No one asks for more blogging software

Why I wrote this program.

2020-05-04 09:00AM PDTJohn

But I did want to write some posts. I wanted to finally host my own site and show my own work.

I was begged to use a common static site generator or repository integration. But I thought: blogs can be more transparent to the author. Writing a blog need not imply submitting markup to an opaque binary program - with one big caveat. Theming and styling can be transparent and alive to the author. Hosting and runtime, similarly, might be closer to the author's fingertips.

Moreover: the humble blog presents an opportunity to bring a problem into a language.

Gabriel Gonzalez says the following. Also - note the lack of quote formatting, I'll come back to that.

Your average developer will struggle with things not built into their programming language,
 including operations, security, packaging, and distributed systems

Many engineering problems are actually programming language problems in disguise

Why bring a problem into a language? Good question. Sounds like a good blog post, journal article even.

So ignoring any justification for the language-oriented approach, I set out with two design goals for this blog. I have definitely not met these goals yet. What follows is my experience so far - what went well and what could use improvement.

Goal 1 - Make as many tasks as possible language constructs

If I can write blog posts in plain text or a markup language, great! But, those posts should be managed by the language. I do not want to have to do so much extra work to validate, load, and ensure their existence in a static directory at runtime. For example if posts were written in a markdown format I would not want to place them in a directory and let blog software parse and render them at runtime. Ideally, the blog post should be as much a language construct as possible. I would love if styling, spell checking, comments, and deployment could all be expressed in this language.

Disclaimers 1 and 2:

1. Racket is a language oriented programming language that is elegant and wonderful and everyone should totally try it one day. I like types and I am already familiar with Haskell, though, so I chose it.

2. I am using Template Haskell (a library for manipulating Haskell source code). Template Haskell is Badâ„¢ because it is slow to compile, not type-safe and more. It also elegantly brings more things into the language that I want. Admittedly, language oriented design makes the most sense when the language has simple and proven constructs (like lambda calculus). Again, that's a different blog post. This is my first blog post. If Template Haskell does not work, I will try something else.

One big caveat

"Submitting a markup language to an opaque binary" is exactly what one does when sending source code through a compiler.

Early success

Template Haskell, file-embed and org-mode give me hooks to extend Haskell with org files. At compile time I parse each org file into a valid Post. This grants me extra parsing that can be hooked into the Haskell compiler. If I leave out the #+description metadata, for instance, the post won't parse. Then the Show instance for DecodeError lets me write my own error messages.

embedPosts :: FilePath -> Q Exp
embedPosts fp = do
  typ <- [t|[(FilePath, Post)]|]
  orgFiles <- either (fail . show) pure =<< runIO (postsDir fp)
  traverse_ (addDependentFile . (\p -> fp <> "/" <> p) . fst) orgFiles
  e <- ListE <$> traverse lift orgFiles
  pure $ SigE e typ

postsDir :: FilePath -> IO (Either DecodeError [(FilePath, Post)])
postsDir fp = do
  files <- getDir fp
  pure $ traverse (traverse decodePost) files

decodePost :: ByteString -> Either DecodeError Post
decodePost bs = do
  txt <- first UnicodeError $ decodeUtf8' bs
  org <- maybe (Left ImproperOrgFile) pure $ org txt
  first ImproperPost $ orgToPost org

embedPosts is a only a slight modification of embedDir from file-embed. There are almost certainly some issues with the above code. I'm not sure addDependentFile correctly registers the org files as dependencies, for instance.

There are a couple immediate benefits here. To check my post is formatted the way I want, I just recompile the program. The language user specifies the posts directory in the program as opposed to a runtime flag. At some point this could cause problems but for one contributor and one post it will do.

The Haskell org-mode library even has a library to generate Html: org-mode-lucid. One of my newfound problems as a compiler writer has been solved for me.

Possible improvements

The org file parsing library is incomplete, as is to be expected. Org mode has many features including spreadsheets, agendas, and literate programming. One simple fix I will look into submitting will be to add parsing for quote blocks. They are currently missing which is why I used a code block for the quote above.

Syntax highlighting and fonts are currently provided outside of the blog language. There are some haskell libraries for syntax highlighting but I need to look more closely into them. CSS, likewise is put in a separate file. While it might be nice to have separate CSS files, I would love if the stylesheets were also parsed by GHC like posts. The same issues also go for fonts.

Goal 2 - Be polymorphic in runtime and hosting

Early failure

I wrote a whole library to run in a popular function as a service environment. Implementing the runtime for the environment went smoothly. After the initial elation over implementing the runtime, I found I would have to implement a webserver on top of it. I searched - maybe incompletely - for WAI (the Web Application Interface) implementations that might shed some light on my situation. That search seemed to justify my desire for polymorphism over runtime. Primary implementations of WAI involve very concrete socket management. Making an alternative would probably require more work from the ground up.

I moved the function as a service implementation aside and decided to use a more standard web server runtime. This version allowed me to get up and going quickly, with many of the same benefits as the FAAS. The downsides to both of these current solutions is that they are neither transparent to the author nor easily integrated into a language.

Clear polymorphism wins

The Servant Haskell library provides a language to express a web api. looks like this:

type Club =
  Get '[HTML] About
    :<|> "posts" :> Get '[HTML] AllPosts
    :<|> "post" :> Capture "slug" Text :> Get '[HTML] Post
    :<|> Raw

No references to sockets (or servers!) in sight. One Servant API definition specifies a server, client, documentation, or more. Plus, Servant provides a standard function to turn your API type into a WAI application.

Possible improvements

Servant's general use case is json APIs, even though it has library support for html, xml, websockets and more. There exists at least one library to generate api definition formats (like Swagger) from a Servant api. I would love to generate a sitemap for an api, too.

A functional package manager could play a role in the future of the blog. Most functional package managers express polymorphism in runtime quite nicely, have deployment options, and have extension languages. The problem is that languages aside from their official extension language are not yet supported. If operations is to be brought into the blog language, more tooling will be necessary.

A more achievable near-term goal would just be more WAI implementations. Just one lightweight implementation that wasn't "industrial grade" would do a lot to help my polymorphism story.


Ideally the kernel of this blogging language would be simple and proven. Though there is probably nothing as elegant as lambda calculus underpinning the org format or a blog post. For this reason I find Template Haskell fits my use case very well. Template Haskell has so far allowed me to extend GHC with whatever nonesense I saw fit without bringing any of my nonesense into GHC. The common refrain from various lisp communities is that the language should be as simple as possible but extendable by the user and I find Template Haskell suits that goal nicely.

Recent languages try to express distributed systems in one concrete syntax. Some new languages (like darklang) seem like Template Haskell to the lambda calculus of Unison and Erlang. One of my all-time favorite languages - Ur/Web - provides one (or two) languages to express full web applications complete with databases and remote calls between javascript and server. The more difficult challenges to writing the blog language will also be distribution and operations. Instead of creating a large software stack for the purpose, I would prefer smaller pieces. For instance: a WAI implementation with few dependencies that makes few assumptions would be a welcome starting point.

Hosting will also be a problem for the blog language. The various platforms I have tried do not allow expression in a language with the exception of yaml. I will need to research how to embed hosting options since I have mostly been focused on blog posts and starting up. The other problem with hosting options is they are neither simple, proven, or transparent to the blogger.

With those challenges in mind I have really enjoyed working on this blog software. It has been a good set of yaks to shave. Plus I feel like I can make digestible improvements to existing libraries. And of course, I am again reminded of how flexible GHC Haskell is.

Full source from snippets:, commit 6dbe644ad7b06746cb62885a74cf1ba8ba395e6b