Fix interaction of top-level divisions `part` or `chapter` with
unnumbered headers when emitting LaTeX. Headers are ensured to be
written using stared commands (like `\subsection*{}`).
Fixes: #3272
Previously setting writerStandalone = True did nothing unless
a template was provided in writerTemplate. Now a fragment
will be generated if writerTemplate is Nothing; otherwise,
the specified template will be used and standalone output
generated. [API change]
+ Removed Text.Pandoc.Readers.Docx.Fonts
+ Moved its code to texmath; we now use (from texmath 0.9)
Text.TeXMath.Unicode.Fonts
+ Use texmath 0.9 (currently from git).
+ Updated epub tests because texmath now handles more mathml.
The "default" option is no longer represented as `Nothing` but via a new
type constructor, making the `Maybe` wrapper superfluous.
The default behavior of using heuristics can now be enabled explicitly
by setting `--top-level-division=default`.
API change (`Text.Pandoc.Options`): The `Division` type was renamed to
`TopLevelDivision`. The `Section`, `Chapter`, and `Part` constructors
were renamed to `TopLevelSection`, `TopLevelChapter`, and
`TopLevelPart`, respectively. An additional `TopLevelDefault`
constructor was added, which is now also the new default value of the
`writerTopLevelDivision` field in `WriterOptions`.
We now check explicitly for non-1 rowspan or colspan
attributes, and fail when we encounter them. Previously
we checked that each row had the same number of cells,
but that could be true even with rowspans/colspans.
And there are cases where it isn't true in tables that
we can handle fine -- e.g. when a tr element is empty.
So now we just pad rows with empty cells when needed.
Closes#3027.
This reverts commit fee0b913c5.
The previous commit did not provide a good way to get increased
indentation for nested block quotes.
Rolling it back for now. @jjsheets feel free to submit something
that handles multiple levels of block quote smoothly, if you like.
* Open Document writer: set first level of blockquotes to not use indent
Nested blockquotes start using indents like before. Quotation style is
still in use, so the style's indent settings take effect on the first
level of blockquotes.
* Removed list construction to improve pull request to fix#2747
Latex doesn't like when hypertargets or images are
put in the options list of the section. They are not
lost since they were actually duplicated and present
also in the second argument list.
Note on the implementation:
I had to inline the definiton of 'foldMap' since it is
not implemented in every version of Haskell that Pandoc
supports.
ODT's reader always put empty captions for the parsed
tables. This commit
1) checks paragraphs that follow the table definition
2) treats specially a paragraph with a style named 'Table'
3) does some postprocessing of the paragraphs that combines
tables followed immediately by captions
The ODT writer used 'TableCaption' style name for the caption
paragraph. This commit follows the open office approach which
allows for appending captions to table but uses a built-in style
named 'Table' instead of 'TableCaption'. Any users of odt format
(both writer and reader) are therefore required to change the
style's name to 'Table', if necessary.
Pandoc uses heuristics to determine the most resonable top-level
division type when emitting LaTeX or Docbook markup. It is now possible
to overwrite this implicitly set top-level division via the
`top-level-division` command line parameter.
API change (`Text.Pandoc.Options`): the type of the
`writerTopLevelDivision` field in of the `WriterOptions` data type is
altered from `Division` to `Maybe Division`. The field's default value
is changed from `Section` to `Nothing`.
Closes: #3197
Table column properties can optionally specify a column's width with
which it is displayed in the buffer. Some exporters, notably the ODT
exporter in org-mode v9.0, use these values to calculate relative column
widths. The org reader now implements the same behavior.
Note that the org-mode LaTeX and HTML exporters in Emacs don't support
this feature yet, which should be kept in mind by users who use the
column widths parameters.
Closes: #3246
This also fixes excessive CPU and memory usage for tables
when --columns is set in such a way that cells must be very
tiny.
Now cells are guaranteed to be big enough so that single
words don't need to line break, even if this pushes the
line length above the column width.
Closes#1911.
We can now parse all of the tables emitted by pandoc in
our tests.
The only thing we don't get yet are alignments and
column widths in more complex tables.
See #2669.
Reader can now parse simple LaTeX tables such as those
generated by pandoc itself.
We still can't handle pandoc multiline tables which involve
minipages and column widths.
Partially addresses #2669.
* Markdown reader: modify bracketedSpan to check small caps
* MANUAL.txt: add description on the use of `bracketed_spans` in small cap
* Improve markdown readers: bracketedSpan function EXACTLY as spanHtml
Previously a tight bullet sublist got rendered with
a blank line after, while a tight ordered sublist did
not. Now we don't get the blank line in either case.
Closes#3232.
We wrap `[CHART]` in a `<span class="chart">`. Note that it maps to
inlines because, in docx, anything in a drawing tag can be part of a
larger paragraph.
We not only want "w:drawing", because that could also include
charts. Now we specify "w:drawing"//"pic:pic". This shouldn't change
behavior at all, but it's a first step toward allowing other sorts of
drawing data as well.
Images which are the only element in a paragraph can still be given HTML
attributes, even if the image does not have a caption and is hence not a figure.
The following will add set the `width` attribute of the image to `50%`:
#+ATTR_HTML: :width 50%
[[file:image.jpg]]
Closes: #3222
When a piece of text has a text 'Source_Text' then
we assume that this is a piece of the document
that represents a code that needs to be inlined.
Addapted an odt writer to also reflect that change;
previously it was just writing a 'preformatted' text using
a non-distinguishable font style.
Code blocks are still not recognized by the ODT reader.
That's a separate issue.
...if citations extension disabled. Example: in
[link text][@a]
[@a]: url
`link text` isn't hyperlinked because `[@a]` is parsed as a citation.
Previously this happened whether or not the `citations` extension was
enabled. Now it happens only if the `citations` extension is enabled.
Closes#3209.
Previously, if given an empty namespace:
(elemName ns "" "foo")
`elemName` would output a QName with a `Just ""` namespace. This is
never what we want. Now we output a `Nothing`. If someone *does* want a
`Just ""` in the namespace, they can enter the QName value explicitly.
unless something else is explicitly specified in xmlns.
Provided it parses as MathML, of course.
Also fixed default which should be to inline math if no
display attribute is used.
ODT reader simply provided an empty header list
which meant that the contents of the whole table,
even if not empty, was simply ignored.
While we still do not infer headers we at least have
to provide default properties of columns.
Special blocks (i.e. blocks with unrecognized names) can be prefixed
with an `ATTR_HTML` block attribute. The attributes defined in that
meta-directive are added to the `Div` which is used to represent the
special block.
Closes: #3182
The `todo` export option allows to toggle the inclusion of TODO keywords
in the output. Setting this to `nil` causes TODO keywords to be dropped
from headlines. The default is to include the keywords.
Headlines can have optional todo-markers which can be controlled via the
`#+TODO`, `#+SEQ_TODO`, or `#+TYP_TODO` meta directive. Multiple such
directives can be given, each adding a new set of recognized todo-markers.
If no custom todo-markers are defined, the default `TODO` and `DONE`
markers are used.
Todo-markers are conceptually separate from headline text and are hence
excluded when autogenerating headline IDs.
The markers are rendered as spans and labelled with two classes: One
class is the markers name, the other signals the todo-state of the
marker (either `todo` or `done`).
Technically `**@user` is a valid email address, but if we
allow things like this, we get bad results in markdown flavors
that autolink raw email addresses. (See #2940.)
So we exclude a few valid email addresses in order to
avoid these more common bad cases.
Closes#2940.
This is needed because github flavored Markdown has a slightly
different set of escapable symbols than original Markdown;
it includes angle brackets.
Closes#2846.
The `--chapters` option is replaced with `--top-level-division` which allows
users to specify the type as which top-level headers should be output. Possible
values are `section` (the default), `chapter`, or `part`.
The formats LaTeX, ConTeXt, and Docbook allow `part` as top-level division, TEI
only allows to set the `type` attribute on `div` containers. The writers are
altered to respect this option in a sensible way.
@tarleb this is an interesting one, see the build log in
https://travis-ci.org/jgm/pandoc/jobs/168612017
It only failed on ghc 7.8; I think this must have to do with
the change making Monad a superclass of Applicative, hence
this change.
Frame can contain other frames with the text boxes.
This is something that has not been considered before
and meant that the whole construction of images was
broken in those cases. Also the captions were fixed/ignored.
RST requires a space before a footnote marker. We discard those spaces
so that footnotes will be adjacent to the text that comes before
it. This is in line with what rst2latex does. rst2html does not discard
the space, but its html output is different than pandoc's, so this seems
the most semantically correct approach.
Closes#3163
A `#+CAPTION` attribute before an image is enough to turn an image into a
figure. This wasn't the case because the `parseFromString` function, which
processes the caption value, would fail on empty values. Adding a newline
character to the caption value fixes this.
Fixes: #3161
Review revealed that we didn't handle the case
when the starting point is an empty string. While
this is not a valid .odt file, we simply added
a special case to deal with it.
Also added tests for the new feature.
This reverts commit 3f82471355.
We might want to revert the requirement of http-client 0.5,
as this is not yet in Stackage and that is starting to
cause problems. I can't recall why it is there.
Line blocks are allowed to contain empty lines and should be parsed as a
single block in that case. Previously an empty (line block) line would
have terminated parsing of the line block element.
Markup-features focusing on lines as distinctive part of the markup are read
into `LineBlock` elements. This currently means line blocks in reStructuredText
and Markdown (the latter only if the `line_block` extension is enabled), the
`linegroup`/`line` combination from the Docbook 5.1 working draft, and Org-mode
`VERSE` blocks.
The following markup features are used to output the lines of the `LineBlock`
element:
- AsciiDoc: a `[verse]` block,
- ConTeXt: text surrounded by `\startlines` and `\endlines`,
- HTML: `div` with an per-element style setting to interpret the content as
pre-wrapped,
- Markdown: line blocks if the `line_blocks` extension is enabled, a simple
paragraph with hard linebreaks otherwise,
- Org: VERSE block,
- RST: a line block, and
- all other formats: a paragraph, containing hard linebreaks between lines.
Custom lua writers should be updated to use the `LineBlock` element.
The `linesToBlock` function takes a list of lines and combines them by appending
a hard `LineBreak` to each line and concatenating the result, putting the result
it into a `Para`. This is most useful when dealing when converting `LineBlock`
elements.
Previously the starting value of the lists' items has been
hardcoded to 1. In reality ODT's list style definition can
provide a new starting value in one of its attributes.
Writers already handle the modified start value so no need
to change anything in that area.
This allows footnotes and refs to be placed at the end of blocks and
sections. Note that we only place them at the end of blocks that are at
the top level and before headers that are the top level. We add an
environment variable to keep track of this. Because we clear the
footnotes and refs when we use them, we also add a state variable to
keep track of the starting number.
Finally, note that we still add any remaining footnotes at the end. This
takes care of the final section, if we are placing at the end of a
section, and will always come after a final block as well.
Previously we'd emit raw HTML tables even if the `raw_html`
extension was disabled.
Now we just emit `[TABLE]` if no table formats are enabled
and raw HTML is not enabled.
We also check for the `raw_html` extension before emiting
a raw HTML block.
Closes#3154.
Now RTL is turned and off by a general function, `withDirection`
wrapping `inlineToOpenXML` and `blockToOpenXML`. This acts according to
the `envRTL` variable. This means we can just set the environment at the
outset, and change the environment with `local` as need be.
Note that this requires making the `inlineToOpenXML` and
`blockToOpenXML` functions into wrappers around
primed-versions (`{inline,block}ToOpenXML`) where the real work takes
place.
In general, we want things that are either:
1. unchanging environment variables, or
2. environment variables that will change for a the scope of
a function and then pop back
to be in the reader monad. This is safer for (1), since we won't
accidentally change it, and easier for (2), since we can use `local`
instad of setting the old value and then resetting.
We keep the StateT monad for values that we will want to accumulate or
change and then use later.
We had to use this because we set the env, which means that setRTL
wouldn't do anything at the top level. We now don't set the env (it will
always be false at the outset), which means the toplevel setRTL will
work if necessary.
This will allow us to add text and paragraph properties depending on if
rtl is already set or not.
(It would probably be cleaner and safer to move the paraprops and
textprops to this part of the stack in the future.)
In AsciiDoc, you must use a special form of emphasis (double `__`)
for intraword emphasis. Pandoc was previously using this more
than necessary.
Closes#3068.
Backticks in verbatim environments are converted to
open-single-quotes. This change makes them appear as backticks. This
corresponds to how we treat `'' in verbatim environments (with
\textquotesingle{}).
Add --parts command line argument.
This only effects LaTeX writer, and only for non-beamer output formats.
It changes the output levels so the top level is 'part', the next
'chapter' and then into sections.
This tests for a min value >= 0.5. But we have a lower bound of 0.5 in
pandoc.cabal, so the test will always pass.
(If we bump the lower bound to 0.5.1, we can remove a conditional in the
HTML writer as well.)
Because time 1.4 is a boot library for GHC 7.8, we will support the
compatibility module as long as we support 7.8. But we should be clear
about when we will no longer need it.
We already lower-bound tagsoup at 0.13.7, which means we were always
running the compatibility layer (it was conditional on min value
0.13). Better to just use `lookupEntity` from the library directly, and
convert a string to a char if need be.
Instead, emit the alt text, emphasized. This accords with what
the ODT writer currently does.
The user will still get a warning about a nonexistent image,
but will no longer get a LaTeX crash.
Closes#3100.
Sections the `unnumbered` property should, as the name implies, be
excluded from the automatic numbering of section provided by some output
formats. The Pandoc convention for this is to add an "unnumbered" class
to the header. The reader treats properties as key-value pairs per
default, so a special case is added to translate the above property to a
class instead.
Closes#3095.
The `creator` option controls whether the creator meta-field should be
included in the final markup. Setting `#+OPTIONS: creator:nil` will
drop the creator field from the final meta-data output.
Org-mode recognizes the special value `comment` for this field, causing
the creator to be included in a comment. This is difficult to translate
to Pandoc internals and is hence interpreted the same as other truish
values (i.e. the meta field is kept if it's present).
The `email` option controls whether the email meta-field should be
included in the final markup. Setting `#+OPTIONS: email:nil` will drop
the email field from the final meta-data output.
The `author` option controls whether the author should be included in
the final markup. Setting `#+OPTIONS: author:nil` will drop the author
from the final meta-data output.
HTML-specific head content can be defined in `#+HTML_head` lines. They
are parsed as format-specific inlines to ensure that they will only show
up in HTML output.
LaTeX-specific header commands can be defined in `#+LaTeX_header` lines.
They are parsed as format-specific inlines to ensure that they will only
show up in LaTeX output.
The last meta-line of any given type is the significant line.
Previously the value of the first line was kept, even if more lines of
the same type were encounterd.
Previously we only used the first anchor span to affect header ids. This
allows us to use all the anchor spans in a header, whether they're
nested or not.
Along with 62882f97, this closes#3088.
Previously we always generated an id for headers (since they wouldn't
bring one from Docx). Now we let it use an existing one if
possible. This should allow us to recurs through anchor spans.
Previously, we would only be able to figure out internal links to a
header in a docx if the anchor span was empty. We change that to read
the inlines out of the first anchor span in a header.
This still leaves another problem: what to do if there are multiple
anchor spans in a header. That will be addressed in a future commit.
Pandoc and Org-mode use different programming language identifiers. An
additional translation between those identifiers is added to avoid
unexpected behavior. This fixes a problem where language specific
source code would sometimes be output as example code.
Org-mode treats links as document internal searches unless the link
target looks like a URL or file path, either relative or absolute. This
change ensures that this is always the case.
An Org-mode figure should be surrounded by blank lines. The figure
would be recognized regardless, but images in the following line would
unintentionally be treated as figures as well.
This enables dynamic styling on spans. It uses the same prefix as we
used on divs ("docx-style" for the moment). It does not yet inject the
style into styles.xml.
The functions `isElem` and `elemName` (defined in Docx/Util.hs) make the
code a lot cleaner than the original XML.Light functions, but they had
been used inconsistently. This puts them in wherever applicable.
Image sources as those in plain images, image links, or figures, must be
proper URIs or relative file paths to be recognized as images. This
restriction is now enforced for all image sources.
This also fixes the reader's usage of uncleaned image sources, leading
to `file:` prefixes not being deleted from figure
images (e.g. `[[file:image.jpg]]` leading to a broken image `<img
src="file:image.jpg"/>)
Thanks to @bsag for noticing this bug.
Previously these yielded strings of alternating Code and Space
elements; we now incorporate the spaces into the Code. Emphasis
etc. is still possible inside these.
Closes#3055.
Previously an unquoted attribute value in a table row
could cause parsing problems.
Fixes#3053 (well, proper rowspans and colspans aren't
created, but that's a bigger limitation with the current
Pandoc document model for tables).
In the latex parser when includes are processed, the text of the
included file is directly included into the parse stream. This caused
problems when there was an error in the included file (and the included
file was longer than the original file) as the error would be reported
at this position.
The error handling tries to display the line and position where the
error occured. It works by including a copy of the input and finding the
place in the input when given the position of the error. In the
previously described scenario, the input file would be the original
source file but the error position would be the position of the error in
the included file.
The fix is to not try to show the exact line when it would cause an
out-of-bounds error.
The starred variants don't exist.
This helps with part of #3058...it gets rid of the spurious *s.
But we still have numbers on the 4th and 5th level headers.
Previously blockquotes were used. Now a Div is used
with class `admonition` and (if relevant) one of the
following: `attention`, `caution`, `danger`, `error`,
`hint`, `important`, `note`, `tip`, `warning`.
`sidebar` is also put into a Div.
Note: This will change rendering of RST documents!
It should provide much more flexibility.
Closes#3031.
We now handle cell and row attributes, mostly by skipping
them. However, alignments are now handled properly.
Since in pandoc alignment is per-column, not per-cell, we
try to devine column alignments from cell alignments.
Table captions are also now parsed, and textile indicators
for thead and tfoot no longer cause parse failure. (However,
a row designated as tfoot will just be a regular row in pandoc.)
Previously, we had used the user-supplied date, if available, for Word's
document creation metadata. This could lead to weird results, as in
cases where the user post-dates a document (so the modification might be
prior to the creation). Here we use the actual computer time to set the
document creation.
Previously we parsed a list of dates, took the first one, and then
tested its year range. That meant that if the first one failed, we
returned nothing, regardless of what the others did. Now we test for
sanity before running `msum` over the list of Maybe values. Anything
failing the test will be Nothing, so will not be a candidate.
We only allow years between 1601 and 9999, inclusive. The ISO 8601
actually says that years are supposed to start with 1583, but MS Word
only allows 1601-9999. This should stop corrupted word files if the date
is out of that range, or is parsed incorrectly.
We want to avoid illegal dates -- in particular years with greater than
four digits. We attempt to parse series of digits first as `%Y%m%d`, then
`%Y%m`, and finally `%Y`.
