Repository navigation
This release contains security fixes and a few behavior changes, see below. There will be no further 3.4.x release.
Changelog
- Fixed a further
SAFE_FOR_TEMPLATESbypass where nested tags could reassemble expressions on reparse, thanks @e1abrador - Fixed several ways for DOM-built input to get a template expression past
SAFE_FOR_TEMPLATES: entity-encoded text in astyleelement nested in atextareaortitle, text hidden in the children of a void element, NUL characters splitting a marker, and comments that end early on reparse - Fixed several gaps in the
IN_PLACEfail-closed handling: a rejected root is now neutralized inside its open shadow roots, a failed call no longer leaves text or comments in the root, and a removed node that a hook or custom element puts back is sanitized again - Fixed a case where
IN_PLACEnever returned because a custom element or hook kept re-inserting a removed node, the call now fails closed - Fixed custom elements kept via
CUSTOM_ELEMENT_HANDLINGskipping the namespace check - Fixed a crash in the attribute loop when an
uponSanitizeAttributehook removed several attributes of the current node - Fixed a
sanitize()call made from inside a hook or config callback replacing the config of the outer call - Fixed the types for
RETURN_DOM, see #1728, thanks @YevheniiKotyrlo - Added in-depth hardening against clobbering for
IN_PLACE - Added regression tests for all of the above
- Reworked the CI setup: one always-on workflow with a single required check, a read-only build job for the SLSA provenance, and a check that the tagged
dist/matches a clean build - Extended the Sigstore signatures to all four bundles covered by the provenance
- Updated the README, SECURITY.md and wiki
- Updated several dependencies
Behavioral changes
SAFE_FOR_TEMPLATESis guaranteed for string input. For DOM node input andIN_PLACEit is best effort, because a tree the HTML parser did not build can change shape when it is parsed again- A name passing
CUSTOM_ELEMENT_HANDLING.tagNameCheckis only kept in HTML. Inside SVG or MathML it is not a custom element and is now removed - When a DOM node is passed in, a
textareaortitleholding an element child and a void element holding children are removed. The HTML parser never produces these, so string input is not affected - A
sanitize()call made from a hook no longer changes how the rest of the outer call behaves. Code that relied on the old side effect needs to pass the config explicitly IN_PLACEnow throws when a hook or custom element keeps putting a removed node back, instead of returning the node or running indefinitely