleGroup::Nested_14::Nested_1:0x107b6e808>
# ./spec/unit/puppet/parser/functions/get_module_path_spec.rb:21
# ./spec/unit/puppet/parser/functions/get_module_path_spec.rb:40
Finished in 1.53 seconds
326 examples, 3 failures, 1 pending
Paired-with: Andrew Parker <andy@puppetlabs.com>
This patch switches the spec tests for the get_module_path function to
use mock objects. The underlying Puppet::Module.find method has
reasonable test coverage inside of Puppet core so we might as well break
the tight dependency while we're fixing up the specs to use the new
parser scope.
The behavior of the parser function itself should still have complete
coverage even though the tests have switched to mock the implementation
inside of Puppet.
This patch is the same approach as the one that want into 2.2.x. It
covers the functions in 2.3.x that do not exist in 2.2.x.
Without this patch all of the spec tests for parser functions in stdlib
would instantiate their own scope instances. This is a problem because
the standard library is tightly coupled with the internal behavior of
Puppet. Tight coupling like this creates failures when we change the
internal behavior of Puppet. This is exactly what happened recently
when we changed the method signature for the initializer of
Puppet::Parser::Scope instances.
This patch fixes the problem by creating scope instances using the
puppet labs spec helper. The specific method that provides scope
instances in Puppet-version-independent way is something like this:
let(:scope) { PuppetlabsSpec::PuppetInternals.scope }
This patch simply implements this across the board.
The module PuppetlabsSpec::PuppetSeams has been renamed in the
puppetlabs_spec_helper gem to PuppetlabsSpec::PuppetInternals.
The method to obtain a scope object has also changed slightly. Without
this patch the spec tests will fail because the stdlib module is not
aligned with the spec helper gem. This patch fixes the problem by
matching up messages with their receivers in the spec helper library.
Paired-with: Andrew Parker <andy@puppetlabs.com>
Without this patch all of the spec tests for parser functions in stdlib
would instantiate their own scope instances. This is a problem because
the standard library is tightly coupled with the internal behavior of
Puppet. Tight coupling like this creates failures when we change the
internal behavior of Puppet. This is exactly what happened recently
when we changed the method signature for the initializer of
Puppet::Parser::Scope instances.
This patch fixes the problem by creating scope instances using the
puppet labs spec helper. The specific method that provides scope
instances in Puppet-version-independent way is something like this:
let(:scope) { PuppetlabsSpec::PuppetSeams.parser_scope }
This patch simply implements this across the board.
This time around I actually know why I'm doing this thanks to the
reminder from Nick Lewis.
Ruby will replace itself in memory with the executable listed in the
interpreter line if the string "ruby" is not in there.
Since /usr/bin/env rspec doesn't contain the substring "ruby", you can't
actually run ruby -W1 or whatever on the file.
This patch fixes the problem by making sure "ruby" is present,
preventing ruby from replacing itself in memory.
This patch back ports the file from the master branch. The spec tests
fail without this patch applied. This should make it easier to setup
Puppet settings using the puppet_spec_helper project.
The examples in the file_line resource documentation state the following
resource should work:
file_line { 'sudo_rule':
path => '/etc/sudoers',
line => '%sudo ALL=(ALL) ALL',
}
Without this patch the example does not work because ensure is not set
to present.
This patch fixes the problem by setting the default value of ensure to
present.
This patch back ports the file from the master branch. The spec tests
fail without this patch applied. This should make it easier to setup
Puppet settings using the puppet_spec_helper project.
When using MRI 1.9.x the stdlib spec helper does not invoke because
Puppet.settings.private_methods returns symbols instead of strings.
This is a problem because we need to set default configuration settings
like Puppet[:vardir] when using the compiler.
This patch fixes the issue by simply checking the Puppet version. This
seems a better choice than rescuing NoMethodError since the method might
be renamed or removed in the future.
Without this patch the spec_helper sends a message named
initialize_everything_for_tests to Puppet.settings. This is a problem
because Puppet 2.6.x does not have this method, only Puppet 2.7.x and
Puppet master have this method at this time and we're getting false
positive test failures.
This patch fixes the problem by looking before we leap. We test if the
private method exists before calling it. This works with Ruby 1.8.5 and
onwards and Puppet 2.6, 2.7 and master.
This should fix all of the failures I've caused in Jenkins today.
Without this patch the specified behavior of strings that are numeric
only and zero padded is unclear and untested in the spec tests. This is
a problem because it's not clear that range('00', '10') will actually
return [ "0", "1", ..., "10" ] instead of [ "00", "01", ..., "10" ]
This patch addresses the issue by providing explicit test coverage. If
the string conversion behavior of puppet changes, this test will begin
to fail.
Without this patch, the previous change set to the
validate_absolute_path() parser function contains Puppet 2.6
incompatible changes. stdlib 2.x is compatible with Puppet 2.6. These
changes are a problem because we cannot introduce backwards incompatible
changes in a minor release.
This patch fixes the problem by back porting the implementation of the
`Puppet::Util.absolute_path?` from 2.7.x to the function block itself.
The function block tests to see if `Puppet::Util.absolute_path?` will
respond and if not, falls back to the inline back ported implementation.
The spec tests have been updated to simulate the behavior of Puppet 2.6
even when running with Puppet 2.7.
I've seen a number of times the following error displayed to the end
user:
validate_re(): "" does not match "^true$|^false$" at /p/t/f.pp:40
This is an absolutely horrific error message. I'm to blame for it.
Users stumble over this quite often and they shouldn't have to go read
the code to sort out what's happening.
This patch makes an effort to fix the problem by adding a third,
optional, argument to validate_re(). This third argument will be the
message thrown back in the exception which will be displayed to the end
user.
This sets the stage for nicer error messages coming from modules we
write.
This patch is backwards compatible but is a new feature.
This patch adds a new function to validate if a string is an absolute
filesystem path or not.
The intent of this is to make this functionality generic and reusable.
Josh left a comment in another pull request I had:
If node_installdir or $node_vardir is not defined, then we should
raise an error, otherwise we may create a scheduled task to an
untrusted directory.
One solution to this comment is to validate the Puppet variable is an
absolute path.
Examples of this function look like:
function_validate_absolute_path
Using Puppet::Parser::Scope.new
Garbage inputs
validate_absolute_path(nil) should fail
validate_absolute_path([nil]) should fail
validate_absolute_path({"foo"=>"bar"}) should fail
validate_absolute_path({}) should fail
validate_absolute_path("") should fail
relative paths
validate_absolute_path("relative1") should fail
validate_absolute_path(".") should fail
validate_absolute_path("..") should fail
validate_absolute_path("./foo") should fail
validate_absolute_path("../foo") should fail
validate_absolute_path("etc/puppetlabs/puppet") should fail
validate_absolute_path("opt/puppet/bin") should fail
absolute paths
validate_absolute_path("C:/") should not fail
validate_absolute_path("C:\\") should not fail
validate_absolute_path("C:\\WINDOWS\\System32") should not fail
validate_absolute_path("C:/windows/system32") should not fail
validate_absolute_path("X:/foo/bar") should not fail
validate_absolute_path("X:\\foo\\bar") should not fail
validate_absolute_path("/var/tmp") should not fail
validate_absolute_path("/var/lib/puppet") should not fail
validate_absolute_path("/var/opt/../lib/puppet") should not fail
validate_absolute_path("C:\\Program Files (x86)\\Puppet Labs\\Puppet Enterprise") should not fail
validate_absolute_path("C:/Program Files (x86)/Puppet Labs/Puppet Enterprise") should not fail
Finished in 0.05637 seconds
23 examples, 0 failures
Without this patch every rspec run prints out the directory of the
spec_helper script.
I think this was just a debugging line or whatever that accidentally got
added.
Without this patch the PE modules don't have a way to identify a
filesystem path where it's OK to place variable data related to managing
the target node. This is a problem when a module like pe_compliance
needs to write a wrapper script to the node's filesystem.
This patch addresses the problem by exposing the node's Puppet[:vardir]
setting as a Facter fact.
This fact value will be set to `nil` if Puppet is not loaded into
memory. If Puppet is loaded, e.g. using `facter --puppet` or using
`puppet agent` or `puppet apply` then the fact will automatically set
the value to Puppet[:vardir]
The value of this setting is subject to Puppet's run_mode.
This patch implements a new utility method in the standard library
module named `Facter::Util::PuppetSettings.with_puppet`. The method
accepts a block and will only invoke the block if the Puppet library is
loaded into the Ruby process. If Puppet is not loaded, the method
always returns nil. This makes it easy to define Facter facts that only
give values if Puppet is loaded in memory.
Without this patch the root_home fact fails on windows. This patch
fixes the problem by only calling methods on the object returned by the
`getent passwd root` command if the object evaluates to true.
Because there is no root account on Windows the code block simply
returns `nil` which makes the Facter fact undefined on Windows
platforms.
The root cause of the failure is that we always expected the command to
succeed and return something useful, and it may not on all supported
platforms.
This function is used to validate a string is less than a maximum length. The
string, or array of strings, is passed as the first argument to the function.
The maximum length of the string is passed as the second argument.
It is useful to validate, for example, that Puppet is not sending a username
to a downstream system that the system cannot cope with, but that might not
cause an error message - for example, MySQL will not accept a username of
more than 16 characters. This enables a Puppet administrator to validate
the data that it may have been passed from upstream through, for example,
Hiera.
* Implement a simple destroy method.
* Add tests for it
* Refactor code, so file is actually read only once. However, due
to the nature how provider tests are run, we need to ensure that
the file is read before we open it to write it.
Without this patch some valid domain names are not covered in the spec
tests as Stig Sandbeck Mathisen <ssm@debian.org> points out. This patch
adds spec tests for the domains "." and "x.com" which are both valid.
This is an opportunity improvement since I'm in the code. Get rid of
instance variables in the spec test and replace them with a memo let
method block.
Between Ruby 1.8.7 p352 and p357 the way arrays were returned when using
keys and values in Ruby changed, and due to assumption about the
ordering our tests are now failing.
This patch fixes the issue by using the =~ operator matcher in rspec.
This matcher is implemented as RSpec::Matchers::MatchArray and performs
multiset equality matching of arrays. Order doesn't matter, but
duplicate values do.
This patch also switches @scope instance variables to memoized let
methods for clarity in the code.
Original Author: Ken Barber
Reviewed-by: Nick Lewis
This commit closes GH-29
The rspec code for the time function was trying to match the type to be a
'Fixnum'. Ruby will sometimes make this a 'Bignum' depending on its internals
and we can't rely on this to be true all the time.
This patch just makes sure the type is an integer instead.
OS X 10.7 introduced salted-SHA512 password hashes as opposed to the
older LANMAN + SHA1 hashes. To assist in generating properly-formatted
password hashes, this commit adds the str2saltedsha512() function which
accepts a single string argument (the password) and returns a
salted-SHA512 password hash which can be fed as the password attribute
of a user resource in OS X 10.7.
Spec tests are also added to ensure that functionality isn't broken with
future commits.
