=================
 Docutils_ Notes
=================
:Author: David Goodger (with input from many)
:Contact: goodger@users.sourceforge.net
:Date: $Date: 2004/05/09 13:45:44 $
:Revision: $Revision: 1.201 $
:Copyright: This document has been placed in the public domain.

.. _Docutils: http://docutils.sourceforge.net/

.. contents::


To Do
=====

Priority items are marked with "@" symbols.  The more @s, the higher
the priority.  Items in question form (containing "?") are ideas which
require more thought and debate; they are potential to-do's.

Many of these items are awaiting champions.  If you see something
you'd like to tackle, please do!


Bugs
----

Also see the `SourceForge Bug Tracker`__.

__ http://sourceforge.net/tracker/?group_id=38414&atid=422030

* The parser doesn't know anything about double-width characters such
  as Chinese hanza & Japanese kanji/kana.  Also, it's dependent on
  whitespace and punctuation as markup delimiters, which may not be
  applicable in these languages.

* In text inserted by the "include" directive, errors are often not
  reported with the correct "source" or "line" numbers.  Perhaps all
  Reporter calls need "source" and "line" keyword arguments.
  Elements' .line assignments should be checked.  (Assign to .source
  too?  Add a set_info method?  To what?)  There's a test in
  test/test_parsers/test_rst/test_directives/test_include.py.

* Explicit targets are sometimes mis-located.  In particular, placing
  a target before a section header puts the target at the end of the
  previous section instead of the start of the next section.  The code
  in docutils.transforms.misc.ClassAttribute could be used to fix
  this.  (Reported by David Priest.)

* Upon reviewing RFC 2396, I see that asterisks are valid URL
  characters, sometimes actually used.  Beni Cherniavsky found one in
  mid-November and fixed it by modifying the text.  There's a conflict
  with emphasis, but backslash escapes should overcome that; they
  don't though.  I consider it a bug in the parser that escaped
  asterisks in URLs aren't recognized.  Here's the URL that broke::

      http://cvs.sf.net/viewcvs.py/*checkout*/emu/speech_tools/scripts/tex_to_images.prl?rev=HEAD

  We should be able to escape the first asterisk like this::

      http://cvs.sf.net/viewcvs.py/\*checkout*/emu/speech_tools/scripts/tex_to_images.prl?rev=HEAD

* Document title should grow an implicit target.

* David Abrahams pointed out that doubly-indirect substitutions have a
  bug, but only when there's multiple references::

      |substitute| my coke for gin
      |substitute| you for my mum
      at least I'll get my washing done

      .. |substitute| replace:: |replace|
      .. |replace| replace:: swap

  This is tricky.  Substitutions have to propagate back completely.

* Another bug from David Abrahams (run with ``rst2html.py --traceback``)::

      .. [#crtp] See |runtime| 

      foo [#tag_dispatching]_

      .. [#tag_dispatching] See  |runtime|  

      .. |runtime| replace:: 7__
      __ runtime.html

      __ reference/__main.html

  Change the references.Substitutions tranform's priority from 220 to
  680, so it happens after reference resolution?  Then we have to deal
  with multiple IDs.  Perhaps the Substitution transform should remove
  all IDs from definitions after the first substitution reference is
  processed.


General
-------

* Refactor

  - Rename methods & variables according to the `Python coding
    conventions`_ below.

  - The name-to-id conversion and hyperlink resolution code needs to be
    checked for correctness and refactored.  I'm afraid it's a bit of
    a spaghetti mess now.

* Add validation?  See http://pytrex.sourceforge.net, RELAX NG, pyRXP.

* Ask Python-dev for opinions (GvR for a pronouncement) on special
  variables (__author__, __version__, etc.): convenience vs. namespace
  pollution.  Ask opinions on whether or not Docutils should recognize
  & use them.

* In ``docutils.readers.get_reader_class`` (& ``parsers`` &
  ``writers`` too), should we be importing "standalone" or
  "docutils.readers.standalone"?  (This would avoid importing
  top-level modules if the module name is not in docutils/readers.
  Potential nastiness.)

* Perhaps store a _`name-to-id mapping file`?  This could be stored
  permanently, read by subsequent processing runs, and updated with
  new entries.  ("Persistent ID mapping"?)

* Need a Unicode to HTML entities codec for HTML writer?  No, the
  "xmlcharrefreplace" output encoding error handler is sufficient.
  Make it the default for HTML & XML writers?

* Perhaps the ``Component.supports`` method should deal with
  individual features ("meta" etc.) instead of formats ("html" etc.)?

* Standalone Reader: Implement an option to turn off the DocTitle
  transform?

* Add /usr/etc/docutils.conf to config file list?  System-wide,
  whereas /etc/docutils.conf is machine-specific.
  /usr/local/etc/docutils.conf too?  See the `Filesystem Hierarchy
  Standard`_.

  .. _Filesystem Hierarchy Standard: http://www.pathname.com/fhs.

* Add _`object numbering and object references` (tables & figures).
  These would be the equivalent of DocBook's "formal" elements.

  We may need _`persistent sequences`, such as chapter numbers.  See
  `OpenOffice.org XML`_ "fields".  Should the sequences be automatic
  or manual (user-specifyable)?

  We need to name the objects:

  - "name" option for the "figure" directive? ::

        .. figure:: image.png
           :name: image's name

    Same for the "table" directive::

        .. table:: optional title here
           :name: table's name

           =====  =====
             x    not x
           =====  =====
           True   False
           False  True
           =====  =====

    This would also allow other options to be set, like border
    styles.  The same technique could be used for other objects.

    A preliminary "table" directive has been implemented, supporting
    table titles.  Perhaps the name should derive from the title.

  - The object could also be done this way::

        .. _figure name:

        .. figure:: image.png

    This may be a more general solution, equally applicable to tables.
    However, explicit naming using an option seems simpler to users.

  We'll also need syntax for object references.  See `OpenOffice.org
  XML`_ "reference fields":

  - Parameterized substitutions?  For example::

        See |figure (figure name)| on |page (figure name)|.

        .. |figure (name)| figure-ref:: (name)
        .. |page (name)| page-ref:: (name)

    The result would be::

        See figure 3.11 on page 157.

    But this would require substitution directives to be processed at
    reference-time, not at definition-time as they are now.  Or,
    perhaps the directives could just leave ``pending`` elements
    behind, and the transforms do the work?  How to pass the data
    through?  Too complicated.

  - An interpreted text approach is simpler and better::

        See :figure:`figure name` on :page:`figure name`.

    The "figure" and "page" roles could generate appropriate
    boilerplate text.  The position of the role (prefix or suffix)
    could also be utilized.

    See `Interpreted Text`_ below.

  .. _OpenOffice.org XML: http://xml.openoffice.org/

* Think about _`large documents` made up of multiple subdocument
  files.  Issues: continuity (`persistent sequences`_ above),
  cross-references (`name-to-id mapping file`_ above and `targets in
  other documents`_ below).

  When writing a book, the author probably wants to split it up into
  files, perhaps one per chapter (but perhaps even more detailed).
  However, we'd like to be able to have references from one chapter to
  another, and have continuous numbering (pages and chapters, as
  applicable).  Of course, none of this is implemented yet.  There has
  been some thought put into some aspects; see `the "include"
  directive`__ and the `Reference Merging`_ transform below.

  When I was working with SGML in Japan, we had a system where there
  was a top-level coordinating file, book.sgml, which contained the
  top-level structure of a book: the <book> element, containing the
  book <title> and empty component elements (<preface>, <chapter>,
  <appendix>, etc.), each with filename attributes pointing to the
  actual source for the component.  Something like this::

      <book id="bk01">
      <title>Title of the Book</title>
      <preface inrefid="pr01"></preface>
      <chapter inrefid="ch01"></chapter>
      <chapter inrefid="ch02"></chapter>
      <chapter inrefid="ch03"></chapter>
      <appendix inrefid="ap01"></appendix>
      </book>

  (The "inrefid" attribute stood for "insertion reference ID".)

  The processing system would process each component separately, but
  it would recognize and use the book file to coordinate chapter and
  page numbering, and keep a persistent ID to (title, page number)
  mapping database for cross-references.  Docutils could use a similar
  system for large-scale, multipart documents.

  __ rst/directives.html#including-an-external-document-fragment

  Aahz's idea:

      First the ToC::

          .. ToC-list::
              Introduction.txt
              Objects.txt
              Data.txt
              Control.txt

      Then a sample use::

          .. include:: ToC.txt

          As I said earlier in chapter :chapter:`Objects.txt`, the
          reference count gets increased every time a binding is made.

      Which produces::

          As I said earlier in chapter 2, the
          reference count gets increased every time a binding is made.

      The ToC in this form doesn't even need to be references to actual
      reST documents; I'm simply doing it that way for a minimum of
      future-proofing, in case I do want to add the ability to pick up
      references within external chapters.

  Perhaps, instead of ToC (which would overload the "contents"
  directive concept already in use), we could use "manifest".  A
  "manifest" directive might associate local reference names with
  files::

      .. manifest::
         intro: Introduction.txt
         objects: Objects.txt
         data: Data.txt
         control: Control.txt

  Then the sample becomes::

      .. include:: manifest.txt

      As I said earlier in chapter :chapter:`objects`, the
      reference count gets increased every time a binding is made.

* Add functional testing to Docutils: Readers, Writers, front ends.

* Changes to sandbox/davidg/infrastructure/docutils-update?

  - Modify the script to only update the snapshots if files have
    actually changed in CVS (saving some SourceForge server cycles).

  - Make passing the test suite a prerequisite to snapshot update,
    but only if the process is completely automatic.

  - Rewrite in Python?

* Publisher: "Ordinary setup" shouldn't requre specific ordering; at
  the very least, there ought to be error checking higher up in the
  call chain.  [Aahz]

  ``Publisher.get_settings`` requires that all components be set up
  before it's called.  Perhaps the I/O *objects* shouldn't be set, but
  I/O *classes*.  Then options are set up (``.set_options``), and
  ``Publisher.set_io`` (or equivalent code) is called with source &
  destination paths, creating the I/O objects.

  Perhaps I/O objects shouldn't be instantiated until required.  For
  split output, the Writer may be called multiple times, once for each
  doctree, and each doctree should have a separate Output object (with
  a different path).  Is the "Builder" pattern applicable here?

* Perhaps I/O objects should become full-fledged components (i.e.
  subclasses of ``docutils.Component``, as are Readers, Parsers, and
  Writers now), and thus have associated option/setting specs and
  transforms.

* Multiple file I/O suggestion from Michael Hudson: use a file-like
  object or something you can iterate over to get file-like objects.

* Language modules: in accented languages it may be useful to have
  both accented and unaccented entries in the ``bibliographic_fields``
  mapping for versatility.

* Add a "--strict-language" option & setting: no English fallback for
  language-dependent features.

* Add an "--input-language" option & setting?  Specify a different
  language module for input (bibliographic fields, directives) than
  for output.  The "--language" option would set both input & output
  languages.

* Auto-generate reference tables for language-dependent features?
  Could be generated from the source modules.  A special command-line
  option could be added to Docutils front ends to do this.  (Idea from
  Engelbert Gruber.)

* Change the "class" attribute of elements (set with
  Element.set_class) to a list?

* Enable feedback of some kind from internal decisions, such as
  reporting the successful input encoding.  Modify runtime settings?
  System message?  Simple stderr output?

* Rationalize Writer settings (HTML/LaTeX/PEP) -- share settings.

* The "docutils.conf" included with Docutils should become complete,
  with examples of every setting (many/most commented out).  It's
  currently sparse, requiring doc lookups.

* Merge doc/latex.txt info into tools.txt and config.txt.

* Add internationalization to _`footer boilerplate text` (resulting
  from "--generator", "--source-link", and "--date" etc.), allowing
  translations.

* Add an "--include file" command-line option (config setting too?),
  equivalent to ".. include:: file" as the first line of the doc text?
  Especially useful for character entity sets, text transform specs,
  boilerplate, etc.

* Parameterize the Reporter object or class?  See the `2004-02-18
  "rest checking and source path"`_ thread.

  .. _2004-02-18 "rest checking and source path":
     http://thread.gmane.org/gmane.text.docutils.user/1112

* Add a "disable_transforms" setting?  And a dummy Writer subclass
  that does nothing when its .write() method is called?  Would allow
  for easy syntax checking.  See the `2004-02-18 "rest checking and
  source path"`_ thread.

* Add a generic meta-stylesheet mechanism?  An external file could
  associate style names ("class" attributes) with specific elements.
  Could be generalized to arbitrary output attributes; useful for HTML
  & XMLs.  Aahz implemented something like this in
  sandbox/aahz/Effective/EffMap.py.


Documentation
-------------

* User docs.  What's needed?

* Add an "examples" directory, beside "tools" and "docs", for
  interesting examples of Docutils usage?  Have a top-level README.txt
  file and a subdirectory for each example.  (Martin Blais)


Implementation Docs
```````````````````

* Internal module documentation (docstrings).

* spec/doctree.txt: DTD element structural relationships, semantics,
  and attributes.  In progress; element descriptions to be completed.

* How-to docs: In spec/howto/.

  - How a Writer works & how to write one

  - Transforms

* Document the ``pending`` elements, how they're generated and what
  they do.

* Document the transforms (perhaps in docstrings?): how they're used,
  what they do, dependencies & order considerations.

* Document the HTML classes used by html4css1.py.


Specification
`````````````

* Complete PEP 258 Docutils Design Specification.

  - Fill in the blanks in API details.

  - Specify the nodes.py internal data structure implementation?

        [Tibs:] Eventually we need to have direct documentation in
        there on how it all hangs together - the DTD is not enough
        (indeed, is it still meant to be correct?  [Yes, it is.
        --DG]).

* Rework PEP 257, separating style from spec from tools, wrt Docutils?
  See Doc-SIG from 2001-06-19/20.


Python Source Reader
--------------------

General:

* Analyze Tony Ibbs' PySource code.

* Analyze Doug Hellmann's HappyDoc project.

* Investigate how POD handles literate programming.

* Take the best ideas and integrate them into Docutils 0.3.

Miscellaneous ideas:

* If we can detect that a comment block begins with ``##``, a la
  JavaDoc, it might be useful to indicate interspersed section headers
  & explanatory text in a module.  For example::

      """Module docstring."""

      ##
      # Constants
      # =========

      a = 1
      b = 2

      ##
      # Exception Classes
      # =================

      class MyException(Exception): pass

      # etc.

* Should standalone strings also become (module/class) docstrings?
  Under what conditions?  We want to prevent arbitrary strings from
  becomming docstrings of prior attribute assignments etc.  Assume
  that there must be no blank lines between attributes and attribute
  docstrings?  (Use lineno of NEWLINE token.)

  Triple-quotes are sometimes used for multi-line comments (such as
  commenting out blocks of code).  How to reconcile?

* HappyDoc's idea of using comment blocks when there's no docstring
  may be useful to get around the conflict between `additional
  docstrings`_ and ``from __future__ import`` for module docstrings.
  A module could begin like this::

      #!/usr/bin/env python
      # :Author: Me
      # :Copyright: whatever

      """This is the public module docstring (``__doc__``)."""

      # More docs, in comments.
      # All comments at the beginning of a module could be
      # accumulated as docstrings.
      # We can't have another docstring here, because of the
      # ``__future__`` statement.

      from __future__ import division

  Using the JavaDoc convention of a doc-comment block beginning with
  ``##`` is useful though.  It allows doc-comments and implementation
  comments.

  .. _additional docstrings: pep-0258.html#additional-docstrings

* HappyDoc uses an initial comment block to set "parser configuration
  values".  Do the same thing for Docutils, to set runtime settings on
  a per-module basis?  I.e.::

      # Docutils:setting=value

  Could be used to turn on/off function parameter comment recognition
  & other marginal features.  Could be used as a general mechanism to
  augment config files and command-line options (but which takes
  precedence?).

* Multi-file output should be divisible at arbitrary level.

* Support all forms of ``import`` statements:

  - ``import module``: listed as "module"
  - ``import module as alias``: "alias (module)"
  - ``from module import identifier``: "identifier (from module)"
  - ``from module import identifier as alias``: "alias (identifier
    from module)"
  - ``from module import *``: "all identifiers (``*``) from module"

* Have links to colorized Python source files from API docs?  And
  vice-versa: backlinks from the colorized source files to the API
  docs!

* In summaries, use the first *sentence* of a docstring if the first
  line is not followed by a blank line.


reStructuredText Parser
-----------------------

Also see the `... Or Not To Do?`__ list.

__ rst/alternatives.html#or-not-to-do

* Clean up the code; refactor as required.

* Add motivation sections for constructs in spec.

* Allow very long titles (on two or more lines)?

* And for the sake of completeness, should definition list terms be
  allowed to be very long (two or more lines) also?

* Support generic hyperlink references to _`targets in other
  documents`?  Not in an HTML-centric way, though (it's trivial to say
  ``http://www.example.com/doc#name``, and useless in non-HTML
  contexts).  XLink/XPointer?  ``.. baseref::``?  See Doc-SIG
  2001-08-10.

* .. _adaptable file extensions:

  In target URLs, it would be useful to not explicitly specify the
  file extension.  If we're generating HTML, then ".html" is
  appropriate; if PDF, then ".pdf"; etc.  How about using ".*" to
  indicate "choose the most appropriate filename extension?  For
  example::

      .. _Another Document: another.*

  Should the choice be from among existing files only?  Documents
  only, or objects (images, etc.) also?  (How to differentiate?
  Element context [within "image"]?)

  This may not be just a parser issue though; it may need framework
  support.

* Implement the header row separator modification to table.el.  (Wrote
  to Takaaki Ota & the table.el mailing list on 2001-08-12, suggesting
  support for "=====" header rows.  On 2001-08-17 he replied, saying
  he'd put it on his to-do list, but "don't hold your breath".)

* Tony says inline markup rule 7 could do with a *little* more
  exposition in the spec, to make clear what is going on for people
  with head colds.

* @@ Fix the parser's indentation handling to conform with the
  stricter definition in the spec.  (Explicit markup blocks should be
  strict or forgiving?)

* @@ Tighten up the spec for indentation of "constructs using complex
  markers": field lists and option lists?  Bodies may begin on the
  same line as the marker or on a subsequent line (with blank lines
  optional).  Require that for bodies beginning on the same line as
  the marker, all lines be in strict alignment.  Currently, this is
  acceptable::

      :Field-name-of-medium-length: Field body beginning on the same
          line as the field name.

  This proposal would make the above example illegal, instead
  requiring strict alignment.  A field body may either begin on the
  same line::

      :Field-name-of-medium-length: Field body beginning on the same
                                    line as the field name.

  Or it may begin on a subsequent line::

      :Field-name-of-medium-length:
          Field body beginning on a line subsequent to that of the
          field name.

  This would be especially relevant in degenerate cases like this::

      :Number-of-African-swallows-requried-to-carry-a-coconut:
          It would be very difficult to align the field body with
          the left edge of the first line if it began on the same
          line as the field name.

* Allow for variant styles by interpreting _`indented lists` as if
  they weren't indented?  For example, currently the list below will
  be parsed as a list within a block quote::

      paragraph

        * list item 1
        * list item 2

  But a lot of people seem to write that way, and HTML browsers make
  it look as if that's the way it should be.  The parser could check
  the contents of block quotes, and if they contain only a single
  list, remove the block quote wrapper.  There would be two problems:

  1. What if we actually *do* want a list inside a block quote?

  2. What if such a list comes immediately after an indented
     construct, such as a literal block?

  Both could be solved using empty comments (problem 2 already exists
  for a block quote after a literal block).  But that's a hack.

  Perhaps a runtime setting, allowing or disabling this convenience,
  would be appropriate.  But that raises issues too:

      User A, who writes lists indented (and their config file is set
      up to allow it), sends a file to user B, who doesn't (and their
      config file disables indented lists).  The result of processing
      by the two users will be different.

  It may seem minor, but it adds ambiguity to the parser, which is
  bad.

  See the `Doc-SIG discussion starting 2001-04-18`__ with Ed Loper's
  "Structuring: a summary; and an attempt at EBNF", item 4 (and
  follow-ups, here__ and here__).  Also `docutils-users,
  2003-02-17`__ and `beginning 2003-08-04`__.

  __ http://mail.python.org/pipermail/doc-sig/2001-April/001776.html
  __ http://mail.python.org/pipermail/doc-sig/2001-April/001789.html
  __ http://mail.python.org/pipermail/doc-sig/2001-April/001793.html
  __ http://sourceforge.net/mailarchive/message.php?msg_id=3838913
  __ http://sf.net/mailarchive/forum.php?thread_id=2957175&forum_id=11444

* Make the parser modular.  Allow syntax constructs to be added or
  disabled at run-time.  Or is subclassing enough?

* Continue to report (info, level 1) enumerated lists whose start
  value is not ordinal-1?

* Generalize the "doctest block" construct (which is overly
  Python-centric) to other interactive sessions?  "Doctest block"
  could be renamed to "I/O block" or "interactive block", and each of
  these could also be recognized as such by the parser:

  - Shell sessions::

        $ cat example1.txt
        A block beginning with a "$ " prompt is interpreted as a shell
        session interactive block.  As with Doctest blocks, the
        interactive block ends with the first blank line, and wouldn't
        have to be indented.

  - Root shell sessions::

        # cat example2.txt
        A block beginning with a "# " prompt is interpreted as a root
        shell session (the user is or has to be logged in as root)
        interactive block.  Again, the block ends with a blank line.

  Other standard (and unambiguous) interactive session prompts could
  easily be added (such as "> " for WinDOS).

  Tony Ibbs spoke out against this idea (2002-06-14 Doc-SIG thread
  "docutils feedback").

* Should the "doctest" element go away, and the construct simply be a
  front-end to generic literal blocks?

* Add support for pragma (syntax-altering) directives.

  Some pragma directives could be local-scope unless explicitly
  specified as global/pragma using ":global:" options.

* Remove leading numbers from section titles for implicit link names?
  A section titled "3. Conclusion" could then be referred to by
  "``Conclusion_``" (i.e., without the "3.").

* `Syntax for line blocks`__ (currently a directive)?

  __ rst/alternatives.html#syntax-for-line-blocks

* Implement auto-enumerated lists?  See `Auto-Enumerated Lists`__.

  __ rst/alternatives.html#auto-enumerated-lists

* Support whitespace in angle-bracketed standalone URLs according to
  Appendix E ("Recommendations for Delimiting URI in Context") of `RFC
  2396`_.

  .. _RFC 2396: http://www.rfc-editor.org/rfc/rfc2396.txt

* Use the vertical spacing of the source text to determine the
  corresponding vertical spacing of the output?

* [From Mark Nodine]  For cells in simple tables that comprise a
  single line, the justification can be inferred according to the
  following rules:

  1. If the text begins at the leftmost column of the cell,
     then left justification, ELSE
  2. If the text begins at the rightmost column of the cell,
     then right justification, ELSE
  3. Center justification.

  The onus is on the author to make the text unambiguous by adding
  blank columns as necessary.  There should be a parser setting to
  turn off justification-recognition (normally on would be fine).

  Decimal justification?

* Make enumerated list parsing more strict, so that this would parse
  as a paragraph with an info message::

      1. line one
      3. line two

* Line numbers in system messages are inconsistent in the parser.
  Fix?

* Generalize the "target-notes" directive into a command-line option
  somehow?  See docutils-develop 2003-02-13.

* Include the _`character entity substitution definition files`
  `temporarily stored here <tmp/charents>`__, perhaps in a
  ``docutils/parsers/rst/includes/`` directory.  See `misc.include`_
  below.

* Should ^L (or something else in reST) be defined to mean
  force/suggest page breaks in whatever output we have?

  A "break" or "page-break" directive would be easy to add.  A new
  doctree element would be required though (perhaps "break").  The
  final behavior would be up to the Writer.  The directive argument
  could be one of page/column/recto/verso for added flexibility.

  Currently ^L (Python's "\f") characters are treated as whitespace.
  They're converted to single spaces, actually, as are vertical tabs
  (^K, Python's "\v").  It would be possible to recognize form feeds
  as markup, but it requires some thought and discussion first.  Are
  there any downsides?  Many editing environments do not allow the
  insertion of control characters.  Will it cause any harm?  It would
  be useful as a shorthand for the directive.

  It's common practice to use ^L before Emacs "Local Variables"
  lists::

      ^L
      ..
         Local Variables:
         mode: indented-text
         indent-tabs-mode: nil
         sentence-end-double-space: t
         fill-column: 70
         End:

  These are already present in many PEPs and Docutils project
  documents.  From the Emacs manual (info):

      A "local variables list" goes near the end of the file, in the
      last page.  (It is often best to put it on a page by itself.)

  It would be unfortunate if this construct caused a final blank page
  to be generated (for those Writers that recognize the page breaks).
  We'll have to add a transform that looks for a "break" plus zero or
  more comments at the end of a document, and removes them.

* Could the "break" concept above be extended to inline forms?
  E.g. "^L" in the middle of a sentence could cause a line break.
  Only recognize it at the end of a line (i.e., "\f\n")?

  Or is formfeed inappropriate?  Perhaps vertical tab ("\v"), but even
  that's a stretch.  Can't use carriage returns, since they're
  commonly used for line endings.

* Allow a "::"-only paragraph (first line, actually) to introduce a
  literal block without a blank line?  (Idea from Paul Moore.) ::

      ::
          This is a literal block

  Is indentation enough to make the separation between a paragraph
  which contains just a ``::`` and the literal text unambiguous?
  There's one problem with this concession.  What if one wants a
  definition list item which defines the term "::"?  We'd have to
  escape it.  Currenty, "\::" doesn't work (although it should;
  **bug**), and ":\:" is misinterpreted as a field name (name "\";
  also a **bug**).  Assuming these bugs are squashed, I suppose it's a
  useful special case.  It would only be reasonable to apply it to
  "::"-only paragraphs though.  I think the blank line is visually
  necessary if there's text before the "::"::

      The text in this paragraph needs separation
      from the literal block following::
          This doesn't look right.

  Another idea.  Would it be worthwhile to allow literal blocks to
  begin without a newline after the "::"?  Example::

      ::  while True:
              print 'hello world'

  Perhaps.  Perhaps not.

* Add new syntax for _`nested inline markup`?  Or extend the parser to
  parse nested inline markup somehow?  See the `collected notes
  <http://docutils.sf.net/spec/rst/alternatives.html#nested-inline-markup>`__.

* Drop the backticks from embedded URIs with omitted reference text?
  Should the angle brackets be kept in the output or not? ::

      <file_name>_

  Probably not worth the trouble.

* Add ``^superscript^`` inline markup?  The only common non-markup
  uses of "^" I can think of are as short hand for "superscript"
  itself and for describing control characters ("^C to cancel").  The
  former supports the proposed syntax, and it could be argued that the
  latter ought to be literal text anyhow (e.g. "``^C`` to cancel").

* Add _`math markup`.  We should try for a general solution, that's
  applicable to any output format.  Using a standard, such as MathML_,
  would be best.  TeX (or itex_) would be acceptable as a *front-end*
  to MathML.  See `the culmination of a relevant discussion
  <http://article.gmane.org/gmane.text.docutils.user/118>`__.

  Both a directive and an interpreted text role will be necessary (for
  each markup).  Directive example::

      .. itex::
         \alpha_t(i) = P(O_1, O_2, \dots O_t, q_t = S_i \lambda)

  The same thing inline::

      The equation in question is :itex:`\alpha_t(i) = P(O_1, O_2,
      \dots O_t, q_t = S_i \lambda)`.

  .. _MathML: http://www.w3.org/TR/MathML2/
  .. _itex: http://pear.math.pitt.edu/mathzilla/itex2mmlItex.html

* How about a syntax for alternative hyperlink behavior, such as "open
  in a new window" (as in HTML's ``<a target="_blank">``)?  Double
  angle brackets might work for inline targets::

      The `reference docs <<url>>`__ may be handy.

  But what about explicit targets?

  The MoinMoin wiki uses a caret ("^") at the beginning of the URL
  ("^" is not a legal URI character).  That could work for both inline
  and explicit targets::

      The `reference docs <^url>`__ may be handy.

      .. _name: ^url

* Allow transitions between sections?  A transition immediately
  preceding a section title could be interpreted as a sibling of that
  section.  The context would obviate the need for a special
  directive.  The document model would have to accommodate the
  change::

      <!ENTITY % structure.model
        " ( ( (%body.elements; | topic | sidebar)+,
              (transition, (%body.elements; | topic | sidebar)+ )*,
              (transition?, (%section.elements;))* )
          | (transition?, (%section.elements;))+ ) ">


Directives
``````````

Directives below are often referred to as "module.directive", the
directive function.  The "module." is not part of the directive name
when used in a document.

* Allow directives to be added at run-time?

* Use the language module for directive option names?

* Add "substitution_only" and "substitution_ok" function attributes,
  and automate context checking?

* Change directive functions to directive classes?  Superclass'
  ``__init__()`` could handle all the bookkeeping.

* Implement options on existing directives:

  - Add a "name" option to directives, to set an author-supplied
    identifier?

  - All directives that produce titled elements should grow implicit
    reference names based on the titles.

  - @ _`body.table`: Document, add tests.

  - _`images.image`: "border"?

    _`Units of measure`?  (See `docutils-users, 2003-03-02
    <http://article.gmane.org/gmane.text.docutils.user/154>`__, and
    `docutils-develop, 2004-04-29
    <http://article.gmane.org/gmane.text.docutils.devel/1439>`_.)

  - _`images.figure`: "title" and "number", to indicate a formal
    figure?

  - _`parts.sectnum`: "local"?, "start", "refnum"

    A "local" option could enable numbering for sections from a
    certain point down, and sections in the rest of the document are
    not numbered.  For example, a reference section of a manual might
    be numbered, but not the rest.  OTOH, an all-or-nothing approach
    would probably be enough.

    The "start" option will specify the sequence set to use at the
    same time as the starting value, for the first part of the section
    number (i.e., section, not subsection).  For example::

        .. sectnum: :start: 1

        .. sectnum: :start: A

        .. sectnum: :start: 5

        .. sectnum: :start: I

    The first one is the default: start at 1, numbered.  The second
    one specifies letters, and start at "A".  The third specifies
    numbers, start at 5.  The last example could signal Roman
    numerals, although I don't know if they'd be applicable here.
    Enumerated lists already do all this; perhaps that code could be
    reused.

    Here comes the tricky part.  The "sectnum" directive should be
    usable multiple times in a single document.  For example, in a
    long document with "chapter" and "appendix" sections, there could
    be a second "sectnum" before the first appendix, changing the
    sequence used (from 1,2,3... to A,B,C...).  This is where the
    "local" concept comes in.  This part of the implementation can be
    left for later.

    A "refnum" option (better name?) would insert reference names
    (targets) consisting of the reference number.  Then a URL could be
    of the form ``http://host/document.html#2.5`` (or "2-5"?).  Allow
    internal references by number?  Allow name-based *and*
    number-based ids at the same time, or only one or the other (which
    would the table of contents use)?  Usage issue: altering the
    section structure of a document could render hyperlinks invalid.

  - _`parts.contents`: Add a "suppress" or "prune" option?  It would
    suppress contents display for sections in a branch from that point
    down.  Or a new directive, like "prune-contents"?

    Add an option to include topics in the TOC?  Another for sidebars?
    See docutils-develop 2003-01-29.

  - _`misc.include`:

    - @@@ Allow whitespace in paths (just not at line boundaries).

    - "encoding" option?  Take default from runtime settings.  Use
      Input component to read it in?

    - Option to select a range of lines?

    - Option to label lines?

    - Default directory for "built-in includes", using the C syntax
      ``#include <name>``?

          Use C-preprocessor semantics for locating include files?
          E.g., ``.. include:: file.txt`` will read another file into
          the current one, relative to the current file's directory,
          and ``.. include:: <standard>`` will read a standard include
          file from ``docutils/include/``.  (Should "quotes" be
          required around non-standard include files?)

          -- http://sf.net/mailarchive/message.php?msg_id=1938401

      I now think that ``docutils/parsers/rst/include/`` is a better
      place for these files, since they're reStructuredText-specific.

      Keeping standard data files together with the package code makes
      sense to me.  It seems much less complex to implement than a
      separate system data directory, such as ``/usr/share/docutils``.
      Any reason a system data directory should be used?  How does
      Distutils handle data files?

      How about an environment variable, say RSTINCLUDEPATH or
      RSTPATH?  This could be combined with a setting/option to allow
      user-defined include directories.

      For a specific application, see the discussion of `character
      entity substitution definition files`_ above.

  - _`misc.class`: @@@ Allow multiple classes (whitespace-separated)
    to be declared.

* Implement directives.  Each of the list items below begins with an
  identifier of the form, "module_name.directive_function_name".  The
  directive name itself could be the same as the
  directive_function_name, or it could differ.

  - _`html.imagemap` (Useful outside of HTML?  If not, replace with
    image only in non-HTML writers?)

  - _`parts.endnotes` (or "footnotes"): See `Footnote & Citation Gathering`_.

  - _`parts.citations`: See `Footnote & Citation Gathering`_.

  - _`misc.exec`: Execute Python code & insert the results.  Perhaps
    dangerous?  Call it "python" to allow for other languages?

  - _`misc.system`?: Execute an ``os.system()`` call, and insert the
    results (possibly as a literal block).  Definitely dangerous!  How
    to make it safe?  Perhaps such processing  should be left outside
    of the document, in the user's production system (a makefile or a
    script or whatever).  Or, the directive could be disabled by
    default and only enabled with an explicit command-line option or
    config file setting.  Even then, an interactive prompt may be
    useful, such as:

        The file.txt document you are processing contains a "system"
        directive requesting that the ``sudo rm -rf /`` command be
        executed.  Allow it to execute?  (y/N)

  - _`misc.eval`: Evaluate an expression & insert the text.  At parse
    time or at substitution time?  Dangerous?  Perhaps limit to canned
    macros; see text.date_ below.

  - _`misc.encoding`: Specify the character encoding of the input
    data.  But there are problems:

    - When it sees the directive, the parser will already have read
      the input data, and encoding determination will already have
      been done.

    - If a file with an "encoding" directive is edited and saved with
      a different encoding, the directive may cause data corruption.

  - _`misc.language`: Specify the language of a document.  There is a
    problem similar to the first problem listed for misc.encoding_,
    although to a lesser degree.

  - _`misc.settings`: Set any Docutils runtime setting from within a
    document?

  - _`misc.charents`: Equivalent to::

        .. include:: {includepath}/charents.txt

  - Docutils already has the ability to say "use this content for
    Writer X" (via the "raw" directive), but it doesn't have the
    ability to say "use this content for any Writer other than X".  It
    wouldn't be difficult to add this ability though.

    My first idea would be to add a set of _`conditional directives`.
    Let's call them "writer-is" and "writer-is-not" for discussion
    purposes (don't worry about implemention details).  We might
    have::

         .. writer-is:: text-only

            ::

                +----------+
                |   SNMP   |
                +----------+
                |   UDP    |
                +----------+
                |    IP    |
                +----------+
                | Ethernet |
                +----------+

         .. writer-is:: pdf

            .. figure:: protocol_stack.eps

         .. writer-is-not:: text-only pdf

            .. figure:: protocol_stack.png

    This could be an interface to the Filter transform
    (docutils.transforms.components.Filter).

    The ideas in `adaptable file extensions`_ above may also be
    applicable here.

    Here's an example of a directive that could produce multiple
    outputs (*both* raw troff pass-through *and* a GIF, for example)
    and allow the Writer to select. ::

        .. eqn::

           .EQ
           delim %%
           .EN
           %sum from i=o to inf c sup i~=~lim from {m -> inf}
           sum from i=0 to m sup i%
           .EQ
           delim off
           .EN

  - _`body.qa` (directive a.k.a. "faq", "questions"): Questions &
    Answers.  Implement as a generic two-column marked list?  As a
    standalone (non-directive) construct?  (Is the markup ambiguous?)
    Add support to parts.contents.

    New elements would be required.  Perhaps::

        <!ELEMENT question_list (question_list_item+)>
        <!ATTLIST question_list
            numbering  (none | local | global)
                                #IMPLIED
            start     NUMBER    #IMPLIED>
        <!ELEMENT question_list_item (question, answer*)>
        <!ELEMENT question %text.model;>
        <!ELEMENT answer (%body.elements;)+>

    Originally I thought of implementing a Q&A list with special
    syntax::

        Q: What am I?

        A: You are a question-and-answer
           list.

        Q: What are you?

        A: I am the omniscient "we".

    Where each "Q" and "A" could also be numbered (e.g., "Q1").
    However, a simple enumerated or bulleted list will do just fine
    for syntax.  A directive could treat the list specially; e.g. the
    first paragraph could be treated as a question, the remainder as
    the answer (multiple answers could be represented by nested
    lists).  Without special syntax, this directive becomes low
    priority.

  - _`body.example`: Examples; suggested by Simon Hefti.  Semantics as
    per Docbook's "example"; admonition-style, numbered, reference,
    with a caption/title.

  - _`body.index`: Index targets.

    Were I writing a book with an index, I guess I'd need two
    different kinds of index targets: inline/implicit and
    out-of-line/explicit.  For example::

        In this `paragraph`:index:, several words are being
        `marked`:index: inline as implicit `index`:index:
        entries.

        .. index:: markup
        .. index:: syntax

        The explicit index directives above would refer to
        this paragraph.  It might also make sense to allow multiple
        entries in an ``index`` directive:

        .. index::
            markup
            syntax

    The words "paragraph", "marked", and "index" would become index
    entries pointing at the words in the first paragraph.  The index
    entry words appear verbatim in the text.  (Don't worry about the
    ugly ":index:" part; if indexing is the only/main application of
    interpreted text in your documents, it can be implicit and
    omitted.)  The two directives provide manual indexing, where the
    index entry words ("markup" and "syntax") do not appear in the
    main text.  We could combine the two directives into one::

        .. index:: markup; syntax

    Semicolons instead of commas because commas could *be* part of the
    index target, like::

        .. index:: van Rossum, Guido

    Another reason for index directives is because other inline markup
    wouldn't be possible within inline index targets.

    Sometimes index entries have multiple levels.  Given::

        .. index:: statement syntax: expression statements

    In a hypothetical index, combined with other entries, it might
    look like this::

        statement syntax
            expression statements ..... 56
            assignment ................ 57
            simple statements ......... 58
            compound statements ....... 60

    Inline multi-level index targets could be done too.  Perhaps
    something like::

        When dealing with `expression statements <statement syntax:>`,
        we must remember ...

    The opposite sense could also be possible::

        When dealing with `index entries <:multi-level>`, there are
        many permutations to consider.

    Also "see / see also" index entries.

    Given::

        Here's a paragraph.

        .. index:: paragraph

    (The "index" directive above actually targets the *preceding*
    object.)  The directive should produce something like this XML::

        <paragraph>
        <index_entry text="paragraph"/>
        Here's a paragraph.
        </paragraph>

    This kind of content model would also allow true inline
    index-entries::

        Here's a `paragraph`:index:.

    If the "index" role were the default for the application, it could be
    dropped::

        Here's a `paragraph`.

    Both of these would result in this XML::

        <paragraph>
        Here's a <index_entry>paragraph</index_entry>.
        </paragraph>

  - _`body.literal`: Literal block, possibly "formal" (see `object
    numbering and object references`_ above).  Possible options:

	- "highlight" a range of lines

	- "number" or "line-numbers"

    - "styled" could indicate that the directive should check for
      style comments at the end of lines to indicate styling or
      markup.

      Specific derivatives (i.e., a "python-interactive" directive)
      could interpret style based on cues, like the ">>> " prompt and
      "input()"/"raw_input()" calls.

	See docutils-users 2003-03-03.

  - _`body.sidebar`: Add to the already implemented directive.  Allow
    internal section structure, with adornment styles independent of
    the main document.

  - _`body.list-table`: The idea came from Dylan Jay:

        ... to use a two level bulleted list with something to
        indicate it should be rendered as a table ...

    It's an interesting idea.  It could be implemented in as a
    directive which transforms a uniform two-level bullet list into a
    table.  Using a directive would allow the author to explicitly set
    the table's orientation (by column or by row), the presence of row
    headers, etc.

  - _`colorize.python`: Colorize Python code.  Fine for HTML output,
    but what about other formats?  Revert to a literal block?  Do we
    need some kind of "alternate" mechanism?  Perhaps use a "pending"
    transform, which could switch its output based on the "format" in
    use.  Use a factory function "transformFF()" which returns either
    "HTMLTransform()" instance or "GenericTransform" instance?

    If we take a Python-to-HTML pretty-printer and make it output a
    Docutils internal doctree (as per nodes.py) instead of HTML, then
    each output format's stylesheet (or equivalent) mechanism could
    take care of the rest.  The pretty-printer code could turn this
    doctree fragment::

         <literal_block xml:space="preserve">
         print 'This is Python code.'
         for i in range(10):
             print i
         </literal_block>

    into something like this ("</>" is end-tag shorthand)::

         <literal_block xml:space="preserve" class="python">
         <keyword>print</> <string>'This is Python code.'</>
         <keyword>for</> <identifier>i</> <keyword
         >in</> <expression>range(10)</>:
             <keyword>print</> <expression>i</>
         </literal_block>

    But I'm leaning toward adding a single new general-purpose
    element, "phrase", equivalent to HTML's <span>.  Here's the
    example rewritten using the generic "phrase"::

        <literal_block xml:space="preserve" class="python">
        <phrase class="keyword">print</> <phrase
         class="string">'This is Python code.'</>
        <phrase class="keyword">for</> <phrase
         class="identifier">i</> <phrase class="keyword">in</> <phrase
         class="expression">range(10)</>:
            <phrase class="keyword">print</> <phrase
             class="expression">i</>
        </literal_block>

    It's more verbose but more easily extensible and more appropriate
    for the case at hand.  It allows us to edit style sheets to add
    support for new formats, not the Docutils code itself.

    Perhaps a single directive with a format parameter would be
    better::

        .. colorize:: python

           print 'This is Python code.'
           for i in range(10):
               print i

    But directives can have synonyms for convenience.  "format::
    python" was suggested, but "format" seems too generic.

  - _`text.date`: Datestamp.  For substitutions.  The directive could
    be followed by a formatting string, using strftime codes.  Default
    is "%Y-%m-%d" (ISO 8601 date), but time fields can also be used.

    - Combined with the "include" directive, implement canned macros?
      E.g.::

          .. include:: <macros>

          Today's date is |date|.

      Where "macros" contains ``.. |date| date::``, among others.

  - _`text.time`: Timestamp.  For substitutions.  Shortcut for
    ``.. date:: %H:%M``.  Date fields can also be used.

  - _`pysource.usage`: Extract a usage message from the program,
    either by running it at the command line with a ``--help`` option
    or through an exposed API.  [Suggestion for Optik.]


Interpreted Text
````````````````

Interpreted text is entirely a reStructuredText markup construct, a
way to get around built-in limitations of the medium.  Some roles are
intended to introduce new doctree elements, such as "title-reference".
Others are merely convenience features, like "RFC".

All supported interpreted text roles must already be known to the
Parser when they are encountered in a document.  Whether pre-defined
in core/client code, or in the document, doesn't matter; the roles
just need to have already been declared.  Adding a new role often
involves adding a new element to the DTD and may require extensive
support, therefore such additions should be well thought-out.  There
should be a limited number of roles.

The only place where no limit is placed on variation is at the start,
at the Reader/Parser interface.  Transforms are inserted by the Reader
into the Transformer's queue, where non-standard elements are
converted.  Once past the Transformer, no variation from the standard
Docutils doctree is possible.

An example is the Python Source Reader, which will use interpreted
text extensively.  The default role will be "Python identifier", which
will be further interpreted by namespace context into <class>,
<method>, <module>, <attribute>, etc. elements (see
spec/pysource.dtd), which will be transformed into standard hyperlink
references, which will be processed by the various Writers.  No Writer
will need to have any knowledge of the Python-Reader origin of these
elements.

* Alan Jaffray suggested (and I agree) that it would be sensible to:

  - have a directive and/or command-line option to specify a default
    role for interpreted text
  - allow the reST processor to take an argument for the default role
    (this will be subsumed by the above via the runtime settings
    mechanism)
  - issue a warning when processing documents with no default role
    which contain interpreted text with no explicitly specified role
    (there will always be a default role, so this won't happen)

* Add explicit interpreted text roles for the rest of the implicit
  inline markup constructs: named-reference, anonymous-reference,
  footnote-reference, citation-reference, substitution-reference,
  target, uri-riference (& synonyms).

* Add directives for each role as well?  This would allow indirect
  nested markup::

      This text contains |nested inline markup|.

      .. |nested inline markup| emphasis::

         nested ``inline`` markup

* Implement roles:

  - "acronym" and "abbreviation": Associate the full text with a short
    form.  Jason Diamond's description:

        I want to translate ```reST`:acronym:`` into ``<acronym
        title='reStructuredText'>reST</acronym>``.  The value of the
        title attribute has to be defined out-of-band since you can't
        parameterize interpreted text.  Right now I have them in a
        separate file but I'm experimenting with creating a directive
        that will use some form of reST syntax to let you define them.

    Should Docutils complain about undefined acronyms or
    abbreviations?

    What to do if there are multiple definitions?  How to
    differentiate between CSS (Content Scrambling System) and CSS
    (Cascading Style Sheets) in a single document?  David Priest
    responds,

        The short answer is: you don't.  Anyone who did such a thing
        would be writing very poor documentation indeed.  (Though I
        note that `somewhere else in the docs`__, there's mention of
        allowing replacement text to be associated with the
        abbreviation.  That takes care of the duplicate
        acronyms/abbreviations problem, though a writer would be
        foolish to ever need it.)

        __ `inline parameter syntax`_

    How to define the full text?  Possibilities:

    1. With a directive and a definition list? ::

           .. acronyms::

              reST
                  reStructuredText
              DPS
                  Docstring Processing System

       Would this list remain in the document as a glossary, or would
       it simply build an internal lookup table?  A "glossary"
       directive could be used to make the intention clear.
       Acronyms/abbreviations and glossaries could work together.

       Then again, a glossary could be formed by gathering individual
       definitions from around the document.

    2. Some kind of `inline parameter syntax`_? ::

           `reST <reStructuredText>`:acronym: is `WYSIWYG <what you
           see is what you get>`:acronym: plaintext markup.

       .. _inline parameter syntax:
          rst/alternatives.html#parameterized-interpreted-text

    3. A combination of 1 & 2?

       The multiple definitions issue could be handled by establishing
       rules of priority.  For example, directive-based lookup tables
       have highest priority, followed by the first inline definition.
       Multiple definitions in directive-based lookup tables would
       trigger warnings, similar to the rules of `implicit hyperlink
       targets`__.

       __ rst/reStructuredText.html#implicit-hyperlink-targets

  - "annotation": The equivalent of the HTML "title" attribute.  This
    is secondary information that may "pop up" when the pointer hovers
    over the main text.  A corresponding directive would be required
    to associate annotations with the original text (by name, or
    positionally as in anonymous targets?).

  - "figure", "table", "listing", "chapter", "page", etc: See `object
    numbering and object references`_ above.

  - "term"?: Unfamiliar or specialized terminology.

  - "glossary-term": This would establish a link to a glossary.  It
    would require an associated "glossary-entry" directive, whose
    contents could be a definition list::

        .. glossary-entry::

           term1
               definition1
           term2
               definition2

    This would allow entries to be defined anywhere in the document,
    and collected (via a "glossary" directive perhaps) at one point.


Unimplemented Transforms
------------------------

Footnote & Citation Gathering
`````````````````````````````

Collect and move footnotes & citations to the end of a document.
(Separate transforms.)


Hyperlink Target Gathering
``````````````````````````

It probably comes in two phases, because in a Python context we need
to *resolve* them on a per-docstring basis [do we? --DG], but if the
user is trying to do the callout form of presentation, they would
then want to group them all at the end of the document.


Reference Merging
`````````````````

When merging two or more subdocuments (such as docstrings),
conflicting references may need to be resolved.  There may be:

* duplicate reference and/or substitution names that need to be made
  unique; and/or
* duplicate footnote numbers that need to be renumbered.

Should this be done before or after reference-resolving transforms
are applied?  What about references from within one subdocument to
inside another?


Document Splitting
``````````````````

If the processed document is written to multiple files (possibly in a
directory tree), it will need to be split up.  Internal references
will have to be adjusted.

(HTML only?  Initially, yes.  Eventually, anything should be
splittable.)

Idea: insert a "destination" attribute into the root element of each
split-out document, containing the path/filename.  The Output object
or Writer will recognize this attribute and split out the files
accordingly.  Must allow for common headers & footers, prev/next,
breadcrumbs, etc.


Navigation
``````````

If a document is split up, each segment will need navigation links:
parent, children (small TOC), previous (preorder), next (preorder).
Part of `Document Splitting`_?


List of System Messages
```````````````````````

The ``system_message`` elements are inserted into the document tree,
adjacent to the problems themselves where possible.  Some (those
generated post-parse) are kept until later, in ``document.messages``,
and added as a special final section, "Docutils System Messages".

Docutils could be made to generate hyperlinks to all known
system_messages and add them to the document, perhaps to the end of
the "Docutils System Messages" section.

Fred L. Drake, Jr. wrote:

    I'd like to propose that both parse- and transformation-time
    messages are included in the "Docutils System Messages" section.
    If there are no objections, I can make the change.

The advantage of the current way of doing things is that parse-time
system messages don't require a transform; they're already in the
document.  This is valuable for testing (unit tests,
tools/quicktest.py).  So if we do decide to make a change, I think the
insertion of parse-time system messages ought to remain as-is and the
Messages transform ought to move all parse-time system messages
(remove from their originally inserted positions, insert in System
Messages section).


Others
``````

Index


HTML Writer
-----------

* Construct a _`templating system`, as in ht2html/yaptu, using
  directives and substitutions for dynamic stuff?  Or a specialized
  writer to generate .ht & links.h files for ht2html?

  This is an overly generic, ambitious, and unnecessary idea.  The two
  systems, ht2html & yaptu, represent different approaches to
  templating.  Ht2html stiches bits of HTML together to make unified
  web pages.  Yaptu is for generic programmable text replacement.
  Templating should simply be left to other systems.  Users can choose
  a templating system to apply to their reStructuredText documents as
  best serves their interests.

* Add more support for <link> elements, especially for navigation
  bars.

* Make the admonitions more distinctive and varied.

* Make the "class" attributes optional?  Implies no stylesheet?

* Base list compaction on the spacing of source list?  Would require
  parser support.  (Idea: fantasai, 16 Dec 2002, doc-sig.)

* Add a tool tip ("title" attribute?) to footnote back-links
  identifying them as such.  Text in Docutils language module.

* Add an option to restrict the document title to <head><title> only,
  and not include it in the document body.  Subtitle?

* Insert a comment at the top of HTML files that describes how to deal
  with the broken servers w.r.t. encodings?  Perhaps something like
  this:

      <!--
      If your browser is showing gibberish, the server may be broken.
      Try manually setting the character coding to "UTF-8".  In
      Mozilla/Firefox, do ...  In Internet Explorer, do ...
      For details, see <URL>.
      -->


Front-End Tools
---------------

* What about if we don't know which Reader and/or Writer we are
  going to use?  If the Reader/Writer is specified on the
  command-line?  (Will this ever happen?)

  Perhaps have different types of front ends:

  a) _`Fully qualified`: Reader and Writer are hard-coded into the
     front end (e.g. ``pep2html [options]``, ``pysource2pdf
     [options]``).

  b) _`Partially qualified`: Reader is hard-coded, and the Writer is
     specified a sub-command (e.g. ``pep2 html [options]``,
     ``pysource2 pdf [options]``).  The Writer is known before option
     processing happens, allowing the OptionParser to be built
     dynamically.  Alternatively, the Writer could be hard-coded and
     the Reader specified as a sub-command (e.g. ``htmlfrom pep
     [options]``).

  c) _`Unqualified`: Reader and Writer are specified as subcommands
     (e.g. ``publish pep html [options]``, ``publish pysource pdf
     [options]``).  A single front end would be sufficient, but
     probably only useful for testing purposes.

  d) _`Dynamic`: Reader and/or Writer are specified by options, with
     defaults if unspecified (e.g. ``publish --writer pdf
     [options]``).  Is this possible?  The option parser would have
     to be told about new options it needs to handle, on the fly.
     Component-specific options would have to be specified *after*
     the component-specifying option.

  Allow common options before subcommands, as in CVS?  Or group all
  options together?  In the case of the `fully qualified`_
  front ends, all the options will have to be grouped together
  anyway, so there's no advantage (we can't use it to avoid
  conflicts) to splitting common and component-specific options
  apart.

* Parameterize help text & defaults somehow?  Perhaps a callback?  Or
  initialize ``settings_spec`` in ``__init__`` or ``init_options``?

* Disable common options that don't apply?

* Implement the "sectnum" directive as a command-line option also?

* @@@ Come up with better names for the most-used tools, and install
  them as scripts.

* Create a single dynamic_ or unqualified_ front end that can be
  installed?


Project Policies
================

A few quotes sum up the policies of the Docutils project.  The IETF's
classic credo (by MIT professor Dave Clark) is an ideal we can aspire
to:

    We reject: kings, presidents, and voting.  We believe in: rough
    consensus and running code.

As architect, chief cook and bottle-washer, I currently function as
BDFN (Benevolent Dictator For Now), but I would happily abdicate the
throne given a suitable candidate.  Any takers?

Eric S. Raymond, anthropologist of the hacker subculture, writes in
his essay `The Magic Cauldron`_:

    The number of contributors [to] projects is strongly and inversely
    correlated with the number of hoops each project makes a user go
    through to contribute.

    .. _The Magic Cauldron:
       http://www.tuxedo.org/~esr/writings/magic-cauldron/

Therefore, we will endeavour to keep the barrier to entry as low as
possible.  The policies below should not be thought of as barriers,
but merely as a codification of experience to date.  These are "best
practices", not absolutes; exceptions are expected, tolerated, and
used as a source of improvement.

As for control issues, Emmett Plant (CEO of the Xiph.org Foundation,
originators of Ogg Vorbis) put it well when he said:

    Open source dictates that you lose a certain amount of control
    over your codebase, and that's okay with us.


Python Coding Conventions
-------------------------

These are the conventions I use in my own code.  Contributed code will
not be refused merely because it does not strictly adhere to these
conditions; as long as it's internally consistent, clean, and correct,
it probably will be accepted.  But don't be surprised if the
"offending" code gets fiddled over time to conform to these
conventions.

The Docutils project shall follow the generic coding conventions as
specified in the `Style Guide for Python Code`_ and `Docstring
Conventions`_ PEPs, with the following clarifications (from most to
least important):

* 4 spaces per indentation level.  No tabs.

* Use only ASCII, no 8-bit strings.  See `Docutils
  Internationalization`_.

* No one-liner compound statements (i.e., no ``if x: return``: use two
  lines & indentation), except for degenerate class or method
  definitions (i.e., ``class X: pass`` is O.K.).

* Lines should be no more than 78 characters long.

* Use "StudlyCaps" for class names (except for element classes in
  docutils.nodes).

* Use "lowercase" or "lowercase_with_underscores" for function,
  method, and variable names.  For short names, maximum two words,
  joined lowercase may be used (e.g. "tagname").  For long names with
  three or more words, or where it's hard to parse the split between
  two words, use lowercase_with_underscores (e.g.,
  "note_explicit_target", "explicit_target").  If in doubt, use
  underscores.

* Use 'single quotes' for string literals, and """triple double
  quotes""" for docstrings.

.. _Style Guide for Python Code:
   http://www.python.org/peps/pep-0008.html
.. _Docstring Conventions: http://www.python.org/peps/pep-0257.html
.. _Docutils Internationalization: howto/i18n.html#python-code


Copyrights and Licensing
------------------------

The majority of the Docutils project code and documentation has been
placed in the public domain.  Unless clearly and explicitly indicated
otherwise, any patches (modifications to existing files) submitted to
the project for inclusion (via CVS, SourceForge trackers, mailing
lists, or private email) are assumed to be in the public domain as
well.

Any new files contributed to the project should clearly state their
intentions regarding copyright, in one of the following ways:

* Public domain (preferred): include the statement "This
  module/document has been placed in the public domain."

* Copyright & open source license: include a copyright notice, along
  with either an embedded license statement, a reference to an
  accompanying license file, or a license URL.

One of the goals of the Docutils project, once complete, is to be
incorporated into the Python standard library.  At that time copyright
of the Docutils code will be assumed by or transferred to the Python
Software Foundation (PSF), and will be released under Python's
license.  If the copyright/license option is chosen for new files, the
license should be compatible with Python's current license, and the
author(s) of the files should be willing to assign copyright to the
PSF.


CVS Check-ins
-------------

Instructions for CVS access can be found at
http://sourceforge.net/cvs/?group_id=38414.  Anyone can access the CVS
repository anonymously.  Only project developers can make changes.
See `Setting Up For Docutils Development`_ below for more info.

Unless you really *really* know what you're doing, please limit your
CVS commands to ``cvs checkout``, ``cvs commit/checkin``, and ``cvs
add``.  Do **NOT** use ``cvs import`` unless you're absolutely sure
you know what you're doing.  Even then, grab a copy of the `nightly
CVS tarball <http://cvs.sf.net/cvstarballs/docutils-cvsroot.tar.gz>`_,
set it up on your own machine, and experiment *there* first.

The `main source tree`_ ("docutils" CVS module) should always be kept
in a stable state (usable and as problem-free as possible).  The
Docutils project shall follow the `Python Check-in Policies`_ (as
applicable), with particular emphasis as follows:

* Before checking in any changes, run the entire Docutils test suite
  to be sure that you haven't broken anything.  From a shell::

      cd docutils/test
      alltests.py

* When adding new functionality (or fixing bugs), be sure to add test
  cases to the test suite.  Practise test-first programming; it's fun,
  it's addictive, and it works!

* The `sandbox CVS directory`_ is the place to put new, incomplete or
  experimental code.  See `Additions to Docutils`_ and `The Sandbox`_
  below.

* For bugs or omissions that have an obvious fix and can't possibly
  mess up anything else, go right ahead and check it in directly.

* For larger changes, use your best judgement.  If you're unsure of
  the impact, or feel that you require advice or approval, patches or
  `the sandbox`_ are the way to go.

Docutils will pursue an open and trusting policy for as long as
possible, and deal with any aberrations if (and hopefully not when)
they happen.  I'd rather see a torrent of loose contributions than
just a trickle of perfect-as-they-stand changes.  The occasional
mistake is easy to fix.  That's what CVS is for.

.. _main source tree:
   http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/docutils/docutils/
.. _Python Check-in Policies: http://www.python.org/dev/tools.html
.. _sandbox CVS directory:
   http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/docutils/sandbox/


Additions to Docutils
`````````````````````

Additions to the project, such as new components, should be developed
in the `sandbox CVS directory`_ until they're in `good shape`_,
usable_, and `reasonably complete`_.  Adding to the `main source
tree`_ or to a `parallel project`_ implies a commitment to the
Docutils user community.

* Why the sandbox?

  Developers should be able to try out new components while they're
  being developed for addition to main source tree.  See `The
  Sandbox`_ below.

* _`Good shape` means that the component code is clean, readable, and
  free of junk code (unused legacy code; by analogy with "junk DNA").

* _`Usable` means that the code does what it claims to do.  An "XYZ
  Writer" should produce reasonable XYZ.

* _`Reasonably complete` means that the code must handle all input.
  Here "handle" means that no input can cause the code to fail (cause
  an exception, or silently and incorrectly produce nothing).
  "Reasonably complete" does not mean "finished" (no work left to be
  done).  For example, a writer must handle every standard element
  from the Docutils document model; for unimplemented elements, it
  must *at the very least* warn that "Output for element X is not yet
  implemented in writer Y".

If you really want to check code into the main source tree, you can,
but you'll have to be prepared to work on it intensively and complete
it quickly.  People will start to use it and they will expect it to
work!  If there are any issues with your code, or if you only have
time for gradual development, you should put it in the sandbox first.
It's easy to move code over to the main source tree once it's closer
to completion.


Setting Up For Docutils Development
```````````````````````````````````

When making changes to the code, good developers always test their
changes.  That means running the code to check that it produces the
expected results, and running the test suite too.  The modified
Docutils code has to be accessible to Python for the tests to have any
meaning.  There are two ways to keep the Docutils code accessible:

* Update your ``PYTHONPATH`` environment variable so that Python picks
  up your local working copy of the code.  This is the recommended
  method.

  For the bash shell and Docutils checked out from CVS in
  ``~/projects/docutils/``, add this to your ``~/.profile``::

      PYTHONPATH=$HOME/projects/docutils/docutils
      PYTHONPATH=$PYTHONPATH:$HOME/projects/docutils/docutils/extras
      export PYTHONPATH

  The first line points to the directory containing the ``docutils``
  package.  The second line adds the directory containing the
  third-party modules Docutils depends on.  The third line exports
  this environment variable.  You may also wish to add the ``tools``
  directory to your ``PATH``::

      PATH=$PATH:$HOME/projects/docutils/docutils/tools

* Before you run anything, every time you make a change, reinstall
  Docutils::

      python setup.py install

  .. CAUTION::

     This method is **not** recommended for day-to-day development;
     it's too easy to forget.  Confusion inevitably ensues.

     If you install Docutils this way, Python will always pick up the
     last-installed copy of the code.  If you ever forget to reinstall
     the "docutils" package, Python won't see your latest changes.


Mailing Lists
-------------

Developers should subscribe to the mailing lists:

* The `Python Documentation Special Interest Group (Doc-SIG) mailing
  list`__ for high-level discussions on syntax, strategy, and design
  (email to Doc-SIG@python.org).
* Docutils-develop__, for implementation discussions
  (email to docutils-develop@lists.sourceforge.net).
* Docutils-checkins__, to monitor CVS checkin messages (automatically
  generated; normally read-only).

__ http://mail.python.org/mailman/listinfo/doc-sig
__ http://lists.sourceforge.net/lists/listinfo/docutils-develop
__ http://lists.sourceforge.net/lists/listinfo/docutils-checkins


The Sandbox
-----------

The `sandbox CVS directory`_ is a place to play around, to try out and
share ideas.  It's a part of the CVS repository but it isn't
distributed as part of Docutils releases.  Feel free to check in code
to the CVS sandbox; that way people can try it out but you won't have
to worry about it working 100% error-free, as is the goal of the `main
source tree`_.  Each developer who wants to play in the sandbox should
create either a project-specific subdirectory or personal subdirectory
(suggested name: SourceForge ID, nickname, or given name + family
initial).  It's OK to make a mess!  But please, play nice.

Please update the `sandbox README`_ file with links and a brief
description of your work.

In order to minimize the work necessary for others to install and try
out new, experimental components, the following sandbox directory
structure is recommended::

    sandbox/
        project_name/ # For a project where you invite contributions.
                      # Structure as in userid/component_name below.
        userid/       # For personal space.
            component_name/ # A verbose name is best.
                README.txt  # Please explain requirements,
                            # purpose/goals, and usage.
                docs/
                    ...
                component.py    # The component is a single module.
                            # *OR* (but *not* both)
                component/  # The component is a package.
                    __init__.py  # Contains the Reader/Writer class.
                    other1.py    # Other modules and data files used
                    data.txt     # by this component.
                    ...
                test/       # Test suite.
                    ...
                tools/      # For front ends etc.
                    ...
                setup.py    # Use Distutils to install the component
                            # code and tools/ files into the right
                            # places in Docutils.

Some sandbox projects are destined to become Docutils components once
completed.  Others, such as add-ons to Docutils or applications of
Docutils, graduate to become `parallel projects`_.

.. _sandbox README: http://docutils.sf.net/sandbox/README.html


.. _parallel project:

Parallel Projects
-----------------

Parallel projects contain useful code that is not central to the
functioning of Docutils.  Examples are specialized add-ons or
plug-ins, and applications of Docutils.  They use Docutils, but
Docutils does not require their presence to function.

An official parallel project will have its own CVS directory beside
(or parallel to) the main Docutils CVS directory.  It can have its own
web page in the docutils.sourceforge.net domain, its own file releases
and downloadable CVS snapshots, and even a mailing list if that proves
useful. However, an official parallel project has implications: it is
expected to be maintained and continue to work with changes to the
core Docutils.

A parallel project requires a project leader, who must commit to
coordinate and maintain the implementation:

* Answer questions from users and developers.
* Review suggestions, bug reports, and patches.
* Monitor changes and ensure the quality of the code and
  documentation.
* Coordinate with Docutils to ensure interoperability.
* Put together official project releases.

Of course, related projects may be created independently of Docutils.
The advantage of a parallel project is that the SourceForge
environment and the developer and user communities are already
established.  Core Docutils developers are available for consultation
and may contribute to the parallel project.  It's easier to keep the
projects in sync when there are changes made to the core Docutils
code.


Web Site
========

The Docutils web site, <http://docutils.sourceforge.net/>, is
maintained automatically by the docutils-update script, run as an
hourly cron job (by user "goodger").  The script will process any .txt
file which is newer than the corresponding .html file in the project's
web directory on SourceForge (``/home/groups/d/do/docutils/htdocs/``).
For a new .txt file, just SSH to SourceForge and ::

    touch filename.html
    chmod g+w filename.html
    sleep 1
    touch filename.txt

The script will take care of the rest within an hour.  Thereafter
whenever the .txt file is modified (checked in to CVS), the .html will
be regenerated automatically.

After adding directories to CVS, allow the script to run once to
create the directories in the filesystem before preparing for HTML
processing as described above.

The docutils-update__ script is in CVS as
``sandbox/davidg/infrastructure/docutils-update``.

__ http://docutils.sf.net/sandbox/davidg/infrastructure/docutils-update


Release Procedure
=================

1. Edit the version number in the following files:

   * docutils:

     - setup.py
     - HISTORY.txt
     - docutils/__init__.py

   * web: index.txt

2. Run the test suite: ``cd test ; alltests.py``.

3. Isolate from outside influence:

   (a) Remove the old installation from site-packages (including
       roman.py, and optparse.py, textwrap.py for pre-2.3
       installations).

   (b) Clear/unset the PYTHONPATH environment variable.

4. Create the release tarball:

   (a) Create a new empty directory and ``cd`` into it.

   (b) Get a clean snapshot of the CVS files::

           cvs -z3 -d:pserver:anonymous@cvs.sf.net:/cvsroot/docutils \
               export -rHEAD docutils

   (c) Use Distutils to create the release tarball::

           cd docutils
           python setup.py sdist

5. Expand and install the release tarball **in isolation** (as per
   step 3 above):

   (a) Expand the tarball in a new location, not over any existing
       files.

   (b) Install from expanded directory::

           cd docutils-X.Y
           python setup.py install

       The "install" command may require root permissions.

6. Run the test suite from the expanded archive directory: ``cd test ;
   alltests.py``.

7. Run ``cd tools ; buildhtml.py ..`` to confirm that there are no
   unexpected issues with the docs.

8. Upload the release tarball::

       $ ftp upload.sourceforge.net
       Connected to osdn.dl.sourceforge.net.
       ...
       Name (upload.sourceforge.net:david): anonymous
       331 Anonymous login ok, send your complete e-mail address as password.
       Password:
       ...
       230 Anonymous access granted, restrictions apply.
       ftp> bin
       200 Type set to I.
       ftp> cd /incoming
       250 CWD command successful.
       ftp> put filename

9. Log in to the SourceForge web interface.

10. Access the file release system on SourceForge (Admin interface).
    Fill in the fields:

        :Package ID:      docutils
        :Release Name:    <use release number only, e.g. 0.3>
        :Release Date:    <today's date>
        :Status:          Active
        :File Name:       <select the file just uploaded>
        :File Type:       Source .gz
        :Processor Type:  Platform-Independent
        :Release Notes:   <insert README.txt file here>
        :Change Log:      <insert summary from announcement>

        Also check the "Preserve my pre-formatted text" box.

11. Wait up to 30 minutes for the file to become available on
    SourceForge.

12. Download the release tarball and verify its integrity by walking
    through an installation, as outlined above (steps 5, 6, & 7).

13. Add a SourceForge News item, with title "Docutils 0.x released"
    and containing the release tarball's download URL.

14. Send announcement email to:

    * docutils-develop@lists.sourceforge.net
    * docutils-users@lists.sourceforge.net
    * doc-sig@python.org
    * python-list@python.org
    * python-announce@python.org

15. Register

    (a) with PyPI (Fill in details.  ``python setup.py register``?
        How to log in?)
    (b) with Vaults of Parnassus
    (c) with FreshMeat?


..
   Local Variables:
   mode: indented-text
   indent-tabs-mode: nil
   sentence-end-double-space: t
   fill-column: 70
   End:
