[stcredzero]

I would say eclipse already does what light table aims to do.

Can you come to a realization in the middle of an Eclipse debugging session, get a great idea for making yourself more productive, take time to script yourself a working and fully debugged extension of Eclipse and continue what you were doing 20 minutes later?

I've done precisely this in Smalltalk on many occasions. You can script something like a custom browser that shows you everything that uses function A and also function B and has a reference to class C (and with complete syntactic accuracy) in around that time.

I suspect that this kind of very small grained toolsmithing is common for really good programmers in dynamic environments. The only problem is that it's been harder than necessary to integrate that with the most commonly used coding tool. (TextMate's ability to export selections to scripts is very nice, but it can go farther than that. Emacs is very powerful for this, but not everyone wants to script in elisp.)

It's not that Light Table is going to do something unprecedented. It's more like what git did with quick access to diffs: a quantitative change can become a qualitative change. Change the cost-benefit of toolsmithing, and you can use it in new ways.

[catenate]

Acme supports constantly improving how you work in little ways, better than tools with plugins or extension languages, by not having many features, and providing a good interface to route text through shell pipelines.

For a simple example, Acme has no paragraph justifier, but it doesn't need one, since fmt exists. To use it, highlight a paragraph of text with the left mouse button, highlight a typed-in |fmt command with the middle mouse button, and release the middle mouse button to run the paragraph through |fmt and replace it in place.

With this I can leverage anything I can call from the command line, in any language, to be part of my editor. No need to rewrite tools in the program's specific extension language, whether Elisp or Smalltalk or Clojure.

As another example, since it's so easy to just run a shell command from Acme (highlight the command text and middle-click it), I'm constantly customizing pre-written shell commands to e.g. grep for lines in code (to the point of creating indexes on the fly for files), and then right-clicking on grep -n output to jump to a specific line in a file.

[TeMPOraL]

You still need to use the mouse. One of the biggest benefits of Vi(m)/Emacs over everything else out there is that you don't need to use mouse for anything. For example, to justify a paragraph in Emacs I highlight it (several quick ways to do it with keyboard) and press C-u M-q. And you can do the general "replace text by running a shell command over it" by highlighting the text, and doing C-u M-| [shell-command-here].

The feature you're describing is awesome and useful, but I wouldn't be happy to have to use mouse for it. Is there a way to invoke it from keyboard in Acme?

[roguecoder]

An Eclipse plugin usually takes me about half a day at this point. It's not good for short scripts, but these examples don't seem to be that either.

For the "20 minutes later" version, I typically abuse JUnit tests, or use search+working sets.

What Eclipse misses more than anything else is performance and generalization. It's sufficient for my Java work: it is insufficient for my four-language code base.

[silverlake]

Couldn't you build a LightTable-like IDE in Squeak right now?

[stcredzero]

Squeak is somewhat LightTable-like already, except everything needs to be refreshed manually. (And not all of it is obvious to the beginner, though all of it is readily available if you know how.)

You actually couldn't have everything update automatically and expect bad things never to happen. Functional programming has an advantage here.

[joesb]

Is Light Table written in Haskell-like language? Because other functional language still allow side effect and IO and running everything in those language can still cause bad things.

[stcredzero]

Well, you can have bad things happen even more easily in Smalltalk, but enough of the library is well behaved that it's workable in development, especially since you can save the image periodically and you have a transactional log for even what you do after your last save.

To make Light Table work, you don't need 100% perfect semantics. You just need 99% good semantics, plus some way of turning off automatic refresh and refreshing manually.