Long looooong ago, I wrote a deep review of the XHTML 2.0 spec that was one of the elements that led to the resuming of the HTML activity at the W3C and the final dismissal of XHTML 2.0. 

Long ago, I started a similar effort on EPUB that led to Dave Cramer's EPUB Zero. It's time (fr-FR) to draw some conclusions.

This document is maintained on GitHub and accepting contributions.

Daniel Glazman

OCF

  1. EPUB publications are not "just a zip". They are zips with special constraints. I question the "mimetype file in first uncompressed position" constraint since I think the vast majority of reading systems don't care (and can't care) because most of the people creating EPUB at least partially by hand don't know/care. The three last contraints (zip container fields) on the ZIP package described in section 4.2 of the spec are usually not implemented by Reading Systems.
  2. the container element of the META-INF/container.xml file has a version attribute that is always "1.0", whatever the EPUB version. That forces editors, filters and reading systems to dive into the default rendition to know the EPUB version and that's clearly one useless expensive step too much.
  3. having multiple renditions per publication reminds me of MIME multipart/alternative. When Borenstein and Freed added it to the draft of RFC 1341 some 25 years ago, Mail User Agents developers (yours truly counted) envisioned and experimented far more than alternatives between text/html and text/plain only. I am under the impression multiple renditions start from the same good will but fail to meet the goals for various reasons:
    1. each additional rendition drastically increases the size of the publication...
    2. most authoring systems, filters and converters on the market don't deal very well with multiple renditions
    3. EPUB 2 defined the default rendition as the first rendition with a application/oebps-package+xml  mimetype while the EPUB 3 family of specs defines it as the first rendition in the container
    4. while a MIME-compliant Mail User Agent will let you compose a message in text/html and output for you the multipart/alternative between that text/html and its text/plain serialization, each Publication rendition must be edited separately.
    5. in the case of multiple renditions, each rendition has its own metadata and it's then legitimate to think the publication needs its own metadata too. But the META-INF/metadata.xml file has always been quoted as "This version of the OCF specification does not define metadata for use in the metadata.xml file. Container-level metadata may be defined in future versions of this specification and in IDPF-defined EPUB extension specifications." by all EPUB specifications. The META-INF/metadata.xml should be dropped.
    6. encryption, rights and signatures are per-publication resources while they should be per-rendition resources.
  4. the full-path attribute on rootfile elements is the only path in a publication that is relative to the publication's directory. All other URIs (for instance in href attributes) are relative to the current document instance. I think full-path should be deprecated in favor of href here, and finally superseded by href for the next major version of EPUB.
  5. we don't need the mimetype attribute on rootfile elements since the prose of EPUB 3.1 says the target of a rootfile must be a Package Document, i.e. an OPF file... If EPUB 2 OCF could directly target for instance a PDF file, it's not the case any more for OCF 3.
  6. absolute URIs (for instance /files/xhtml/index.html with a leading slash, cf. path-absolute construct in RFC 3986) are harmful to EPUB+Web convergence
  7. if multiple renditions are dropped, the META-INF/container.xml file becomes useless and it can be dropped.
  8. the prose for the META-INF/manifest.xml makes me wonder why this still exists: "does not mandate a format", "MUST NOT be used". Don't even mention it! Just say that extra unspecified files inside META-INF directory must be ignored (Cf. OCF section 3.5.1) , and possibly reserve the metadata.xml file name, period. Oh, and a ZIP is also a manifest of files...
  9. I am not an expert of encryption, signatures, XML-ENC Core or XML DSIG Core, so I won't discuss encryption.xml and signatures.xml files
  10. the rights.xml file still has no specified format. Strange. Cf. item 8 above.
  11. Resource obfuscation is so weak and useless it's hilarious. Drop it. It is also painful in EPUB+Web convergence.
  12. RNG schemas are not enough in the OCF spec. Section 3.5.2.1 about container.xml file for instance shouls have models and attribute lists for each element, similarly to the Packages spec.
  13. not sure I have ever seen links and link elements in a container.xml file... (Cf. issue #374). The way these links are processed is unspecified anyway. Why are these elements normatively specified since extra elements are allowed - and explicitely ignored by spec - in the container?

Packages

  1. a Package consists of one rendition only.
  2. I have never understood the need for a manifest of resources inside the package, probably because my own publications don't use Media Overlays or media fallbacks.
  3. fallbacks are an inner mechanism also similar to multipart/alternative for renditions. I would drop it.
  4. I think the whole package should have properties identifying the required technologies for the rendition of the Package (e.g. script), and avoid these properties on a per-item basis that makes no real sense. The feature is either present in the UA for all content documents or for none.
  5. the spine element represent the default reading order of the package. Basically, it's a list. We have lists in html, don't we? Why do we need a painful and complex proprietary xml format here?
  6. the name of the linear attribute, that discriminates between primary and supplementary content, is extremely badly chosen. I always forget what really is linear because of that.
  7. Reading Systems are free, per spec, to completely ignore the linear attribute, making it pointless from an author's point of view.
  8. I have never seen the collection element used and I don't really understand why it contains link elements and not itemref elements
  9. metadata fun, as always with every EPUB spec. Implementing refines in 3.0 was a bit of a hell (despite warnings to the EPUB WG...), and it's gone from 3.1, replaced by new attributes. So no forwards compatbility, no backwards compatibility. Yet another parser and serializer for EPUB-compliant user agents.
  10. the old OPF guide element is now a html landmarks list, proving it's feasible to move OPF features to html
  11. the Navigation Document, an html document, is mandatory... So all the logics mentioned above could be there.
  12. Fixed Layout delenda est. Let's use CSS Fragmentation to make sure there's no orphaned content in the document post-pagination, and if CSS Fragmentation is not enough, make extension contributions to the CSS WG.
  13. without 3.0 refines, there is absolutely nothing any more in 3.1 preventing Package's metadata to be expressed in html; in 3.0, the refines attribute was a blocker, implying an extension of the model of the meta html element or another ugly IDREF mechanism in html.
  14. the prefix attribute on the package element is a good thing and should be preserved
  15. the rendition-flow property is weird, its values being paginated, scrolled-continuous, scrolled-doc and auto. Where is paginated-doc, the simplest paginated mode to implement?
  16. no more NCX, finally...
  17. the Navigation Document is already a html document with nav elements having a special epub:type/role (see issue #941), that's easy to make it contain an equivalent to the spine or more.

Content Documents

  1. let's get rid of the epub namespace, please...

Media Overlays

  1. SMIL support across rendering engines is in very bad shape and the SMIL polyfill does not totally help. Drop Media Overlays for the time being and let's focus on visual content.

Alternate Style Tags

  1. terrible spec... Needed because of Reading Systems' limitations but still absolutely terrible spec...
  2. If we get rid of backwards compatibility, we can drop it. Submit extensions to Media Queries if needed.

On the fly conclusions

  1. backwards compatibility is an enormous burden on the EPUB ecosystem
  2. build a new generation of EPUB that is not backwards-compatible
  3. the mimetype file is useless
  4. file extension of the publication MUST be well-defined
  5. one rendition only per publication and no more links/link elements
  6. in that case, we don't need the container.xml file any more
  7. metadata.xml and manifest.xml files removed
  8. we may still need encryption.xml, signatures.xml and rights.xml inside a META-INF directory (or directly in the package's root after all) to please the industry.
  9. application/oebps-package+xml mimetype is not necessary
  10. the EPUB spec evolution model is/was "we must deal with all cases in the world, we move very fast and we fix the mistakes afterwards". I respectfully suggest a drastic change for a next generation: "let's start from a low-level common ground only and expand slowly but cleanly"
  11. the OPF file is not needed any more. The root of the unique rendition in a package should be the html Navigation Document.
  12. the metadata of the package should be inside the body element of the Navigation Document
  13. the spine of the package can be a new nav element inside the body of the Navigation Document
  14. intermediary summary: let's get rid of both META-INF/container.xml and the OPF file... Let's have the Navigation Document mandatorily named index.xhtml so a directory browsing of the uncompressed publication through http will render the Navigation Document.
  15. let's drop the Alternate Style Tags spec for now. Submission of new Media Queries to CSS WG if needed.
  16. let's drop Media Overlays spec for now.

CONCLUSION

EPUB is a monster, made to address very diverse markets and ecosystems, too many markets and ecosystems. It's weak, complex, a bit messy, disconnected from the reality of the Web it's supposed to be built upon and some claim (link in fr-FR) it's too close to real books to be disruptive and innovative.

I am then suggesting to severe backwards compatibility ties and restart almost from scratch, and entirely and purely from W3C Standards. Here's the proposed result:


1. E0 Publication

A E0 Publication is a ZIP container. Files in the ZIP must be stored as is (no compression) or use the Deflate algorithm. File and directory names must be encoded in UTF-8 and follow some restrictions (see EPUB 3.1 filename restrictions).

The file name of a E0 Publication MUST use the e0 file extension.

A E0 Publication MUST contain a Navigation Document. It MAY contain files encryption.xml, signatures.xml and rights.xml (see OCF 3.1 for more information). All these files must be placed directly inside the root of the E0 Publication.

A E0 Publication can also contain Content Documents and their resources.

Inside a E0 Publication, all internal references (links, anchors, references to replaced elements, etc) MUST be strictly relative. With respect to section 4.2 of RFC 3986, only path-noscheme and path-empty are allowed for IRIs' relative-part. External references are not restricted.

2. E0 Navigation Document

A E0 Navigation Document is a html document. Its file name MUST be index.xhtml if the document is a XML document and index.html if it it is not a XML document. A E0 Publication cannot contain both index.html and index.xhtml files.

A E0 Navigation Document contains at least one header element (for metadata) and at least two nav html elements (for spine and table of contents) in its body element.

2.1. E0 Metadata in the Navigation Document

E0 metadata are designed to be editable in any Wysiwyg html editor, and potentially rendered as regular html content by any Web browser.

E0 metadata are expressed inside a mandatory header html element inside the Navigation Document. That element must carry the "metadata" ID and the vocab attribute with value "http://www.idpf.org/2007/opf". All metadata inside that header element are then expressed using html+RDFa Lite 1.1. E0 metadata reuse EPUB 3.1 metadata and corresponding unicity rules, expressed in a different way.

Refinements of metadata are expressed through nesting of elements.

Example:

<header id="metadata"
        vocab="http://www.idpf.org/2007/opf">
<h1>Reading Order</h1> <ul> <li>Author: <span property="dc:creator">glazou
(<span property="file-as">Glazman, Daniel</span>)</span></li> <li>Title: <span property="dc:title">E0 Publications</span></li> </ul> </header>

The mandatory title element of the Navigation Document, contained in its head element, should have the same text contents than the first "dc:title" metadata inside that header element.

2.2. E0 Spine

The spine of a E0 Publication is expressed in its Navigation Document as a new nav element holding the "spine" ID. The spine nav element is mandatory.

See EPUB 3.1 Navigation Document.

2.3. E0 Table of Contents

The Table of Contents of a E0 Publication is expressed in its Navigation Document as a nav element carrying the "toc" ID. The Table of Contents nav element is mandatory.

See EPUB 3.1 Navigation Document.

2.4. E0 Landmarks

The Landmarks of a E0 Publication is expressed in its Navigation Document as a nav element carrying the "landmarks" ID. The Landmarks nav element is optional.

See EPUB 3.1 Navigation Document.

2.5. Other nav elements

The Navigation Document may include one or more nav elements. These additional nav elements should have an role attribute to provide a machine-readable semantic, and must have a human-readable heading as their first child.

IDs "metadata", "spine" , 'landmarks" and "toc" are reserved in the Navigation Document and must not be used by these extra nav elements.

2.6. Example of a Navigation Document

<!DOCTYPE html>
<html lang="en">
  <head>
    <meta content="text/html; charset=UTF-8" http-equiv="content-type">
    <title>Moby-Dick</title>
  </head>
  <body>
    <header id="metadata"
            vocab="http://www.idpf.org/2007/opf">
      <ul>
        <li>Author:
          <span property="dc:creator">Herman Melville
(<span property="file-as">Melville Herman</span>)</span></li> <li>Title: <span property="dc:title">Moby-Dick</span></li> <li>Identifier: <span property="dc:identifier">glazou.e0.samples.moby-dick</span></li> <li>Language: <span property="dc:language">en-US</span></li> <li>Last modification: <span property="dcterms:modified">2017-01-17T11:16:41Z</span></li> <li>Publisher: <span property="dc:publisher">Harper & Brothers, Publishers</span></li> <li>Contributor: <span property="dc:contributor">Daniel Glazman</span></li> </ul> </header> <nav id="spine"> <h1>Default reading order</h1> <ul> <li><a href="cover.html">Cover</a></li> <li><a href="titlepage.html">Title</a></li> <li><a href="toc-short.html">Brief Table of Contents</a></li> ... </ul> </nav> <nav id="toc" role="doc-toc"> <h1>Table of Contents</h1> <ol> <li><a href="titlepage.html">Moby-Dick</a></li> <li><a href="preface_001.html">Original Transcriber’s Notes:</a></li> <li><a href="introduction_001.html">ETYMOLOGY.</a></li> ... </ol> </nav> </body> </html>

3. Directories

A E0 Publication may contain any number of directories and nested directories.

4. E0 Content Documents

E0 Content Documents are referenced from the Navigation Document. E0 Content Documents are html documents.

E0 Content Documents should contain <link rel="prev"...> and <link rel="next"...> elements in their head element conformant to the reading order of the spine present in the Navigation Document. Content Documents not present in that spine don't need such elements.

The epub:type attribute is superseded by the role attribute and must not be used.

5. E0 Resources

E0 Publications can contain any number of extra resources (CSS stylesheets, images, videos, etc.) referenced from either the Navigation Document or Content Documents.