devdocs/docs/scraper-reference.md

Table of contents:

• Overview
• Configuration
  • Attributes
  • Filter stacks
  • Filter options

Overview

Starting from a root URL, scrapers recursively follow links that match a set of rules, passing each valid response through a chain of filters before writing the file on the local filesystem. They also create an index of the pages' metadata (determined by one filter), which is dumped into a JSON file at the end.

Scrapers rely on the following libraries:

• Typhoeus for making HTTP requests
• HTML::Pipeline for applying filters
• Nokogiri for parsing HTML

There are currently two kinds of scrapers: UrlScraper which downloads files via HTTP and FileScraper which reads them from the local filesystem. They function almost identically (both use URLs), except that FileScraper substitutes the base URL with a local path before reading a file. FileScraper uses the placeholder localhost base URL by default and includes a filter to remove any URL pointing to it at the end.

To be processed, a response must meet the following requirements:

• 200 status code
• HTML content type
• effective URL (after redirection) contained in the base URL (explained below)

(FileScraper only checks if the file exists and is not empty.)

Each URL is requested only once (case-insensitive).

Configuration

Configuration is done via class attributes and divided into three main categories:

• Attributes — essential information such as name, version, URL, etc.
• Filter stacks — the list of filters that will be applied to each page.
• Filter options — the options passed to said filters.

Note: scrapers are located in the lib/docs/scrapers directory. The class's name must be the CamelCase equivalent of the filename.

Attributes

• name [String] Must be unique. Defaults to the class's name.

• slug [String] Must be unique, lowercase, and not include dashes (underscores are ok). Defaults to name lowercased.

• type [String] (required, inherited) Defines the CSS class name (_[type]) and custom JavaScript class (app.views.[Type]Page) that will be added/loaded on each page. Documentations sharing a similar structure (e.g. generated with the same tool or originating from the same website) should use the same type to avoid duplicating the CSS and JS. Must include lowercase letters only.

• release [String] (required) The version of the software at the time the scraper was last run. This is only informational and doesn't affect the scraper's behavior.

• base_url [String] (required in UrlScraper) The documents' location. Only URLs inside the base_url will be scraped. "inside" more or less means "starting with" except that /docs is outside /doc (but /doc/ is inside). Defaults to localhost in FileScraper. (Note: any iframe, image, or skipped link pointing to localhost will be removed by the CleanLocalUrls filter; the value should be overridden if the documents are available online.) Unless root_path is set, the root/initial URL is equal to base_url.

• root_path [String] (inherited) The path from the base_url of the root URL.

• initial_paths [Array] (inherited) A list of paths (from the base_url) to add to the initial queue. Useful for scraping isolated documents. Defaults to []. (Note: the root_path is added to the array at runtime.)

• dir [String] (required, FileScraper only) The absolute path where the files are located on the local filesystem. Note: FileScraper works exactly like UrlScraper (manipulating the same kind of URLs) except that it substitutes base_url with dir in order to read files instead of making HTTP requests.

• params [Hash] (inherited, UrlScraper only) Query string parameters to append to every URL. (e.g. { format: 'raw' } → ?format=raw) Defaults to {}.

• abstract [Boolean] Make the scraper abstract / not runnable. Used for sharing behavior with other scraper classes (e.g. all MDN scrapers inherit from the abstract Mdn class). Defaults to false.

Filter stacks

Each scraper has two filter stacks: html_filters and text_filters. They are combined into a pipeline (using the HTML::Pipeline library) which causes each filter to hand its output to the next filter's input.

HTML filters are executed first and manipulate a parsed version of the document (a Nokogiri node object), whereas text filters manipulate the document as a string. This separation avoids parsing the document multiple times.

Filter stacks are like sorted sets. They can modified using the following methods:

push(*names)                 # append one or more filters at the end
insert_before(index, *names) # insert one filter before another (index can be a name)
insert_after(index, *names)  # insert one filter after another (index can be a name)
replace(index, name)         # replace one filter with another (index can be a name)

"names" are require paths relative to Docs (e.g. jquery/clean_html → Docs::Jquery::CleanHtml).

Default html_filters:

• ContainerFilter — changes the root node of the document (remove everything outside)
• CleanHtmlFilter — removes HTML comments, <script>, <style>, etc.
• NormalizeUrlsFilter — replaces all URLs with their fully qualified counterpart
• InternalUrlsFilter — detects internal URLs (the ones to scrape) and replaces them with their unqualified, relative counterpart
• NormalizePathsFilter — makes the internal paths consistent (e.g. always end with .html)
• CleanLocalUrlsFilter — removes links, iframes and images pointing to localhost (FileScraper only)

Default text_filters:

• InnerHtmlFilter — converts the document to a string
• CleanTextFilter — removes empty nodes
• AttributionFilter — appends the license info and link to the original document

Additionally:

• TitleFilter is a core HTML filter, disabled by default, which prepends the document with a title (<h1>).
• EntriesFilter is an abstract HTML filter that each scraper must implement and responsible for extracting the page's metadata.

Filter options

The filter options are stored in the options Hash. The Hash is inheritable (a recursive copy) and empty by default.

More information about how filters work is available on the Filter Reference page.

• ContainerFilter

  • :container [String or Proc] A CSS selector of the container element. Everything outside of it will be removed and become unavailable to the other filters. If more than one element match the selector, the first one inside the DOM is used. If no elements match the selector, an error is raised. If the value is a Proc, it is called for each page with the filter instance as argument, and should return a selector or nil. The default container is the <body> element. Note: links outside of the container element will not be followed by the scraper. To remove links that should be followed, use a CleanHtml filter later in the stack.

• NormalizeUrlsFilter The following options are used to modify URLs in the pages. They are useful to remove duplicates (when the same page is accessible from multiple URLs) and fix websites that have a bunch of redirections in place (when URLs that should be scraped, aren't, because they are behind a redirection which is outside of the base_url — see the MDN scrapers for examples of this).

  • :replace_urls [Hash] Replaces all instances of a URL with another. Format: { 'original_url' => 'new_url' }
  • :replace_paths [Hash] Replaces all instances of a sub-path (path from the base_url) with another. Format: { 'original_path' => 'new_path' }
  • :fix_urls [Proc] Called with each URL. If the returned value is nil, the URL isn't modified. Otherwise the returned value is used as replacement.

  Note: before these rules are applied, all URLs are converted to their fully qualified counterpart (http://...).

• InternalUrlsFilter

  Internal URLs are the ones inside the scraper's base_url ("inside" more or less means "starting with", except that /docs is outside /doc). They will be scraped unless excluded by one of the following rules. All internal URLs are converted to relative URLs inside the pages.

  • :skip_links [Boolean or Proc] If false, does not convert or follow any internal URL (creating a single-page documentation). If the value is a Proc, it is called for each page with the filter instance as argument.
  • :follow_links [Proc] Called for page with the filter instance as argument. If the returned value is false, does not add internal URLs to the queue.
  • :trailing_slash [Boolean] If true, adds a trailing slash to all internal URLs. If false, removes it. This is another option used to remove duplicate pages.
  • :skip [Array] Ignores internal URLs whose sub-paths (path from the base_url) are in the Array (case-insensitive).
  • :skip_patterns [Array] Ignores internal URLs whose sub-paths match any Regexp in the Array.
  • :only [Array] Ignores internal URLs whose sub-paths aren't in the Array (case-insensitive) and don't match any Regexp in :only_patterns.
  • :only_patterns [Array] Ignores internal URLs whose sub-paths don't match any Regexp in the Array and aren't in :only.

  If the scraper has a root_path, the empty and / paths are automatically skipped. If :only or :only_patterns is set, the root path is automatically added to :only.

  Note: pages can be excluded from the index based on their content using the Entries filter. However, their URLs will still be converted to relative in the other pages and trying to open them will return a 404 error. Although not ideal, this is often better than having to maintain a long list of :skip URLs.

• AttributionFilter

  • :attribution [String] (required) An HTML string with the copyright and license information. See the other scrapers for examples.

• TitleFilter

  • :title [String or Boolean or Proc] Unless the value is false, adds a title to every page. If the value is nil, the title is the name of the page as determined by the Entries filter. Otherwise the title is the String or the value returned by the Proc (called for each page, with the filter instance as argument). If the Proc returns nil or false, no title is added.
  • :root_title [String or Boolean] Overrides the :title option for the root page only.

  Note: this filter is disabled by default.

Keeping scrapers up-to-date

In order to keep scrapers up-to-date the get_latest_version(opts) method should be overridden. If self.release is defined, this should return the latest version of the documentation. If self.release is not defined, it should return the Epoch time when the documentation was last modified. If the documentation will never change, simply return 1.0.0. The result of this method is periodically reported in a "Documentation versions report" issue which helps maintainers keep track of outdated documentations.

To make life easier, there are a few utility methods that you can use in get_latest_version:

• fetch(url, opts)

  Makes a GET request to the url and returns the response body.

  Example: lib/docs/scrapers/bash.rb

• fetch_doc(url, opts)

  Makes a GET request to the url and returns the HTML body converted to a Nokogiri document.

  Example: lib/docs/scrapers/git.rb

• fetch_json(url, opts)

  Makes a GET request to the url and returns the JSON body converted to a dictionary.

  Example: lib/docs/scrapers/mdn/mdn.rb

• get_npm_version(package, opts)

  Returns the latest version of the given npm package.

  Example: lib/docs/scrapers/bower.rb

• get_latest_github_release(owner, repo, opts)

  Returns the tag name of the latest GitHub release of the given repository. If the tag name is preceded by a "v", the "v" will be removed.

  Example: lib/docs/scrapers/jsdoc.rb

• get_github_tags(owner, repo, opts)

  Returns the list of tags on the given repository (format).

  Example: lib/docs/scrapers/liquid.rb

• get_github_file_contents(owner, repo, path, opts)

  Returns the contents of the requested file in the default branch of the given repository.

  Example: lib/docs/scrapers/minitest.rb