Generating this website part 7

Slideshows

22 January, 2016

This is part seven of the “generating this website” series. To read the rest of the series, go to the series index here

Occasionally I give talks for which I want to write up an accompanying blog post. Since Pandoc supports various slideshow formats directly, I can actually use the same system both for my blog posts and for writing my slides – in fact, I can use the same source to produce all three: the post, the slides, and any accompanying code – using literate programming!

Of course, this won’t always work. Sometimes, the way you present things in a talk must flow quite differently from the way you present them in a blog post. But this is a limitation of the presentation media, not a technical limitation, and so I’d like to support both formats. Here’s an outline of what we’re going to try and achieve:

All in all, this set of requirements should be fairly simple; the only potentially tricky part is searching for slides using the posts’ metadata rather than their filename, and even that should just come down to quitting out of the compiler early in case the slides field isn’t present.

Preliminaries

As always, we begin by specifying OverloadedStrings and importing Hakyll.

> {-# LANGUAGE OverloadedStrings #-}
> {-# LANGUAGE UnicodeSyntax #-}
> module Slides where
> import Control.Monad.Unicode
> import Hakyll

Monoid’s mappend operator is also useful to have.

> import Data.Monoid ((<>))

We’ll be making use of a custom pandoc compiler to actually output the slides, so we’ll need to bring the appropriate pandoc options into scope.

> import Text.Pandoc.Options (
WriterOptions (..),
>                             
HTMLSlideVariant(..))

I’m also going to import the Posts module as the compiler I use for generating slideshows will be similar to the one used for generating standard blog posts, so I’ll want to reuse that. Check out the post on posts for more information on what’s in this module.

> import Posts

Locating slideshow files

The first step is to create some Hakyll Rules specifying which files to match and where to put them. The requirements outlined above essentially specify two alternatives:

The easiest of the two is the latter – just match the files and compile them.

> slideshows :: Rules ()
> slideshows = match "slides/*" $ do
>   route $ setExtension ".html"
>   compileSlideshow

Picking out those posts which contain a slides metadata field requires the matchMetadata function which was introduced in hakyll 4.6.4.0. We use version to keep slides distinct from normal posts, allowing both to be built and preventing slideshows from being indexed. After that, we compile the slideshow as above.

> slideshowPosts :: Rules ()
> slideshowPosts =
>   matchMetadata "posts/*" isSlideshow . version "slideshow" $
>   do 
route $ setExtension ".html"
>      
compileSlideshow
>   where
>     isSlideshow = maybe False (const True) . lookupString "slides"

The compileSlideshow rule extracts the slides metadata in order to decide which type of slideshow to generate, then passes it to slidesCompiler.

> compileSlideshow :: Rules ()
> compileSlideshow =
>   compile $ getUnderlying
>= (`getMetadataField` "slides")
>= slidesCompiler . maybe SlidySlides read
>= relativizeUrls

Generating the slideshows

Now that we know which files we should be creating slideshows for, we’re ready to do the actual generation! As usual, we start with the context, which is very simple – just attach a date to the default context.

> slidesCtx :: Context String
> slidesCtx 
= 
dateField "date" "%e %B, %Y"
>          
<> 
defaultContext

Pandoc supports a variety of different slideshow formats, and I haven’t settled on one in particular, so I’ll add support for them individually as I try them out. Each slideshow engine requires a different template, so we’ll make a quick lookup for that.

> slidesTemplate :: HTMLSlideVariant  Identifier
> slidesTemplate RevealJsSlides 
= "templates/slides/reveal.js.html"
> slidesTemplate S5Slides       
= "templates/slides/s5.html"
> slidesTemplate SlidySlides    
= "templates/slides/slidy.html"
> slidesTemplate s              
= error $ "Unsupported slide variant: " ++ show s

The final clause covers the case where I haven’t yet added support for one of Pandoc’s supported slideshow formats.

Another difference between the various HTML slideshow engines is which HTML they expect to be working with! Reveal.js expects HTML 5, while the others I’ve tried all work better with HTML 4 source. I’ve made a quick lookup for that as well.

> slidesExpectHTML5 :: HTMLSlideVariant  Bool
> slidesExpectHTML5 RevealJsSlides 
= True
> slidesExpectHTML5 _              
= False

Finally, the compiler itself! This takes a standard pandoc compiler, adds the readerOptions and writerOptions we defined in the entry on standard posts, and then customizes them with slideshow-specific functionality.

> slidesCompiler :: HTMLSlideVariant  Compiler (Item String)
> slidesCompiler sv =
>   pandocCompilerWith readerOptions slideWriterOpts
>= loadAndApplyTemplate (slidesTemplate sv) slidesCtx
>   where
>     slideWriterOpts 
= writerOptions
>                     
{ writerSlideVariant 
= sv
>                     
, writerHtml5        
= slidesExpectHTML5 sv
>                     
, writerIncremental  
= True
>                     
}

Added extras

That’s really all there is to it on the Hakyll side! To make writing slides and blog posts in the same file easier, I have a couple of CSS div classes set up in the slideshow style files – for instance, anything inside a <div class="notes"> will be shown in the blog post but not in the slideshow. I have yet to add the reverse functionality but I can’t imagine it would be too difficult. All that can be done with a bit of CSS cleverness, though, with no special support from Hakyll itself.