Fixed: Sanitize now strips null bytes (
\u0000) before passing input to Nokogumbo, since they can cause recent versions to crash with a failed assertion in the Gumbo parser.
srcsetto the attribute whitelist for
imgelements in the relaxed config. @ejtttje - #156
- Methods can now be used as transformers. @Skipants - #155
-webkit-font-smoothingto the relaxed CSS config. @louim - #154
Fixed: Nokogumbo >=1.4.9 changed its behavior in a way that allowed invalid doctypes (like
<!DOCTYPE nonsense>) when the
:allow_doctypeconfig setting was
true. Invalid doctypes are now coerced to valid ones as they were prior to this Nokogumbo change.
- Added a new CSS config setting,
:import_url_validator. This is a Proc or other callable object that will be called with each
@importURL, and should return
trueto allow the URL or
falseto remove it. @nikz - #153
- Unpinned the Nokogumbo dependency. @rubys - #141
Potentially breaking changes
Added two new CSS config settings,
:at_rules_with_styles. These allow you to define which at-rules should be allowed to contain properties and which should be allowed to contain style rules. Previously this was hard-coded internally. #111
:at_rulessetting still exists, and defines at-rules that may not have associated blocks, such as
@import. If you have a custom config that contains an
:at_rulessetting, you may need to move rules can have blocks to either
See Sanitize's relaxed config for an example.
Added full support for CSS
@pagerules in the relaxed config, including support for all page-margin box rules (such as
Added the following CSS at-rules to the relaxed config:
Added a whole bunch of CSS properties to the relaxed config. View the complete list here.
Small performance improvements.
Fixed: Upgraded Crass to 1.0.2 to pick up a fix that affected the parsing of CSS
- Fixed: Deleting a node in a custom transformer could trigger a memory leak in Nokogiri if that node's children were later reparented, which the built-in CleanElement transformer did by default. The CleanElement transformer is now careful not to reparent the children of deleted nodes. #129
#fragmentfailed on frozen strings, and could unintentionally modify unfrozen strings if they used an encoding other than UTF-8 or if they contained characters not allowed in HTML. @AnchorCat - #128
Added the following CSS properties to the relaxed config. @ehudc - #120
Updated Nokogumbo to 1.2.0 to pick up a fix for a Gumbo bug where the entity
Æleft its semicolon behind when it was converted to a character during parsing. #119
- Fixed: Harmless whitespace preceding a URL protocol (such as " http://") caused the URL to be removed even when the protocol was whitelisted. @benubois - #126
- Fixed: Some CSS selectors weren't parsed correctly inside the body of a
@mediablock, causing them to be removed even when whitelist rules should have allowed them to remain. #121
- Updated Nokogumbo to 1.1.12, because 1.1.11 silently reverted the change we were trying to pick up in the last release. Now issue #114 is actually fixed.
- Updated Nokogumbo to 1.1.11 to pick up a fix for a Gumbo bug in which certain
HTML character entities, such as
Ö, were parsed incorrectly, leaving the semicolon behind in the output. #114
As of this version, Sanitize adheres strictly to the SemVer 2.0.0 versioning standard. This release contains API and output changes that are incompatible with previous releases, as indicated by the major version increment.
HTML is now parsed using Google's Gumbo HTML5 parser, which adheres to the HTML5 parsing spec and behaves much more like modern browser parsers than the previous libxml2-based parser. As a result, HTML output may differ from that of previous versions of Sanitize.
All transformers now traverse the document from the top down, starting with the first node, then its first child, and so on. The
:transformers_breadthconfig has been removed, and old bottom-up transformers (the previous default) may need to be rewritten.
Sanitize's built-in configs are now deeply frozen to prevent people from modifying them (either accidentally or maliciously). To customize a built-in config, create a new copy using
Sanitize::Config.merge(), like so:
Sanitize.fragment(html, Sanitize::Config.merge(Sanitize::Config::BASIC, :elements => Sanitize::Config::BASIC[:elements] + ['div', 'table'], :remove_contents => true ))
clean_document!methods were removed, since they weren't useful and tended to confuse people.
cleanmethod was renamed to
fragmentto more clearly indicate that its intended use is to sanitize an HTML fragment.
clean_documentmethod was renamed to
clean_node!method was renamed to
documentmethod now raises a
<html>element isn't whitelisted, rather than a
RuntimeError. This error is also now raised regardless of the
:outputconfig has been removed. Output is now always HTML, not XHTML.
:output_encodingconfig has been removed. Output is now always UTF-8.
Added advanced CSS sanitization support using Crass, which is fully compliant with the CSS Syntax Module Level 3 parsing spec. The contents of whitelisted
styleattributes in HTML will be sanitized as CSS, or you can use the
Sanitize::CSSclass to manually sanitize CSS stylesheets or properties.
true, well-formed doctype definitions will be allowed in documents. When
false(the default), doctype definitions will be removed from documents. Doctype definitions are never allowed in fragments, regardless of this setting.
Added the following elements to the relaxed config, in addition to various attributes:
:whitespace_elementsconfig is now a Hash, and allows you to specify the text that should be inserted before and after these elements when they're removed. The old-style Array-based config value is still supported for backwards compatibility. @alperkokmen - #94
Unsuitable Unicode characters are now removed from HTML before it's parsed. #106
Fixed: Non-tag brackets in input like
"1 > 2 and 2 < 1"are now parsed and escaped correctly in accordance with the HTML5 spec, becoming
"1 > 2 and 2 < 1". #83
Fixed: Siblings added after the current node during traversal are now also traversed. In previous versions they were simply skipped. #91
Fixed: Nokogiri has been smacked and instructed to stop adding newlines after certain elements, because if people wanted newlines there they'd have put them there, dammit. #103
Fixed: Added a workaround for a libxml2 bug that caused an undesired content-type meta tag to be added to all documents with
<head>elements. Nokogiri #1008
Added support for whitelisting arbitrary HTML5
data-*attributes. Use the symbol
:datainstead of an attribute name in the
:attributesconfig to indicate that arbitrary data attributes should be allowed on an element.
Added the following elements to the relaxed config:
Fixed: A colon (
:) character in a URL fragment identifier such as
#foo:1was incorrectly treated as a protocol delimiter. @heathd - #87
- Fixed: Version 2.0.5 inadvertently included some work-in-progress changes that shouldn't have made their way into the master branch. This is what happens when I release before coffee instead of after.
- Loosened the Nokogiri dependency back to >= 1.4.4 to allow Sanitize to coexist in newer Rubies with other libraries that restrict Nokogiri to 1.5.x for 1.8.7 compatibility. Sanitize still no longer supports 1.8.7, but this should make life easier for people who need those other libs.
Sanitize.clean_document, which sanitizes a full HTML document rather than just a fragment. [Ben Anderson]
Nokogiri dependency bumped to 1.6.x.
Dropped support for Ruby versions older than 1.9.2.
- Loosened the Nokogiri dependency to allow Nokogiri 1.5.x.
- Fixed a bug in which a protocol like "java\script:" would be translated to "java%5Cscript:" and allowed through the filter when relative URLs were enabled. This didn't actually allow malicious code to run, but it is undesired behavior.
- Updated the protocol regex to anchor at the beginning of the string rather than the beginning of a line. [Eaden McKee]
The environment data passed into transformers and the return values expected from transformers have changed. Old transformers will need to be updated. See the README for details.
Transformers now receive nodes of all types, not just element nodes.
Sanitize's own core filtering logic is now implemented as a set of always-on transformers.
The default value for the
:outputconfig is now
:html. Previously it was
:whitespace_elementsconfig, which specifies elements (such as
<p>) that should be replaced with whitespace when removed in order to preserve readability. See the README for the default list of elements that will be replaced with whitespace when removed.
:transformers_breadthconfig, which may be used to specify transformers that should traverse nodes in a breadth-first mode rather than the default depth-first mode.
varelements to the whitelists for the basic and relaxed configs.
wbrelements to the whitelist for the relaxed config.
titleattributes are now whitelisted for all elements in the relaxed config.
Bumped minimum Nokogiri version to 1.4.4 to avoid a bug in 1.4.2+ (issue #315) that caused
</body></html>to be appended to the CDATA inside unterminated script and style elements.
:remove_contentsconfig setting. If set to
true, Sanitize will remove the contents of all non-whitelisted elements in addition to the elements themselves. If set to an array of element names, Sanitize will remove the contents of only those elements (when filtered), and leave the contents of other filtered elements. [Thanks to Rafael Souza for the array option]
:output_encodingconfig setting to allow the character encoding for HTML output to be specified. The default is utf-8.
The environment hash passed into transformers now includes a
:node_nameitem containing the lowercase name of the current HTML node (e.g. "div").
Returning anything other than a Hash or nil from a transformer will now raise a meaningful
Sanitize::Errorexception rather than an unintended
Requires Nokogiri ~> 1.4.1.
Added support for transformers, which allow you to filter and alter nodes using your own custom logic, on top of (or instead of) Sanitize's core filter. See the README for details and examples.
Sanitize.clean_node!, which sanitizes a
Nokogiri::XML::Nodeand all its children.
<h6>to the Relaxed whitelist. [Suggested by David Reese]
Migrated from Hpricot to Nokogiri. Requires libxml2 >= 2.7.2 [Adam Hooper]
:outputconfig setting to allow the output format to be specified. Supported formats are
:xhtml(the default) and
:html(which outputs HTML4).
Changed protocol regex to ensure Sanitize doesn't kill URLs with colons in path segments. [Peter Cooper]
- Added a workaround for an Hpricot bug that prevents attribute names from being downcased in recent versions of Hpricot. This was exploitable to prevent non-whitelisted protocols from being cleaned. [Reported by Ben Wanicur]
Requires Hpricot 0.8.1+, which is finally compatible with Ruby 1.9.1.
Fixed a bug that caused named character entities containing digits (like
²) to be escaped when they shouldn't have been. [Reported by Sebastian Steinmetz]
Removed htmlentities gem dependency.
Existing well-formed character entity references in the input string are now preserved rather than being decoded and re-encoded.
'character is now encoded as
'to prevent problems in IE6.
You can now specify the symbol
:allin place of an element name in the attributes config hash to allow certain attributes on all elements. [Thanks to Mutwin Kraus]
Fixed a bug introduced in version 1.0.3 that prevented non-whitelisted protocols from being cleaned when relative URLs were allowed. [Reported by Dev Purkayastha]
Fixed "undefined method `parent='" exceptions caused by parser changes in edge Hpricot.
- Fixed a bug that made it possible to sneak a non-whitelisted element through by repeating it several times in a row. All versions of Sanitize prior to 1.0.4 are vulnerable. [Reported by Cristobal]
- Fixed a bug whereby incomplete Unicode or hex entities could be used to prevent non-whitelisted protocols from being cleaned. Since IE6 and Opera still decode the incomplete entities, users of those browsers may be vulnerable to malicious script injection on websites using versions of Sanitize prior to 1.0.3.
- Fixed a bug that caused an exception to be thrown when parsing a valueless attribute that's expected to contain a URL.
You can now specify
:relativein a protocol config array to allow attributes containing relative URLs with no protocol. The Basic and Relaxed configs have been updated to allow relative URLs.
Added a workaround for an Hpricot bug that causes HTML entities for non-ASCII characters to be replaced by question marks, and all other entities to be destructively decoded.
- First release.