If deferred media can't be fetched, we catch the error and warn
instead. We add an internal function for fetching which returns a Maybe
value, and then run catMaybes to only keep the Just's.
Move the downloading/reading-in logic out of fetchItem, so we can use it
to fill the MediaBag. Now when other modules use `fetchItem` it will
fill the MediaBag as expected.
The DeferredMediaBag is now the object that is held in state. It should
not be visible to users, who will still deal with MediaBag through
exported getters and setters.
We now have a function `fetchDeferredMedia` which returns () but
downloads/reads in all of the deferred media.
Note that getMediaBag first fetches all deferred media.
This is a lazy MediaBag, that will only be evaluated (downloaded/read
in) upon demand.
Note that we use fetchItem in getDefferedMedia at the moment to read
in/download. This means that we don't need to distinguish between URIs
and FilePaths. But there is an inefficiency here: `fetchItem` will pull
an item out of the mediaBag if it's already there, and then we'll
reinsert it. We could separate out `fetchItem` into the function that
checks the MediaBag and the underlying downloader/read-inner.
Removed fetchItem and fetchItem'.
Provide fetchItem in PandocMonad (it uses openURL and readFileStrict).
TODO:
- PandocPure instance for openURL.
- Fix places where fetchItem is used so that we trap the
exception instead of checking for a Left value. (At least
in the places where we want a warning rather than a failure.)
* Removed normalize, normalizeInlines, normalizeBlocks
from Text.Pandoc.Shared. These shouldn't now be necessary,
since normalization is handled automatically by the Builder
monoid instance.
* Remove `--normalize` command-line option.
* Don't use normalize in tests.
* A few revisions to readers so they work well without normalize.
* Text.Pandoc.Options.WriterOptions: removed writerReferenceDocx
and writerReferenceODT, replaced them with writerReferenceDoc.
This can hold either an ODT or a Docx. In this way, writerReferenceDoc
is like writerTemplate, which can hold templates of different
formats. [API change]
* Removed `--reference-docx` and `--reference-odt` options.
* Added `--reference-doc` option.
Signature of pipeProcess has changed: the return value is
now IO (ExitCode, ByteString) -- with only stdout. Stderr
is just inherited from the parent.
This means that stderr from filters will now be streamed
as the filters are run.
Closes#2729.
There are two states in PandocPure, but it is only easy to deal with
CommonState. In the past, to do state monad operations on
PureState (the state specific to PandocPure) you had to add (lift
. lift) to the monadic operation and then rewrap in the newtype. This
adds four functions ({get,gets,put,modify}PureState) corresponding to
normal state monad operations. This allows the user to modify
PureState in PandocPure without worrying about where it sits in the
monad stack or rewrapping the newtype.
It now lives in IO and gives a proper message + exit
instead of calling 'error'.
We shouldn't be making it easier for people to raise error on
pure code. And this is better for the main application
in IO.
* Removed handleIncludes from LaTeX reader [API change].
* Now the ordinary LaTeX reader handles includes in a way
that is appropriate to the monad it is run in.
