Haskell Interactive Mode Architecture

This page is about the architecture of Haskell Interactive Mode, which should be helpful for curious users and prospective hackers.

How it works

Your Emacs will contain multiple sessions: one for each Haskell project you're working on. Each session is associated with a connected GHCi process and a REPL buffer.

             Emacs
            /     \
          /         \
    Session          Session
   /       \         /     \
Process  REPL     Process   REPL

When you open a Haskell file in a fresh project directory, it will prompt to start a new session for the enclosing directory containing a .cabal file. It will also start a GHCi process and open up an associated REPL buffer.

Command Protocol

There is a command queue which all communication with the GHCi process must go through:

Module 1 -> |*|
            | |
Module 2 -> |*|
            \_/---------> Process
                            |
Module 3  <--------*--------'

Only one command can run at once, anything else is queued up.

Code architecture

This mode consists of a few modules:

• Sessions: haskell-session.el
• Processes: haskell-process.el
• REPL: haskell-interactive-mode.el

REPL Interaction

The prompt is set to EOT (end-of-transmission), which makes it easier to match when a command is completed and avoid interspersing with a prompt like Prelude> . The following interaction

λ> 1
1
λ>

In the background (and you can view this in the buffer *haskell-process-log*) the communication is actually:

-> "1
"
<- "1
"
<- ""

^D is the notation Emacs uses to refer to codepoint \4. M-x describe-char shows:

Character code properties: customize what to show
  name: <control>
  old-name: END OF TRANSMISSION
  general-category: Cc (Other, Control)
  decomposition: (4) ('')

So, when you're looking in the log, that's what that character is.