See #2252.
This also changes fixDisplayMath from Text.Pandoc.Writers.Shared
so that it no longer produces empty Para as an artifact.
(That was the original reason the writer omitted them.)
We now have the `--strip-empty-paragraphs` option for that,
if you want it. Closes#2252.
Updated docx reader tests.
We use stripEmptyParagraphs to avoid changing too
many tests. We should add new tests for empty paragraphs.
Previously both needed to be specified (unless the image was
being resized to be smaller than its original size).
If height but not width is specified, we now set width to
textwidth (and similarly if width but not height is specified).
Since we have keepaspectratio, this yields the desired result.
Pandoc and Meta elements are now pushed by calling the respective
constructor functions of the pandoc Lua module. This makes serialization
consistent with the way blocks and inlines are pushed to lua and allows
to use List methods with the `blocks` value.
The List module is automatically loaded, but not assigned to a global
variable. It can be included in filters by calling `List = require
'List'`.
Lists of blocks, lists of inlines, and lists of classes are now given
`List` as a metatable, making working with them more convenient. E.g.,
it is now possible to concatenate lists of inlines using Lua's
concatenation operator `..` (requires at least one of the operants to
have `List` as a metatable):
function Emph (emph)
local s = {pandoc.Space(), pandoc.Str 'emphasized'}
return pandoc.Span(emph.content .. s)
end
Closes: #4081
This generates a JSON version of a highlighting style, which can be
saved as a .theme file, modified, and used with `--highlight-style`.
Closes#4106.
Closes#4096.
The `Generic` JSON instances for `Text.Pandoc.App.Opt` seem to tickle a
particulary bad quadratic complexity case (Generics complexity is worse
than quadratic with respect to the number of fields in the datatype).
This is with GHC-8.2.1, I didn't test it using 8.0 but I assume it is
similar.
Using `Generic`, compilation of the `Text.Pandoc.App` module takes
minutes and often gets killed due to out of memory on slower machines
with "only" 8GB of accessible memory. This is particularly annoying to
me since it means I cannot build pandoc on Travis.
TemplateHaskell is a little uglier, but the module seems to compile
within a few seconds, and compilation doesn't take more than 1GB of
memory.
Should I also change the other JSON instances throughout the codebase
for consistency?
