haskell - I taught ghci to compile my StackOverflow posts. Can I make it slicker?

Question

Haskell Stack Overflow layout preprocessor

module StackOverflow where  -- yes, the source of this post compiles as is

Skip down to What to do to get it working if you want to play with this first (1/2 way down).
Skip down to What I would like if I witter on a bit and you just want to find out what help I'm seeking.

TLDR Question summary:

1. Can I get ghci to add filename completion to the :so command I defined in my ghci.conf?
2. Could I somehow define a ghci command that returns code for compilation instead of returning a ghci command, or does ghci instead have a better way for me to plug in Haskell code as a file-extension-specific pre-processor, so :l would work for .hs and .lhs files as usual, but use my handwritten preprocessor for .so files?

Background:

Haskell supports literate programming in .lhs source files, two ways:

• LaTeX style egin{code} and end{code}.
• Bird tracks: Code starts with > , anything else is a comment.
  There must be a blank line between code and comments (to stop trivial accidental misuse of >).

Don't Bird tracks rules sound similar to StackOverflow's code blocks?

References: 1. The .ghci manual 2. GHCi haskellwiki 3. Neil Mitchell blogs about :{ and :} in .ghci

The preprocessor

I like writing SO answers in a text editor, and I like to make a post that consists of code that works, but end up with comment blocks or >s that I have to edit out before posting, which is less fun.

So, I wrote myself a pre-processor.

• If I've pasted some ghci stuff in as a code block, it usually starts with * or :.
• If the line is completely blank, I don't want it treated as code, because otherwise I get accidental code-next-to-comment-line errors because I can't see the 4 spaces I accidentally left on an otherwise blank line.
• If the preceeding line was not code, this line shouldn't be either, so we can cope with StackOverflow's use of indentation for text layout purposes outside code blocks.

At first we don't know (I don't know) whether this line is code or text:

dunnoNow :: [String] -> [String]
dunnoNow [] = []
dunnoNow (line:lines)
  | all (==' ') line = line:dunnoNow lines     -- next line could be either
  | otherwise = let (first4,therest) = splitAt 4 line in 
     if first4 /="    "                 -- 
        || null therest                 -- so the next line won't ever crash
        || head therest `elem` "*:"     -- special chars that don't start lines of code.
     then line:knowNow False lines      -- this isn't code, so the next line isn't either
     else ('>':line):knowNow True lines -- this is code, add > and the next line has to be too

but if we know, we should keep in the same mode until we hit a blank line:

knowNow :: Bool -> [String] -> [String]
knowNow _ [] = []
knowNow itsCode (line:lines) 
  | all (==' ') line = line:dunnoNow lines
  | otherwise = (if itsCode then '>':line else line):knowNow itsCode lines

Getting ghci to use the preprocessor

Now we can take a module name, preprocess that file, and tell ghci to load it:

loadso :: String -> IO String
loadso fn = fmap (unlines.dunnoNow.lines) (readFile $ fn++".so") -- so2bird each line
        >>= writeFile (fn++"_so.lhs")                     -- write to a new file
        >> return (":def! rso (\_ -> return ":so "++ fn ++"")
:load "++fn++"_so.lhs")

I've used silently redefining the :rso command becuase my previous attemts to use let currentStackOverflowFile = .... or currentStackOverflowFile <- return ... didn't get me anywhere.

What to do to get it working

Now I need to put it in my ghci.conf file, i.e. in appdata/ghc/ghci.conf as per the instructions

:{
let dunnoNow [] = []
    dunnoNow (line:lines)
      | all (==' ') line = line:dunnoNow lines     -- next line could be either
      | otherwise = let (first4,therest) = splitAt 4 line in 
         if first4 /="    "                 -- 
            || null therest                 -- so the next line won't ever crash
            || head therest `elem` "*:"     -- special chars that don't start lines of code.
         then line:knowNow False lines      -- this isn't code, so the next line isn't either
         else ('>':line):knowNow True lines -- this is code, add > and the next line has to be too
    knowNow _ [] = []
    knowNow itsCode (line:lines) 
      | all (==' ') line = line:dunnoNow lines
      | otherwise = (if itsCode then '>':line else line):knowNow itsCode lines
    loadso fn = fmap (unlines.dunnoNow.lines) (readFile $ fn++".so") -- convert each line
        >>= writeFile (fn++"_so.lhs")                            -- write to a new file
        >> return (":def! rso (\_ -> return ":so "++ fn ++"")
:load "++fn++"_so.lhs")
:}
:def so loadso

Usage

Now I can save this entire post in LiterateSo.so and do lovely things in ghci like

*Prelude> :so StackOverflow
[1 of 1] Compiling StackOverflow    ( StackOverflow_so.lhs, interpreted )
Ok, modules loaded: StackOverflow.

*StackOverflow> :rso
[1 of 1] Compiling StackOverflow    ( StackOverflow_so.lhs, interpreted )
Ok, modules loaded: StackOverflow.

*StackOverflow>

Hooray!

What I would like:

I would prefer to enable ghci to support this more directly. It would be nice to get rid of the intermediate .lhs file.

Also, it seems ghci does filename completion starting at the shortest substring of :load that determines you're actually doing load, so using :lso instead of :so doesn't fool it.

(I would not like to rewrite my code in C. I also would not like to recompile ghci from source.)

TLDR Question reminder:

1. Can I get ghci to add filename completion to the :so command I defined in my ghci.conf?
2. Could I somehow define a ghci command that returns code for compilation instead of returning a ghci command, or does ghci instead have a better way for me to plug in Haskell code as a file-extension-specific pre-processor, so :l would work for .hs and .lhs files as usual, but use my handwritten preprocessor for .so files?

Answer 1

I would try to make a standalone preprocessor that runs SO preprocessing code or the standard literary preprocessor, depending on file extension. Then just use :set -pgmL SO-preprocessor in ghci.conf.

For the standard literary preprocessor, run the unlit program, or use Distribution.Simple.PreProcess.Unlit.

This way, :load and filename completion just work normally.

GHCI passes 4 arguments to the preprocessor, in order: -h, the label, the source file name, and the destination file name. The preprocessor should read the source and write to the destination. The label is used to output #line pragmas. You can ignore it if you don't alter the line count of the source (i.e. replace "comment" lines with -- comments or blank lines).