Raw File
<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.2 20190208//EN"
<article xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" dtd-version="1.2" article-type="other">
<article-title>Digital Succession Identifiers</article-title>
<contrib contrib-type="author">
<contrib-id contrib-id-type="orcid">https://orcid.org/0000-0002-5014-4809</contrib-id>
<given-names>E. Castedo</given-names>
<pub-date date-type="pub" publication-format="electronic" iso-8601-date="2023-10-07">
<copyright-statement>© 2023, Ellerman et al</copyright-statement>
<copyright-holder>Ellerman et al</copyright-holder>
<license license-type="open-access">
<ali:license_ref xmlns:ali="http://www.niso.org/schemas/ali/1.0/">https://creativecommons.org/licenses/by/4.0/</ali:license_ref>
<license-p>This document is distributed under a Creative Commons Attribution 4.0 International license.</license-p>
<p><bold>DOCUMENT TYPE</bold>: Technical Specification</p>
<p>The Digital Succession Identifier (DSI) is a persistent identifier for bibliographic
references. This specification of DSI is specific to the initial implementation using
<sec id="introduction">
  <p>For a non-technical background and motivation for this technology,
  refer to <ext-link ext-link-type="uri" xlink:href="https://perm.pub/wk1LzCaCSKkIvLAYObAvaoLNGPc">Why Publish Digital Successions</ext-link>.</p>
<sec id="digital-successions">
  <title>Digital successions</title>
  <p>A digital succession contains digital objects,
  which are fixed finite collections of bits.
  Examples of digital objects include computer files and file system directories (folders).
  Both <ext-link ext-link-type="uri" xlink:href="https://en.wikipedia.org/wiki/Git">git</ext-link> <xref alt="1" rid="ref-enwikiU003Agit" ref-type="bibr">1</xref> and <ext-link ext-link-type="uri" xlink:href="https://softwareheritage.org">Software
  Heritage</ext-link> <xref alt="2" rid="ref-cosmo_referencing_2020" ref-type="bibr">2</xref>
  can represent a file system directory as a digital object.
  Each digital object in a digital succession represents a new edition of a previous digital object
  and is assigned a number that determines its order.
  The digital succession expands, but the digital objects within the succession
  remain unchanged.</p>
  <p>In addition, a digital succession is digitally signed.
  In this specification, the digital signature is made using an SSH signing key via
  <ext-link ext-link-type="uri" xlink:href="https://en.wikipedia.org/wiki/Git">git</ext-link> <xref alt="1" rid="ref-enwikiU003Agit" ref-type="bibr">1</xref>.
  When a digital succession is first created, it only consists of a digitally signed
  genesis record and no digital object editions.
  A <italic>Digital Succession Identifier</italic> is an intrinsic identifier
  <xref alt="2" rid="ref-cosmo_referencing_2020" ref-type="bibr">2</xref> <xref alt="3" rid="ref-dicosmoU003Ahal-01865790" ref-type="bibr">3</xref> of the genesis record.</p>
  <sec id="digital-successions-implemented-via-git">
    <title>Digital successions implemented via git</title>
    <p>In the git implementation, the genesis record is a signed initial git commit with an
    empty tree (and no parent). To expand the digital succession, additional git commits
    are made using the same signing key. The git tree of each commit is not a
    digital object in the succession. Instead, it is a record of all the
    digital object editions in the succession. The top level directory consists of
    subdirectories named as non-negative integers. Each subdirectory
    contains either an entry named <monospace>object</monospace> or entries named as non-negative integers that are
    subdirectories. An entry named <monospace>object</monospace> represents a digital object edition in the digital
    succession, which can be a file (git blob) or a directory (git tree).
    For example, when a single file is added as edition 1, the full succession record is a
    directory with the path <monospace>1/object</monospace> leading to a git blob representing edition 1.</p>
<sec id="multilevel-edition-numbering">
  <title>Multilevel edition numbering</title>
  <p>In the simplest scenario, edition numbers are positive integers.
  However, multilevel edition numbering may be used for more advanced usage.
  Multilevel numbering is commonly used in the numbering of
  chapters, sections, and subsections (e.g. chapter 2, section 2.4, subsection 2.4.3)
  as well as software release versions (e.g. software release 2.19.2).</p>
  <p>An edition number prefix, such as <monospace>1</monospace>, can specify either a digital object edition or
  the entire sequence of editions <monospace>1.1</monospace>, <monospace>1.2</monospace>, <monospace>1.3</monospace>, etc…
  An edition number identifies either a digital object edition or a sequence of subordinate
  edition numbers, but not both. Larger integers indicate newer editions
  that obsolete older editions with smaller integers. The DSI specification does not
  assign any semantic meaning to different number levels.</p>
<sec id="textual-representation-of-a-digital-succession-identifier">
  <title>Textual representation of a Digital Succession Identifier</title>
  <p>The textual representation of a Digital Succession Identifier (DSI) is a base
  identifier followed by an optional slash followed by an optional edition number
  (possibly multilevel). The edition number is represented as non-negative
  integers separated by periods.</p>
  <p>The base DSI is calculated from the git initial commit (genesis record) of a digital
  succession. Git-compatible software can calculate a 20-byte binary hash that identifies the digital succession.
  This 20-byte binary hash is usually represented textually as a 40-digit
  hexadecimal number. However, for a DSI, this 20-byte binary hash has a 27-character
  representation in standard base64url format (RFC 4648)<xref alt="4" rid="ref-rfc4648" ref-type="bibr">4</xref>.</p>
  <p>It is worth noting that when a new digital succession (git initial commit) is created,
  a user can choose not to use it and instead immediately create a new one with a different base64url text representation.
  There is little cost in not using empty digital successions and recreating new genesis
  records (git initial commits) until an acceptable base64url identifier is found.</p>
  <sec id="examples">
        <term>Base DSI of this specification:</term>
        <term>DSI of the first edition:</term>
        <term>DSI of the first digital object (subedition) of the first edition:</term>
  <sec id="future-extensions">
    <title>Future extensions</title>
    <p>To support future enhancements,
    there are three paths to extending this textual representation.
    These paths involve using a base DSI where:</p>
    <list list-type="bullet">
        <p>a character is neither a slash (<monospace>/</monospace>) nor one of the 64 base64url characters,</p>
        <p>the number of characters is different than 27, or</p>
        <p>the 27th character is one of the 48 base64url characters that never appear
        as the last character of a base64url encoding of 40 bytes
        (i.e., any base64url character that is not
        <monospace>A</monospace>, <monospace>E</monospace>, <monospace>I</monospace>, <monospace>M</monospace>, <monospace>Q</monospace>, <monospace>U</monospace>, <monospace>Y</monospace>, <monospace>c</monospace>, <monospace>g</monospace>, <monospace>k</monospace>, <monospace>o</monospace>, <monospace>s</monospace>, <monospace>w</monospace>, <monospace>0</monospace>, <monospace>4</monospace>, or <monospace>8</monospace>).</p>
<sec id="design-choice-rationales">
  <title>Design Choice Rationales</title>
  <sec id="separation-of-git-history-from-edition-history">
    <title>Separation of git history from edition history</title>
    <p>A significant design choice is to not directly rely on git commit history
    to determine the succession of editions.
    Git commit history accurately records the actions performed with git, but it can be
    inflexible and confusing for establishing a clean linear history. Software
    Heritage automatically preserves git commits, which compounds the risk that git commits
    do not correspond well, or even prevent, an intended clean linear history.
    There may be situations where having merge commits and non-linear
    git commit history will be convenient.</p>
    <p>Having edition history separate from git history also provides a potential path
    for an enhancement akin to retractions of specific editions.</p>
  <sec id="use-of-git-tree-paths-instead-of-git-tags">
    <title>Use of git tree paths instead of git tags</title>
    <p>Edition numbers are similar to release versions of software projects.
    The usual practice in software development is to use git tags to identify releases.
    In contrast, this specification takes a different approach. Edition numbers are
    recorded with file paths in git trees rather than git tags.</p>
    <p>The main reason for this different approach has to do with how digital successions are
    recorded and copied. A highly desirable feature for digital successions is the ease of
    copying without errors. It is
    convenient to copy digital successions from multiple sources and store them in a single
    git repository. In this specification, a complete digital succession is captured
    by just a chain of commits (initial commit to latest commit).
    In contrast, entire software projects, which often include release tags, are usually
    copied by cloning an entire repository.
    If edition numbers were recorded as git annotated tags, copying digital successions
    properly would be more complicated and error-prone (due to missing tags).</p>
    <p>Git repository branches are a convenient tool for managing digital successions.
    However, branch names are not part of the digital succession record.</p>
  <sec id="use-of-base64url-rather-than-hexadecimal">
    <title>Use of base64url rather than hexadecimal</title>
    <p>The textual identifier is in base64url (Base64 with URL and filename safe alphabet) as
    specified in RFC 4648 <xref alt="4" rid="ref-rfc4648" ref-type="bibr">4</xref>. Both this format and hexadecimal have pros and cons.</p>
    <p>As mentioned in the introduction to the textual representation, there is no obligation
    to use the Digital Succession Identifier (DSI) generated by a brand new digital
    succession. If a DSI is unacceptable to the creator of a new digital succession, a new
    DSI can be easily generated. Helper tools could perform this function for users by
    <p>The main con to base64url is that it can contain characters that are more prone to copy
    errors when transmitted via human sight. Certain fonts and handwriting can make poor
    or no distinction between certain character pairs. For example, some popular sans serif
    fonts make no visual distinction between capital I and lowercase l.</p>
    <p>The main pro to base64url is that it is 27 characters rather than 40.
    Given that DSIs are used in contexts similar to DOIs, it is likely more acceptable to
    adopters to use 27 characters, which is very typical for a long DOI, rather than 40
    Displaying an ID of 27 characters rather than 40 characters also fits better on web
    pages rendered on mobile devices. 27 characters are less likely to be hidden by showing
    ellipsis, horizontal scroll regions, or shrunk into extremely small fonts.</p>
    <p>This design decision is partly made on the belief that the copy-via-human-sight issue is
    increasingly mitigated by a few technology trends:</p>
    <list list-type="order">
        <p>Computer systems are relied upon more and more via more reliable copying mechanisms
        such as clicking on hyperlinks, copy-and-paste, and QR codes instead of visual copying
        and handwriting.</p>
        <p>Websites tend to switch to using appropriate fonts for code like base64url IDs when
        output into HTML. The same applies to tools that automatically produce PDFs from
        computer data.</p>
        <p>Human-to-computer interfaces, such as input fields of Internet-connected software,
        tend to add features like autocomplete and search disambiguation for typos when
        sufficient need arises for such a feature.</p>
  <sec id="use-of-the-term-edition">
    <title>Use of the term edition</title>
    <p>The term edition is used to refer to a digital object in a digital succession.
    Part of the motivation is to avoid confusion with the many uses of the term version.
    For instance, version can be used to refer to versions of a digital object that
    are not added as editions.</p>
    <p>This language is partly due to the initial application of digital successions for
    scientific articles. Although there are rarely
    new versions of scientific articles after they are published, it is well established
    that scientific textbooks are published in multiple <italic>editions</italic>.</p>
    <p>In the initial application of digital successions, documents are the digital object.
    A single edition of a document might be presented in different HTML and PDF <italic>versions</italic>,
    even though they are the same <italic>edition</italic> of a document.</p>
<sec id="minutiae">
  <list list-type="bullet">
      <p>A multilevel edition number should not have more than four levels (components).
      Edition number components (per level) must not be negative and should not exceed four
      decimal digits.</p>
<sec id="acknowledgments">
  <p>Thank you to Valentin Lorentz for questions about design choices
  and pointing out an important shortcoming in how GPG digital signatures were used in the
  initial implementation of the hidos library (version 0.3) <xref alt="5" rid="ref-hidosU003A0.3" ref-type="bibr">5</xref>.</p>
<sec id="further-reading">
  <title>Further Reading</title>
  <list list-type="bullet">
      <p>This specification is heavily influenced by the concept of <italic>intrinsic identifier</italic> and
      related concepts discussed in
      <xref alt="2" rid="ref-cosmo_referencing_2020" ref-type="bibr">2</xref> <xref alt="3" rid="ref-dicosmoU003Ahal-01865790" ref-type="bibr">3</xref>.</p>
      <p>For a discussion on various concepts and proposed terminology about persistent
      identifiers, see <xref alt="6" rid="ref-kunze_persistence_2017" ref-type="bibr">6</xref>. In the proposed terminology, a DSI is a
      persistent identifier (PID) that is “frozen” and “waxing” with both intraversioned
      and extraversioned PIDs depending on the edition number.</p>
<sec id="changes">
  <sec id="from-edition-1.2">
    <title>From edition 1.2</title>
    <list list-type="bullet">
        <p>Reference SSH signing keys instead of GPG/PGP signing keys.</p>
  <sec id="from-edition-0.2">
    <title>From edition 0.2</title>
    <list list-type="bullet">
        <p>Add section about multilevel edition numbering.</p>
        <p>Add design rationale behind git tree paths instead of git tags.</p>
        <p>Note future expansion paths for text representation.</p>
  <ref id="ref-cosmo_referencing_2020">
    <element-citation publication-type="article-journal">
      <person-group person-group-type="author">
        <name><surname>Cosmo</surname><given-names>Roberto Di</given-names></name>
      <article-title>Referencing Source Code Artifacts: A Separate Concern in Software Citation</article-title>
      <source>Computing in Science &amp; Engineering</source>
      <year iso-8601-date="2020-03">2020</year><month>03</month>
      <date-in-citation content-type="access-date"><year iso-8601-date="2022-09-05">2022</year><month>09</month><day>05</day></date-in-citation>
      <issn>1521-9615, 1558-366X</issn>
      <pub-id pub-id-type="doi">10.1109/MCSE.2019.2963148</pub-id>
  <ref id="ref-dicosmoU003Ahal-01865790">
    <element-citation publication-type="paper-conference">
      <person-group person-group-type="author">
        <name><surname>Di Cosmo</surname><given-names>Roberto</given-names></name>
      <article-title>Identifiers for Digital Objects: the Case of Software Source Code Preservation</article-title>
      <source>iPRES 2018 - 15th International Conference on Digital Preservation</source>
      <publisher-loc>Boston, United States</publisher-loc>
      <year iso-8601-date="2018-09">2018</year><month>09</month>
  <ref id="ref-kunze_persistence_2017">
    <element-citation publication-type="article-journal">
      <person-group person-group-type="author">
        <name><surname>DeBarry</surname><given-names>Jeremy D.</given-names></name>
      <article-title>Persistence Statements: Describing Digital Stickiness</article-title>
      <source>Data Science Journal</source>
      <year iso-8601-date="2017-08">2017</year><month>08</month>
      <date-in-citation content-type="access-date"><year iso-8601-date="2022-11-13">2022</year><month>11</month><day>13</day></date-in-citation>
      <pub-id pub-id-type="doi">10.5334/dsj-2017-039</pub-id>
  <ref id="ref-rfc4648">
    <element-citation publication-type="report">
      <person-group person-group-type="author">
      <article-title>The Base16, Base32, and Base64 data encodings</article-title>
      <publisher-name>RFC Editor; Internet Requests for Comments; RFC Editor</publisher-name>
      <year iso-8601-date="2006-10">2006</year><month>10</month>
      <pub-id pub-id-type="doi">10.17487/RFC4648</pub-id>
  <ref id="ref-enwikiU003Agit">
      <person-group person-group-type="author">
        <string-name>Wikipedia contributors</string-name>
      <article-title>Git — Wikipedia, the free encyclopedia</article-title>
      <year iso-8601-date="2023">2023</year>
  <ref id="ref-hidosU003A0.3">
      <person-group person-group-type="author">
        <string-name>Hidos contributors</string-name>
      <article-title>Hidos 0.3</article-title>
      <year iso-8601-date="2022">2022</year>
back to top