HTML-Pipeline

HTML processing filters and utilities. This module is a small framework for defining CSS-based content filters and applying them to user provided content.

Although this project was started at GitHub, they no longer use it. This gem must be considered standalone and independent from GitHub.

Installation

Add this line to your application's Gemfile:

gem 'html-pipeline'

And then execute:

$ bundle

Or install it by yourself as:

$ gem install html-pipeline

Usage

This library provides a handful of chainable HTML filters to transform user content into HTML markup. Each filter does some work, and then hands off the results tothe next filter. A pipeline has several kinds of filters available to use:

  • Multiple TextFilters, which operate a UTF-8 string
  • A ConvertFilter filter, which turns text into HTML (eg., Commonmark/Asciidoc -> HTML)
  • A SanitizationFilter, which remove dangerous/unwanted HTML elements and attributes
  • Multiple NodeFilters, which operate on a UTF-8 HTML document

You can assemble each sequence into a single pipeline, or choose to call each filter individually.

As an example, suppose we want to transform Commonmark source text into Markdown HTML:

Hey there, @gjtorikian

With the content, we also want to:

  • change every instance of Hey to Hello
  • strip undesired HTML
  • linkify @mention

We can construct a pipeline to do all that like this:

require 'html_pipeline'

class HelloJohnnyFilter < HTMLPipelineFilter
  def call
    text.gsub("Hey", "Hello")
  end
end

pipeline = HTMLPipeline.new(
  text_filters: [HelloJohnnyFilter.new]
  convert_filter: HTMLPipeline::ConvertFilter::MarkdownFilter.new,
    # note: next line is not needed as sanitization occurs by default;
    # see below for more info
  sanitization_config: HTMLPipeline::SanitizationFilter::DEFAULT_CONFIG,
  node_filters: [HTMLPipeline::NodeFilter::MentionFilter.new]
)
pipeline.call(user_supplied_text) # recommended: can call pipeline over and over

Filters can be custom ones you create (like HelloJohnnyFilter), and HTMLPipeline additionally provides several helpful ones (detailed below). If you only need a single filter, you can call one individually, too:

filter = HTMLPipeline::ConvertFilter::MarkdownFilter.new
filter.call(text)

Filters combine into a sequential pipeline, and each filter hands its output to the next filter's input. Text filters are processed first, then the convert filter, sanitization filter, and finally, the node filters.

Some filters take optional context and/or result hash(es). These are used to pass around arguments and metadata between filters in a pipeline. For example, if you want to disable footnotes in the MarkdownFilter, you can pass an option in the context hash:

context = { markdown: { extensions: { footnotes: false } } }
filter = HTMLPipeline::ConvertFilter::MarkdownFilter.new(context: context)
filter.call("Hi **world**!")

Alternatively, you can construct a pipeline, and pass in a context during the call:

pipeline = HTMLPipeline.new(
  convert_filter: HTMLPipeline::ConvertFilter::MarkdownFilter.new,
  node_filters: [HTMLPipeline::NodeFilter::MentionFilter.new]
)
pipeline.call(user_supplied_text, context: { markdown: { extensions: { footnotes: false } } })

Please refer to the documentation for each filter to understand what configuration options are available.

More Examples

Different pipelines can be defined for different parts of an app. Here are a few paraphrased snippets to get you started:

# The context hash is how you pass options between different filters.
# See individual filter source for explanation of options.
context = {
  asset_root: "http://your-domain.com/where/your/images/live/icons",
  base_url: "http://your-domain.com"
}

# Pipeline used for user provided content on the web
MarkdownPipeline = HTMLPipeline.new (
  text_filters: [HTMLPipeline::TextFilter::ImageFilter.new],
  convert_filter: HTMLPipeline::ConvertFilter::MarkdownFilter.new,
  node_filters: [
    HTMLPipeline::NodeFilter::HttpsFilter.new,HTMLPipeline::NodeFilter::MentionFilter.new,
  ], context: context)

# Pipelines aren't limited to the web. You can use them for email
# processing also.
HtmlEmailPipeline = HTMLPipeline.new(
  text_filters: [
    PlainTextInputFilter.new,
    ImageFilter.new
  ], {})

Filters

TextFilters

TextFilters must define a method named call which is called on the text. @text, @config, and @result are available to use, and any changes made to these ivars are passed on to the next filter.

  • ImageFilter - converts image url into <img> tag
  • PlainTextInputFilter - html escape text and wrap the result in a <div>

ConvertFilter

The ConvertFilter takes text and turns it into HTML. @text, @config, and @result are available to use. ConvertFilter must defined a method named call, taking one argument, text. call must return a string representing the new HTML document.

  • MarkdownFilter - creates HTML from text using Commonmarker

Sanitization

Because the web can be a scary place, HTML is automatically sanitized after the ConvertFilter runs and before the NodeFilters are processed. This is to prevent malicious or unexpected input from entering the pipeline.

The sanitization process takes a hash configuration of settings. See the Selma documentation for more information on how to configure these settings. Note that users must correctly configure the sanitization configuration if they expect to use it correctly in conjunction with handlers which manipulate HTML.

A default sanitization config is provided by this library (HTMLPipeline::SanitizationFilter::DEFAULT_CONFIG). A sample custom sanitization allowlist might look like this:

ALLOWLIST = {
  elements: ["p", "pre", "code"]
}

pipeline = HTMLPipeline.new \
  text_filters: [
    HTMLPipeline::TextFilter::ImageFilter.new,
  ],
  convert_filter: HTMLPipeline::ConvertFilter::MarkdownFilter.new,
  sanitization_config: ALLOWLIST

result = pipeline.call <<-CODE
This is *great*:

    some_code(:first)

CODE
result[:output].to_s

This would print:

<p>This is great:</p>
<pre><code>some_code(:first)
</code></pre>

Sanitization can be disabled if and only if nil is explicitly passed as the config:

pipeline = HTMLPipeline.new \
  text_filters: [
    HTMLPipeline::TextFilter::ImageFilter.new,
  ],
  convert_filter: HTMLPipeline::ConvertFilter::MarkdownFilter.new,
  sanitization_config: nil

For more examples of customizing the sanitization process to include the tags you want, check out the tests and the FAQ.

NodeFilters

NodeFilterss can operate either on HTML elements or text nodes using CSS selectors. Each NodeFilter must define a method named selector which provides an instance of Selma::Selector. If elements are being manipulated, handle_element must be defined, taking one argument, element; if text nodes are being manipulated, handle_text_chunk must be defined, taking one argument, text_chunk. @config, and @result are available to use, and any changes made to these ivars are passed on to the next filter.

NodeFilter also has an optional method, after_initialize, which is run after the filter initializes. This can be useful in setting up a fresh custom state for result to start from each time the pipeline is called.

Here's an example NodeFilter that adds a base url to images that are root relative:

require 'uri'

class RootRelativeFilter < HTMLPipeline::NodeFilter

  SELECTOR = Selma::Selector.new(match_element: "img")

  def selector
    SELECTOR
  end

  def handle_element(img)
    next if img['src'].nil?
    src = img['src'].strip
    if src.start_with? '/'
      img["src"] = URI.join(context[:base_url], src).to_s
    end
  end
end

For more information on how to write effective NodeFilters, refer to the provided filters, and see the underlying lib, Selma for more information.

  • AbsoluteSourceFilter: replace relative image urls with fully qualified versions
  • AssetProxyFilter: replace image links with an encoded link to an asset server
  • EmojiFilter: converts :<emoji>: to emoji
    • (Note: the included MarkdownFilter will already convert emoji)
  • HttpsFilter: Replacing http urls with https versions
  • ImageMaxWidthFilter: link to full size image for large images
  • MentionFilter: replace @user mentions with links
  • SanitizationFilter: allow sanitize user markup
  • SyntaxHighlightFilter: applies syntax highlighting to pre blocks
    • (Note: the included MarkdownFilter will already apply highlighting)
  • TableOfContentsFilter: anchor headings with name attributes and generate Table of Contents html unordered list linking headings
  • TeamMentionFilter: replace @org/team mentions with links

Dependencies

Since filters can be customized to your heart's content, gem dependencies are not bundled; this project doesn't know which of the default filters you might use, and as such, you must bundle each filter's gem dependencies yourself.

For example, SyntaxHighlightFilter uses rouge to detect and highlight languages; to use the SyntaxHighlightFilter, you must add the following to your Gemfile:

gem "rouge"

Note See the Gemfile :test group for any version requirements.

When developing a custom filter, call HTMLPipeline.require_dependency at the start to ensure that the local machine has the necessary dependency. You can also use HTMLPipeline.require_dependencies to provide a list of dependencies to check.

On a similar note, you must manually require whichever filters you desire:

require "html_pipeline" # must be included
require "html_pipeline/convert_filter/markdown_filter" # included because you want to use this filter
require "html_pipeline/node_filter/mention_filter" # included because you want to use this filter

Documentation

Full reference documentation can be found here.

Instrumenting

Filters and Pipelines can be set up to be instrumented when called. The pipeline must be setup with an ActiveSupport::Notifications compatible service object and a name. New pipeline objects will default to the HTMLPipeline.default_instrumentation_service object.

# the AS::Notifications-compatible service object
service = ActiveSupport::Notifications

# instrument a specific pipeline
pipeline = HTMLPipeline.new [MarkdownFilter], context
pipeline.setup_instrumentation "MarkdownPipeline", service

# or set default instrumentation service for all new pipelines
HTMLPipeline.default_instrumentation_service = service
pipeline = HTMLPipeline.new [MarkdownFilter], context
pipeline.setup_instrumentation "MarkdownPipeline"

Filters are instrumented when they are run through the pipeline. A call_filter.html_pipeline event is published once any filter finishes; call_text_filters and call_node_filters is published when all of the text and node filters are finished, respectively. The payload should include the filter name. Each filter will trigger its own instrumentation call.

service.subscribe "call_filter.html_pipeline" do |event, start, ending, transaction_id, payload|
  payload[:pipeline] #=> "MarkdownPipeline", set with `setup_instrumentation`
  payload[:filter] #=> "MarkdownFilter"
  payload[:context] #=> context Hash
  payload[:result] #=> instance of result class
  payload[:result][:output] #=> output HTML String
end

The full pipeline is also instrumented:

service.subscribe "call_text_filters.html_pipeline" do |event, start, ending, transaction_id, payload|
  payload[:pipeline] #=> "MarkdownPipeline", set with `setup_instrumentation`
  payload[:filters] #=> ["MarkdownFilter"]
  payload[:doc] #=> HTML String
  payload[:context] #=> context Hash
  payload[:result] #=> instance of result class
  payload[:result][:output] #=> output HTML String
end

Third Party Extensions

If you have an idea for a filter, propose it as an issue first. This allows us to discuss whether the filter is a common enough use case to belong in this gem, or should be built as an external gem.

Here are some extensions people have built:

FAQ

1. Why doesn't my pipeline work when there's no root element in the document?

To make a pipeline work on a plain text document, put the PlainTextInputFilter at the end of your text_filters config . This will wrap the content in a div so the filters have a root element to work with. If you're passing in an HTML fragment, but it doesn't have a root element, you can wrap the content in a div yourself.

2. How do I customize an allowlist for SanitizationFilters?

HTMLPipeline::SanitizationFilter::ALLOWLIST is the default allowlist used if no sanitization_config argument is given. The default is a good starting template for you to add additional elements. You can either modify the constant's value, or re-define your own config and pass that in, such as:

config = HTMLPipeline::SanitizationFilter::DEFAULT_CONFIG.deep_dup
config[:elements] << "iframe" # sure, whatever you want

Contributors

Thanks to all of these contributors.

This project is a member of the OSS Manifesto.