1 ==================================================
2 A Record of reStructuredText Syntax Alternatives
3 ==================================================
6 :Contact: goodger@python.org
9 :Copyright: This document has been placed in the public domain.
11 The following are ideas, alternatives, and justifications that were
12 considered for reStructuredText syntax, which did not originate with
13 Setext_ or StructuredText_. For an analysis of constructs which *did*
14 originate with StructuredText or Setext, please see `Problems With
15 StructuredText`_. See the `reStructuredText Markup Specification`_
16 for full details of the established syntax.
18 The ideas are divided into sections:
20 * Implemented_: already done. The issues and alternatives are
21 recorded here for posterity.
23 * `Not Implemented`_: these ideas won't be implemented.
25 * Tabled_: these ideas should be revisited in the future.
27 * `To Do`_: these ideas should be implemented. They're just waiting
28 for a champion to resolve issues and get them done.
30 * `... Or Not To Do?`_: possible but questionable. These probably
31 won't be implemented, but you never know.
33 .. _Setext: http://docutils.sourceforge.net/mirror/setext.html
35 http://www.zope.org/DevHome/Members/jim/StructuredTextWiki/FrontPage
36 .. _Problems with StructuredText: problems.html
37 .. _reStructuredText Markup Specification:
38 ../../ref/rst/restructuredtext.html
50 Prior to the syntax for field lists being finalized, several
51 alternatives were proposed.
53 1. Unadorned RFC822_ everywhere::
58 Advantages: clean, precedent (RFC822-compliant). Disadvantage:
59 ambiguous (these paragraphs are a prime example).
63 2. Special case: use unadorned RFC822_ for the very first or very last
64 text block of a document::
70 The rest of the document...
73 Advantages: clean, precedent (RFC822-compliant). Disadvantages:
74 special case, flat (unnested) field lists only, still ambiguous::
77 Usage: cmdname [options] arg1 arg2 ...
79 We obviously *don't* want the like above to be interpreted as a
80 field list item. Or do we?
83 Conclusion: rejected for the general case, accepted for specific
84 contexts (PEPs, email).
93 Advantages: explicit and unambiguous, RFC822-compliant.
94 Disadvantage: cumbersome.
96 Conclusion: rejected for the general case (but such a directive
97 could certainly be written).
99 4. Use Javadoc-style::
105 Advantages: unambiguous, precedent, flexible. Disadvantages:
106 non-intuitive, ugly, not RFC822-compliant.
108 Conclusion: rejected.
110 5. Use leading colons::
115 Advantages: unambiguous, obvious (*almost* RFC822-compliant),
116 flexible, perhaps even elegant. Disadvantages: no precedent, not
117 quite RFC822-compliant.
119 Conclusion: accepted!
121 6. Use double colons::
126 Advantages: unambiguous, obvious? (*almost* RFC822-compliant),
127 flexible, similar to syntax already used for literal blocks and
128 directives. Disadvantages: no precedent, not quite
129 RFC822-compliant, similar to syntax already used for literal blocks
132 Conclusion: rejected because of the syntax similarity & conflicts.
134 Why is RFC822 compliance important? It's a universal Internet
135 standard, and super obvious. Also, I'd like to support the PEP format
136 (ulterior motive: get PEPs to use reStructuredText as their standard).
137 But it *would* be easy to get used to an alternative (easy even to
138 convert PEPs; probably harder to convert python-deviants ;-).
140 Unfortunately, without well-defined context (such as in email headers:
141 RFC822 only applies before any blank lines), the RFC822 format is
142 ambiguous. It is very common in ordinary text. To implement field
143 lists unambiguously, we need explicit syntax.
145 The following question was posed in a footnote:
147 Should "bibliographic field lists" be defined at the parser level,
148 or at the DPS transformation level? In other words, are they
149 reStructuredText-specific, or would they also be applicable to
150 another (many/every other?) syntax?
152 The answer is that bibliographic fields are a
153 reStructuredText-specific markup convention. Other syntaxes may
154 implement the bibliographic elements explicitly. For example, there
155 would be no need for such a transformation for an XML-based markup
158 .. _RFC822: http://www.rfc-editor.org/rfc/rfc822.txt
161 Interpreted Text "Roles"
162 ========================
164 The original purpose of interpreted text was as a mechanism for
165 descriptive markup, to describe the nature or role of a word or
166 phrase. For example, in XML we could say "<function>len</function>"
167 to mark up "len" as a function. It is envisaged that within Python
168 docstrings (inline documentation in Python module source files, the
169 primary market for reStructuredText) the role of a piece of
170 interpreted text can be inferred implicitly from the context of the
171 docstring within the program source. For other applications, however,
172 the role may have to be indicated explicitly.
174 Interpreted text is enclosed in single backquotes (`).
176 1. Initially, it was proposed that an explicit role could be indicated
177 as a word or phrase within the enclosing backquotes:
179 - As a prefix, separated by a colon and whitespace::
181 `role: interpreted text`
183 - As a suffix, separated by whitespace and a colon::
185 `interpreted text :role`
187 There are problems with the initial approach:
189 - There could be ambiguity with interpreted text containing colons.
190 For example, an index entry of "Mission: Impossible" would
191 require a backslash-escaped colon.
193 - The explicit role is descriptive markup, not content, and will
194 not be visible in the processed output. Putting it inside the
195 backquotes doesn't feel right; the *role* isn't being quoted.
197 2. Tony Ibbs suggested that the role be placed outside the
200 role:`prefix` or `suffix`:role
202 This removes the embedded-colons ambiguity, but limits the role
203 identifier to be a single word (whitespace would be illegal).
204 Since roles are not meant to be visible after processing, the lack
205 of whitespace support is not important.
207 The suggested syntax remains ambiguous with respect to ratios and
208 some writing styles. For example, suppose there is a "signal"
209 identifier, and we write::
211 ...calculate the `signal`:noise ratio.
213 "noise" looks like a role.
215 3. As an improvement on #2, we can bracket the role with colons::
217 :role:`prefix` or `suffix`:role:
219 This syntax is similar to that of field lists, which is fine since
220 both are doing similar things: describing.
222 This is the syntax chosen for reStructuredText.
224 4. Another alternative is two colons instead of one::
226 role::`prefix` or `suffix`::role
228 But this is used for analogies ("A:B::C:D": "A is to B as C is to
231 Both alternative #2 and #4 lack delimiters on both sides of the
232 role, making it difficult to parse (by the reader).
234 5. Some kind of bracketing could be used:
238 (role)`prefix` or `suffix`(role)
242 {role}`prefix` or `suffix`{role}
246 [role]`prefix` or `suffix`[role]
250 <role>`prefix` or `suffix`<role>
252 (The overlap of \*ML tags with angle brackets would be too
253 confusing and precludes their use.)
255 Syntax #3 was chosen for reStructuredText.
261 A problem with comments (actually, with all indented constructs) is
262 that they cannot be followed by an indented block -- a block quote --
263 without swallowing it up.
265 I thought that perhaps comments should be one-liners only. But would
266 this mean that footnotes, hyperlink targets, and directives must then
267 also be one-liners? Not a good solution.
269 Tony Ibbs suggested a "comment" directive. I added that we could
270 limit a comment to a single text block, and that a "multi-block
271 comment" could use "comment-start" and "comment-end" directives. This
272 would remove the indentation incompatibility. A "comment" directive
273 automatically suggests "footnote" and (hyperlink) "target" directives
274 as well. This could go on forever! Bad choice.
276 Garth Kidd suggested that an "empty comment", a ".." explicit markup
277 start with nothing on the first line (except possibly whitespace) and
278 a blank line immediately following, could serve as an "unindent". An
279 empty comment does **not** swallow up indented blocks following it,
280 so block quotes are safe. "A tiny but practical wart." Accepted.
286 Alan Jaffray came up with this idea, along with the following syntax::
288 Search the `Python DOC-SIG mailing list archives`{}_.
290 .. _: http://mail.python.org/pipermail/doc-sig/
292 The idea is sound and useful. I suggested a "double underscore"
295 Search the `Python DOC-SIG mailing list archives`__.
297 .. __: http://mail.python.org/pipermail/doc-sig/
299 But perhaps single underscores are okay? The syntax looks better, but
300 the hyperlink itself doesn't explicitly say "anonymous"::
302 Search the `Python DOC-SIG mailing list archives`_.
304 .. _: http://mail.python.org/pipermail/doc-sig/
306 Mixing anonymous and named hyperlinks becomes confusing. The order of
307 targets is not significant for named hyperlinks, but it is for
308 anonymous hyperlinks::
310 Hyperlinks: anonymous_, named_, and another anonymous_.
316 Without the extra syntax of double underscores, determining which
317 hyperlink references are anonymous may be difficult. We'd have to
318 check which references don't have corresponding targets, and match
319 those up with anonymous targets. Keeping to a simple consistent
320 ordering (as with auto-numbered footnotes) seems simplest.
322 reStructuredText will use the explicit double-underscore syntax for
323 anonymous hyperlinks. An alternative (see `Reworking Explicit Markup
324 (Round 1)`_ below) for the somewhat awkward ".. __:" syntax is "__"::
326 An anonymous__ reference.
331 Reworking Explicit Markup (Round 1)
332 ===================================
334 Alan Jaffray came up with the idea of `anonymous hyperlinks`_, added
335 to reStructuredText. Subsequently it was asserted that hyperlinks
336 (especially anonymous hyperlinks) would play an increasingly important
337 role in reStructuredText documents, and therefore they require a
338 simpler and more concise syntax. This prompted a review of the
339 current and proposed explicit markup syntaxes with regards to
344 .. _blah: internal hyperlink target
345 .. _blah: http://somewhere external hyperlink target
346 .. _blah: blahblah_ indirect hyperlink target
347 .. __: anonymous internal target
348 .. __: http://somewhere anonymous external target
349 .. __: blahblah_ anonymous indirect target
350 .. [blah] http://somewhere footnote
351 .. blah:: http://somewhere directive
352 .. blah: http://somewhere comment
356 The comment text was intentionally made to look like a hyperlink
361 * Except for the colon (a delimiter necessary to allow for
362 phrase-links), hyperlink target ``.. _blah:`` comes from Setext.
363 * Comment syntax from Setext.
364 * Footnote syntax from StructuredText ("named links").
365 * Directives and anonymous hyperlinks original to reStructuredText.
369 + Consistent explicit markup indicator: "..".
370 + Consistent hyperlink syntax: ".. _" & ":".
374 - Anonymous target markup is awkward: ".. __:".
375 - The explicit markup indicator ("..") is excessively overloaded?
376 - Comment text is limited (can't look like a footnote, hyperlink,
377 or directive). But this is probably not important.
379 2. Alan Jaffray's proposed syntax #1::
381 __ _blah internal hyperlink target
382 __ blah: http://somewhere external hyperlink target
383 __ blah: blahblah_ indirect hyperlink target
384 __ anonymous internal target
385 __ http://somewhere anonymous external target
386 __ blahblah_ anonymous indirect target
387 __ [blah] http://somewhere footnote
388 .. blah:: http://somewhere directive
389 .. blah: http://somewhere comment
391 The hyperlink-connoted underscores have become first-level syntax.
395 + Anonymous targets are simpler.
396 + All hyperlink targets are one character shorter.
400 - Inconsistent internal hyperlink targets. Unlike all other named
401 hyperlink targets, there's no colon. There's an extra leading
402 underscore, but we can't drop it because without it, "blah" looks
403 like a relative URI. Unless we restore the colon::
405 __ blah: internal hyperlink target
409 3. Alan Jaffray's proposed syntax #2::
411 .. _blah internal hyperlink target
412 .. blah: http://somewhere external hyperlink target
413 .. blah: blahblah_ indirect hyperlink target
414 .. anonymous internal target
415 .. http://somewhere anonymous external target
416 .. blahblah_ anonymous indirect target
417 .. [blah] http://somewhere footnote
418 !! blah: http://somewhere directive
419 ## blah: http://somewhere comment
421 Leading underscores have been (almost) replaced by "..", while
422 comments and directives have gained their own syntax.
426 + Anonymous hyperlinks are simpler.
427 + Unique syntax for comments. Connotation of "comment" from
428 some programming languages (including our favorite).
429 + Unique syntax for directives. Connotation of "action!".
433 - Inconsistent internal hyperlink targets. Again, unlike all other
434 named hyperlink targets, there's no colon. There's a leading
435 underscore, matching the trailing underscores of references,
436 which no other hyperlink targets have. We can't drop that one
437 leading underscore though: without it, "blah" looks like a
438 relative URI. Again, unless we restore the colon::
440 .. blah: internal hyperlink target
442 - All (except for internal) hyperlink targets lack their leading
443 underscores, losing the "hyperlink" connotation.
445 - Obtrusive syntax for comments. Alternatives::
447 ;; blah: http://somewhere
448 (also comment syntax in Lisp & others)
449 ,, blah: http://somewhere
450 ("comma comma": sounds like "comment"!)
452 - Iffy syntax for directives. Alternatives?
454 4. Tony Ibbs' proposed syntax::
456 .. _blah: internal hyperlink target
457 .. _blah: http://somewhere external hyperlink target
458 .. _blah: blahblah_ indirect hyperlink target
459 .. anonymous internal target
460 .. http://somewhere anonymous external target
461 .. blahblah_ anonymous indirect target
462 .. [blah] http://somewhere footnote
463 .. blah:: http://somewhere directive
464 .. blah: http://somewhere comment
466 This is the same as the current syntax, except for anonymous
467 targets which drop their "__: ".
471 + Anonymous targets are simpler.
475 - Anonymous targets lack their leading underscores, losing the
476 "hyperlink" connotation.
477 - Anonymous targets are almost indistinguishable from comments.
478 (Better to know "up front".)
480 5. David Goodger's proposed syntax: Perhaps going back to one of
481 Alan's earlier suggestions might be the best solution. How about
482 simply adding "__ " as a synonym for ".. __: " in the original
483 syntax? These would become equivalent::
485 .. __: anonymous internal target
486 .. __: http://somewhere anonymous external target
487 .. __: blahblah_ anonymous indirect target
489 __ anonymous internal target
490 __ http://somewhere anonymous external target
491 __ blahblah_ anonymous indirect target
493 Alternative 5 has been adopted.
496 Backquotes in Phrase-Links
497 ==========================
499 [From a 2001-06-05 Doc-SIG post in reply to questions from Doug
502 The first draft of the spec, posted to the Doc-SIG in November 2000,
503 used square brackets for phrase-links. I changed my mind because:
505 1. In the first draft, I had already decided on single-backquotes for
508 2. However, I wanted to minimize the necessity for backslash escapes,
509 for example when quoting Python repr-equivalent syntax that uses
512 3. The processing of identifiers (function/method/attribute/module
513 etc. names) into hyperlinks is a useful feature. PyDoc recognizes
514 identifiers heuristically, but it doesn't take much imagination to
515 come up with counter-examples where PyDoc's heuristics would result
516 in embarassing failure. I wanted to do it deterministically, and
517 that called for syntax. I called this construct "interpreted
520 4. Leveraging off the ``*emphasis*/**strong**`` syntax, lead to the
521 idea of using double-backquotes as syntax.
523 5. I worked out some rules for inline markup recognition.
525 6. In combination with #5, double backquotes lent themselves to inline
526 literals, neatly satisfying #2, minimizing backslash escapes. In
527 fact, the spec says that no interpretation of any kind is done
528 within double-backquote inline literal text; backslashes do *no*
529 escaping within literal text.
531 7. Single backquotes are then freed up for interpreted text.
533 8. I already had square brackets required for footnote references.
535 9. Since interpreted text will typically turn into hyperlinks, it was
536 a natural fit to use backquotes as the phrase-quoting syntax for
537 trailing-underscore hyperlinks.
539 The original inspiration for the trailing underscore hyperlink syntax
540 was Setext. But for phrases Setext used a very cumbersome
541 ``underscores_between_words_like_this_`` syntax.
543 The underscores can be viewed as if they were right-pointing arrows:
544 ``-->``. So ``hyperlink_`` points away from the reference, and
545 ``.. _hyperlink:`` points toward the target.
548 Substitution Mechanism
549 ======================
551 Substitutions arose out of a Doc-SIG thread begun on 2001-10-28 by
552 Alan Jaffray, "reStructuredText inline markup". It reminded me of a
553 missing piece of the reStructuredText puzzle, first referred to in my
554 contribution to "Documentation markup & processing / PEPs" (Doc-SIG
557 Substitutions allow the power and flexibility of directives to be
558 shared by inline text. They are a way to allow arbitrarily complex
559 inline objects, while keeping the details out of the flow of text.
560 They are the equivalent of SGML/XML's named entities. For example, an
561 inline image (using reference syntax alternative 4d (vertical bars)
562 and definition alternative 3, the alternatives chosen for inclusion in
565 The |biohazard| symbol must be used on containers used to dispose
568 .. |biohazard| image:: biohazard.png
571 The ``|biohazard|`` substitution reference will be replaced in-line by
572 whatever the ``.. |biohazard|`` substitution definition generates (in
573 this case, an image). A substitution definition contains the
574 substitution text bracketed with vertical bars, followed by a an
575 embedded inline-compatible directive, such as "image". A transform is
576 required to complete the substitution.
578 Syntax alternatives for the reference:
580 1. Use the existing interpreted text syntax, with a predefined role
583 The `biohazard`:sub: symbol...
585 Advantages: existing syntax, explicit. Disadvantages: verbose,
588 2. Use a variant of the interpreted text syntax, with a new suffix
589 akin to the underscore in phrase-link references::
600 Due to incompatibility with other constructs and ordinary text
601 usage, (f) and (g) are not possible.
603 3. Use interpreted text syntax with a fixed internal format::
619 To avoid ML confusion (k) and (l) are definitely out. Square
620 brackets (j) won't work in the target (the substitution definition
621 would be indistinguishable from a footnote).
623 The ```/name/``` syntax (g) is reminiscent of "s/find/sub"
624 substitution syntax in ed-like languages. However, it may have a
625 misleading association with regexps, and looks like an absolute
626 POSIX path. (i) is visually equivalent and lacking the
629 A disadvantage of all of these is that they limit interpreted text,
630 albeit only slightly.
632 4. Use specialized syntax, something new::
649 "#" (a) and "@" (b) are obtrusive. "/" (c) without backquotes
650 looks just like a POSIX path; it is likely for such usage to appear
653 "|" (d) and "^" (h) are feasible.
655 5. Redefine the trailing underscore syntax. See definition syntax
656 alternative 4, below.
658 Syntax alternatives for the definition:
660 1. Use the existing directive syntax, with a predefined directive such
661 as "sub". It contains a further embedded directive resolving to an
662 inline-compatible object::
665 .. image:: biohazard.png
669 That bird wouldn't *voom* if you put 10,000,000 volts
672 The advantages and disadvantages are the same as in inline
675 2. Use syntax as in #1, but with an embedded directivecompressed::
677 .. sub:: biohazard image:: biohazard.png
680 This is a bit better than alternative 1, but still too much.
682 3. Use a variant of directive syntax, incorporating the substitution
683 text, obviating the need for a special "sub" directive name. If we
684 assume reference alternative 4d (vertical bars), the matching
685 definition would look like this::
687 .. |biohazard| image:: biohazard.png
690 4. (Suggested by Alan Jaffray on Doc-SIG from 2001-11-06.)
692 Instead of adding new syntax, redefine the trailing underscore
693 syntax to mean "substitution reference" instead of "hyperlink
694 reference". Alan's example::
696 I had lunch with Jonathan_ today. We talked about Zope_.
698 .. _Jonathan: lj [user=jhl]
699 .. _Zope: http://www.zope.org/
701 A problem with the proposed syntax is that URIs which look like
702 simple reference names (alphanum plus ".", "-", "_") would be
703 indistinguishable from substitution directive names. A more
704 consistent syntax would be::
706 I had lunch with Jonathan_ today. We talked about Zope_.
708 .. _Jonathan: lj:: user=jhl
709 .. _Zope: http://www.zope.org/
711 (``::`` after ``.. _Jonathan: lj``.)
713 The "Zope" target is a simple external hyperlink, but the
714 "Jonathan" target contains a directive. Alan proposed is that the
715 reference text be replaced by whatever the referenced directive
716 (the "directive target") produces. A directive reference becomes a
717 hyperlink reference if the contents of the directive target resolve
718 to a hyperlink. If the directive target resolves to an icon, the
719 reference is replaced by an inline icon. If the directive target
720 resolves to a hyperlink, the directive reference becomes a
723 This seems too indirect and complicated for easy comprehension.
725 The reference in the text will sometimes become a link, sometimes
726 not. Sometimes the reference text will remain, sometimes not. We
727 don't know *at the reference*::
729 This is a `hyperlink reference`_; its text will remain.
730 This is an `inline icon`_; its text will disappear.
734 The syntax that has been incorporated into the spec and parser is
735 reference alternative 4d with definition alternative 3::
737 The |biohazard| symbol...
739 .. |biohazard| image:: biohazard.png
742 We can also combine substitution references with hyperlink references,
743 by appending a "_" (named hyperlink reference) or "__" (anonymous
744 hyperlink reference) suffix to the substitution reference. This
745 allows us to click on an image-link::
747 The |biohazard|_ symbol...
749 .. |biohazard| image:: biohazard.png
751 .. _biohazard: http://www.cdc.gov/
753 There have been several suggestions for the naming of these
754 constructs, originally called "substitution references" and
757 1. Candidate names for the reference construct:
759 (a) substitution reference
760 (b) tagging reference
761 (c) inline directive reference
762 (d) directive reference
763 (e) indirect inline directive reference
764 (f) inline directive placeholder
765 (g) inline directive insertion reference
766 (h) directive insertion reference
767 (i) insertion reference
768 (j) directive macro reference
770 (l) substitution directive reference
772 2. Candidate names for the definition construct:
775 (b) substitution directive
780 (g) inline directive definition
781 (h) referenced directive
782 (i) indirect directive
783 (j) indirect directive definition
784 (k) directive definition
785 (l) indirect inline directive
786 (m) named directive definition
787 (n) inline directive insertion definition
788 (o) directive insertion definition
789 (p) insertion definition
790 (q) insertion directive
791 (r) substitution definition
792 (s) directive macro definition
794 (u) substitution directive definition
795 (v) substitution definition
797 "Inline directive reference" (1c) seems to be an appropriate term at
798 first, but the term "inline" is redundant in the case of the
799 reference. Its counterpart "inline directive definition" (2g) is
800 awkward, because the directive definition itself is not inline.
802 "Directive reference" (1d) and "directive definition" (2k) are too
803 vague. "Directive definition" could be used to refer to any
804 directive, not just those used for inline substitutions.
806 One meaning of the term "macro" (1k, 2s, 2t) is too
807 programming-language-specific. Also, macros are typically simple text
808 substitution mechanisms: the text is substituted first and evaluated
809 later. reStructuredText substitution definitions are evaluated in
810 place at parse time and substituted afterwards.
812 "Insertion" (1h, 1i, 2n-2q) is almost right, but it implies that
813 something new is getting added rather than one construct being
816 Which brings us back to "substitution". The overall best names are
817 "substitution reference" (1a) and "substitution definition" (2v). A
818 long way to go to add one word!
821 Inline External Targets
822 =======================
824 Currently reStructuredText has two hyperlink syntax variations:
828 This is a named reference_ of one word ("reference"). Here is
829 a `phrase reference`_. Phrase references may even cross `line
832 .. _reference: http://www.example.org/reference/
833 .. _phrase reference: http://www.example.org/phrase_reference/
834 .. _line boundaries: http://www.example.org/line_boundaries/
838 - The plaintext is readable.
839 - Each target may be reused multiple times (e.g., just write
840 ``"reference_"`` again).
841 - No syncronized ordering of references and targets is necessary.
845 - The reference text must be repeated as target names; could lead
847 - The target URLs may be located far from the references, and hard
848 to find in the plaintext.
850 * Anonymous hyperlinks (in current reStructuredText)::
852 This is an anonymous reference__. Here is an anonymous
853 `phrase reference`__. Phrase references may even cross `line
856 __ http://www.example.org/reference/
857 __ http://www.example.org/phrase_reference/
858 __ http://www.example.org/line_boundaries/
862 - The plaintext is readable.
863 - The reference text does not have to be repeated.
867 - References and targets must be kept in sync.
868 - Targets cannot be reused.
869 - The target URLs may be located far from the references.
871 For comparison and historical background, StructuredText also has two
872 syntaxes for hyperlinks:
874 * First, ``"reference text":URL``::
876 This is a "reference":http://www.example.org/reference/
877 of one word ("reference"). Here is a "phrase
878 reference":http://www.example.org/phrase_reference/.
880 * Second, ``"reference text", http://example.com/absolute_URL``::
882 This is a "reference", http://www.example.org/reference/
883 of one word ("reference"). Here is a "phrase reference",
884 http://www.example.org/phrase_reference/.
886 Both syntaxes share advantages and disadvantages:
890 - The target is specified immediately adjacent to the reference.
894 - Poor plaintext readability.
895 - Targets cannot be reused.
896 - Both syntaxes use double quotes, common in ordinary text.
897 - In the first syntax, the URL and the last word are stuck
898 together, exacerbating the line wrap problem.
899 - The second syntax is too magical; text could easily be written
900 that way by accident (although only absolute URLs are recognized
901 here, perhaps because of the potential for ambiguity).
903 A new type of "inline external hyperlink" has been proposed.
905 1. On 2002-06-28, Simon Budig proposed__ a new syntax for
906 reStructuredText hyperlinks::
908 This is a reference_(http://www.example.org/reference/) of one
909 word ("reference"). Here is a `phrase
910 reference`_(http://www.example.org/phrase_reference/). Are
911 these examples, (single-underscore), named? If so, `anonymous
912 references`__(http://www.example.org/anonymous/) using two
913 underscores would probably be preferable.
915 __ http://mail.python.org/pipermail/doc-sig/2002-June/002648.html
917 The syntax, advantages, and disadvantages are similar to those of
922 - The target is specified immediately adjacent to the reference.
926 - Poor plaintext readability.
927 - Targets cannot be reused (unless named, but the semantics are
932 - The ``"`ref`_(URL)"`` syntax forces the last word of the
933 reference text to be joined to the URL, making a potentially
934 very long word that can't be wrapped (URLs can be very long).
935 The reference and the URL should be separate. This is a
936 symptom of the following point:
938 - The syntax produces a single compound construct made up of two
939 equally important parts, *with syntax in the middle*, *between*
940 the reference and the target. This is unprecedented in
943 - The "inline hyperlink" text is *not* a named reference (there's
944 no lookup by name), so it shouldn't look like one.
946 - According to the IETF standards RFC 2396 and RFC 2732,
947 parentheses are legal URI characters and curly braces are legal
948 email characters, making their use prohibitively difficult.
950 - The named/anonymous semantics are unclear.
952 2. After an analysis__ of the syntax of (1) above, we came up with the
953 following compromise syntax::
955 This is an anonymous reference__
956 __<http://www.example.org/reference/> of one word
957 ("reference"). Here is a `phrase reference`__
958 __<http://www.example.org/phrase_reference/>. `Named
959 references`_ _<http://www.example.org/anonymous/> use single
962 __ http://mail.python.org/pipermail/doc-sig/2002-July/002670.html
964 The syntax builds on that of the existing "inline internal
965 targets": ``an _`inline internal target`.``
969 - The target is specified immediately adjacent to the reference,
970 improving maintainability:
972 - References and targets are easily kept in sync.
973 - The reference text does not have to be repeated.
975 - The construct is executed in two parts: references identical to
976 existing references, and targets that are new but not too big a
977 stretch from current syntax.
979 - There's overwhelming precedent for quoting URLs with angle
984 - Poor plaintext readability.
985 - Lots of "line noise".
986 - Targets cannot be reused (unless named; see below).
988 To alleviate the readability issue slightly, we could allow the
989 target to appear later, such as after the end of the sentence::
991 This is a named reference__ of one word ("reference").
992 __<http://www.example.org/reference/> Here is a `phrase
993 reference`__. __<http://www.example.org/phrase_reference/>
995 Problem: this could only work for one reference at a time
996 (reference/target pairs must be proximate [refA trgA refB trgB],
997 not interleaved [refA refB trgA trgB] or nested [refA refB trgB
998 trgA]). This variation is too problematic; references and inline
999 external targets will have to be kept imediately adjacent (see (3)
1002 The ``"reference__ __<target>"`` syntax is actually for "anonymous
1003 inline external targets", emphasized by the double underscores. It
1004 follows that single trailing and leading underscores would lead to
1005 *implicitly named* inline external targets. This would allow the
1006 reuse of targets by name. So after ``"reference_ _<target>"``,
1007 another ``"reference_"`` would point to the same target.
1010 From RFC 2396 (URI syntax):
1012 The angle-bracket "<" and ">" and double-quote (")
1013 characters are excluded [from URIs] because they are often
1014 used as the delimiters around URI in text documents and
1017 Using <> angle brackets around each URI is especially
1018 recommended as a delimiting style for URI that contain
1021 From RFC 822 (email headers):
1023 Angle brackets ("<" and ">") are generally used to indicate
1024 the presence of a one machine-usable reference (e.g.,
1025 delimiting mailboxes), possibly including source-routing to
1028 3. If it is best for references and inline external targets to be
1029 immediately adjacent, then they might as well be integrated.
1030 Here's an alternative syntax embedding the target URL in the
1033 This is an anonymous `reference <http://www.example.org
1034 /reference/>`__ of one word ("reference"). Here is a `phrase
1035 reference <http://www.example.org/phrase_reference/>`__.
1037 Advantages and disadvantages are similar to those in (2).
1038 Readability is still an issue, but the syntax is a bit less
1039 heavyweight (reduced line noise). Backquotes are required, even
1040 for one-word references; the target URL is included within the
1041 reference text, forcing a phrase context.
1043 We'll call this variant "embedded URIs".
1045 Problem: how to refer to a title like "HTML Anchors: <a>" (which
1046 ends with an HTML/SGML/XML tag)? We could either require more
1047 syntax on the target (like ``"`reference text
1048 __<http://example.com/>`__"``), or require the odd conflicting
1049 title to be escaped (like ``"`HTML Anchors: \<a>`__"``). The
1050 latter seems preferable, and not too onerous.
1052 Similarly to (2) above, a single trailing underscore would convert
1053 the reference & inline external target from anonymous to implicitly
1054 named, allowing reuse of targets by name.
1056 I think this is the least objectionable of the syntax alternatives.
1058 Other syntax variations have been proposed (by Brett Cannon and Benja
1061 `phrase reference`->http://www.example.com
1063 `phrase reference`@http://www.example.com
1065 `phrase reference`__ ->http://www.example.com
1067 `phrase reference` [-> http://www.example.com]
1069 `phrase reference`__ [-> http://www.example.com]
1071 `phrase reference` <http://www.example.com>_
1073 None of these variations are clearly superior to #3 above. Some have
1074 problems that exclude their use.
1076 With any kind of inline external target syntax it comes down to the
1077 conflict between maintainability and plaintext readability. I don't
1078 see a major problem with reStructuredText's maintainability, and I
1079 don't want to sacrifice plaintext readability to "improve" it.
1081 The proponents of inline external targets want them for easily
1082 maintainable web pages. The arguments go something like this:
1084 - Named hyperlinks are difficult to maintain because the reference
1085 text is duplicated as the target name.
1087 To which I said, "So use anonymous hyperlinks."
1089 - Anonymous hyperlinks are difficult to maintain becuase the
1090 references and targets have to be kept in sync.
1092 "So keep the targets close to the references, grouped after each
1093 paragraph. Maintenance is trivial."
1095 - But targets grouped after paragraphs break the flow of text.
1097 "Surely less than URLs embedded in the text! And if the intent is
1098 to produce web pages, not readable plaintext, then who cares about
1101 Many participants have voiced their objections to the proposed syntax:
1103 Garth Kidd: "I strongly prefer the current way of doing it.
1104 Inline is spectactularly messy, IMHO."
1106 Tony Ibbs: "I vehemently agree... that the inline alternatives
1107 being suggested look messy - there are/were good reasons they've
1108 been taken out... I don't believe I would gain from the new
1111 Paul Moore: "I agree as well. The proposed syntax is far too
1112 punctuation-heavy, and any of the alternatives discussed are
1113 ambiguous or too subtle."
1115 Others have voiced their support:
1117 fantasai: "I agree with Simon. In many cases, though certainly
1118 not in all, I find parenthesizing the url in plain text flows
1119 better than relegating it to a footnote."
1121 Ken Manheimer: "I'd like to weigh in requesting some kind of easy,
1122 direct inline reference link."
1124 (Interesting that those *against* the proposal have been using
1125 reStructuredText for a while, and those *for* the proposal are either
1126 new to the list ["fantasai", background unknown] or longtime
1127 StructuredText users [Ken Manheimer].)
1129 I was initially ambivalent/against the proposed "inline external
1130 targets". I value reStructuredText's readability very highly, and
1131 although the proposed syntax offers convenience, I don't know if the
1132 convenience is worth the cost in ugliness. Does the proposed syntax
1133 compromise readability too much, or should the choice be left up to
1134 the author? Perhaps if the syntax is *allowed* but its use strongly
1135 *discouraged*, for aesthetic/readability reasons?
1137 After a great deal of thought and much input from users, I've decided
1138 that there are reasonable use cases for this construct. The
1139 documentation should strongly caution against its use in most
1140 situations, recommending independent block-level targets instead.
1141 Syntax #3 above ("embedded URIs") will be used.
1144 Doctree Representation of Transitions
1145 =====================================
1147 (Although not reStructuredText-specific, this section fits best in
1150 Having added the "horizontal rule" construct to the `reStructuredText
1151 Markup Specification`_, a decision had to be made as to how to reflect
1152 the construct in the implementation of the document tree. Given this
1164 The horizontal rule indicates a "transition" (in prose terms) or the
1165 start of a new "division". Before implementation, the parsed document
1169 <section names="document">
1174 -------- <--- error here
1178 There are several possibilities for the implementation:
1180 1. Implement horizontal rules as "divisions" or segments. A
1181 "division" is a title-less, non-hierarchical section. The first
1182 try at an implementation looked like this::
1185 <section names="document">
1194 But the two paragraphs are really at the same level; they shouldn't
1195 appear to be at different levels. There's really an invisible
1196 "first division". The horizontal rule splits the document body
1197 into two segments, which should be treated uniformly.
1199 2. Treating "divisions" uniformly brings us to the second
1203 <section names="document">
1213 With this change, documents and sections will directly contain
1214 divisions and sections, but not body elements. Only divisions will
1215 directly contain body elements. Even without a horizontal rule
1216 anywhere, the body elements of a document or section would be
1217 contained within a division element. This makes the document tree
1218 deeper. This is similar to the way HTML_ treats document contents:
1219 grouped within a ``<body>`` element.
1221 3. Implement them as "transitions", empty elements::
1224 <section names="document">
1233 A transition would be a "point element", not containing anything,
1234 only identifying a point within the document structure. This keeps
1235 the document tree flatter, but the idea of a "point element" like
1236 "transition" smells bad. A transition isn't a thing itself, it's
1237 the space between two divisions. However, transitions are a
1240 Solution 3 was chosen for incorporation into the document tree model.
1242 .. _HTML: http://www.w3.org/MarkUp/
1245 Syntax for Line Blocks
1246 ======================
1248 * An early idea: How about a literal-block-like prefix, perhaps
1249 "``;;``"? (It is, after all, a *semi-literal* literal block, no?)
1252 Take it away, Eric the Orchestra Leader! ;;
1254 A one, two, a one two three four
1256 Half a bee, philosophically,
1257 must, *ipso facto*, half not be.
1258 But half the bee has got to be,
1259 *vis a vis* its entity. D'you see?
1261 But can a bee be said to be
1262 or not to be an entire bee,
1263 when half the bee is not a bee,
1264 due to some ancient injury?
1270 * Another idea: in an ordinary paragraph, if the first line ends with
1271 a backslash (escaping the newline), interpret the entire paragraph
1272 as a verse block? For example::
1274 Add just one backslash\
1275 And this paragraph becomes
1278 (Awful, and arguably invalid, since in Japanese the word "haiku"
1279 contains three syllables not two.)
1281 This idea was superceded by the rules for escaped whitespace, useful
1282 for `character-level inline markup`_.
1284 * In a `2004-02-22 docutils-develop message`__, Jarno Elonen proposed
1285 a "plain list" syntax (and also provided a patch)::
1288 | President, SuperDuper Corp.
1291 __ http://thread.gmane.org/gmane.text.docutils.devel/1187
1293 This syntax is very natural. However, these "plain lists" seem very
1294 similar to line blocks, and I see so little intrinsic "list-ness"
1295 that I'm loathe to add a new object. I used the term "blurbs" to
1296 remove the "list" connotation from the originally proposed name.
1297 Perhaps line blocks could be refined to add the two properties they
1300 A) long lines wrap nicely
1301 B) HTML output doesn't look like program code in non-CSS web
1304 (A) is an issue of all 3 aspects of Docutils: syntax (construct
1305 behaviour), internal representation, and output. (B) is partly an
1306 issue of internal representation but mostly of output.
1308 ReStructuredText will redefine line blocks with the "|"-quoting
1309 syntax. The following is my current thinking.
1315 Perhaps line block syntax like this would do::
1319 | IMF: not decided yet, but probably one of the following:
1325 Note that the "nested" list does not have nested syntax (the "|" are
1326 not further indented); the leading whitespace would still be
1327 significant somehow (more below). As for long lines in the input,
1328 this could suffice::
1331 | Founder, President, Chief Executive Officer, Cook, Bottle
1332 Washer, and All-Round Great Guy
1336 The lack of "|" on the third line indicates that it's a continuation
1337 of the second line, wrapped.
1339 I don't see much point in allowing arbitrary nested content. Multiple
1340 paragraphs or bullet lists inside a "blurb" doesn't make sense to me.
1341 Simple nested line blocks should suffice.
1344 Internal Representation
1345 -----------------------
1347 Line blocks are currently represented as text blobs as follows::
1349 <!ELEMENT line_block %text.model;>
1350 <!ATTLIST line_block
1354 Instead, we could represent each line by a separate element::
1356 <!ELEMENT line_block (line+)>
1357 <!ATTLIST line_block %basic.atts;>
1359 <!ELEMENT line %text.model;>
1360 <!ATTLIST line %basic.atts;>
1362 We'd keep the significance of the leading whitespace of each line
1363 either by converting it to non-breaking spaces at output, or with a
1364 per-line margin. Non-breaking spaces are simpler (for HTML, anyway)
1365 but kludgey, and wouldn't support indented long lines that wrap. But
1366 should inter-word whitespace (i.e., not leading whitespace) be
1367 preserved? Currently it is preserved in line blocks.
1369 Representing a more complex line block may be tricky::
1371 | But can a bee be said to be
1372 | or not to be an entire bee,
1373 | when half the bee is not a bee,
1374 | due to some ancient injury?
1376 Perhaps the representation could allow for nested line blocks::
1378 <!ELEMENT line_block (line | line_block)+>
1380 With this model, leading whitespace would no longer be significant.
1381 Instead, left margins are implied by the nesting. The example above
1382 could be represented as follows::
1386 But can a bee be said to be
1389 or not to be an entire bee,
1392 when half the bee is not a bee,
1395 due to some ancient injury?
1397 I wasn't sure what to do about even more complex line blocks::
1405 How should that be parsed and nested? Should the first line have
1406 the same nesting level (== indentation in the output) as the fourth
1407 line, or the same as the last line? Mark Nodine suggested that such
1408 line blocks be parsed similarly to complexly-nested block quotes,
1409 which seems reasonable. In the example above, this would result in
1410 the nesting of first line matching the last line's nesting. In
1411 other words, the nesting would be relative to neighboring lines
1418 In HTML, line blocks are currently output as "<pre>" blocks, which
1419 gives us significant whitespace and line breaks, but doesn't allow
1420 long lines to wrap and causes monospaced output without stylesheets.
1421 Instead, we could output "<div>" elements parallelling the
1422 representation above, where each nested <div class="line_block"> would
1423 have an increased left margin (specified in the stylesheet).
1425 Jarno suggested the following HTML output::
1427 <div class="line_block">
1428 <span class="line">First, top level line</span><br class="hidden"/>
1429 <div class="line_block"><span class="hidden"> </span>
1430 <span class="line">Second, once nested</span><br class="hidden"/>
1431 <span class="line">Third, once nested</span><br class="hidden"/>
1437 The ``<br class="hidden" />`` and ``<span
1438 class="hidden"> </span>`` are meant to support non-CSS and
1439 non-graphical browsers. I understand the case for "br", but I'm not
1440 so sure about hidden " ". I question how much effort should be
1441 put toward supporting non-graphical and especially non-CSS browsers,
1442 at least for html4css1.py output.
1444 Should the lines themselves be ``<span>`` or ``<div>``? I don't like
1445 mixing inline and block-level elements.
1451 We'll leave the old implementation in place (via the "line-block"
1452 directive only) until all Writers have been updated to support the new
1453 syntax & implementation. The "line-block" directive can then be
1454 updated to use the new internal representation, and its documentation
1455 will be updated to recommend the new syntax.
1461 The original idea came from Dylan Jay:
1463 ... to use a two level bulleted list with something to
1464 indicate it should be rendered as a table ...
1466 It's an interesting idea. It could be implemented in as a directive
1467 which transforms a uniform two-level list into a table. Using a
1468 directive would allow the author to explicitly set the table's
1469 orientation (by column or by row), the presence of row headers, etc.
1473 1. (Implemented in Docutils 0.3.8).
1475 Bullet-list-tables might look like this::
1487 - If we took the bones out, it wouldn't be crunchy,
1493 This list must be written in two levels. This wouldn't work::
1510 * If we took the bones out...
1512 The above is a single list of 12 items. The blank lines are not
1513 significant to the markup. We'd have to explicitly specify how
1514 many columns or rows to use, which isn't a good idea.
1516 2. Beni Cherniavsky suggested a field list alternative. It could look
1519 .. field-list-table::
1526 - :treat: Albatross!
1530 - :treat: Crunchy Frog!
1532 :descr: If we took the bones out, it wouldn't be
1533 crunchy, now would it?
1535 Column order is determined from the order of fields in the first
1536 row. Field order in all other rows is ignored. As a side-effect,
1537 this allows trivial re-arrangement of columns. By using named
1538 fields, it becomes possible to omit fields in some rows without
1539 losing track of things, which is important for spans.
1541 3. An alternative to two-level bullet lists would be to use enumerated
1542 lists for the table cells::
1554 3. If we took the bones out, it wouldn't be crunchy,
1557 That provides better correspondence between cells in the same
1558 column than does bullet-list syntax, but not as good as field list
1559 syntax. I think that were only field-list-tables available, a lot
1560 of users would use the equivalent degenerate case::
1562 .. field-list-table::
1568 4. Another natural variant is to allow a description list with field
1569 lists as descriptions::
1582 :descr: If we took the bones out, it wouldn't be
1583 crunchy, now would it?
1585 This would make the whole first column a header column ("stub").
1586 It's limited to a single column and a single paragraph fitting on
1587 one source line. Also it wouldn't allow for empty cells or row
1588 spans in the first column. But these are limitations that we could
1589 live with, like those of simple tables.
1591 The List-driven table feature could be done in many ways. Each user
1592 will have their preferred usage. Perhaps a single "list-table"
1593 directive could handle them all, depending on which options and
1594 content are present.
1598 * How to indicate that there's 1 header row? Perhaps two lists? ::
1610 This is probably too subtle though. Better would be a directive
1611 option, like ``:headrows: 1``. An early suggestion for the header
1612 row(s) was to use a directive option::
1614 .. field-list-table::
1619 - :treat: Albatross!
1623 But the table data is at two levels and looks inconsistent.
1625 In general, we cannot extract the header row from field lists' field
1626 names because field names cannot contain everything one might put in
1627 a table cell. A separate header row also allows shorter field names
1628 and doesn't force one to rewrite the whole table when the header
1629 text changes. But for simpler cases, we can offer a ":header:
1630 fields" option, which does extract header cells from field names::
1632 .. field-list-table::
1635 - :Treat: Albatross!
1637 :Description: On a stick!
1639 * How to indicate the column widths? A directive option? ::
1644 Automatic defaults from the text used?
1646 * How to handle row and/or column spans?
1648 In a field list, column-spans can be indicated by specifying the
1649 first and last fields, separated by space-dash-space or ellipsis::
1652 - :foo ... baz: quuux
1654 Commas were proposed for column spans::
1658 But non-adjacent columns become problematic. Should we report an
1659 error, or duplicate the value into each span of adjacent columns (as
1660 was suggested)? The latter suggestion is appealing but may be too
1661 clever. Best perhaps to simply specify the two ends.
1663 It was suggested that comma syntax should be allowed, too, in order
1664 to allow the user to avoid trouble when changing the column order.
1665 But changing the column order of a table with spans is not trivial;
1666 we shouldn't make it easier to mess up.
1668 One possible syntax for row-spans is to simply treat any row where a
1669 field is missing as a row-span from the last row where it appeared.
1670 Leaving a field empty would still be possible by writing a field
1671 with empty content. But this is too implicit.
1673 Another way would be to require an explicit continuation marker
1674 (``...``/``-"-``/``"``?) in all but the first row of a spanned
1675 field. Empty comments could work (".."). If implemented, the same
1676 marker could also be supported in simple tables, which lack
1677 row-spanning abilities.
1679 Explicit markup like ":rowspan:" and ":colspan:" was also suggested.
1681 Sometimes in a table, the first header row contains spans. It may
1682 be necessary to provide a way to specify the column field names
1683 independently of data rows. A directive option would do it.
1685 * We could specify "column-wise" or "row-wise" ordering, with the same
1686 markup structure. For example, with definition data::
1699 - If we took the bones out, it wouldn't be
1700 crunchy, now would it?
1702 * A syntax for _`stubs in grid tables` is easy to imagine::
1704 +------------------------++------------+----------+
1705 | Header row, column 1 || Header 2 | Header 3 |
1706 +========================++============+==========+
1707 | body row 1, column 1 || column 2 | column 3 |
1708 +------------------------++------------+----------+
1710 Or this idea from Nick Moffitt::
1721 Auto-Enumerated Lists
1722 =====================
1724 Implemented 2005-03-24: combination of variation 1 & 2.
1726 The advantage of auto-numbered enumerated lists would be similar to
1727 that of auto-numbered footnotes: lists could be written and rearranged
1728 without having to manually renumber them. The disadvantages are also
1729 the same: input and output wouldn't match exactly; the markup may be
1730 ugly or confusing (depending on which alternative is chosen).
1732 1. Use the "#" symbol. Example::
1738 Advantages: simple, explicit. Disadvantage: enumeration sequence
1739 cannot be specified (limited to arabic numerals); ugly.
1741 2. As a variation on #1, first initialize the enumeration sequence?
1748 Advantages: simple, explicit, any enumeration sequence possible.
1749 Disadvantages: ugly; perhaps confusing with mixed concrete/abstract
1752 3. Alternative suggested by Fred Bremmer, from experience with MoinMoin::
1758 Advantages: enumeration sequence is explicit (could be multiple
1759 "a." or "(I)" tokens). Disadvantages: perhaps confusing; otherwise
1760 erroneous input (e.g., a duplicate item "1.") would pass silently,
1761 either causing a problem later in the list (if no blank lines
1762 between items) or creating two lists (with blanks).
1764 Take this input for example::
1768 1. Unintentional duplicate of item 1.
1772 Currently the parser will produce two list, "1" and "1,2" (no
1773 warnings, because of the presence of blank lines). Using Fred's
1774 notation, the current behavior is "1,1,2 -> 1 1,2" (without blank
1775 lines between items, it would be "1,1,2 -> 1 [WARNING] 1,2"). What
1776 should the behavior be with auto-numbering?
1778 Fred has produced a patch__, whose initial behavior is as follows::
1783 1,2,2,3 -> 1,2,3 [WARNING] 3
1784 1,1,2 -> 1,2 [WARNING] 2
1786 (After the "[WARNING]", the "3" would begin a new list.)
1788 I have mixed feelings about adding this functionality to the spec &
1789 parser. It would certainly be useful to some users (myself
1790 included; I often have to renumber lists). Perhaps it's too
1791 clever, asking the parser to guess too much. What if you *do* want
1792 three one-item lists in a row, each beginning with "1."? You'd
1793 have to use empty comments to force breaks. Also, I question
1794 whether "1,2,2 -> 1,2,3" is optimal behavior.
1796 In response, Fred came up with "a stricter and more explicit rule
1797 [which] would be to only auto-number silently if *all* the
1798 enumerators of a list were identical". In that case::
1801 1,2,2 -> 1,2 [WARNING] 2
1803 1,2,2,3 -> 1,2 [WARNING] 2,3
1804 1,1,2 -> 1,2 [WARNING] 2
1806 Should any start-value be allowed ("3,3,3"), or should
1807 auto-numbered lists be limited to begin with ordinal-1 ("1", "A",
1810 __ http://sourceforge.net/tracker/index.php?func=detail&aid=548802
1811 &group_id=38414&atid=422032
1813 4. Alternative proposed by Tony Ibbs::
1816 #3. Aha - I edited this in later.
1819 The initial proposal required unique enumerators within a list, but
1820 this limits the convenience of a feature of already limited
1821 applicability and convenience. Not a useful requirement; dropped.
1823 Instead, simply prepend a "#" to a standard list enumerator to
1824 indicate auto-enumeration. The numbers (or letters) of the
1825 enumerators themselves are not significant, except:
1827 - as a sequence indicator (arabic, roman, alphabetic; upper/lower),
1829 - and perhaps as a start value (first list item).
1831 Advantages: explicit, any enumeration sequence possible.
1832 Disadvantages: a bit ugly.
1835 Adjacent citation references
1836 ============================
1838 A special case for inline markup was proposed and implemented:
1839 multiple citation references could be joined into one::
1841 [cite1]_[cite2]_ instead of requiring [cite1]_ [cite2]_
1843 However, this was rejected as an unwarranted exception to the rules
1845 (The main motivation for the proposal, grouping citations in the latex writer,
1846 was implemented by recognising the second group in the example above and
1847 transforming it into ``\cite{cite1,cite2}``.)
1853 Using a standard, such as MathML_\ [#]_ seems a good start. However,
1855 MathML in its full XML form was never thought to be usable as an
1856 input format: for a start it is far too verbose as necessitated by
1857 its intention to be expressive.
1859 -- http://www.w3.org/Math/Roadmap/
1861 Like for non-mathematical content, a different input format is needed.
1863 .. [#] For an overview of MathML implementations and tests, see e.g.
1864 the `mathweb wiki`_ or the `ConTeXT MathML page`_.
1866 .. _MathML: http://www.w3.org/TR/MathML2/
1867 .. _mathweb wiki: http://www.mathweb.org/wiki/MathML
1868 .. _ConTeXT MathML page: http://wiki.contextgarden.net/MathML
1874 * Intended for input by a human, widely used, and well documented.
1876 * Can be used for both, input and internal storage.
1878 * Convertible to all supported output formats (building on
1879 existing extensions like `latex_math`_ (LaTeX2MathML) or the math
1882 * Used by the `Math support in Sphinx`_.
1884 * Unicode input similar to the "unicode-math" package for
1885 XeTeX/LuaTeX provides for improved readability of the source.
1887 * The backtick (GRAVE ACCENT, 0x60) cannot be used in a math role.
1888 Fortunately, it is not required in LaTeX math mode, as the
1889 ``\grave`` macro is used instead of ``\```.
1891 .. _latex_math: ../../../sandbox/jensj/latex_math/
1892 .. _Math support in Sphinx: http://sphinx.pocoo.org/ext/math.html
1896 Simple, ASCII based math input language (see also `ASCIIMath tutorial`_).
1898 * The Python module ASCIIMathML_ translates a string with ASCIIMath into a
1899 MathML tree. Used, e.g., by MultiMarkdown__.
1901 * For conversion to LaTeX, there is a JavaScript script at
1902 http://dlippman.imathas.com/asciimathtex/ASCIIMath2TeX.js
1904 .. _ASCIIMath: http://www1.chapman.edu/~jipsen/mathml/asciimath.html
1905 .. _ASCIIMath tutorial:
1906 http://www.wjagray.co.uk/maths/ASCIIMathTutorial.html
1907 .. _ASCIIMathML: http://pypi.python.org/pypi/asciimathml/
1908 __ http://fletcherpenney.net/multimarkdown/
1910 `Unicode Nearly Plain Text Encoding of Mathematics`_
1911 format for lightly marked-up representation of mathematical
1912 expressions in Unicode.
1914 (Unicode Technical Note. Sole responsibility for its contents rests
1915 with the author(s). Publication does not imply any endorsement by
1916 the Unicode Consortium.)
1918 .. _Unicode Nearly Plain Text Encoding of Mathematics:
1919 http://www.unicode.org/notes/tn28/
1922 See `the culmination of a relevant discussion in 2003
1923 <http://article.gmane.org/gmane.text.docutils.user/118>`__.
1926 Since Docutils 0.8, a "math" role and directive using LaTeX math
1927 syntax as input format is part of reStructuredText.
1937 As a further wrinkle (see `Reworking Explicit Markup (Round 1)`_
1938 above), in the wee hours of 2002-02-28 I posted several ideas for
1939 changes to footnote syntax:
1941 - Change footnote syntax from ``.. [1]`` to ``_[1]``? ...
1942 - Differentiate (with new DTD elements) author-date "citations"
1943 (``[GVR2002]``) from numbered footnotes? ...
1944 - Render footnote references as superscripts without "[]"? ...
1946 These ideas are all related, and suggest changes in the
1947 reStructuredText syntax as well as the docutils tree model.
1949 The footnote has been used for both true footnotes (asides expanding
1950 on points or defining terms) and for citations (references to external
1951 works). Rather than dealing with one amalgam construct, we could
1952 separate the current footnote concept into strict footnotes and
1953 citations. Citations could be interpreted and treated differently
1954 from footnotes. Footnotes would be limited to numerical labels:
1955 manual ("1") and auto-numbered (anonymous "#", named "#label").
1957 The footnote is the only explicit markup construct (starts with ".. ")
1958 that directly translates to a visible body element. I've always been
1959 a little bit uncomfortable with the ".. " marker for footnotes because
1960 of this; ".. " has a connotation of "special", but footnotes aren't
1961 especially "special". Printed texts often put footnotes at the bottom
1962 of the page where the reference occurs (thus "foot note"). Some HTML
1963 designs would leave footnotes to be rendered the same positions where
1964 they're defined. Other online and printed designs will gather
1965 footnotes into a section near the end of the document, converting them
1966 to "endnotes" (perhaps using a directive in our case); but this
1967 "special processing" is not an intrinsic property of the footnote
1968 itself, but a decision made by the document author or processing
1971 Citations are almost invariably collected in a section at the end of a
1972 document or section. Citations "disappear" from where they are
1973 defined and are magically reinserted at some well-defined point.
1974 There's more of a connection to the "special" connotation of the ".. "
1975 syntax. The point at which the list of citations is inserted could be
1976 defined manually by a directive (e.g., ".. citations::"), and/or have
1977 default behavior (e.g., a section automatically inserted at the end of
1978 the document) that might be influenced by options to the Writer.
1987 .. [#] Auto-numbered footnote.
1988 .. [#label] Auto-labeled footnote.
1990 - The syntax proposed in the original 2002-02-28 Doc-SIG post:
1991 remove the ".. ", prefix a "_"::
1994 _[#] Auto-numbered footnote.
1995 _[#label] Auto-labeled footnote.
1997 The leading underscore syntax (earlier dropped because
1998 ``.. _[1]:`` was too verbose) is a useful reminder that footnotes
1999 are hyperlink targets.
2001 - Minimal syntax: remove the ".. [" and "]", prefix a "_", and
2005 _#. Auto-numbered footnote.
2006 _#label. Auto-labeled footnote.
2008 ``_1.``, ``_#.``, and ``_#label.`` are markers,
2011 Footnotes could be rendered something like this in HTML
2013 | 1. This is a footnote. The brackets could be dropped
2014 | from the label, and a vertical bar could set them
2015 | off from the rest of the document in the HTML.
2017 Two-way hyperlinks on the footnote marker ("1." above) would also
2018 help to differentiate footnotes from enumerated lists.
2020 If converted to endnotes (by a directive/transform), a horizontal
2021 half-line might be used instead. Page-oriented output formats
2022 would typically use the horizontal line for true footnotes.
2024 + Footnote references:
2028 [1]_, [#]_, [#label]_
2030 - Minimal syntax to match the minimal footnote syntax above::
2034 As a consequence, pure-numeric hyperlink references would not be
2035 possible; they'd be interpreted as footnote references.
2037 + Citation references: no change is proposed from the current footnote
2044 - Current syntax (footnote syntax)::
2046 .. [GVR2001] Python Documentation; van Rossum, Drake, et al.;
2047 http://www.python.org/doc/
2049 - Possible new syntax::
2051 _[GVR2001] Python Documentation; van Rossum, Drake, et al.;
2052 http://www.python.org/doc/
2055 Docutils: Python Documentation Utilities project; Goodger
2056 et al.; http://docutils.sourceforge.net/
2058 Without the ".. " marker, subsequent lines would either have to
2059 align as in one of the above, or we'd have to allow loose
2060 alignment (I'd rather not)::
2062 _[GVR2001] Python Documentation; van Rossum, Drake, et al.;
2063 http://www.python.org/doc/
2065 I proposed adopting the "minimal" syntax for footnotes and footnote
2066 references, and adding citations and citation references to
2067 reStructuredText's repertoire. The current footnote syntax for
2068 citations is better than the alternatives given.
2070 From a reply by Tony Ibbs on 2002-03-01:
2072 However, I think easier with examples, so let's create one::
2074 Fans of Terry Pratchett are perhaps more likely to use
2075 footnotes [1]_ in their own writings than other people
2076 [2]_. Of course, in *general*, one only sees footnotes
2077 in academic or technical writing - it's use in fiction
2078 and letter writing is not normally considered good
2079 style [4]_, particularly in emails (not a medium that
2080 lends itself to footnotes).
2082 .. [1] That is, little bits of referenced text at the
2084 .. [2] Because Terry himself does, of course [3]_.
2085 .. [3] Although he has the distinction of being
2086 *funny* when he does it, and his fans don't always
2088 .. [4] Presumably because it detracts from linear
2089 reading of the text - this is, of course, the point.
2091 and look at it with the second syntax proposal::
2093 Fans of Terry Pratchett are perhaps more likely to use
2094 footnotes [1]_ in their own writings than other people
2095 [2]_. Of course, in *general*, one only sees footnotes
2096 in academic or technical writing - it's use in fiction
2097 and letter writing is not normally considered good
2098 style [4]_, particularly in emails (not a medium that
2099 lends itself to footnotes).
2101 _[1] That is, little bits of referenced text at the
2103 _[2] Because Terry himself does, of course [3]_.
2104 _[3] Although he has the distinction of being
2105 *funny* when he does it, and his fans don't always
2107 _[4] Presumably because it detracts from linear
2108 reading of the text - this is, of course, the point.
2110 (I note here that if I have gotten the indentation of the
2111 footnotes themselves correct, this is clearly not as nice. And if
2112 the indentation should be to the left margin instead, I like that
2115 and the third (new) proposal::
2117 Fans of Terry Pratchett are perhaps more likely to use
2118 footnotes 1_ in their own writings than other people
2119 2_. Of course, in *general*, one only sees footnotes
2120 in academic or technical writing - it's use in fiction
2121 and letter writing is not normally considered good
2122 style 4_, particularly in emails (not a medium that
2123 lends itself to footnotes).
2125 _1. That is, little bits of referenced text at the
2127 _2. Because Terry himself does, of course 3_.
2128 _3. Although he has the distinction of being
2129 *funny* when he does it, and his fans don't always
2131 _4. Presumably because it detracts from linear
2132 reading of the text - this is, of course, the point.
2134 I think I don't, in practice, mind the targets too much (the use
2135 of a dot after the number helps a lot here), but I do have a
2136 problem with the body text, in that I don't naturally separate out
2137 the footnotes as different than the rest of the text - instead I
2138 keep wondering why there are numbers interspered in the text. The
2139 use of brackets around the numbers ([ and ]) made me somehow parse
2140 the footnote references as "odd" - i.e., not part of the body text
2141 - and thus both easier to skip, and also (paradoxically) easier to
2142 pick out so that I could follow them.
2144 Thus, for the moment (and as always susceptable to argument), I'd
2145 say -1 on the new form of footnote reference (i.e., I much prefer
2146 the existing ``[1]_`` over the proposed ``1_``), and ambivalent
2147 over the proposed target change.
2149 That leaves David's problem of wanting to distinguish footnotes
2150 and citations - and the only thing I can propose there is that
2151 footnotes are numeric or # and citations are not (which, as a
2152 human being, I can probably cope with!).
2154 From a reply by Paul Moore on 2002-03-01:
2156 I think the current footnote syntax ``[1]_`` is *exactly* the
2157 right balance of distinctness vs unobtrusiveness. I very
2158 definitely don't think this should change.
2160 On the target change, it doesn't matter much to me.
2162 From a further reply by Tony Ibbs on 2002-03-01, referring to the
2163 "[1]" form and actual usage in email:
2165 Clearly this is a form people are used to, and thus we should
2166 consider it strongly (in the same way that the usage of ``*..*``
2167 to mean emphasis was taken partly from email practise).
2169 Equally clearly, there is something "magical" for people in the
2170 use of a similar form (i.e., ``[1]``) for both footnote reference
2171 and footnote target - it seems natural to keep them similar.
2175 I think that this established plaintext usage leads me to strongly
2176 believe we should retain square brackets at both ends of a
2177 footnote. The markup of the reference end (a single trailing
2178 underscore) seems about as minimal as we can get away with. The
2179 markup of the target end depends on how one envisages the thing -
2180 if ".." means "I am a target" (as I tend to see it), then that's
2181 good, but one can also argue that the "_[1]" syntax has a neat
2182 symmetry with the footnote reference itself, if one wishes (in
2183 which case ".." presumably means "hidden/special" as David seems
2184 to think, which is why one needs a ".." *and* a leading underline
2185 for hyperlink targets.
2187 Given the persuading arguments voiced, we'll leave footnote & footnote
2188 reference syntax alone. Except that these discussions gave rise to
2189 the "auto-symbol footnote" concept, which has been added. Citations
2190 and citation references have also been added.
2193 Syntax for Questions & Answers
2194 ==============================
2196 Implement as a generic two-column marked list? As a standalone
2197 (non-directive) construct? (Is the markup ambiguous?) Add support to
2200 New elements would be required. Perhaps::
2202 <!ELEMENT question_list (question_list_item+)>
2203 <!ATTLIST question_list
2204 numbering (none | local | global)
2206 start NUMBER #IMPLIED>
2207 <!ELEMENT question_list_item (question, answer*)>
2208 <!ELEMENT question %text.model;>
2209 <!ELEMENT answer (%body.elements;)+>
2211 Originally I thought of implementing a Q&A list with special syntax::
2215 A: You are a question-and-answer
2220 A: I am the omniscient "we".
2222 Where each "Q" and "A" could also be numbered (e.g., "Q1"). However,
2223 a simple enumerated or bulleted list will do just fine for syntax. A
2224 directive could treat the list specially; e.g. the first paragraph
2225 could be treated as a question, the remainder as the answer (multiple
2226 answers could be represented by nested lists). Without special
2227 syntax, this directive becomes low priority.
2229 As described in the FAQ__, no special syntax or directive is needed
2230 for this application.
2232 __ http://docutils.sf.net/FAQ.html
2233 #how-can-i-mark-up-a-faq-or-other-list-of-questions-answers
2240 Reworking Explicit Markup (Round 2)
2241 ===================================
2243 See `Reworking Explicit Markup (Round 1)`_ for an earlier discussion.
2245 In April 2004, a new thread becan on docutils-develop: `Inconsistency
2246 in RST markup`__. Several arguments were made; the first argument
2247 begat later arguments. Below, the arguments are paraphrased "in
2248 quotes", with responses.
2250 __ http://thread.gmane.org/gmane.text.docutils.devel/1386
2252 1. References and targets take this form::
2256 .. _targetname: stuff
2258 But footnotes, "which generate links just like targets do", are
2265 "Footnotes should be written as"::
2271 But they're not the same type of animal. That's not a "footnote
2272 target", it's a *footnote*. Being a target is not a footnote's
2273 primary purpose (an arguable point). It just happens to grow a
2274 target automatically, for convenience. Just as a section title::
2279 isn't a "title target", it's a *title*, which happens to grow a
2280 target automatically. The consistency is there, it's just deeper
2281 than at first glance.
2283 Also, ".. [1]" was chosen for footnote syntax because it closely
2284 resembles one form of actual footnote rendering. ".. _[1]:" is too
2285 verbose; excessive punctuation is required to get the job done.
2287 For more of the reasoning behind the syntax, see `Problems With
2288 StructuredText (Hyperlinks) <problems.html#hyperlinks>`__ and
2289 `Reworking Footnotes`_.
2291 2. "I expect directives to also look like ``.. this:`` [one colon]
2292 because that also closely parallels the link and footnote target
2295 There are good reasons for the two-colon syntax:
2297 Two colons are used after the directive type for these reasons:
2299 - Two colons are distinctive, and unlikely to be used in common
2302 - Two colons avoids clashes with common comment text like::
2304 .. Danger: modify at your own risk!
2306 - If an implementation of reStructuredText does not recognize a
2307 directive (i.e., the directive-handler is not installed), a
2308 level-3 (error) system message is generated, and the entire
2309 directive block (including the directive itself) will be
2310 included as a literal block. Thus "::" is a natural choice.
2312 -- `restructuredtext.html#directives
2313 <../../ref/rst/restructuredtext.html#directives>`__
2315 The last reason is not particularly compelling; it's more of a
2316 convenient coincidence or mnemonic.
2318 3. "Comments always seemed too easy. I almost never write comments.
2319 I'd have no problem writing '.. comment:' in front of my comments.
2320 In fact, it would probably be more readable, as comments *should*
2321 be set off strongly, because they are very different from normal
2324 Many people do use comments though, and some applications of
2325 reStructuredText require it. For example, all reStructuredText
2326 PEPs (and this document!) have an Emacs stanza at the bottom, in a
2327 comment. Having to write ".. comment::" would be very obtrusive.
2329 Comments *should* be dirt-easy to do. It should be easy to
2330 "comment out" a block of text. Comments in programming languages
2331 and other markup languages are invariably easy.
2333 Any author is welcome to preface their comments with "Comment:" or
2334 "Do Not Print" or "Note to Editor" or anything they like. A
2335 "comment" directive could easily be implemented. It might be
2336 confused with admonition directives, like "note" and "caution"
2337 though. In unrelated (and unpublished and unfinished) work, adding
2338 a "comment" directive as a true document element was considered::
2340 If structure is necessary, we could use a "comment" directive
2341 (to avoid nonsensical DTD changes, the "comment" directive
2342 could produce an untitled topic element).
2344 4. "One of the goals of reStructuredText is to be *readable* by people
2345 who don't know it. This construction violates that: it is not at
2346 all obvious to the uninitiated that text marked by '..' is a
2347 comment. On the other hand, '.. comment:' would be totally
2350 Totally transparent, perhaps, but also very obtrusive. Another of
2351 `reStructuredText's goals`_ is to be unobtrusive, and
2352 ".. comment::" would violate that. The goals of reStructuredText
2353 are many, and they conflict. Determining the right set of goals
2354 and finding solutions that best fit is done on a case-by-case
2357 Even readability is has two aspects. Being readable without any
2358 prior knowledge is one. Being as easily read in raw form as in
2359 processed form is the other. ".." may not contribute to the former
2360 aspect, but ".. comment::" would certainly detract from the latter.
2363 .. _reStructuredText's goals: ../../ref/rst/introduction.html#goals
2365 5. "Recently I sent someone an rst document, and they got confused; I
2366 had to explain to them that '..' marks comments, *unless* it's a
2369 The explanation of directives *is* roundabout, defining comments in
2370 terms of not being other things. That's definitely a wart.
2372 6. "Under the current system, a mistyped directive (with ':' instead
2373 of '::') will be silently ignored. This is an error that could
2374 easily go unnoticed."
2376 A parser option/setting like "--comments-on-stderr" would help.
2378 7. "I'd prefer to see double-dot-space / command / double-colon as the
2379 standard Docutils markup-marker. It's unusual enough to avoid
2380 being accidently used. Everything that starts with a double-dot
2381 should end with a double-colon."
2383 That would increase the punctuation verbosity of some constructs
2386 8. Edward Loper proposed the following plan for backwards
2389 1. ".. foo" will generate a deprecation warning to stderr, and
2390 nothing in the output (no system messages).
2391 2. ".. foo: bar" will be treated as a directive foo. If there
2392 is no foo directive, then do the normal error output.
2393 3. ".. foo:: bar" will generate a deprecation warning to
2394 stderr, and be treated as a directive. Or leave it valid?
2396 So some existing documents might start printing deprecation
2397 warnings, but the only existing documents that would *break*
2398 would be ones that say something like::
2400 .. warning: this should be a comment
2404 .. warning:: this should be a comment
2406 Here, we're trading fairly common a silent error (directive
2407 falsely treated as a comment) for a fairly uncommon explicitly
2408 flagged error (comment falsely treated as directive). To make
2409 things even easier, we could add a sentence to the
2410 unknown-directive error. Something like "If you intended to
2411 create a comment, please use '.. comment:' instead".
2413 On one hand, I understand and sympathize with the points raised. On
2414 the other hand, I think the current syntax strikes the right balance
2415 (but I acknowledge a possible lack of objectivity). On the gripping
2416 hand, the comment and directive syntax has become well established, so
2417 even if it's a wart, it may be a wart we have to live with.
2419 Making any of these changes would cause a lot of breakage or at least
2420 deprecation warnings. I'm not sure the benefit is worth the cost.
2422 For now, we'll treat this as an unresolved legacy issue.
2429 Nested Inline Markup
2430 ====================
2432 These are collected notes on a long-discussed issue. The original
2433 mailing list messages should be referred to for details.
2435 * In a 2001-10-31 discussion I wrote:
2437 Try, for example, `Ed Loper's 2001-03-21 post`_, which details
2438 some rules for nested inline markup. I think the complexity is
2439 prohibitive for the marginal benefit. (And if you can understand
2440 that tree without going mad, you're a better man than I. ;-)
2442 Inline markup is already fragile. Allowing nested inline markup
2443 would only be asking for trouble IMHO. If it proves absolutely
2444 necessary, it can be added later. The rules for what can appear
2445 inside what must be well thought out first though.
2447 .. _Ed Loper's 2001-03-21 post:
2448 http://mail.python.org/pipermail/doc-sig/2001-March/001487.html
2450 -- http://mail.python.org/pipermail/doc-sig/2001-October/002354.html
2452 * In a 2001-11-09 Doc-SIG post, I wrote:
2454 The problem is that in the
2455 what-you-see-is-more-or-less-what-you-get markup language that
2456 is reStructuredText, the symbols used for inline markup ("*",
2457 "**", "`", "``", etc.) may preclude nesting.
2459 I've rethought this position. Nested markup is not precluded, just
2460 tricky. People and software parse "double and 'single' quotes" all
2461 the time. Continuing,
2463 I've thought over how we might implement nested inline
2464 markup. The first algorithm ("first identify the outer inline
2465 markup as we do now, then recursively scan for nested inline
2466 markup") won't work; counterexamples were given in my `last post
2467 <http://mail.python.org/pipermail/doc-sig/2001-November/002363.html>`__.
2469 The second algorithm makes my head hurt::
2472 scan for start-string
2475 scan for start or end string
2476 if new start string found:
2478 elif matching end string found:
2480 elif non-matching end string found:
2481 if its a markup error:
2483 elif the initial start-string was misinterpreted:
2484 # e.g. in this case: ***strong** in emphasis*
2485 restart with the other interpretation
2486 # but it might be several layers back ...
2489 This is similar to how the parser does section title
2490 recognition, but sections are much more regular and
2493 Bottom line is, I don't think the benefits are worth the effort,
2494 even if it is possible. I'm not going to try to write the code,
2495 at least not now. If somebody codes up a consistent, working,
2496 general solution, I'll be happy to consider it.
2498 -- http://mail.python.org/pipermail/doc-sig/2001-November/002388.html
2500 * In a `2003-05-06 Docutils-Users post`__ Paul Tremblay proposed a new
2501 syntax to allow for easier nesting. It eventually evolved into
2506 The duplication with the existing interpreted text syntax is
2509 __ http://article.gmane.org/gmane.text.docutils.user/317
2511 * Could the parser be extended to parse nested interpreted text? ::
2513 :emphasis:`Some emphasized text with :strong:`some more
2514 emphasized text` in it and **perhaps** :reference:`a link``
2516 * In a `2003-06-18 Docutils-Develop post`__, Mark Nodine reported on
2517 his implementation of a form of nested inline markup in his
2518 Perl-based parser (unpublished). He brought up some interesting
2519 ideas. The implementation was flawed, however, by the change in
2520 semantics required for backslash escapes.
2522 __ http://article.gmane.org/gmane.text.docutils.devel/795
2524 * Docutils-develop threads between David Abrahams, David Goodger, and
2525 Mark Nodine (beginning 2004-01-16__ and 2004-01-19__) hashed out
2526 many of the details of a potentially successful implementation, as
2527 described below. David Abrahams checked in code to the "nesting"
2528 branch of CVS, awaiting thorough review.
2530 __ http://thread.gmane.org/gmane.text.docutils.devel/1102
2531 __ http://thread.gmane.org/gmane.text.docutils.devel/1125
2533 It may be possible to accomplish nested inline markup in general with
2534 a more powerful inline markup parser. There may be some issues, but
2535 I'm not averse to the idea of nested inline markup in general. I just
2536 don't have the time or inclination to write a new parser now. Of
2537 course, a good patch would be welcome!
2539 I envisage something like this. Explicit-role interpreted text must
2540 be nestable. Prefix-based is probably preferred, since suffix-based
2541 will look like inline literals::
2543 ``text`:role1:`:role2:
2545 But it can be disambiguated, so it ought to be left up to the author::
2547 `\ `text`:role1:`:role2:
2549 In addition, other forms of inline markup may be nested if
2552 *emphasized ``literal`` and |substitution ref| and link_*
2554 IOW, the parser ought to be as permissive as possible.
2557 Index Entries & Indexes
2558 =======================
2560 Were I writing a book with an index, I guess I'd need two
2561 different kinds of index targets: inline/implicit and
2562 out-of-line/explicit. For example::
2564 In this `paragraph`:index:, several words are being
2565 `marked`:index: inline as implicit `index`:index:
2571 The explicit index directives above would refer to
2572 this paragraph. It might also make sense to allow multiple
2573 entries in an ``index`` directive:
2579 The words "paragraph", "marked", and "index" would become index
2580 entries pointing at the words in the first paragraph. The index
2581 entry words appear verbatim in the text. (Don't worry about the
2582 ugly ":index:" part; if indexing is the only/main application of
2583 interpreted text in your documents, it can be implicit and
2584 omitted.) The two directives provide manual indexing, where the
2585 index entry words ("markup" and "syntax") do not appear in the
2586 main text. We could combine the two directives into one::
2588 .. index:: markup; syntax
2590 Semicolons instead of commas because commas could *be* part of the
2591 index target, like::
2593 .. index:: van Rossum, Guido
2595 Another reason for index directives is because other inline markup
2596 wouldn't be possible within inline index targets.
2598 Sometimes index entries have multiple levels. Given::
2600 .. index:: statement syntax: expression statements
2602 In a hypothetical index, combined with other entries, it might
2606 expression statements ..... 56
2607 assignment ................ 57
2608 simple statements ......... 58
2609 compound statements ....... 60
2611 Inline multi-level index targets could be done too. Perhaps
2614 When dealing with `expression statements <statement syntax:>`,
2615 we must remember ...
2617 The opposite sense could also be possible::
2619 When dealing with `index entries <:multi-level>`, there are
2620 many permutations to consider.
2622 Also "see / see also" index entries.
2628 .. index:: paragraph
2630 (The "index" directive above actually targets the *preceding*
2631 object.) The directive should produce something like this XML::
2634 <index_entry text="paragraph"/>
2638 This kind of content model would also allow true inline
2641 Here's a `paragraph`:index:.
2643 If the "index" role were the default for the application, it could be
2646 Here's a `paragraph`.
2648 Both of these would result in this XML::
2651 Here's a <index_entry>paragraph</index_entry>.
2655 from 2002-06-24 docutils-develop posts
2656 --------------------------------------
2658 If all of your index entries will appear verbatim in the text,
2659 this should be sufficient. If not (e.g., if you want "Van Rossum,
2660 Guido" in the index but "Guido van Rossum" in the text), we'll
2661 have to figure out a supplemental mechanism, perhaps using
2664 I've thought a bit more on this, and I came up with two possibilities:
2666 1. Using interpreted text, embed the index entry text within the
2669 ... by `Guido van Rossum [Van Rossum, Guido]` ...
2671 The problem with this is obvious: the text becomes cluttered and
2672 hard to read. The processed output would drop the text in
2673 brackets, which goes against the spirit of interpreted text.
2675 2. Use substitutions::
2677 ... by |Guido van Rossum| ...
2679 .. |Guido van Rossum| index:: Van Rossum, Guido
2681 A problem with this is that each substitution definition must have
2682 a unique name. A subsequent ``.. |Guido van Rossum| index:: BDFL``
2683 would be illegal. Some kind of anonymous substitution definition
2684 mechanism would be required, but I think that's going too far.
2686 Both of these alternatives are flawed. Any other ideas?
2693 This is the realm of the possible but questionably probable. These
2694 ideas are kept here as a record of what has been proposed, for
2695 posterity and in case any of them prove to be useful.
2698 Compound Enumerated Lists
2699 =========================
2701 Allow for compound enumerators, such as "1.1." or "1.a." or "1(a)", to
2702 allow for nested enumerated lists without indentation?
2708 Allow for variant styles by interpreting indented lists as if they
2709 weren't indented? For example, currently the list below will be
2710 parsed as a list within a block quote::
2717 But a lot of people seem to write that way, and HTML browsers make it
2718 look as if that's the way it should be. The parser could check the
2719 contents of block quotes, and if they contain only a single list,
2720 remove the block quote wrapper. There would be two problems:
2722 1. What if we actually *do* want a list inside a block quote?
2724 2. What if such a list comes immediately after an indented construct,
2725 such as a literal block?
2727 Both could be solved using empty comments (problem 2 already exists
2728 for a block quote after a literal block). But that's a hack.
2730 Perhaps a runtime setting, allowing or disabling this convenience,
2731 would be appropriate. But that raises issues too:
2733 User A, who writes lists indented (and their config file is set up
2734 to allow it), sends a file to user B, who doesn't (and their
2735 config file disables indented lists). The result of processing by
2736 the two users will be different.
2738 It may seem minor, but it adds ambiguity to the parser, which is bad.
2740 See the `Doc-SIG discussion starting 2001-04-18`__ with Ed Loper's
2741 "Structuring: a summary; and an attempt at EBNF", item 4 (and
2742 follow-ups, here__ and here__). Also `docutils-users, 2003-02-17`__
2743 and `beginning 2003-08-04`__.
2745 __ http://mail.python.org/pipermail/doc-sig/2001-April/001776.html
2746 __ http://mail.python.org/pipermail/doc-sig/2001-April/001789.html
2747 __ http://mail.python.org/pipermail/doc-sig/2001-April/001793.html
2748 __ http://sourceforge.net/mailarchive/message.php?msg_id=3838913
2749 __ http://sf.net/mailarchive/forum.php?thread_id=2957175&forum_id=11444
2752 Sloppy Indentation of List Items
2753 ================================
2755 Perhaps the indentation shouldn't be so strict. Currently, this is
2761 Anything wrong with this? ::
2770 Block quote. (no good: requires some indent relative to first
2775 2. Have to carefully define where the literal block ends::
2781 Hmm... Non-strict indentation isn't such a good idea.
2784 Lazy Indentation of List Items
2785 ==============================
2787 Another approach: Going back to the first draft of reStructuredText
2788 (2000-11-27 post to Doc-SIG)::
2790 - This is the fourth item of the main list (no blank line above).
2791 The second line of this item is not indented relative to the
2792 bullet, which precludes it from having a second paragraph.
2794 Change that to *require* a blank line above and below, to reduce
2795 ambiguity. This "loosening" may be added later, once the parser's
2796 been nailed down. However, a serious drawback of this approach is to
2797 limit the content of each list item to a single paragraph.
2800 David's Idea for Lazy Indentation
2801 ---------------------------------
2803 Consider a paragraph in a word processor. It is a single logical line
2804 of text which ends with a newline, soft-wrapped arbitrarily at the
2805 right edge of the page or screen. We can think of a plaintext
2806 paragraph in the same way, as a single logical line of text, ending
2807 with two newlines (a blank line) instead of one, and which may contain
2808 arbitrary line breaks (newlines) where it was accidentally
2809 hard-wrapped by an application. We can compensate for the accidental
2810 hard-wrapping by "unwrapping" every unindented second and subsequent
2811 line. The indentation of the first line of a paragraph or list item
2812 would determine the indentation for the entire element. Blank lines
2813 would be required between list items when using lazy indentation.
2815 The following example shows the lazy indentation of multiple body
2818 - This is the first paragraph
2819 of the first list item.
2821 Here is the second paragraph
2822 of the first list item.
2824 - This is the first paragraph
2825 of the second list item.
2827 Here is the second paragraph
2828 of the second list item.
2830 A more complex example shows the limitations of lazy indentation::
2832 - This is the first paragraph
2833 of the first list item.
2835 Next is a definition list item:
2838 Definition. The indentation of the term is
2839 required, as is the indentation of the definition's
2842 When the definition extends to more than
2843 one line, lazy indentation may occur. (This is the second
2844 paragraph of the definition.)
2846 - This is the first paragraph
2847 of the second list item.
2849 - Here is the first paragraph of
2850 the first item of a nested list.
2852 So this paragraph would be outside of the nested list,
2853 but inside the second list item of the outer list.
2855 But this paragraph is not part of the list at all.
2857 And the ambiguity remains::
2859 - Look at the hyphen at the beginning of the next line
2860 - is it a second list item marker, or a dash in the text?
2862 Similarly, we may want to refer to numbers inside enumerated
2865 1. How many socks in a pair? There are
2866 2. How many pants in a pair? Exactly
2869 Literal blocks and block quotes would still require consistent
2870 indentation for all their lines. For block quotes, we might be able
2871 to get away with only requiring that the first line of each contained
2872 element be indented. For example::
2876 This is a paragraph inside a block quote.
2877 Second and subsequent lines need not be indented at all.
2879 - A bullet list inside
2882 Second paragraph of the
2883 bullet list inside the block quote.
2885 Although feasible, this form of lazy indentation has problems. The
2886 document structure and hierarchy is not obvious from the indentation,
2887 making the source plaintext difficult to read. This will also make
2888 keeping track of the indentation while writing difficult and
2889 error-prone. However, these problems may be acceptable for Wikis and
2890 email mode, where we may be able to rely on less complex structure
2891 (few nested lists, for example).
2894 Multiple Roles in Interpreted Text
2895 ==================================
2897 In reStructuredText, inline markup cannot be nested (yet; `see
2898 above`__). This also applies to interpreted text. In order to
2899 simultaneously combine multiple roles for a single piece of text, a
2900 syntax extension would be necessary. Ideas:
2904 `interpreted text`:role1,role2:
2906 2. Suggested by Jason Diamond::
2908 `interpreted text`:role1:role2:
2910 If a document is so complex as to require nested inline markup,
2911 perhaps another markup system should be considered. By design,
2912 reStructuredText does not have the flexibility of XML.
2914 __ `Nested Inline Markup`_
2917 Parameterized Interpreted Text
2918 ==============================
2920 In some cases it may be expedient to pass parameters to interpreted
2921 text, analogous to function calls. Ideas:
2923 1. Parameterize the interpreted text role itself (suggested by Jason
2926 `interpreted text`:role1(foo=bar):
2928 Positional parameters could also be supported::
2930 `CSS`:acronym(Cascading Style Sheets): is used for HTML, and
2931 `CSS`:acronym(Content Scrambling System): is used for DVDs.
2933 Technical problem: current interpreted text syntax does not
2934 recognize roles containing whitespace. Design problem: this smells
2935 like programming language syntax, but reStructuredText is not a
2936 programming language.
2938 2. Put the parameters inside the interpreted text::
2940 `CSS (Cascading Style Sheets)`:acronym: is used for HTML, and
2941 `CSS (Content Scrambling System)`:acronym: is used for DVDs.
2943 Although this could be defined on an individual basis (per role),
2944 we ought to have a standard. Hyperlinks with embedded URIs already
2945 use angle brackets; perhaps they could be used here too::
2947 `CSS <Cascading Style Sheets>`:acronym: is used for HTML, and
2948 `CSS <Content Scrambling System>`:acronym: is used for DVDs.
2950 Do angle brackets connote URLs too much for this to be acceptable?
2951 How about the "tag" connotation -- does it save them or doom them?
2953 3. `Nested inline markup`_ could prove useful here::
2955 `CSS :def:`Cascading Style Sheets``:acronym: is used for HTML,
2956 and `CSS :def:`Content Scrambling System``:acronym: is used for
2959 Inline markup roles could even define the default roles of nested
2960 inline markup, allowing this cleaner syntax::
2962 `CSS `Cascading Style Sheets``:acronym: is used for HTML, and
2963 `CSS `Content Scrambling System``:acronym: is used for DVDs.
2965 Does this push inline markup too far? Readability becomes a serious
2966 issue. Substitutions may provide a better alternative (at the expense
2967 of verbosity and duplication) by pulling the details out of the text
2970 |CSS| is used for HTML, and |CSS-DVD| is used for DVDs.
2972 .. |CSS| acronym:: Cascading Style Sheets
2973 .. |CSS-DVD| acronym:: Content Scrambling System
2976 ----------------------------------------------------------------------
2978 This whole idea may be going beyond the scope of reStructuredText.
2979 Documents requiring this functionality may be better off using XML or
2980 another markup system.
2982 This argument comes up regularly when pushing the envelope of
2983 reStructuredText syntax. I think it's a useful argument in that it
2984 provides a check on creeping featurism. In many cases, the resulting
2985 verbosity produces such unreadable plaintext that there's a natural
2986 desire *not* to use it unless absolutely necessary. It's a matter of
2987 finding the right balance.
2990 Syntax for Interpreted Text Role Bindings
2991 =========================================
2993 The following syntax (idea from Jeffrey C. Jacobs) could be used to
2994 associate directives with roles::
2996 .. :rewrite: class:: rewrite
2998 `She wore ribbons in her hair and it lay with streaks of
3001 The syntax is similar to that of substitution declarations, and the
3002 directive/role association may resolve implementation issues. The
3003 semantics, ramifications, and implementation details would need to be
3006 The example above would implement the "rewrite" role as adding a
3007 ``class="rewrite"`` attribute to the interpreted text ("inline"
3008 element). The stylesheet would then pick up on the "class" attribute
3009 to do the actual formatting.
3011 The advantage of the new syntax would be flexibility. Uses other than
3012 "class" may present themselves. The disadvantage is complexity:
3013 having to implement new syntax for a relatively specialized operation,
3014 and having new semantics in existing directives ("class::" would do
3015 something different).
3017 The `"role" directive`__ has been implemented.
3019 __ ../../ref/rst/directives.html#role
3022 Character Processing
3023 ====================
3025 Several people have suggested adding some form of character processing
3026 to reStructuredText:
3028 * Some sort of automated replacement of ASCII sequences:
3030 - ``--`` to em-dash (or ``--`` to en-dash, and ``---`` to em-dash).
3031 - Convert quotes to curly quote entities. (Essentially impossible
3032 for HTML? Unnecessary for TeX.)
3033 - Various forms of ``:-)`` to smiley icons.
3034 - ``"\ "`` to . Problem with line-wrapping though: it could
3035 end up escaping the newline.
3036 - Escaped newlines to <BR>.
3037 - Escaped period or quote or dash as a disappearing catalyst to
3038 allow character-level inline markup?
3040 * XML-style character entities, such as "©" for the copyright
3043 Docutils has no need of a character entity subsystem. Supporting
3044 Unicode and text encodings, character entities should be directly
3045 represented in the text: a copyright symbol should be represented by
3046 the copyright symbol character. If this is not possible in an
3047 authoring environment, a pre-processing stage can be added, or a table
3048 of substitution definitions can be devised.
3050 A "unicode" directive has been implemented to allow direct
3051 specification of esoteric characters. In combination with the
3052 substitution construct, "include" files defining common sets of
3053 character entities can be defined and used. `A set of character
3054 entity set definition files have been defined`__ (`tarball`__).
3055 There's also `a description and instructions for use`__.
3057 __ http://docutils.sf.net/tmp/charents/
3058 __ http://docutils.sf.net/tmp/charents.tgz
3059 __ http://docutils.sf.net/tmp/charents/README.html
3061 To allow for `character-level inline markup`_, a limited form of
3062 character processing has been added to the spec and parser: escaped
3063 whitespace characters are removed from the processed document. Any
3064 further character processing will be of this functional type, rather
3065 than of the character-encoding type.
3067 .. _character-level inline markup:
3068 ../../ref/rst/restructuredtext.html#character-level-inline-markup
3072 .. text-replace:: "pattern" "replacement"
3074 - Support Unicode "U+XXXX" codes.
3075 - Support regexps, perhaps with alternative "regexp-replace"
3077 - Flags for regexps; ":flags:" option, or individuals.
3078 - Specifically, should the default be case-sensistive or
3085 * Should ^L (or something else in reST) be defined to mean
3086 force/suggest page breaks in whatever output we have?
3088 A "break" or "page-break" directive would be easy to add. A new
3089 doctree element would be required though (perhaps "break"). The
3090 final behavior would be up to the Writer. The directive argument
3091 could be one of page/column/recto/verso for added flexibility.
3093 Currently ^L (Python's ``\f``) characters are treated as whitespace.
3094 They're converted to single spaces, actually, as are vertical tabs
3095 (^K, Python's ``\v``). It would be possible to recognize form feeds
3096 as markup, but it requires some thought and discussion first. Are
3097 there any downsides? Many editing environments do not allow the
3098 insertion of control characters. Will it cause any harm? It would
3099 be useful as a shorthand for the directive.
3101 It's common practice to use ^L before Emacs "Local Variables"
3108 indent-tabs-mode: nil
3109 sentence-end-double-space: t
3113 These are already present in many PEPs and Docutils project
3114 documents. From the Emacs manual (info):
3116 A "local variables list" goes near the end of the file, in the
3117 last page. (It is often best to put it on a page by itself.)
3119 It would be unfortunate if this construct caused a final blank page
3120 to be generated (for those Writers that recognize the page breaks).
3121 We'll have to add a transform that looks for a "break" plus zero or
3122 more comments at the end of a document, and removes them.
3124 Probably a bad idea because there is no such thing as a page in a
3125 generic document format.
3127 * Could the "break" concept above be extended to inline forms?
3128 E.g. "^L" in the middle of a sentence could cause a line break.
3129 Only recognize it at the end of a line (i.e., ``\f\n``)?
3131 Or is formfeed inappropriate? Perhaps vertical tab (``\v``), but
3132 even that's a stretch. Can't use carriage returns, since they're
3133 commonly used for line endings.
3135 Probably a bad idea as well because we do not want to use control
3136 characters for well-readable and well-writable markup, and after all
3137 we have the line block syntax for line breaks.
3143 Add ``^superscript^`` inline markup? The only common non-markup uses
3144 of "^" I can think of are as short hand for "superscript" itself and
3145 for describing control characters ("^C to cancel"). The former
3146 supports the proposed syntax, and it could be argued that the latter
3147 ought to be literal text anyhow (e.g. "``^C`` to cancel").
3149 However, superscripts are seldom needed, and new syntax would break
3150 existing documents. When it's needed, the ``:superscript:``
3151 (``:sup:``) role can we used as well.
3157 Add the following directives?
3159 - "exec": Execute Python code & insert the results. Call it
3160 "python" to allow for other languages?
3162 - "system": Execute an ``os.system()`` call, and insert the results
3163 (possibly as a literal block). Definitely dangerous! How to make
3164 it safe? Perhaps such processing should be left outside of the
3165 document, in the user's production system (a makefile or a script or
3166 whatever). Or, the directive could be disabled by default and only
3167 enabled with an explicit command-line option or config file setting.
3168 Even then, an interactive prompt may be useful, such as:
3170 The file.txt document you are processing contains a "system"
3171 directive requesting that the ``sudo rm -rf /`` command be
3172 executed. Allow it to execute? (y/N)
3174 - "eval": Evaluate an expression & insert the text. At parse
3175 time or at substitution time? Dangerous? Perhaps limit to canned
3176 macros; see text.date_.
3178 .. _text.date: ../todo.html#text-date
3180 It's too dangerous (or too complicated in the case of "eval"). We do
3181 not want to have such things in the core.
3184 ``encoding`` Directive
3185 ======================
3187 Add an "encoding" directive to specify the character encoding of the
3188 input data? Not a good idea for the following reasons:
3190 - When it sees the directive, the parser will already have read the
3191 input data, and encoding determination will already have been done.
3193 - If a file with an "encoding" directive is edited and saved with
3194 a different encoding, the directive may cause data corruption.
3197 Support for Annotations
3198 =======================
3200 Add an "annotation" role, as the equivalent of the HTML "title"
3201 attribute? This is secondary information that may "pop up" when the
3202 pointer hovers over the main text. A corresponding directive would be
3203 required to associate annotations with the original text (by name, or
3204 positionally as in anonymous targets?).
3206 There have not been many requests for such feature, though. Also,
3207 cluttering WYSIWYG plaintext with annotations may not seem like a good
3208 idea, and there is no "tool tip" in formats other than HTML.
3214 Add a "term" role for unfamiliar or specialized terminology? Probably
3215 not; there is no real use case, and emphasis is enough for most cases.
3221 We need syntax for `object references`_.
3223 - Parameterized substitutions? For example::
3225 See |figure (figure name)| on |page (figure name)|.
3227 .. |figure (name)| figure-ref:: (name)
3228 .. |page (name)| page-ref:: (name)
3230 The result would be::
3232 See figure 3.11 on page 157.
3234 But this would require substitution directives to be processed at
3235 reference-time, not at definition-time as they are now. Or,
3236 perhaps the directives could just leave ``pending`` elements
3237 behind, and the transforms do the work? How to pass the data
3238 through? Too complicated. Use interpreted text roles.
3240 .. _object references:
3241 ../todo.html#object-numbering-and-object-references
3248 indent-tabs-mode: nil
3249 sentence-end-double-space: t