Instead rely on the markdown writer with appropriate extensions.
Export writeCommonMark variant from Markdown writer.
This changes a few small things in rendering markdown,
e.g. w/r/t requiring backslashes before spaces inside
super/subscripts.
Screen readers read an image's `alt` attribute and the figure caption,
both of which come from the same source in pandoc. The figure caption is
hidden from screen readers with the `aria-hidden` attribute. This
improves accessibility.
For HTML4, where `aria-hidden` is not allowed, pandoc still uses an
empty `alt` attribute to avoid duplicate contents.
Closes: #6491
Closes#6385. (The summary element needs to be the first
child of details and should not be enclosed by p tags.)
NOTE: you need to include a blank line before the closing
`</details>`, if you want the last part of the content to
be parsed as a paragraph.
Some CSS to ensure that display math is
displayed centered and on a new line is now included
in the default HTML-based templates; this may be
overridden if the user wants a different behavior.
This reverts a change in the last release; the Div is
no longer needed, because we can now put the id right in
the Table's attributes. However, writers may still need
to be modified to do something with the id in a Table
(e.g. create an anchor), so in the short term we may lose
the ability to link to tables in some writers.
The Builder.simpleTable now only adds a row to the TableHead when the
given header row is not null. This uncovered an inconsistency in the
readers: some would unconditionally emit a header filled with empty
cells, even if the header was not present. Now every reader has the
conditional behaviour. Only the XWiki writer depended on the header
row being always present; it now pads its head as necessary.
- Writers.Native is now adapted to the new Table type.
- Inline captions should now be conditionally wrapped in a Plain, not
a Para block.
- The toLegacyTable function now lives in Writers.Shared.
- SmallCaps instead of Span for the part after the initial capital.
- Ensure that both arguments are parsed, so that in Markdown both
are treated as raw LateX. (Closes #6258.)
This reverts commit 3493d6afaa.
This might be worth considering in the future, but let's not do
it yet...the additional complexity needs a better justification.
This is experimental. Normally metadata values are interpreted
as markdown, but if the !!literal tag is used they will be interpreted
as plain strings.
We need to consider whether this can still be implemented if
we switch back from HsYAML to yaml for performance reasons.
Closes#5849. Previously biblatex citations were only grouped if
there was no prefix. This patch allows them to be grouped in
subgroups split by prefixes and suffixes, which allows better citation
sorting.
If parsing fails in a raw environment (e.g. due to special
characters like unescaped `_`), try again as a verbatim
environment, which is less sensitive to special characters.
This allows us to capture special environments that change
catcodes as raw tex when `-f latex+raw_tex` is used.
Closes#6034.
The fix to #6030 actually changed behavior, so that the
2D nesting occurred at slide level N-1 and N, instead of
at the top-level section. This commit restores the 2.7.3 behavior.
If there are more than 2 levels, the top level is horizontal
and the rest are collapsed to vertical.
Closes#6032.
With positive heading shifts, starting in 2.8 this option caused
metadata titles to be removed and changed to regular headings.
This behavior is incompatible with the old behavior of
`--base-header-level` and breaks old workflows, so with this
commit we are rolling back this change.
Now, there is an asymmetry in positive and negative heading
level shifts:
+ With positive shifts, the metadata title stays the same and
does not get changed to a heading in the body.
+ With negative shifts, a heading can be converted into the
metadata title.
I think this is a desirable combination of features, despite
the asymmetry. One might, e.g., want to have a document
with level-1 section headigs, but render it to HTML with
level-2 headings, retaining the metadata title (which pandoc
will render as a level-1 heading with the default template).
Closes#5957.
Revises #5615.
If a simple table would be too wide, we use a grid table.
The code for generating grid tables has been adjusted to
give more intelligent column widths when widths aren't
given. (This also affects the markdown writer.)
Closes#5899.
`Nothing` means: nothing specified.
`Just []` means: an empty list specified (e.g. in defaults).
Potentially these could lead to different behavior: see #5888.
Superscripts and subscripts cannot contain spaces,
but newlines were previously allowed (unintentionally).
This led to bad interactions in some cases with footnotes.
E.g.
```
foo^[note]
bar^[note]
```
With this change newlines are also not allowed inside
super/subscripts.
Closes#5878.
* Add HTML Reader support for `<dfn>`, parsing this as a Span with class `dfn`.
* Change `htmlSpanLikeElements` implementation to retain classes,
attributes and inline content.
Previously, if a document contained two YAML metadata blocks
that set the same field, the conflict would be resolved in favor
of the first. Now it is resolved in favor of the second (due to
a change in pandoc-types).
This makes the behavior more uniform with other things in pandoc
(such as reference links and `--metadata-file`).
When a div surrounds multiple sections at the same level,
or a section of highre level followed by one of lower level,
then we just leave it as a div and create a new div for the
section.
Closes#5846, closes#5761.
Parse <mark> elements from HTML as HTML span like elements, with a
single class matching the tag name `mark`. Mark elements are rendered to
HTML using the native <mark> element.
Fixes https://github.com/jgm/pandoc/issues/5797.
* Text.Pandoc.Shared: export `htmlSpanLikeElements` [API change]
This commit also introduces a mapping of HTML span like elements that
are internally represented as a Span with a single class, but that are
converted back to the original element by the html writer. As of now,
only the kbd element is handled this way. Ideally these elements should
be handled as plain AST values, but since that would be a breaking
change with a large impact, we revert to this stop-gap solution.
Fixes https://github.com/jgm/pandoc/issues/5796.
...even when we must lose width information.
All in all this seems to be people's preferred behavior, even though it
is slightly lossier.
Closes#2608.
Closes#4497.
on conflicting fields. This changes earlier behavior (but not in
a release), where first took precedence.
Note that this may seem inconsistent with the behavior of
multiple YAML blocks within a document, where the first takes
precedence. Still, it is convenient to be able to override
defaults with options later on the command line.
If this is present on a heading with the 'unnumbered' class,
the heading won't appear in the TOC. This class has no
effect if 'unnumbered' is not also specified.
This affects HTML-based writers (including slide shows
and epub), LateX (including beamer), RTF, and PowerPoint.
Other writers do not yet support `unlisted`.
Closes#1762.
We were requiring consistent indentation, but this
isn't required by RST, as long as each nonblank
line of the block has *some* indentation.
Closes#5753.
Previously we used the following Project Gutenberg conventions
for plain output:
- extra space before and after level 1 and 2 headings
- all-caps for strong emphasis `LIKE THIS`
- underscores surrounding regular emphasis `_like this_`
This commit makes `plain` output plainer. Strong and Emph
inlines are rendered without special formatting. Headings
are also rendered without special formatting, and with only
one blank line following.
To restore the former behavior, use `-t plain+gutenberg`.
API change: Add `Ext_gutenberg` constructor to `Extension`.
See #5741.
Deprecate --base-heading-level.
The new option does everything the old one does, but also
allows negative shifts. It also promotes the document
metadata (if not null) to a level-1 heading with a +1 shift,
and demotes an initial level-1 heading to document metadata
with a -1 shift. This supports converting documents that
use an initial level-1 heading for the document title.
Closes#5615.
* Org reader: allow the `-i` switch to ignore leading spaces.
* Org reader: handle awkwardly-aligned code blocks within lists.
Code blocks in Org lists must have their #+BEGIN_ aligned in a
reasonable way, but their other components can be positioned otherwise.
Text.Pandoc.Shared:
+ Remove `Element` type [API change]
+ Remove `makeHierarchicalize` [API change]
+ Add `makeSections` [API change]
+ Export `deLink` [API change]
Now that we have Divs, we can use them to represent the structure
of sections, and we don't need a special Element type.
`makeSections` reorganizes a block list, adding Divs with
class `section` around sections, and adding numbering
if needed.
This change also fixes some longstanding issues recognizing
section structure when the document contains Divs.
Closes#3057, see also #997.
All writers have been changed to use `makeSections`.
Note that in the process we have reverted the change
c1d058aeb1
made in response to #5168, which I'm not completely
sure was a good idea.
Lua modules have also been adjusted accordingly.
Existing lua filters that use `hierarchicalize` will
need to be rewritten to use `make_sections`.
Revert "hierarchicalize: ensure that sections get ids..."
This reverts commit 212406a61d.
Revert "Improve detection of headings in Divs by hierarchicalize."
This reverts commit 6e2cfd6c97.
Revert "Shared.hierarchicalize: improve handling of div and section structure."
This reverts commit 345b33762e.
The structure
```
<h1>one</h1>
<div>
<h1>two</h1>
</div>
```
should create two coordinate sections, not a section with
a subsection. Now it does.
Extends #3057.
Previously Divs were opaque to hierarchicalize, so headings
inside divs didn't get into the table of contents, for
example (#3057).
Now hierarchicalize treats Divs as sections when appropriate.
For example, these structures both yield a section and a
subsection:
``` html
<div>
<h1>one</h1>
<div>
<h2>two</h2>
</div>
</div>
```
``` html
<div>
<h1>one</h1>
<div>
<h1>two</h1>
</div>
</div>
```
Note that
``` html
<h1>one</h1>
<div>
<h2>two</h2>
</div>
<h1>three</h1>
```
gets parsed as the structure
one
two
three
which may not always be desirable.
Closes#3057.
Previously we used named character references with html5 output.
But these aren't valid XML, and we aim to produce html5 that is
also valid XHTML (polyglot markup). (This is also needed for
epub3.)
Closes#5718.
+ Remove Text.Pandoc.Pretty; use doclayout instead. [API change]
+ Text.Pandoc.Writers.Shared: remove metaToJSON, metaToJSON'
[API change].
+ Text.Pandoc.Writers.Shared: modify `addVariablesToContext`,
`defField`, `setField`, `getField`, `resetField` to work with
Context rather than JSON values. [API change]
+ Text.Pandoc.Writers.Shared: export new function `endsWithPlain` [API
change].
+ Use new templates and doclayout in writers.
+ Use Doc-based templates in all writers.
+ Adjust three tests for minor template rendering differences.
+ Added indentation to body in docbook4, docbook5 templates.
The main impact of this change is better reflowing of content
interpolated into templates. Previously, interpolated variables
were rendered independently and intepolated as strings, which could lead
to overly long lines. Now the templates interpolated as Doc values
which may include breaking spaces, and reflowing occurs
after template interpolation rather than before.
Changed optMetadataFile from `Maybe FilePath` to `[FilePath]`. This allows
for multiple YAML metadata files to be added. The new default value has
been changed from `Nothing` to `[]`.
To account for this change in `Text.Pandoc.App`, `metaDataFromFile` now
operates on two `mapM` calls (for `readFileLazy` and `yamlToMeta`) and a fold.
Added a test (command/5700.md) which tests this functionality and
updated MANUAL.txt, as per the contributing guidelines.
With the current behavior, using `foldr1 (<>)`, values within files
specified first will be used over those in later files. (If the reverse
of this behavior would be preferred, it should be fixed by changing
foldr1 to foldl1.)
the token string is modified by a parser (e.g. accent when
it only takes part of a Word token).
Closes#5686. Still not ideal, because we get the whole
`\t0BAR` and not just `\t0` as a raw latex inline command.
But I'm willing to let this be an edge case, since you
can easily work around this by inserting a space, braces,
or raw attribute. The important thing is that we no longer
drop the rest of the document after a raw latex inline
command that gobbles only part of a Word token!
The web service passed in to `--webtex` may render formulas using inline
or display style by default. Prefixing formulas with the appropriate
command ensures they are rendered correctly.
This is a followup to the discussion in #5656.
Previously we just omitted these. Now we render them
using `\hfill\break` instead of `\\`. This is a revision
of a PR by @sabine (#5591) who should be credited with the
idea.
Closes#3324.
The `raw_attribute` will be used to mark raw bits, even HTML
and LaTeX, and even when `raw_html` and `raw_tex` are enabled,
as they are by default.
To get the old behavior, disable `raw_attribute` in the writer.
Closes#4311.
Now we boldface code but not other things. This matches the
most common style in man pages (particularly option lists).
Also, remove a regression in the last commit in which 'nowrap'
was removed.
Previously if the language was not in the list of listings-
supported languages, it would not be added as a class, so
custom syntax highlighting could not be used.
Closes#5540.
Change is in rawLaTeXInline in LaTeX reader, but
it affects the markdown reader and other readers
that allow raw LaTeX.
Previously, trailing `{}` would be included for
unknown commands, but not for known commands.
However, they are sometimes used to avoid a trailing
space after the command. The chances that a `{}`
after a LaTeX command is not part of the command
are very small.
Closes#5439.
Move up the pattern match to be reachable, closes#5517.
Previously `file:/` URLs were handled wrongly and pandoc attempted
to make HTTP requests, which failed.
Previously if labels had integer names, it could produce a conflict
with auto-labeled reference links. Now we test for a conflict and find
the next available integer.
Note that this involves adding a new state variable `stPrevRefs` to
keep track of refs used in other document parts when using
`--reference-location=block|section`
Closes#5495
Emacs always uses two spaces when indenting the content of src blocks,
e.g., when exiting a `C-c '` edit-buffer. Pandoc used to indent contents
by the space-equivalent of one tab, but now always uses two spaces, too.
Closes: #5440
iOS chooses to render a number of Unicode entities,
including '↩', as big colorful emoji. This can be
defeated by appending Unicode
VARIATION SELECTOR-15'/'VARIATION SELECTOR-16'.
So we now append this character when escaping
strings, for both '↩' and '↔'.
If other characters prove problematic, they can
simply be added to needsVariationSelector.
Closes#5469.
This fixes a bug wherein footnotes appeared in the wrong
order, and with duplicate numbers, when in table captions
and cells.
We now use regular `\footnote` commands, even in the table
caption and the minipages containing cells. Apparently
longtable knows how to handle this.
Closes#5367.
Instead of `$HOME/.pandoc`, the default user data directory is
now `$XDG_DATA_HOME/pandoc`, where `XDG_DATA_HOME` defaults to
`$HOME/.local/share` but can be overridden by setting the environment
variable.
If this directory is missing, then `$HOME/.pandoc` is searched
instead, for backwards compatibility. However, we recommend
moving local pandoc data files from `$HOME/.pandoc` to
`$HOME/.local/share/pandoc`.
On Windows the default user data directory remains the same.
Closes#3582.
This improves on the original fix to #5285 by preventing
other mixed lists (lists with a mix of Plain and Para
elements) that were allowed given the original fix.
This ensures that a figure containing a single image
is parsed as a pandoc "implicit figure" (i.e., a
Para with a single Image whose title attribute begins
with `fig:`). More complex figures will still be parsed
as divs.
Closes#5321.
Previously parsing would break if the code block
contained a string of backticks of sufficient length
followed by something other than end of line.
Closes#5304.
Closes#5285. Previously the algorithm allowed list items
with a mix of Para and Plain, which is never wanted.
compactify in T.P.Shared has been modified so that, if
a list's items contain (at the top level) Para elements
(aside from perhaps at the very end), ALL Plains are
converted to Paras.
Otherwise last block gets parsed as a Plain rather than
a Para.
This is a regression in pandoc 2.x. This patch restores
pandoc 1.19 behavior.
Closes#5271.
`\ldots{}.` doesn't behave as well as `\ldots.` with the latex
ellipsis package. This patch causes pandoc to avoid emitting
the `{}` when it is not necessary. Now `\ldots` and other
control sequences used in escaping will be followed by either
a `{}`, a space, or nothing, depending on context.
Thanks to Elliott Slaughter for the suggestion.
* These were added by the RST reader and, for literate Haskell,
by the Markdown and LaTeX readers. There is no point to
this class, and it is not applied consistently by all readers.
See #5047.
* Reverse order of `literate` and `haskell` classes on code blocks
when parsing literate Haskell. Better if `haskell` comes first.
Previously we used HsYAML's decodeStrict to recognize
boolean values (treating everything else as a string).
This caused problems relating to hvr/HsYAML#7.
We now just check for the recognized boolean values
`true|True|TRUE|false|False|FALSE`, and avoid using
HsYAML.
Closes#5177.
Starting with pandoc 2.4, citations and quoted inlines
were no longer recognized after parentheses. This is
because of commit 9b0bd4ec6f,
which is reverted here.
The point of that commit was to allow relocation of
soft line breaks to before an abbreviation, so that
a nonbreaking space could be added after the
abbreviation. Now we simply leave the soft line
break in place, even though this means that
we won't get a nonbreaking space after "Mr."
at the end of a line (and in LaTeX this may
result in a longer intersentential space).
Those who care about this issue should take care
not to end lines with an abbreviation, or to
insert nonbreaking spaces manually.
Closes#5099.
The parameter is Extensions. This allows these functions to
be sensitive to the settings of `Ext_gfm_auto_identifiers` and
`Ext_ascii_identifiers`.
This allows us to use `uniqueIdent` in the CommonMark reader,
replacing some custom code.
It also means that `gfm_auto_identifiers` can now be used
in all formats.
Semantically, `gfm_auto_identifiers` is now a modifier of
`auto_identifiers`; for identifiers to be set, `auto_identifiers`
must be turned on, and then the type of identifier produced
depends on `gfm_auto_identifiers` and `ascii_identifiers` are set.
Closes#5057.
* Lua: allow access to pandoc state
Lua filters and custom writers now have read-only access to most fields
of pandoc's internal state via the global variable `PANDOC_STATE`.
* Lua: allow iterating through fields of PANDOC_STATE
* Lua filters doc: describe CommonState
* Lua filters doc: mention global variable PANDOC_STATE
* Lua: add access to logs
Log messages can currently only be printed, but not decomposed.
T.P.GroffChar: replaced `essentialEscapes` with `manEscapes`,
which includes all the escapes mentioned in the groff_man manual.
T.P.Writers.Groff: removed escapeCode; changed parameter on
escapeString from Bool to new type `EscapeMode`.
Rewrote `escapeString`.
- `--ascii` is now turned on automatically for man output, for
portability. All man output will be escaped to ASCII.
- In T.P.Writers.Groff, `escapeChar`, `escapeString`, and
`escapeCode` now take a boolean parameter that selects
ascii-only output. This is used by the Ms writer for
`--ascii`, instead of doing an extra pass after writing
the document.
- In ms output without `--ascii`, unicode is used whenever
possible (e.g. for double quotes).
- A few escapes are changed: e.g. `\[rs]` instead of `\\` for
backslash, and `\ga]` instead of `` \` `` for backtick.
(unexported module). These are used in both the man and ms
writers.
Moved groffEscape out of Text.Pandoc.Writers.Shared [cancels earlier
API change from adding it, which was after last release].
This fixes strong/code combination on man (should be `\f[CB]` not
`\f[BC]`), mentioned in #4973.
Updated tests.
Closes#4975.
The default is `-raw_tex`, so no raw tex should result
unless we explicitly say `+raw_tex`. Previously some
raw commands did make it through.
Closes#4527.
Previously, the writer would unconditionally emit HTMLish output for
subscripts, superscripts, strikeouts (if the strikeout extension is
disabled) and small caps, even with raw_html disabled.
Now there are plain-text (and, where possible, fancy Unicode)
fallbacks for all of these corresponding (mostly) to the Markdown
fallbacks, and the HTMLish output is only used when raw_html is
enabled.
This commit adds exported functions `toSuperscript` and
`toSubscript` to `Text.Pandoc.Writers.Shared`. [API change]
Closes#4528.
Closes#4284.
Headers with the corresponding tags should not appear in the output.
If one or more of the specified tags contains a non-tag character
like `+`, Org-mode will not treat that as a valid tag, but will
nonetheless continue scanning for valid tags. That behavior is not
replicated in this patch; entering `cat+dog` as one of the entries in
`#+EXCLUDE_TAGS` and running the file through Pandoc will cause the
parser to fail and result in the only excluded tag being the default, `noexport`.
Now the `write*` functions for Docbook, HTML, ICML, JATS,
Man, Ms, OPML are sensitive to `writerPreferAscii`. Previously
the to-ascii translation was done in Text.Pandoc.App, and
thus not available to those using the writer functions
directly.
In addition, the LaTeX writer is now sensitive to
`writerPreferAscii` and to `--ascii`. 100% ASCII
output can't be guaranteed, but the writer will use
commands like `\"{a}` and `\l` whenever possible,
to avoid emiting a non-ASCII character.
A new unexported module, Text.Pandoc.Groff, has been
added to store functions used in the different groff-based
writers.
* Add support for multiprenote and multipostnote arguments.
The multiprenotes occur before the first prefix of a
multicite, and the multipostnotes follow the last suffix.
* Add test for multiprenote and multipostnote.
With this change, autolinks are parsed as Links with
the `uri` class. (The same is true for bare links, if
the `autolink_bare_uris` extension is enabled.) Email
autolinks are parsed as Links with the `email` class.
This allows the distinction to be represented in the
URI.
Formerly the `uri` class was added to autolinks by
the HTML writer, but it had to guess what was an autolink
and could not distinguish `[http://example.com](http://example.com)`
from `<http://example.com>`. It also incorrectly recognized
`[pandoc](pandoc)` as an autolink. Now the HTML writer
simply passes through the `uri` attribute if it is present,
but does not add anything.
The Textile writer has been modified so that the `uri`
class is not explicitly added for autolinks, even if it
is present.
Closes#4913.
This now allows raw LaTeX environments, `\ref`, and `\eqref` to
be parsed (which is helpful for translation HTML documents using
MathJaX).
Closes#1126.
Also foreigncblockquote, hyphenblockquote, hyphencblockquote.
Closes#4848. But note: currently foreignquote will be
parsed as a regular Quoted inline (not using the quotes
appropriate to the foreign language).
Add support for `\|`, `\b`, `\G`, `\h`, `\d`, `\f`,
`\r`, `\t`, `\U`, `\i`, `\j`, `\newtie`, `\textcircled`.
Also fall back to combining characters when composed
characters are not available.
Closes#4652.
...they should only be recognized in siunitx contexts.
For example, `\l` outside of an siunitx context should be l-slash,
not l (for liter)!
Closes#4842.
We can't always tell if it's LaTeX, ConTeXt, or plain TeX.
Better just to use "tex" always.
Also changed:
ConTeXt writer: now outputs raw "tex" blocks as well as "context".
(Closes#969).
RST writer: uses ".. raw:: latex" for "tex" content.
(RST doesn't support raw context anyway.)
Note that if "context" or "latex" specifically is desired,
you can still force that in a markdown document by using
the raw attribute (see MANUAL.txt):
```{=latex}
\foo
```
Note that this change may affect some filters, if they assume that raw
tex parsed by the Markdown reader will be RawBlock (Format "latex").
In most cases it should be trivial to modify the filters to accept
"tex" as well.
For example: `\def\foo#1[#2]{#1 and #2}`.
Closes#4768. Also fixes#4771.
API change: in Text.Pandoc.Readers.LaTeX.Types,
new type ArgSpec added. Second parameter of Macro
constructor is now `[ArgSpec]` instead of `Int`.
This fixes a regression in 2.2.3, which cause boolean values to
be parsed as MetaInlines instead of MetaBool.
Note also an undocumented (but desirable) change in 2.2.3:
numbers are now parsed as MetaInlines rather than MetaString.
Closes#4819.
* Use a Span with class "title-reference" for the default
title-reference role.
* Use B.text to split up contents into Spaces, SoftBreaks, and Strs
for title-reference.
* Use Code with class "interpreted-text" instead of Span and Str for
unknown roles. (The RST writer has also been modified to round-trip
this properly.)
* Disallow blank lines in interpreted text.
* Backslash-escape now works in interpreted text.
* Backticks followed by alphanumerics no longer end interpreted text.
Closes#4811.
RST does not allow nested emphasis, links, or other inline
constructs.
Closes#4581, double parsing of links with URLs as
link text. This supersedes the earlier fix for #4581
in 6419819b46.
Fixes#4561, a bug parsing with URLs inside emphasis.
Closes#4792.
Starting in 2.2.2, everything after an `\input` (or `\include`)
in a markdown file would be parsed as raw LaTeX.
This commit fixes the issue and adds a regression test.
Closes#4781.
