Some word functions -- especially graphics -- give various choices for
content so there can be backwards compatibility. This follows the
largely undocumented feature by working through the choices until we
find one that works.
Note that we had to split out the processing of child elems of runs into
a separate function so we can recurse properly. Any processing of an
element *within* a run (other than a plain run) should go into
`childElemToRun`.
Traditionally pandoc operates on multiple files by first concetenating
them (around extra line breaks) and then processing the joined file. So
it only parses a multi-file document at the document scope. This has the
benefit that footnotes and links can be in different files, but it also
introduces a couple of difficulties:
- it is difficult to join files with footnotes without some sort of
preprocessing, which makes it difficult to write academic documents
in small pieces.
- it makes it impossible to process multiple binary input files, which
can't be catted.
- it makes it impossible to process files from different input
formats.
This commit introduces alternative method. Instead of catting the files
first, it parses the files first, and then combines the parsed
output. This makes it impossible to have links across multiple files,
and auto-identified headers won't work correctly if headers in multiple
files have the same name. On the other hand, footnotes across multiple
files will work correctly and will allow more freedom for input formats.
Since ByteStringReaders can currently only read one binary file, and
will ignore subsequent files, we also changes the behavior to
automatically parse before combining if using the ByteStringReader. If
we use one file, it will work as normal. If there is more than one file
it will combine them after parsing (assuming that the format is the
same).
Note that this is intended to be an optional method, defaulting to
off. Turn it on with `--file-scope`.
In order to be able to collect warnings during parsing, we add a state
monad transformer to the D monad. At the moment, this only includes a
list of warning strings (nothing currently triggers them, however). We
use StateT instead of WriterT to correspond more closely with the
warnings behavior in T.P.Parsing.
The docx reader used to use a Modifiable typeclass to combine both
Blocks and Inlines. But all the work was in the inlines. So most of the
generality was wasted, at the expense of making the code harder to
understand. This gets rid of the generality, and adds functions for
Blocks and Inlines. It should be a bit easier to work with going forward.
We currently treat all memoir templates as books. This means that pandoc
will infer the `--chapters` argument, even if the `article` iption is
set for memoir.
This commit makes pandoc treats the document as an article if there is
an article option (i.e., `\documentclass[12pt,article]{memoir}`).
Note that this refactors out the parsec parsers for document class and
options, to make it a little clearer what's going on.
The line causes an error with stack 1.0.2:
```
Could not parse '/pandoc-build/pandoc/stack.full.yaml':
AesonException "Error in $.extra-deps: failed to parse field 'extra-deps': failed to parse field extra-deps: expected [a], encountered Null"
See http://docs.haskellstack.org/en/stable/yaml_configuration.html.
```
