Selma

Selma selects and matches HTML nodes using CSS rules. (It can also reject/delete nodes, but then the name isn't as cool.) It's mostly an idiomatic wrapper around Cloudflare's lol-html project.

Selma's strength (aside from being backed by Rust) is that HTML content is parsed once and can be manipulated multiple times.

Installation

Add this line to your application's Gemfile:

gem 'selma'

And then execute:

$ bundle install

Or install it yourself as:

$ gem install selma

Usage

Selma can perform two different actions, either independently or together:

Sanitize HTML, through a Sanitize-like allowlist syntax; and
Select HTML using CSS rules, and manipulate elements and text nodes along the way.

It does this through two kwargs: sanitizer and handlers. The basic API for Selma looks like this:

sanitizer_config = {
   elements: ["b", "em", "i", "strong", "u"],
}
sanitizer = Selma::Sanitizer.new(sanitizer_config)
rewriter = Selma::Rewriter.new(sanitizer: sanitizer, handlers: [MatchElementRewrite.new, MatchTextRewrite.new])
# removes any element that is not  ["b", "em", "i", "strong", "u"];
# then calls `MatchElementRewrite` and `MatchTextRewrite` on matching HTML elements
rewriter.rewrite(html)

Here's a look at each individual part.

Sanitization config

Selma sanitizes by default. That is, even if the sanitizer kwarg is not passed in, sanitization occurs. If you truly want to disable HTML sanitization (for some reason), pass nil:

Selma::Rewriter.new(sanitizer: nil) # dangerous and ill-advised

The configuration for the sanitization process is based on the follow key-value hash allowlist:

# Whether or not to allow HTML comments.
allow_comments: false,

# Whether or not to allow well-formed HTML doctype declarations such as
# "<!DOCTYPE html>" when sanitizing a document.
allow_doctype: false,

# HTML elements to allow. By default, no elements are allowed (which means
# that all HTML will be stripped).
elements: ["a", "b", "img", ],

# HTML attributes to allow in specific elements. The key is the name of the element,
# and the value is an array of allowed attributes. By default, no attributes
# are allowed.
attributes: {
    "a" => ["href"],
    "img" => ["src"],
},

# URL handling protocols to allow in specific attributes. By default, no
# protocols are allowed. Use :relative in place of a protocol if you want
# to allow relative URLs sans protocol. Set to `:all` to allow any protocol.
protocols: {
    "a" => { "href" => ["http", "https", "mailto", :relative] },
    "img" => { "href" => ["http", "https"] },
},

# An Array of element names whose contents will be removed. The contents
# of all other filtered elements will be left behind.
remove_contents: ["iframe", "math", "noembed", "noframes", "noscript"],

# Elements which, when removed, should have their contents surrounded by
# whitespace.
whitespace_elements: ["blockquote", "h1", "h2", "h3", "h4", "h5", "h6", ]

Defining handlers

The real power in Selma comes in its use of handlers. A handler is simply an object with various methods defined:

selector, a method which MUST return instance of Selma::Selector which defines the CSS classes to match
handle_element, a method that's call on each matched element
handle_text_chunk, a method that's called on each matched text node

Here's an example which rewrites the href attribute on a and the src attribute on img to be https rather than http.

class MatchAttribute
  SELECTOR = Selma::Selector.new(match_element: %(a[href^="http:"], img[src^="http:"]"))

  def selector
    SELECTOR
  end

  def handle_element(element)
    if element.tag_name == "a"
      element["href"] = rename_http(element["href"])
    elsif element.tag_name == "img"
      element["src"] = rename_http(element["src"])
    end
  end

  private def rename_http(link)
    link.sub("http", "https")
  end
end

rewriter = Selma::Rewriter.new(handlers: [MatchAttribute.new])

The Selma::Selector object has three possible kwargs:

match_element: any element which matches this CSS rule will be passed on to handle_element
match_text_within: any text_chunk which matches this CSS rule will be passed on to handle_text_chunk
ignore_text_within: this is an array of element names whose text contents will be ignored

Here's an example for handle_text_chunk which changes strings in various elements which are not pre or code:

class MatchText
  SELECTOR = Selma::Selector.new(match_text_within: "*", ignore_text_within: ["pre", "code"])

  def selector
    SELECTOR
  end

  def handle_text_chunk(text)
    string.sub(/@.+/, "<a href=\"www.yetto.app/#{Regexp.last_match}\">")
  end
end

rewriter = Selma::Rewriter.new(handlers: [MatchText.new])

`element` methods

The element argument in handle_element has the following methods:

tag_name: Gets the element's name
tag_name=: Sets the element's name
self_closing?: A bool which identifies whether or not the element is self-closing
[]: Get an attribute
[]=: Set an attribute
remove_attribute: Remove an attribute
has_attribute?: A bool which identifies whether or not the element has an attribute
attributes: List all the attributes
ancestors: List all of an element's ancestors as an array of strings
before(content, as: content_type): Inserts content before the element. content_type is either :text or :html and determines how the content will be applied.
after(content, as: content_type): Inserts content after the element. content_type is either :text or :html and determines how the content will be applied.
prepend(content, as: content_type): prepends content to the element's inner content, i.e. inserts content right after the element's start tag. content_type is either :text or :html and determines how the content will be applied.
append(content, as: content_type): appends content to the element's inner content, i.e. inserts content right before the element's end tag. content_type is either :text or :html and determines how the content will be applied.
set_inner_content: Replaces inner content of the element with content. content_type is either :text or :html and determines how the content will be applied.
remove: Removes the element and its inner content.
remove_and_keep_content: Removes the element, but keeps its content. I.e. remove start and end tags of the element.
removed?: A bool which identifies if the element has been removed or replaced with some content.

`text_chunk` methods

to_s / .content: Gets the text node's content
text_type: identifies the type of text in the text node
before(content, as: content_type): Inserts content before the text. content_type is either :text or :html and determines how the content will be applied.
after(content, as: content_type): Inserts content after the text. content_type is either :text or :html and determines how the content will be applied.
replace(content, as: content_type): Replaces the text node with content. content_type is either :text or :html and determines how the content will be applied.

Security

Theoretically, a malicious user can provide a very large document for processing, which can exhaust the memory of the host machine. To set a limit on how much string content is processed at once, you can provide memory options:

Selma::Rewriter.new(options: { memory: { max_allowed_memory_usage: 1_000_000 } }) # ~1MB

The structure of the memory options looks like this:

{
  memory: {
    max_allowed_memory_usage: 1000,
    preallocated_parsing_buffer_size: 100,
  }
}

Note that preallocated_parsing_buffer_size must always be less than max_allowed_memory_usage. See thelol_html project documentation to learn more about the default values.

Benchmarks

When bundle exec rake benchmark, two different benchmarks are calculated. Here are those results on my machine.

Benchmarks for just the sanitization process

Comparing Selma against popular Ruby sanitization gems:

input size = 25309 bytes, 0.03 MB

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] Warming up -------------------------------------- sanitize-sm 15.000 i/100ms selma-sm 127.000 i/100ms Calculating ------------------------------------- sanitize-sm 157.643 (± 1.9%) i/s - 4.740k in 30.077172s selma-sm 1.278k (± 1.5%) i/s - 38.354k in 30.019722s

Comparison: selma-sm: 1277.9 i/s sanitize-sm: 157.6 i/s - 8.11x slower

input size = 86686 bytes, 0.09 MB

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] Warming up -------------------------------------- sanitize-md 4.000 i/100ms selma-md 33.000 i/100ms Calculating ------------------------------------- sanitize-md 40.034 (± 5.0%) i/s - 1.200k in 30.043322s selma-md 332.959 (± 2.1%) i/s - 9.999k in 30.045733s

Comparison: selma-md: 333.0 i/s sanitize-md: 40.0 i/s - 8.32x slower

input size = 7172510 bytes, 7.17 MB

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] Warming up -------------------------------------- sanitize-lg 1.000 i/100ms selma-lg 1.000 i/100ms Calculating ------------------------------------- sanitize-lg 0.141 (± 0.0%) i/s - 5.000 in 35.426127s selma-lg 3.963 (± 0.0%) i/s - 119.000 in 30.037386s

Comparison: selma-lg: 4.0 i/s sanitize-lg: 0.1 i/s - 28.03x slower

Benchmarks for just the rewriting process

Comparing Selma against popular Ruby HTML parsing gems:

input size = 25309 bytes, 0.03 MB

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] Warming up -------------------------------------- nokogiri-sm 79.000 i/100ms nokolexbor-sm 295.000 i/100ms selma-sm 237.000 i/100ms Calculating ------------------------------------- nokogiri-sm 800.531 (± 2.2%) i/s - 24.016k in 30.016056s nokolexbor-sm 3.033k (± 3.6%) i/s - 91.155k in 30.094884s selma-sm 2.386k (± 1.6%) i/s - 71.574k in 30.001701s

Comparison: nokolexbor-sm: 3033.1 i/s selma-sm: 2386.3 i/s - 1.27x slower nokogiri-sm: 800.5 i/s - 3.79x slower

input size = 86686 bytes, 0.09 MB

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] Warming up -------------------------------------- nokogiri-md 8.000 i/100ms nokolexbor-md 43.000 i/100ms selma-md 38.000 i/100ms Calculating ------------------------------------- nokogiri-md 85.013 (± 8.2%) i/s - 2.024k in 52.257472s nokolexbor-md 416.074 (±11.1%) i/s - 12.341k in 30.111613s selma-md 361.471 (± 4.7%) i/s - 10.830k in 30.033997s

Comparison: nokolexbor-md: 416.1 i/s selma-md: 361.5 i/s - same-ish: difference falls within error nokogiri-md: 85.0 i/s - 4.89x slower

input size = 7172510 bytes, 7.17 MB

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin23] Warming up -------------------------------------- nokogiri-lg 1.000 i/100ms nokolexbor-lg 1.000 i/100ms selma-lg 1.000 i/100ms Calculating ------------------------------------- nokogiri-lg 0.805 (± 0.0%) i/s - 25.000 in 31.148730s nokolexbor-lg 2.194 (± 0.0%) i/s - 66.000 in 30.278108s selma-lg 5.541 (± 0.0%) i/s - 166.000 in 30.037197s

Comparison: selma-lg: 5.5 i/s nokolexbor-lg: 2.2 i/s - 2.53x slower nokogiri-lg: 0.8 i/s - 6.88x slower

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/gjtorikian/selma. This project is a safe, welcoming space for collaboration.

Acknowledgements

https://github.com/flavorjones/ruby-c-extensions-explained#strategy-3-precompiled and Nokogiri for hints on how to ship precompiled cross-platform gems
@vmg for his work at GitHub on goomba, from which some design patterns were learned
sanitize for a comprehensive configuration API and test suite

License

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

selma

Development

Runtime

Selma

Installation

Usage

Sanitization config

Defining handlers

`element` methods

`text_chunk` methods

Security

Benchmarks

Benchmarks for just the sanitization process

Benchmarks for just the rewriting process

Contributing

Acknowledgements

License

selma

Development

Runtime

Selma

Installation

Usage

Sanitization config

Defining handlers

element methods

text_chunk methods

Security

Benchmarks

Benchmarks for just the sanitization process

Benchmarks for just the rewriting process

Contributing

Acknowledgements

License

`element` methods

`text_chunk` methods