v4.0.0

Added

• Additional functionality to the JSON FFI including parsing utilities (#1079)
  and conversion between the Cedar and JSON formats (#1087)
• (*) Schema JSON syntax now accepts a type EntityOrCommon representing a
  typename that can resolve to either an entity or common type, matching the
  behavior of typenames written in the human-readable (Cedar) syntax. (#1060, as
  part of resolving #579)

Changed

• (*) Implemented RFC 70.
  In both the Cedar and JSON schema syntaxes, it is now illegal to define the
  same entity name, common type name, or action name in both the empty namespace
  and a nonempty namespace.
• (*) Implemented RFC 52.
  Names containing __cedar (e.g., __cedar, A::__cedar, __cedar::A, and
  A::__cedar::B) are now invalid. (#969)
• The API around Request::new has changed to remove the Options
  around the entity type arguments. See RFC 55.
• Replaced uses of "natural", "human", "human-readable", and "custom" with "Cedar" (#1114).
  APIs with these names are changed accordingly. E.g., Schema::from_str_natural to Schema::from_cedarschema_str.
  Moreover, the FromStr implementations of Schema and SchemaFragment
  now parse strings in the Cedar schema format. Use Schema::from_json_str and SchemaFragment::from_json_str
  to parse strings in the JSON schema format.
• Significantly reworked all public-facing error types to address some issues
  and improve consistency. See issue #745.
• Finalized the ffi module and cedar-wasm crate which were preview-released
  in 3.2.0. This involved API breaking changes in both. See #757 and #854.
• (*) Changed policy validation to reject comparisons and conditionals between
  record types that differ in whether an attribute is required or optional. (#769)
• (*) Improved consistency between JSON and Cedar schema formats. Both now
  forbid using Bool, Boolean, Entity, Extension, Long, Record, Set,
  and String as common type names. (#1150, resolving #1139)
• Changed the FFI to error on typos or unexpected fields in the input JSON (#1041)
• Changed Policy::parse and Template::parse to accept an Option<PolicyId>
  instead of Option<String> to set the policy id (#1055, resolving #1049)
• PolicySet::template_annotation now returns Option<&str> as opposed to
  Option<String> in the previous version (#1131, resolving #1116)
• Moved <PolicyId as FromStr>::Err to Infallible (#588, resolving #551)
• Removed unnecessary lifetimes from some validation related structs (#715)
• Marked errors/warnings related to parsing and validation as non_exhaustive,
  allowing future variants to be added without a breaking change. (#1137, #1169)

Removed

• (*) Removed unspecified entity type. See RFC 55.
• Removed integration testing harness from the cedar-policy crate. It is now
  in an internal crate, allowing us to make semver incompatible changes. (#857)
• Removed the (deprecated) frontend module in favor of the new ffi module
  introduced in 3.2.0. See #757.
• Removed ParseErrors::errors_as_strings. Callers should consider examining
  the rich data provided by miette::Diagnostic, for instance .help() and
  labels(). Callers can continue using the same behavior by calling
  .iter().map(ToString::to_string). (#882, resolving #543)
• Removed ParseError::primary_source_span. Callers should use the location
  information provided by miette::Diagnostic via .labels() and
  .source_code() instead. (#908)
• Removed Display impl for EntityId in favor of explicit .escaped() and
  .as_ref() for escaped and unescaped representations (respectively) of the
  EntityId; see note there (#921, resolving #884)

Fixed

• (*) JSON format Cedar schemas will now fail to parse if they reference an unknown
  extension type. This was already an error for human-readable schema syntax. (#890, resolving #875)
• (*) Schemas can now reference entity and common types defined in the empty namespace,
  even in contexts occurring in a non-empty namespace. (#1060, resolving #579)