Reasons:
- Performance: HsYAML is around 20 times slower in parsing
large YAML bibliographies (#6084).
- An issue was submitted to HsYAML, but it hasn't gotten
any attention. HsYAML seems borderline unmaintained; it hasn't
had a commit in over a year.
- Unfortunately this goes back on our attempts to free ourselves
from C dependencies (#4535). But I don't see a better alternative
until a better pure Haskell parser is available.
Closes#6084.
Notes:
- We've removed the FromYAML instances for all types that had
them, since this is a HsYAML-specific typeclass [API change].
(The yaml package just uses From/ToJSON.)
- Unlike HsYAML (in the configuration we were using), yaml
parses 'Y', 'N', 'Yes', 'No', 'On', 'Off' as boolean values.
Users may need to quote these when they are meant to be
interpreted as strings. Similarly, 'null' is parsed as
a YAML null value (and will be treated as an empty string
by pandoc rather than the string 'null'). Quoting it will
force it to be interpreted as a string.
- Some tests had to be adjusted accordingly.
- Pandoc now behaves better when the YAML metadata contains
escaping errors: instead of just falling back on treating
the section as a table, it raises a YAML parsing error.
Properties of Block values are marshalled lazily, which generally
improves performance considerably. Script users may also notice the
following differences:
- Block element properties can no longer be accessed by numerical
indexing of the `.c` field. The `.c` property now serves as an alias
for `.content`, so some filter that used this undocumented method
for property access may continue to work, while others will need to
be updated and use proper property names.
- The marshalled Block elements now have a `show` method, and a
`__tostring` metamethod. Both return the Haskell string
representation of the element.
- Block values now have the Lua type `userdata` instead of `table`.
- Adds a new `pandoc.AttributeList()` constructor, which creates the
associative attribute list that is used as the third component of
`Attr` values. Values of this type can often be passed to constructors
instead of `Attr` values.
- `AttributeList` values can no longer be indexed numerically.
The new HsLua version takes a somewhat different approach to marshalling
and unmarshalling, relying less on typeclasses and more on specialized
types. This allows for better performance and improved error messages.
Furthermore, new abstractions allow to document the code and exposed
functions.
Previously pandoc would parse
[link to (@a)](url)
as a citation; similarly
[(@a)]{#ident}
This is undesirable. One should be able to use example references
in citations, and even if `@a` is not defined as an example
reference, `[@a](url)` should be a link containing an author-in-text
citation rather than a normal citation followed by literal `(url)`.
Closes#7632.
Some fields only have an instrText and no content, Pandoc didn't
understand these, causing other fields to be misunderstood because it
seemed like a field was still open when it wasn't.
Fields delimited by fldChar elements can contain other fields. Before,
the nested fields would be ignored, except for the end, which would be
considered the end of the parent field.
To fix this issue, fields needed to be considered containing ParParts
instead of Runs, since a Run can't represent complex enough structures.
This also impacted Hyperlinks since they can originate from a field.
This commit changes the `marL` and `indent` values used for plain
paragraphs and numbered lists, and changes the spacing defined in the
reference doc master for bulleted lists.
For paragraphs, there is now a left-indent taken from the `otherStyle`
in the master. For numbered lists, the number is positioned where the
text would be if this were a plain paragraph, and the text is indented
to the next level. This means that continuation paragraphs line up
nicely with numbered lists.
It also /mostly/ matches the observed PowerPoint behaviour when
inserting paragraphs and numbered lists: the only difference is that
PowerPoint was using a different margin value for the first level
numbered lists – I’ve changed this to match the other levels, as I don’t
think it makes the spacing unappealing and it allows continuation
paragraphs at any level to line up.
With bulleted lists, I’m keeping the observed PowerPoint behaviour of
specifying only a level, letting `marL` and `indent` be automatically
taken from `bodyStyle`. To that end, this commit changes the `bodyStyle`
spacing in the master of the default reference doc, to:
- line up the text of the first paragraph in each bullet with any
continuation paragraphs
- line up nested bullet markers in any continuation paragraphs with the
first paragraph, matching lists and plain paragraphs
This does mean the continuation paragraphs still won’t line up for
anyone using their own reference doc where they haven’t matched the
`otherStyle` and `bodyStyle` indent levels, but I think people in that
situation will be able to troubleshoot.
In PowerPoint, the content of a top-level list is at the same level as
the content of a top-level paragraph – the only difference is that a
list style has been applied.
At the moment, the pptx writer increments the paragraph level on each
list, turning what should be top-level lists into second-level lists.
This commit changes that logic, only incrementing the paragraph level on
continuation paragraphs of lists.
- Fixes https://github.com/jgm/pandoc/issues/4828
- Fixes https://github.com/jgm/pandoc/issues/4663
This fixes a regression in #7604, which modernized
babel usage but omitted to load babel for pdflatex,
with the result that even simple documents could no
longer be produced.
Closes#7627.
AsciiDoctor allows to request line numbering on code blocks by
using a switch on the `source` block, such as in:
```
[source%linesnum,haskell]
----
some Haskell code here
----
```
When a paragraph has an indentation different from the parent (named)
style, it used to be considered a blockquote. But this only makes sense
when the paragraph has more indentation. So this commit adds a check
for the indentation of the parent style.
When I added the tests for moved layouts and deleted layouts, I added
them to all tests. However, this doesn’t really give a lot more info
than having single tests, and the extra tests take up time and disk
space.
This commit removes the moved-layouts and deleted-layouts tests, in
favour of a single test for each of those scenarios.
Update tests.
Reason: it turns out that the native output generated by
pretty-simple isn't always readable by the native reader.
According to https://github.com/cdepillabout/pretty-simple/issues/99
it is not a design goal of the library that the rendered values
be readable using 'read'. This makes it unsuitable for our
purposes.
pretty-show is a bit slower and it uses 4-space indents
(non-configurable), but it doesn't have this serious drawback.
Previously we used our own homespun formatting. But this
produces over-long lines that aren't ideal for diffs in tests.
Easier to use something off-the-shelf and standard.
Closes#7580.
Performance is slower by about a factor of 10, but this isn't
really a problem because native isn't suitable as a serialization
format. (For serialization you should use json, because the reader
is so much faster than native.)
Albert Krewinkel