We used to extract the string from the source code. We should have extracted
the content of the string. This is the difference between `"String"` and
`String`.
Puppet defined types would print the string representation of the ruby object,
and not representation you would see in puppet. This is the difference between
`Puppet::Pops::Types::PStringType` and `String`.
If a provider and a type with the same name conflict strings will overwrite the
documentation of one of them with the other. That is if both a provider and a
type are named apt_key, strings will write the type to doc/apt_key.html and the
provider to doc/apt_key.html. The fix is to write the provider to
doc/apt_key_provider.html and the type to doc/apt_key_type.html.
Parameter directives with no types will have `tag.type` set to nil. We cannot
append nil to an array, so instead, place it in the array and flatten the
array. If the property is empty the second entry in the array will be nil, as
we expect. If the `tag.type` is an array of types we will have and array with
the form:
[tag.text, type1, type2, type3, ...]
In a Puppet type if the user documents a parameter with the directive, then
Yard will read the parameter and think there is an empty docstring there. This
will overwrite the docstring we extracted.
The fix is simple -- overwrite our docstring with theirs, if our docstring
exists at all.
* Dunno, I just plowed through a bunch of features
* Expect puppet provider in stats output
* Fetch default values for Type params and props
* Detect allowed values
* Add allowed values to test
* htmlify scrubbed text
* Add features to Type html output
* Add infrastructure for types
* Add methods for generating lists, etc.
* Add provider code object
* Add provider handler
* Generate list for puppet provider dropdown
* Add puppet provider template
* Add provider details to puppet type template
* Get description properly for types
Adding a provider page and menu
* Add categories to html search bar
* Require provider handler and object classes
* Fetch provider code objects from registry
* Add function to generate the provider list
* Fetch providers from registry in monkey patches
* Add provider templates
* Add provider code object
* Add provider handler
* Add erb file to populate the provider list
* Don't emit type information for providers in html
* Add tests for provider handler
Refactor heredoc:
* Remove heredoc annotations
* Move heredoc functions into a heredoc helper
* Add heredoc helper class
Since we threw away all of Yard's warnings we are no longer checking that the
parameter names match for Ruby methods. Thus we need to override Yard's
method_details template with our own to trigger our warning function. However,
there's a catch. If this ruby method is in a Puppet 4x function, we don't want
our warning function to trigger because the user has already been warned. Look
in the registry to see if there is already a Puppet 4x function with the same
name registered.
Also, print errors to stderr instead of using log.warn.
Before this commit:
packages (Array[String])- A list of packages to be installed
With this commit:
packages (Array[String]) - A list of packages to be installed
Redirect Yard command line warnings to a log file called `.yardwarns`.
Yard warnings may be irrelevant, spurious, or may not conform with our
styling and UX design. They are also printed on stdout by default.
You can pass a block to the constructor of the hostclass. For some reason Yard
is yielding the block twice. Our code is not yielding the block because we only
use yield in one place which is unrelated. I don't think this is intended
behavior on yard's part, it certainly seems odd, but we can dodge the question
and not have to wait for an upstream patch by extracting the logic from the
block and running it after the object has been initialized.
I forgot to include the code which extracts the type information for the code
object in the defined types handler. Without this we do not extract type
information from defined types so warnings cannot be emitted and types in the
documentation are listed as Unknown.
Subclasses have their superclass as part of their namespace and objects in the
subclass's namespace must be registered first.
Reverse the list of parsed objects so
that the classes will be handled and registered in the correct order.
Addresses @hlindberg's comment after PR #35 was closed.
Use the error message from the exception which is actually exactly what I
wanted to say anyway.
The transformer comment matching regex matched all whitespace after a comment.
Lines which were effectively "blank" and just had a comment would be erased.
Markdown lists need to end with a blank line. This messed up markdown
formatting.
The culprit is the regex /^\s*#\s/ which matches all of the comment ' #\n',
however we want to leave the newline alone for markdown to parse.
We replace the regex with /^\s*#[ \t]/ which will only match tabs or spaces
after the hash and leave our beloved newline alone.
The previous iteration eagerly grabbed all parameters when any puppet function
was created. We did not retrieve type information and would grab parameters
from any helper functions!
* Parse the Ruby AST for dispatch blocks which specify type information. Parse
the commands and arguments in those blocks.
* If there are no dispatch blocks, parse the AST for a ruby function with the
same name as the puppet function being created.
Traverse the ruby AST to get method arguments and attach them to the code
object.
* Fixes spurious warnings issued by yard.
* Obviates the need for ugly regex code in the templates to retrieve
parameters from the source code.
Print the name of the parameter, the file, and the line number to stderr if the
parameter name does not match the name specified in the docs.
In order for this to be useful we need to present the user with the file name
and line number of the relevant parameters. This information needs to be
extracted from the code object and passed to the extract_param_details method.
Prior to this commit, puppet namespaces would be totally mangled when
processed by YARD since YARD is looking for Ruby namespaces where the
first letter of each segment must be uppercase.
In order to fix this, patch the relevant regular expression constants
in YARD so that they will accept segments that begin with uppercase
and lowercase letters.
An upstream pull request has been accepted into YARD to add another
regex constant which is necessary to complete this patch. However until
the next version of the YARD gem is released, we must include a temporary
hack to fix the problematic method in question (as it is not yet using
the added variable). Once the new version is released, the conditional
statement in code_ext/yard.rb can be removed entirely and all we will
need to do is patch the two regex constants.
Prior to this commit, the fix for the 3x function issue was trying
to create an empty hash by calling Hash[ [[]] ] which in Ruby 1.9.3
produced and empty hash. However this is not the case in Ruby 2.0 and
up. Therefor, fix up the code so that it does not rely on Hash[ [[]] ]
creating and empty hash.
Prior to this commit, the 3x function handler assumed that at
least two arguments (the name and one or more additional arguments)
were passed into newfunction when creating a 3x function. However
that is not actually required, the only argument needed is the name.
Update the 3x function handler so that it will not throw and exception
if only one argument is given.
Ian Kronquist