This page is meant to gather ideas people have for Yi, an extensible editor written in Haskell.
Coming from an Emacs background, the current version of Yi lacks a few things I think are essential, mainly the introspection capabilities of Emacs. One of the main problems is that Yi is based on purely compiled code --- there is little or no interaction with the run-time system.
Ideally, the next version of Yi would be based on a (modified?) version of GHCi, maybe taking advantage of package GHC.
The following are things I like about Emacs, as an extensible environment:
- Really good online documentation
- Emacs can tell you a lot about a function or variable with a keypress--- the current value, where it is declared, and a hypertext formation string
- All (good) apps allow users to extend, through, e.g., hooks --- a list of functions that are run before/after some event (like saving a file)
- It is really easy in Emacs to have one package interact with another. Thus, I can, e.g., insert a new appointment from my mail app into the diary.
- Everything is One Language
- Ignoring the actual language (Lisp!), everything is handled in a uniform language --- from binding keys to writing apps.
- Easy to start hacking
- I can start playing with the system from the second I start up, and things pretty much work as expected. I.e., I can type a bit of code in, execute it, and the result is displayed in the minibuffer. The good docs help immeasurably.
- Written for the frequent user
- Lots of key shortcuts (and famous for it). There are still menus, for those who like em, but you aren't forced to pretend you just started using it.
- A tonne of code
- Well, Haskell has this to some degree. Haskell is (IMHO) much easier to write than ELisp, so maybe people will be encouraged to contribute.
So, why replace it?:
- Dynamically scoped, Dynamically typed, ugly, old. 'Nuff said
- What's a Parser?
- A lot of apps in emacs do stuff with text, usually text that is in some language. There is no standard parser (like, e.g. parsec), so a lot of it is ugly handwritten spaghetti. This also means that adding analysis tools isn't really done (or done nicely).
- ELisp again
- Haskell is a lot cleaner to write, especially because of the large number of libraries.
Emacs maybeness (?)
Some things that are sometimes bad, sometimes good:
- Everything is a buffer
- Makes some sense, but sometimes doesn't. It is nice to have uniform key bindings do the right thing (e.g., C-Space sets the mark, and the region can then be used, e.g. to delete a sequence of emails in Wl) Sometimes, however, you just want some sort of GUI widget.
- OTOH, having the minibuffer be a special kind of buffer is a good idea.
- It is possible to associate arbitrary properties with symbols. This means you can annotate a symbol and then use that information at a later date
What about vi? I believe we want Yi to subsume vi as well.
Annoying stuff with the current implementation:
- Does not support all terminals properly (e.g. Meta on gnome-terminal)
- Is quite dependent on terminal (gtk port not easy)
- System of windows is strange. A lot of the cursor-logic is spread in many modules
- Throwing exception for switching keymap is ugly
- Yi should include GHCi
- like emacs includes a elisp interpreter.
- An extension to GHCi to support documentation of symbols.
- This seems to be (reasonably) straightforward, as GHCi already has :info. It would mean hacking the type environment (what about values?) to add documentation information. The main problem would seem to be populating this --- maybe hack haddock to produce something from the library docs? I assume that using package GHC uses the parent RTS (package GHC seems to be the way to go, but more investigation is required --- don?)
- Intermixed compiled/interpreted code
- (for speed/hacking)
- GUI abstraction
- want it to work on terminals as well as X
- Views on data
- Rather than just editing a file, you would open a view onto the file, i.e. there is no longer a 1-1 correspondence between buffers and files. Why? Well, for aggregate buffers (i.e., editing multiple files in the one view), or for multiple views of a file (e.g. AST and source-level). There would be some primitive ops for editing a buffer (insertChar, delete, etc.), which would then call update functions on anything observing that file.
- Remote attach so I can work from home, but still use a remote machine
- Haddock documentation
- (no brainer), maybe associate with .hi files for binaries.
- A class MiniBufferRead (or PromptingRead) which allows the user to
invoke a function similar to M-x in Emacs, but without requiring (interactive)
- -- This is incomprehensibe, would the original author elaborate?
- Maybe a class YiShow, which all config items must be a member of? This is to emulate describe-variable
- Per mode/file/buffer/whatever Monads, or reload/recompile? Or some hybrid? How does this interact with the documentation aspects? Do we want to have separate sorts of symbols a la emacs (describe-function, describe-variable), or is everything a function? I would think that configuration info doesn't change that frequently --- is this globally true though?
- We can probably use a GHCi-like "let". Rebinding a function would then be synonym to assign a variable, thereby achieve unification between functions and variables.
- Interface to the runtime
- The scheduler, docs, etc.
- Introspection of e.g. what processes are running.
- There are already libraries in Haskell for processes, but they don't give Yi any extra information --- we really want a layer on top.
Sjw 09:15, 2 June 2006 (UTC)