This document is a companion to the IIIF Presentation API Specification, Version 2.0. It describes the significant changes to the API since Version 1.0. The changes are broken into two groups: Breaking Changes, i.e. those that are not backwards compatible from either a client or server perspective (or both); Other Changes, i.e. those that are backwards compatible. A third section, Deferred Proposals, lists proposals that have been discussed but did not make it into this version of the specification.
In addition to changes in the API, the specification documents have been changed as follows:
- The use of RFC 2119 keywords has been made more consistent.
- Language has been adjusted to make the document less focused on paged, digitized objects.
- Semantic Versioning will be used to enumerate releases.
Table of Contents
- Breaking Changes
- Other Changes
- Deferred Proposals
Name and URIs Changed
The name for the API has been changed from “Metadata API” to “Presentation API” with version 2.0. This was to avoid the implication that the description was semantic, bibliographic information which is not, and has never been, the intent. This resulted in the documentation moving to a different place, along with all of the auxilliary technical files like JSON-LD context and frames.
Ranges Easier to Implement
Ranges were changed to follow the top-down pattern of a list of included ranges and canvases, rather than bottom-up where they would assert which parent range they were within. This makes it significantly easier to implement, and follows the rest of the API’s patterns of listed containership.
In order to support different page turning modalities, additional requirements were added to manifests that claim to be
paged. Notably the first canvas is to be shown by itself, and then subsequent canvases can be assumed to be left/right pairs (depending on the
viewingDirection). An additional value of
non-paged was added to the
viewingHint enumeration to assert that a particular Canvas is not part of the paging sequence.
Viewing Hints as URIs
Following the features in the Image API, viewing hints are now URIs defined in the JSON-LD context document. The JSON representation is identical and hence this is a not a breaking change, however extensions must now use URIs not plain string literals.
URI Requirements and Recommendations
The recommended URI patterns were changed to be more inline with the best practice of Cool URIs and no longer have
.json on the end. This is not forbidden, and is still the easiest for the simplest server of just files on disk, but we should not recommend bad practice, no matter how easy. Secondly, the recommendation for Canvas URIs being HTTP was upgraded to a requirement, as we rely on media fragments, which are only defined in terms of HTTP URIs.
Section 7.6 was added to describe collections of manifests, and sub-collections. This allows discovery in a pragmatic and simple way, that is easy to understand and follows the same structure and approach as the rest of the Presentation API. Collections may embed other collections, in the same way that Manifests embed Sequences, but may not embed Manifests.
Several new fields were added:
thumbnail(recommended for Canvas, Manifest and Collection)
related(although was used in practice in 1.0)
Services Clarified and Extended
In order to manage requests for features that are not universally applicable, but still useful, the service construction that was previously under-specified has been extended to allow additional external specifications to be embedded or referenced. Services are now listed in an annex document, and include the oft-discussed geo-tagging and physical dimensions features.
Server-side Image Rotation Option
Added and described an Open Annotation Selector object that allows specifying the parameters for an Image API URI separately. The original use case was server side rotation of a segment image, however all of the parameters could be useful in different situations.
Image Annotation Requirements Reduced
In order to reduce unnecessary repetition of information, some of the requirements for Image Annotations were dropped, including the need for height, width and format.
A new relationship (
startCanvas) was added to Sequences to allow specifying the URI of the canvas resource which should initially be displayed to the user. This is to enable viewers to skip past non-content canvases such as empty fly-leaves, the table of contents, and so forth.
An additional value of
top was added to the
viewingHint value enumeration, to be used on a range which is the top-most level in a table of contents or other structure.
Use of HTML Clarified
The use of HTML in the values of fields was permitted but inadequately specified in 1.0. This version makes it much clearer and easier to determine the content of the value, and provides solid guidance as to which HTML features are allowed.
In 1.0 it was not possible to add most of the fields to content resources. This was not for any good, or even known, reason and the restrictions were lifted.
Clarified that any reference to a location outside of the dimensions of a Canvas is an error.
Defined Layer Representation
The representation of a Layer was undefined, and hence implementers could not know what to return if its URI was dereferenced. A structure was defined following the pattern of other container objects.
The underlying Shared Canvas data model has an additional construction called a Zone. At least one institution has implemented this construction and is unable to adopt the Presentation API until Zones are included. At least one further institution recognizes that they have requirements that are solved by zones. The issue was deferred until 2.1 as the solution is not easy to implement, and can be overlaid without adding backwards incompatibility.
Dynamic Annotation List Services
An outstanding issue is the ability to associate AnnotationLists with a canvas dynamically rather than through explicit references in the Manifest. The requirements were decided to be not clearly enough defined at this stage. The major use cases also involved Authentication and Authorization, which have been deferred as well. With the change to services, it is expected that this request can be experimented with and either blessed as a service or promoted to a full feature in the future without backwards incompatibility.
Annotations on non-Canvas Resources
Only canvases may list their annotations, and thus it is impossible to refer to annotations on the placeholder for the physical object (the manifest). It is not impossible to annotate manifests, it is just that there isn’t a link from the manifest to its annotations. Without a set of use cases to justify the addition of the feature, it was decided to defer this and experiment with services in the same manner as for Dynamic Annotation Lists.
Particularly in the teaching and learning domain, it may be useful to specify the intended audience of resources. For commentary annotations this is important, and not covered in the base Open Annotation specification. It was decided to defer this until the Open Annotation community can determine a resolution to be adopted, but to promote use cases and existing solutions such as the IDPF use of schema.org’s Audience classes. Also, it is backwards compatible and no clients are ready to use it yet.
The Image API has a very well formed set of compliance requirements. The Presentation API, conversely, does not have any. This was not seen as a requirement that needed to be solved for 2.0, and was deferred for future work.
The Image API in 2.0 has a protocol field that makes the assertion that the
info.json document is part of the IIIF Image API protocol. The Presentation API does not have this field. There were no features or implementations identified that would make use of the field, and as it will be backwards compatible it was deferred until a requirement is expressed.
Minimum/Maximum Size Bounds for Annotation Rendering
The content or commentary resources linked via annotations to a Canvas may be only useful to render at certain sizes, such as trying to render an image below 10 pixels width or text at less than 4 points. While theoretically useful, no real world use cases have been presented that would justify its inclusion. As a backwards compatible new feature, it was deferred until a requirement is expressed.