Tests Coverage Status

Kitchen lets you modify the structure and content of XML files. You create a Recipe with instructions and bake it in the Oven.

Full documentation at


Add this line to your application's Gemfile:

gem 'openstax_kitchen'

And then execute:

$ bundle install

Or install it yourself as:

$ gem install openstax_kitchen

Two Ways to Use Kitchen

There are two ways to use Kitchen: the "generic" way and the "book" way. The generic way provides mechanisms for traversing and modifying an XML document. The book way extends the generic way by adding mechanisms that are specific to the book content XML produced at OpenStax (e.g. the book way knows about chapters and pages, figures and terms, etc, whereas the generic way does not have this knowledge).

We'll first talk about the generic way since those tools are also available in the book way.

Generic Usage

Kitchen lets you modify the structure and content of XML files. You create a Recipe and bake it in the Oven:

require "openstax_kitchen"

recipe = do |document|"div.section").each do |element| = "section"

  input_file: "some_file.xhtml",
  recipes: recipe,
  output_file: "some_other_file.xhtml"

The above example changes all <div class="section"> tags to <section>.

The document above is a Kitchen::Document and the element is a Kitchen::Element. Both have methods for reading and manipulating the XML. You can of course name the block argument whatever you want (see examples below).

The search method and enumerators

search takes one or more CSS and XPath selectors and returns an enumerator that iterates over the matching elements inside the document or element that search is called on.

The enumerator that is returned is an ElementEnumerator which is a subclass of Ruby's Enumerator. Enumerators are also Enumerable which gives you a bunch of methods you can call on enumerators like:

  • count - get the number of matching elements
  • map - form a new array using the matching elements
  • each - do something with each matching element
  • first - return the first matching element
  • etc etc

Here's an example calling search on a document and then calling each on its result:"div.example").each do |div| # find all "div.example" elements in the document
  div.add_class("foo")                  # add a class to each of those elements"p").each do |p|           # find all "p" elements inside the "div.example" elements = "div"                      # change them to "div" tags

Clipboards, cut, copy, and paste

When baking our content, we often want to move content around or make copies of content to reuse elsewhere in the document. Kitchen provides clipboard functionality to help with this.

Every document holds a set of named clipboards. You can cut and copy to these named clipboards:"div.example").each do |div|
  div.cut(to: :my_special_clipboard)
doc.first("p").copy to: :foo

And then in some code where you are building up a string of HTML to insert you can

new_html = doc.clipboard(name: :my_special_clipboard).paste

cut puts the element on the clipboard and removes the original from the document. copy leaves the element in the document and puts a copy of the element on the clipboard.

Instead of using named clipboards, you can also pass any Clipboard object to these methods:

my_clipboard ="div.example").each do |div|
  div.cut(to: my_clipboard)

new_html = my_clipboard.paste

This is often the better way to go because if you use the named ones in the document you have to remember to clear them before you use it to not get stuck with whatever you left there the last time you used it.

ElementEnumerator also provides extra clipboard-related methods to make your life easier. Instead of writing"div.example").each do |div|
  div.cut(to: :my_special_clipboard)

You can say"div.example").cut(to: :my_special_clipboard)

The same applies to copy and these methods also work with passed-in Clipboard objects. If you don't pass in a clipboard name or a Clipboard object, these methods return a new Clipboard containing the cut or copied content:

a_new_clipboard ="div.example").cut

Clipboards are also Enumerable so you can call the enumerable methods (count, each, etc) on them.

When elements that were copied are pasted (or when elements that were cut are pasted more than once), Kitchen will update the IDs of pasted elements to keep them unique. Kitchen adds _copy_1, _copy_2, etc to IDs to make this happen. The _copy_ prefix is configurable (or at least close to it).

If you want to remove an element (or all elements matched by an enumerator) but NOT put those elements on a clipboard, you can use the trash method:



A document also gives you access to named pantries. A pantry is a place to store items that you can label for later retrieval by that label. "some text", label: "some label"
doc.pantry.get("some label") # => "some text"

The above uses the :default pantry. You can also use named pantries:

doc.pantry(name: :figure_titles).store "Moon landing", label: "id42"


Oftentimes we need to count things in a document, for example to number chapters and pages. A document provides named counters:


See book-oriented usage for a better way of counting elements.

Adding content

In kitchen we can prepend or append element children or siblings:

# <div><span>Hi</span></div> => <div><span><br/>Hi</span></div>"span").first.prepend(child: "<br/>")
# <div><span>Hi</span></div> => <div><div></div><span>Hi</span></div>"span").first.prepend(sibling: "<div>")
# <div><span>Hi</span></div> => <div><span>Hi<br/></span></div>"span").first.append(child: "<br/>")
# <div><span>Hi</span></div> => <div><span>Hi</span><p/></div>"span").first.append(sibling: "<p/>")

We can also replace all children:

# <div><span>Hi</span></div> => <div><span><p>Howdy</p></span></div>"span").first.replace_children with: "<p>Howdy</p>"

And we can wrap an element with another element:

# <div><span>Hi</span></div> => <div><span class="other"><span>Hi</span></span></div>"span").first.wrap("<span class='other'>")

or wrap an element's children:

# <div><span>Hi</span></div> => <div><span><span class="other" data-type="foo">Hi</span></span></div>"span").first.wrap_children('span', class: 'other', data_type: 'foo')

Checking for elements

You can see if an element contains an element matching a selector:

my_element.contains?(".title") #=> true or false


  • ElementEnumerator also provides a first! method that is like the standard first except it raises an error if there is no matching first element to return.

Using raw to get at underlying Nokogiri objects.

Kitchen uses the Nokogiri gem to parse and manipulate XML documents. Document objects wraps a Nokogiri::XML::Document object, and Element objects wrap a Nokogiri::XML::Node object. If you want to do something wild and crazy you can access these underlying objects using the raw method on Document and Element. Note that many of the methods on the underlying objects are exposed on the Kitchen object, e.g. instead of saying my_element.raw['data-type'] you can say my_element['data-type'].

Book-Oriented Usage

All of the above works, but it is generic and we have a specific problem handling books that use a specific schema. To that end, Kitchen also includes a BookDocument to use in place of Document as well as elements and enumerators specific to this schema, e.g. BookElement, ChapterElement, PageElement, TableElement, FigureElement, NoteElement, ExampleElement. BookDocument has a method called book that returns a BookElement that wraps the top-level html element. All of these elements have methods on them for searching for other of these specific elements, so that instead of"[data-type='page']")

we can say

In the generic usage, you can chain search methods:"[data-type='page']").search("figure")

will find all figure elements inside pages inside my_chapter.

In the book-oriented usage, you can chain specific search methods to achieve the same effect:

This chaining of enumerators gives other benefits. The above search for figures will yield figures that know the page they were found in as well as their numerical position within that page. So you could do something like this: do |figure|
    "<span class='os-number'>Figure #{figure.count_in(:chapter)}.#{figure.count_in(:page)}</span>" \
    "<span class='os-title'>A figure in chapter #{figure.ancestor(:chapter).title}</span>"

This finds all figures that are in pages that are in chapters in the book. The count_in methods on the figure give the number position of the figure element within the chapter or page so we can form a figure number like "2.13". And as seen here, chapter elements (instances of ChapterElement) have a title method that returns the title text for the chapter. Figures have a caption element, etc.

The CSS for these specific search methods is hidden away so you don't have to deal with it. But if you want to customize that CSS you can. You can pass an overriding CSS selector to these methods, and if you use the $ character in that argument the search method will replace it with the normal CSS selector, e.g. if you wanted to get rid of all of the table elements that have the "unnumbered" class you could say:"$.unnumbered").cut

Sometimes, it is difficult to setup a search using CSS. In such cases, you can also pass only and except arguments to search methods, e.g.: :subfigure?)

only and except can be the names of methods (that return truthy/falsy values) on the element being iterated over, as shown above, or they can be lambdas or procs as shown here: ->(fig) { fig.children.count == 2 })

Obviously this is a somewhat contrived example, but the idea is that by passing a callable you can do complex searches.

Overriding Default Book-Oriented Selectors

Book-oriented methods like book.pages.figures hide from us the CSS or XPath selectors that let us find child elements like .pages. But sometimes, the default selector we have isn't what is used in a certain book. In these cases, we can override the selector once in the recipe and still continue to use the book-oriented usage. For example, a page summary is normally found using the CSS section.summary. But some books use a .section-summary class. For these books, we can override the selectors in their recipes:

recipe = do |doc|
    page_summary: ".section-summary"


All of the above talks about the how to search through the XML file and perform basic operations on that file. Our recipes will be combinations of all of the above: search for elements; cut, copy and paste them; count them; rework them; etc.

One recipe for processing a book probably does 10-30 different kinds of operations: format and number tables, same for figures and examples, number and organize exercises and their solutions in different parts of the book, build chapter glossaries, build an index, build a table of contents, etc, etc.

We're not going to want to write out all of those steps in every receipe. Instead it'd be a better idea to write out each step in its own little piece of code. With the steps isolated from each other we'll be dealing with less code all at once and it'll be much easier to write tests to exercise that code.

In Kitchen, we've started the process of writing out these steps and we've put them in a directions folder (which is also a Directions module). E.g. Kitchen::Directions::BakeChapterSummary modifies a provided chapter to have a chapter summary at the end.

It is probably true that the BakeChapterSummary code will work for some number of books, but other books might have different requirements. As such we can expect that there will be different variants of the chapter summary baking step. To anticipate this, our first implementation of this step lives in a method named v1 (so to run it you call BakeChapterSummary.v1(chapter: some_chapter)). Later if there's a tweak needed that can't fit into v1's approach, we can make a v2 method that could live in its own file. This may or may not be the right approach to handle this kind of code variation, but it is at least a place to start.

Internationalization (I18n)

Recognizing that our books will be translated into multiple languages, Kitchen has support for internationalization (I18n). There's a spot for translation files in the locales directory, in which there is currently one en.yml translation file for English. Within our directions code you'll see uses of it like here to title an Example:

<span class="os-title-label">#{I18n.t(:example)} </span>

Building HTML strings

There are a number of valid ways of building up HTML strings to insert into documents.

Maybe you have a tiny bit of HTML to add and you can use vanilla Ruby strings:

some_element.append(child: "<br/>")

You can continue doing this with multiline strings but it gets to be a pain -- you have to add your own newlines (\n) and line continuation symbols (\):

some_element.append(child: "first line\n" \
                           "second line")

Ruby has a much better way of handling multiline strings: heredocs. The best of these is the "squiggly" heredoc, which captures string content between <<~SOME_ARBITRARY_TEXT and SOME_ARBITRARY_TEXT:

some_element.append(sibling: <<~HTML
    <div class="os-caption-container">
      <span class="os-caption">Awesomeness</span>

The squiggly heredoc removes the shortest leading indentation from each line. It lets you use single and double quotes inside the string without escaping them. And at least in certain editors when you use HTML as your "some arbitrary text", you'll get HTML syntax highlighting. You can also do interpolate variables into the string using #{some_variable} The above example is equivalent to this Ruby string written in the earlier approach:

"<div class=\"os-caption-container\">\n" \
"  <span class=\"os-caption\">Awesomeness</span>\n" \

The big downside to all of these approaches is that for more complicated strings, we often need to use some Ruby logic to build up different parts of the string, and the techniques above don't allow for that.

Let's invent an example of needing to build some HTML that had a listing of all chapter titles in a bulleted list and then their page titles within them in a nested bulleted list (kind of like a table of contents). This is an example of what we'd be shooting for:

    <span>Chapter 1 Title</span>
      <li><span>Page 1.1 Title</span></li>
      <li><span>Page 1.2 Title</span></li>
    ... etc etc

Here's one way we could build up this string using squiggly heredocs:

class SomethingThatBakes
  def bake(doc)
    @book =

    chapter_bullets_array = do |chapter|
      page_bullets_array = do


    final_string = <<~HTML

    # do something with that final_string

The above works but it is a little fragmented to read. We have to build up parts of the bulleted lists in arrays, then join them together with newlines and embed them in other strings (some of which are also collected in an array and then later substituted and joined).

For these more complex strings we have another option: ERB (Embedded RuBy). ERB is part of standard Ruby and had its heyday when Rails came out in the 2000s. ERB lets us make a separate HTML file with Ruby sprinkled within it. Let's call this file blah.html.erb:

  <% @book.chapters.each do |chapter| %>
      <span><%= chapter.title.text</span>
        <% chapter.pages.each do |page| %>
          <li><span><%= page.title.text %></span></li>
        <% end %>
  <% end %>

In our Ruby class doing the generation, we add a renderable statement at the top code we can then say:

class SomethingThatBakes

  def bake(doc)
    final_string = render(file: 'blah.html.erb')

    # do something with that final_string

This ERB approach is a lot easier to read -- you can see the nesting structure directly in the template file. The Ruby code in the ERB template will have access to any instance variable in the code that called it, i.e. the variables that start with @.

The render method takes a file argument that is a string file path. If the path is relative, it is relative to the directory in which the render call is made.

If you want to make relative file paths be relative to a different directory, you can pass a directory string to the renderable statement: renderable dir: '/Some/other/directory'.

Again, all these techniques work and there are times to use them all.

One-file scripts

Want to make a one-file script to do some baking? Use the "inline" form of bundler:

#!/usr/bin/env ruby

require "bundler/inline"

gemfile do
  gem 'openstax_kitchen', '2.0.0'

require "openstax_kitchen"

recipe = do |doc|
  # ... recipe steps here

  input_file: "some_file.xhtml",
  recipes: recipe,
  output_file: "some_other_file.xhtml")

Incidentally, the bake method returns timing information, if you puts its result you'll see it.

Recipe (and Gem) Development


You can use Docker for your development environment. To build the image:

$> ./docker/build

To drop into the running container:

$> ./docker/bash

To run specs (or something else) from the host:

$> ./docker/run rspec


After checking out the repo, run bin/setup to install dependencies. If you want to install this gem onto your local machine, run bundle exec rake install.


You can also run bin/console for an interactive prompt that will allow you to experiment.


There are some tutorials you can work through in the tutorials directory. Each tutorial is in a separated numbered subdirectory, e.g. tutorials/01. Each tutorial directory contains a raw.html file that is your starting point (along with some instructions in comments at the top), an expected_baked.html file that is what you're trying to get to when your recipe is applied to the input file, as well as some number of solution files (don't look at those unless you get stuck!!). To get started, run:

$> ./setup_my_recipes

in the tutorials directory. That will make a blank my_recipe.rb file in each of the numbered tutorial subdirectories. This is where you'll do your work. The first "Hello world!" tutorial ("00") asks you to make a recipe that changes

<div class="hello">


<h1 class="hello">

There's an included script to check to see if your recipe achieves the desired transformation:

$> ./check_it 00

Will check to see if your tutorials/00/my_recipe.rb file does what is needed. If it does, you'll see a "way to go" message. If it doesn't, you'll see a diff between the expected output and the actual output. E.g. if you run ./check_it 00 without having done any work yet, you'll see:

The actual output does not match the expected output.
- = actual output
+ = expected output

@@ -1,4 +1,4 @@
-<div class="hello">
-  <span>Planet?</span>
+<h1 class="hello">
+  <span>World!</span>

The check_it script can also check the solutions. E.g. if you say

$> ./check_it 00 solution_1

you'll see

The actual output matches the expected output! Way to go!

There is normally more than one way to achieve the desired output, so feel free to diverge from what is shown in the solution files. Note that the my_recipe.rb files and all actual_baked.html files are ignored by Git.

Important: If things aren't working in your tutorial work (or actually in any recipe work), use the debugger! Just add a debugger line anywhere in your code to stop execution there so you can poke around. You can print variables by typing out their name, run methods on objects, say s to step into function calls, n ("next") to step over function calls, b 97 to set a new breakpoint at line 97, and c to continue to the next debugger statement, breakpoint, or the end of the script.

Error Messages

Kitchen tries to give helpful error messages instead of huge stack traces, e.g.:

The recipe has an error: undefined method `bleach' for main:Object
at or near the following highlighted line

-----+ ./my_work/test.rb -----
   18|   doc.chapters.each do |chapter|
   19|     chapter.bleach("div.exercise") do |elem|
   20|       elem.first("h3").trash
   21|       elem.cut to: :review_questions

Encountered on line 64 in the input document on element:
<div data-type="chapter">...</div>

If you'd still like the huge stack trace, you can set the VERBOSE environment variable to anything, e.g.

$> VERBOSE=1 ./my_work/test.rb

Comparing old recipe output to new

When comparing new baking output to legacy baking output, I have found it useful to prepare the files before applying a standard diff:

  1. Parse the documents ignoring blank elements. (this one may not be that important actually)
  2. Traverse the documents, sorting their attributes alphabetically by attribute name.
  3. Writing the output back out with a standardized indentation scheme.

I have some code to do this, I'll try to get it into this repo.


To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to


Documentation is handled via YARD. The Solargraph gem can be used in popular editors for code completion.

Run yard server --reload to watch for changes in your local codebase everytime you refresh the page.

Navidate to http://localhost:8808/ to view documentation in your browser.

Use the inch gem to get feedback on where documentation is lacking bundle exec inch (add --help for more options).


Run bundle exec rspec to run the specs. rake rspec probably does the same thing.

Spec offer two ways to compare expected XML output to actual output.

match_normalized_html gets rid of extra blanks, sorts all tag attributes alphabetically by attribute name (e.g. sorts "" to "" so that attribute order doesn't impact a match), prints the HTML back out with a standard indent, and then does a normal string diff.

expect(book_1).to match_normalized_html("some string of HTML here")

match_html_nodes does a node-by-node diff using the nokogiri-diff gem. It gives more specific node diff data but is also not quite as clear.

expect(book_1).to match_html_nodes("some string of HTML here")


If you set the PROFILE environment variable to something before you run specs or a recipe, search query profile data will be collected and printed, e.g.

%> PROFILE=1 rspec


There's a low-level CSS query caching tool that saves repeated queries. In some tests, it saves 15% of query time. It is disabled by default (because we aren't super sure that it is completely safe) but can be turned on with

doc.config.enable_search_cache = true


  1. Visit vscode:extension/ms-vscode-remote.remote-containers in a browser
  2. It'll open VSCode and bring you to an extension install screen, click "Install"
  3. Click the remote button now in the bottom left hand corner.
  4. Click "Remote-Containers: Open Folder in Container"
  5. Select the cloned kitchen folder.

This (assuming you have Docker installed) will launch a docker container for Kitchen, install Ruby and needed libraries, and then let you edit the code running in that container through VSCode. Solargraph will work (code completion and inline documentation) as will Rubocop for linting.


Rubocop is good for helping us keep our code style standardized, but it isn't the end-all be-all of things. We can disable certain checks within a file, e.g.

# rubocop:disable Style/NumericPredicate

or we can disable or change global settings in the .rubocop.yml file.

Rubocop is setup to run within the VSCode dev container (see above).

The lefthook is included in the Docker build. When you push your code to GitHub, lefthook runs Rubocop on all the files you have changed. It won't let you push if you have Rubocop errors. You'll have to fix the errors or make changes to the .rubocop.yml files to bypass the errors. You can also run lefthook directly with

$ /code> lefthook run pre-push

Misc References


  • Fix up tutorials and describe how to use them here


Bug reports and pull requests are welcome on GitHub at This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the code of conduct.


Helpful scripts in the bin directory:

  • normalize - Normalizes content files to make it easier to compare them. E.g. if you want to compare kitchen baked output to cnx-recipes baked output, you should normalize the files first. normalize somefile.xhtml produces somefile.normalized.xhtml which has its attributes sorted by attribute name, copied element ID numbers masked (because they change based on order of operations in recipes, but their values are not important), and some errors in legacy baked files removed (e.g. unnumbered tables get a summary attribute with a bogus number).


The gem is available as open source under the terms of the MIT License.


  • Specs galore :-)
  • Think up and handle a bunch more recipe errors, test they all raise some kind of RecipeError.
  • Encapsulate numbering schemes (e.g. chapter pages are "5.2", appendix pages are "D7") and maybe set on book document? Right now we are doing inline things like *('A'..'Z')][page.count_in(:book)-1]}#{table.count_in(:page) which is ugly.
  • Control I18n language in Oven.
  • README: element_children, .only, selectors, config files
  • Use ERB for more readable string building?


When Kitchen writes out HTML containing unicode characters it uses the hexadecimal form, whereas current CE baking uses the decimal form. I haven't found an internal way to change how Kitchen's underlying library writes these characters, so if you need to do a new-to-old comparison, you can use a few lines of ruby to do a search and replace:

original_output ="kitchen_output.xhtml")
modified_output = original_output.gsub(/&#x([0-9A-F]+);/){"&##{$1.hex};"}"kitchen_output.xhtml", "w") {|file| file.puts modified_output}

If this difference matters (if we need the decimal version), we can do more work to figure out a better implementation.


  • Use tmux for real-time evaluation of recipes to see output within one split terminal (source XML in one pane, recipe in middle, output on right).

Code of Conduct

Everyone interacting in the Kitchen project's codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.