Devcards is a ClojureScript library that helps developers interactively lift code examples out of their source files into an organized set of cards in the browser.
If you have to have the goods now, skip ahead and watch the video.
Code examples and feedback
A REPL allows us to interactively try different code examples and quickly confirm that we understand their behavior. For programmers who are used to this workflow, this interactive validation of expectations starts to become a cornerstone of how they strategize and arrive at the solution to a particular problem. However, when you are coding UIs for the browser you often want to verify code that has a certain display characteristic and in this case the REPL doesn’t help.
The current workflow for verifying visual behavior in the browser is normally constrained to editing, reloading and then manually manipulating the main application into a particular state. For example: we are writing a game, and we just changed some behavior of the game and need to verify that the change worked. We will have to interact with the game in order to put it into the specific state which will help us validate our change worked.
We are normally constrained to working within ONE instance of the application at hand. It doesn’t have to be this way, but currently the cost of displaying different code examples in different states is often higher than just manually manipulating the main application instance into state we are wanting to check.
Thus constrained, we are less likely to freely experiment but rather continually run a cost-benefit analysis in our heads as to whether trying to validate a certain piece of code is practical in our current application environment. We end up writing longer stretches of code without the valuable feedback that a REPL provides. I would venture that this alters the code we write. We become prejudiced towards conservative, proven, patterns that will reduce the likely pain of having to repeatedly manipulate the main application into a certain state over and over again.
This is an extreme divergence from developing with a REPL where we can where we can experiment with different code examples at low cost and with relative ease.
I am proposing a straightforward solution to this problem, a library that allows us to easily create code examples in our source files which are then immediately presented to us in the browser. This library is intended to bring the interactive nature of developing using a REPL to problems that are graphical in nature.
For example, this library would make it easy to interactively surface several 2048 boards in different starting states. Feel free to interact with these examples.
Interacting with these boards quickly verifies that tiles are combining correctly and animations are phasing correctly in these particular starting states. Given that 2048 game progression is random, being able to look at specific examples like this can be very helpful in nailing down the behavior of shifting tile rows.
Seeing examples like these side by side is a luxury that we are not accustomed to. Now imagine being able to surface code examples with ease directly from the the source file your are working in.
I have created Devcards as one possible solution to this problem. Devcards provides an interface that organizes a set of code examples, where each code example is represented by a card. Devcards allows you to define cards in-line in your source file like so:
(defcard my-first-card (react-card (sablono/html [:div [:h1 "I'm a Devcard"]])))
Devcards is written with a great deal of attention towards live code reloading. When you save the file that holds the definition above, a card containing the rendered template will instantly appear in the Devcards interface. Cards are organized by their namespace and will respond to changes as you continue to code.
As you progress through a problem, you can create a set of cards that will all appear in the Devcards interface. This enables you to surface a set of examples that are all responding in real time to your code changes. At the end of a coding session you will potentially have a set of valuable artifacts (cards) that help you and others understand your approach to the problem.
You can see an example of the Devcards interface here.
Devcards derives its interactivity from the live reloading Leinigen plugin: lein-figwheel. Figwheel is a code reloading server/client combination that continually reloads compiled code into the browser as you change your ClojureScript sources.
Developing with Devcards
If you are curious about how this looks in reality, check out the demonstration video.
Feedback - the ultimate programming tool
Working with a page of visual code examples that are all responding to the code I am writing, provides an unprecedented level of feedback that can increase my awareness of the changes propagated due to the code I am writing. Tests start to fail, entities start disappearing, I can go back and interact with components in specific states and see how they are responding to recent code changes. I can then move to other pages of cards and see if they are still working as expected.
This experience has driven home for me the paucity of feedback in our current development workflows. We are still in the cave looking at shadows. Bret Victor has eloquently expressed this several times. Figwheel and Devcards have both the increased the amount of feedback I receive while I code by a very large factor. Once you experience this your eyes will be opened and you will not freely give up this way of coding, just as you would be loath to give up your REPL now.
Pages of examples FTW
As we all know, tests are not enough. Integration tests run in a black box and take forever (preventing real time feedback) and can’t catch the most obvious of display errors. Unit tests live in an isolated world away from the real complexity where things actually go wrong.
Be clear that I am not saying that this solves everything, I am just saying that this is another tool that can be potentially very helpful.
Anyone can implement their own cards easily. If you can create a React or Om component you can create a card. So the potential for having a library of very helpful cards at your fingertips as you program is very high.
We can create cards with history management and backtracking built in. There can also be cards that are very targeted, like a regular expression card which would present an interface to try different strings and allow you to change the regex.
This is new territory and there are probably many creative helpful applications.
Give it a try
These are still the very early days for Devcards but I sincerely hope you give it a try.
The readme on Github provides instructions for getting started.
- Figwheel introductory blog post
- 2048 game
- Learnable Programming
- Media for Thinking the Unthinkable
Special thanks to @willsommers for taking time to review and correct the copy in this post.