It’s a worrisome sign when people start to talk about tools. No real work to report on?
With that out of the way, let’s talk about tools :) – programming tools.
Everyone has their favourite programmer’s editor and operating system. Mine happens to be Vim (MacVim) and Mac OSX. Yours will likely be different. Whatever works, right?
Having said that, I found myself a bit between a rock and a hard place lately, while trying out ClojureScript, that Lisp’y programming language I mentioned last week. The thing is that Lispers tend to use something called the REPL – constantly so, during editing in fact.
What’s a REPL for?
Most programming languages use a form of development based on frequent restarts: edit your code, save it, then re-run the app, re-run the test suite, or refresh the browser. Some development setups have turned this into a very streamlined and convenient fine art. This works well – after all, why else would everybody be doing things this way, right?
But there’s a drawback: when you have to stop the world and restart it, it takes some effort to get back to the exact context you’re working on right now. Either by creating a good set of tests, with “mocks” and “spies” to isolate and analyse the context, or by repeating the steps to get to that specific state in case of interactive GUI- or browser-based apps.
Another workaround, depending on the programming language support for it, is to use a debugger, with “breakpoints” and “watchpoints” set to stop the code just where you want it.
But what if you could keep your application running – assuming it hasn’t locked up, that is? So it’s still running, but just not yet doing what it should. What if we could change a few lines of code and see if that fixes the issue? What if we could edit inside a running app?
What if we could in fact build an app from scratch this way? Take a small empty app, define a function, load it in, see if it works, perhaps call the function from a console-style session running inside the application? And then iterate, extend, tweak, fix, add code… live?
This is what people have been doing with Lisp for over half a century. With a “REPL”:
A similar approach has been possible for some time in a few other languages (such as Tcl). But it’s unfortunately not mainstream. It can take quite some machinery to make it work.
While a traditional edit-save-run cycle takes a few seconds, REPL-based coding is instant.
A nice example of this in action is in Tim Baldridge’s videos about Clojure. He never starts up an application in fact: he just fires up the REPL in an editor window, and then starts writing little pieces of code. To try it out, he hits a key combination which sends the parenthesised form currently under the cursor to the REPL, and that’s it. Errors in the code can be fixed and resent at will. Definitions, but also little test calls, anything.
More substantial bits of code are “require’d” in as needed. So what you end up, is keeping a REPL context running at all times, and loading stuff into it. This isn’t limited to server-side code, it also works in the browser: enter “
(js/alert "Hello")” and up pops a dialog. All it takes is the REPL to be running inside the browser, and some websocket magic. In the browser, it’s a bit like typing everything into the developer console, but unlike that setup, you get to keep all the code and trials you write – in the editor, with all its conveniences.
Another recent development in ClojureScript land is Figwheel by Bruce Hauman. There’s a 6-min video showing an example of use, and a very nice 45-min video where he goes into things in a lot more detail.
In essence, Figwheel is a file-driven hot reloader: you edit some code in your editor, you save the file, and Figwheel forces the browser (or node.js) to reload the code of just that file. The implementation is very different, but the effect is similar to Dan Abramov’s React Hot Reloader – which works for JavaSript in the browser, when combined with React.
There are some limitations for what you can do in both the REPL-based and the Figwheel approach, but if all else fails you can always restart things and have a clean slate again.
The impact of these two approaches on the development process are hard to understate: it’s as if you’re inside the app, looking at things and tweaking it as it runs. App restarts are far less common, which means server-side code can just keep running as you develop pieces of it further. Likewise, browser side, you can navigate to a specific page and context, and change the code while staying on that page and in that context. Even a scroll position or the contents of an input box will stay the same as you edit and reload code.
For an example Figwheel + REPL setup running both in the browser and in node.js at the same time, see this interesting project on GitHub. It’s able to do hot reloads on the server as well as on (any number of) browsers – whenever code changes. Here’s a running setup:
And here’s what I see when typing “(fig-status)” into Figwheel’s REPL:
Figwheel System Status ---------------------------------------------------- Autobuilder running? : true Focusing on build ids: app, server Client Connections server: 1 connection app: 1 connection ----------------------------------------------------
This uses two processes: a Figwheel-based REPL (JVM), and a node-based server app (v8). And then of course a browser, and an editor for actual development. Both Node.js and the browser(s) connect into the Figwheel JVM, which also lets you type in ClojureScript.
So what do we need to work in this way? Well, for one, the language needs to support it and someone needs to have implemented this “hot reload” or “live code injection” mechanism.
For Figwheel, that’s about it. You need to write your code files in a certain way, allowing it to reload what matters without messing up the current state – “defonce” does most of this.
But the real gem is the REPL: having a window into a running app, and peeking and poking at its innards while in flight. If “REPL” sounds funny, then just think of it as “interactive command prompt”. Several scripting languages support this. Not C, C++, or Go, alas.
For this, the editor should offer some kind of support, so that a few keystrokes will let you push code into the app. Whether a function definition or a printf-type call, whatever.
And that’s where vim felt a bit inadequate: there are a few plugins which try to address this, but they all have to work around the limitation that vim has no built-in terminal.
In Emacs-land, there has always been “SLIME” for traditional Lisp languages, and now there is “CIDER” for Clojure (hey, I didn’t make up those names, I just report them!). In a long-ago past, I once tried to learn Emacs for a very intense month, but I gave up. The multi-key acrobatics is not for me, and I have tons of vim key shortcuts stashed into muscle memory by now. Some people even point to research to say that vim’s way works better.
For an idea of what people can do when they practically live inside their Emacs editor, see this 18-min video. Bit hard to follow, but you can see why some people call Emacs an OS…
Anyway, I’m not willing to unlearn those decades of vim conventions by now. I have used many other editors over the years (including TextMate, Sublime Text, and recently Atom), but I always end up going back. The mouse has no place in editing, and no matter how hard some editors try to offer a “vim emulation mode”, they all fail in very awkward ways.
And then I stumbled upon this thing. All I can say is: “Vim, reloaded”.
Wow – a 99% complete emulation, perhaps one or two keystrokes which work differently. And then it adds a whole new set of commands (based on the space bar, hence the name), incredibly nice pop-up help as you type the shortcuts, and… underneath, it’s… Emacs ???
Spacemacs comes with a ton of nice default configuration settings and plugins. Other than some font changes, some extra language bindings, I hardly change it. My biggest config tweak so far has been to make it start up with a fixed position and window size.