FW/1 - The Invisible Framework

Framework One and friends...

Using Clojure With CFML

Clojure and CFML Sitting in a tree

Back in 2010, I started to learn Clojure. It’s a version of the Lisp programming language that runs on the JVM. It’s a mostly pure functional programming language. It’s very simple – very little syntax – but very powerful and expressive. It has a thriving community and ecosystem. Being a functional programming language it favors immutable data structures and higher order functions (map, filter, reduce, etc) which makes programs easy to reason about and it also makes concurrent programming pretty easy. No more thread safety issues!

Although Clojure can be compiled to JVM bytecode and deployed as a JAR or WAR file, it can also be treated like a dynamic scripting language and compiled on-the-fly to JVM bytecode, just like CFML, giving you that very fast edit-reload-test cycle that you’re used to with CFML. In fact, if you use the REPL (Read-Eval-Print-Loop) you can pretty much reduce that to just edit-test since Clojure code is compiled as you type it in and can be evaluated immediately.

Because I was still writing a lot of CFML back in 2010, I figured out an easy way to allow you, as a CFML developer, to load and run Clojure code within your CFML application. That functionality has been available via the standalone cfmljure library for several years, and it has become core to how the Internet dating platform works at my company (World Singles Networks). We’ve had cfmljure in production since Spring 2011 and, in 2014, we declared Clojure to be our primary language and nearly all new development happens in Clojure, hosted within three ColdBox applications and two FW/1 applications. Clojure powers all of our standalone processes as well.

As we’ve come to depend on FW/1 more and more, I’ve wanted to streamline the integration between FW/1 and cfmljure so that we can write services in Clojure and have them autowired into our controllers, just like our CFML services are, as well as have the option to write controllers in Clojure (if all the code they need to call is Clojure, not some mix of CFML and Clojure).

That’s why, in FW/1 3.5, I’ve made cfmljure part of the standard distribution and provided an extension to DI/1 to allow Clojure code to be discovered and autowired into your CFML code, as well as a way to write FW/1 controllers in pure Clojure.

Some Important Caveats (and System Requirements)

Before you get started, there are a handful of things you need to be aware of:

  • The Clojure integration works best on Lucee (4.5 or later) or Railo (4.2). ColdFusion 11 is supported too but interop between CFML and Clojure can get pretty ugly (see the ColdFusion-specific examples in the cfmljure project repo for more details). I doubt it will run on ColdFusion 10 or earlier. I developed exclusively on Railo from 2009 to 2015 and now I develop on Lucee. Thank Andrew Myers for the ColdFusion 11 support!
  • You should be running Java 8 at this point. The Clojure integration will not work on Java 6. It has not been tested on Java 7. The Java 6 EOL notice was February 2011 and public updates stopped in February 2011. The Java 7 EOL notice was March 2014 and public updates stopped in April 2015.
  • Clojure developers are pretty comfortable at the command line, so the build tool you’ll need to install for this is a command line tool: be prepared to work in Terminal / Console quite a bit!
  • Clojure developers use editors that auto-close parentheses, brackets, and braces. Most of those editors also support “structural editing” which means they understand expressions enclosed in matched pairs of parentheses (or brackets or braces) and let you easily manipulate expressions rather than just text. There are plugins for Eclipse, Sublime Text, IntelliJ and several others. Most Clojure developers use Emacs or Vim. I use Emacs for all my editing, CFML included.
  • In order to use Clojure with CFML, you must not run your CFML server under the root account (on Mac or Linux). It’s something you should never do in production anyway for security reasons, and there’s really no reason to do it on your development machine either. cfmljure checks this for you and refuses to initialize (and throws a helpful exception).
  • For ease of debugging any problems that may occur, I strongly recommend you always keep a Terminal / Console window open containing a tail of the console output from your CFML server. If you’re using Lucee or a Railo installation based on Tomcat, you’ll want to find the catalina.out log file and tail that. If you’re using ColdFusion 11, even tho’ it is notionally based on Tomcat, it’s completely non-standard and you’ll want to tail cfusion/logs/coldfusion-out.log.
  • Finally, cfmljure uses a lock file to ensure that a couple of things it does during initialization are single-threaded across the whole server, no matter how your CFML server and application instances are configured. On Mac and Linux, it creates (and deletes) /tmp/cfmljure.lock. On Windows, it creates (and deletes) cfmljure.lock in your TEMP folder which is normally in the AppData folder off your local accounts home folder – check your TEMP environment variable for the exact location. cfmljure tries really hard to clean up after itself but if something goes badly wrong during initialization, it’s possible the lock file will be left behind and you will need to manually delete it. cfmljure will try its best to let you know if it needs you to delete the lock file – but if you’re tailing the logs, you’ll see messages as the initialization proceeds.

Getting Started with Clojure and CFML

Clojure has two standard build tools, called Leiningen and Boot. The former has been around longer and is currently the most popular. The latter is more general purpose. FW/1, via cfmljure, supports both build tools. These tools manage all of your project dependencies (e.g., automatically downloading and installing any libraries you need) as well as providing a REPL for interactive development, running your tests, packaging applications into JAR files, deploying them to standard repositories and so on.

cfmljure leverages Leiningen or Boot to figure out the complete classpath that your Clojure code needs – including any libraries you use – so that it can easily load all the JAR files (and Clojure source code) into your CFML server.

The first step to using FW/1 and Clojure together is to install one (or both!) of these build tools and make sure it is working. We’re going to look at both build tools – support for Boot is new in release 4.0.


In addition to managing your project dependencies, Leiningen also provides a template system for new projects, so it’s the most convenient way to get up and running right now. Since we want to create a test project without needing to manually set up various files, I recommend starting with Leiningen – and looking at Boot later on. For what it’s worth, World Singles has just switched from Leiningen to Boot, in order to leverage the increased flexibility that tool offers.

If you’re on Mac or Linux, follow the Leiningen Install steps for the lein shell script.

If you’re on Windows, I recommend using the Installer for Windows users mentioned near the end of that Install section.

Once you have it installed, open up a new Terminal / Console window and navigate to directory where you can create a project to test that it is working properly, then do this:

lein new app myapp

It will probably download a bunch of libraries the first time you run it, but it should (eventually) respond:

Generating a project called myapp based on the 'app' template.

Now go into that project and run it:

cd myapp
lein run

It should say:

Hello, World!

Congratulations! Leiningen is installed and working!

BTW, Leiningen supports unit testing out of the box, so in addition to creating an application skeleton for myapp, it also generates a test skeleton which you can run like this:

lein test

Because the test skeleton doesn’t have any valid tests – just one deliberate failure – you should see:

lein test myapp.core-test

lein test :only myapp.core-test/a-test

FAIL in (a-test) (core_test.clj:7)
FIXME, I fail.
expected: (= 0 1)
  actual: (not (= 0 1))

Ran 1 tests containing 1 assertions.
1 failures, 0 errors.
Tests failed.

When you start developing services and controllers in Clojure, you’ll find it handy to write unit tests as you go and run them with lein test!


Follow the Boot Install steps for either “Unix, Linux, OSX” or “Windows”. You’ll notice the Windows installation is much simpler than for Leiningen.

You should go through the Getting Started section to verify that Boot is installed and working (boot -h, boot repl).

Boot has an external task to create new projects but it works almost exactly like the new task that Leiningen has built-in:

boot -d seancorfield/boot-new new -t app -n myapp

This specifies an external dependency (on my boot-new task library), then the name of the task (new) then the arguments to that task (the type of the template and name of project to create).

If you want to get going quickly and you’ve already installed Leiningen and gone through the project creation and testing steps above, you can just create a build.boot file in that (Leiningen-created) project and run code and tests like this:

;; build.boot
(set-env! :resource-paths #{"src"}
          :dependencies '[[org.clojure/clojure "1.8.0"]])

(require 'myapp.core)
(deftask run []

Like project.clj under Leiningen, we specify the project dependencies. We also specify where our source code is – there’s no default – using :resource-paths. Next we load the main namespace of our new application and define a task called run that calls the main function in that application. Whilst Leiningen has a built-in task called run, you still have to declare the main namespace (or specify it as a command line argument). Boot takes the approach that this is “just code” so you can define whatever tasks you want, to run whatever Clojure code you want.

Accordingly, Boot does not have a built-in task to run tests, but there are readily available libraries to provide this functionality. We’ll update build.boot as follows:

;; build.boot
(set-env! :resource-paths #{"src"}
          :source-paths #{"test"}
          :dependencies '[[org.clojure/clojure "1.8.0"]
                          [adzerk/boot-test "1.0.7"]])

(require '[adzerk.boot-test :refer [test]])

(require 'myapp.core)
(deftask run []

We tell Boot where our test code is, we add a dependency on boot-test and we load that library and import the test task. Now we can run our tests:

boot test

The output will be somewhat different to that produced by lein test above but you will see the same failing test, followed by an exception that indicates the build failed (this is common to Boot tasks as a way to report build failures).

A note on :resource-paths and :source-paths: Boot considers files that are part of the classpath and also intended to be an output of the project (such as contents of a JAR file) to be “resources”; files that are part of the classpath but not intended to be an output of the project are “sources”. That seems a bit confusing at first (source files are resources, test files are sources), but as you work more with Boot you get used to more things being resources, not just source files, so it soon makes a lot of sense.

Now we’ll move on to using FW/1 and Clojure together, and show how both Leiningen and Boot can be used.

Testing FW/1 4.0 and Clojure

At this point you should be able to start your CFML server (remember: not under the root account!) and, with FW/1 4.0 installed (and possibly with a /framework mapping set up, or with the framework folder moved into your webroot), you should be able to run all of the FW/1 examples. Make sure examples 1-5 work, then try 6helloclojure.

If it doesn’t seem to run correctly, and there’s no obvious exception information in the browser or useful information in the log file, you can try to re-run the 6helloclojure example with ?cfmljure=abortOnFailure added to the URL. This should dump a whole bunch of information to the browser as well as any exception encountered.

If that doesn’t help, add diConfig : { debug : true } to the variables.framework configuration in Application.cfc for that example and try again. This will output more information to the CFML server’s console log (which you’re tailing, right?).

If you can’t figure out the problem from all this extra debugging information, at least you’ll have it all available when you post to the FW/1 mailing list, asking for help!

At this point, however, I hope you got the example to run and you were able to try out the various links and see what it was doing in the trace output in the browser, and perhaps by looking at the code, which we’ll go over next.

By default, 6helloclojure relies on Leiningen. To use Boot instead, add diConfig : { boot : "boot" } to the variables.framework configuration (or merge it with your debug setting). If Boot isn’t on the PATH used by your CFML server, you can specify the full file path to the Boot shell script or executable using that boot configuration (instead of just "boot").

The 6helloclojure Example Explained

If you look in the 6helloclojure folder (in examples/subsystems/), you’ll see a combination of things you expect to see in a FW/1 application and files and folders that would see in your myapp Clojure test project above:

Application.cfc MyApplication.cfc index.cfm views
.gitignore LICENSE README.md doc resources target
boot.properties build.boot project.clj src test

The files shown in the first line are for FW/1. You might later add a controllers folder and a model folder if you write any of those pieces in CFML.

The files shown in the second line are generated by Leiningen and you can pretty much ignore them. You’ll see that by default a Leiningen-generated Clojure project is assumed to live under version control with git, have an open source software license (Eclipse Public License 1.0 by default), have a README (markdown) file that explains what the project is for and how to use it, a documentation folder, possibly some non-source code resource files (configuration files or perhaps assets for a web application), and you may see a target folder which is used by Leiningen for compiling code and generating JAR files etc.

The files shown in the third line are the important parts of the Clojure code:

  • boot.properties (for Boot) defines the configuration of Boot itself: the specific version of Boot, Clojure, and so on.
  • build.boot (for Boot) defines the dependencies of your project (the libraries it needs), as well as any development or test tooling (as tasks) etc. The license, description, and URL information listed in project.clj can be specified for Boot but that is out of the scope of this documentation.
  • project.clj (for Leiningen) defines the dependencies of your project (the libraries it needs), as well as any development or test tooling (as plugins) and several other important aspects of how to run, test, and package your code. You’ll notice that it also has a description, a URL for where to find the project (e.g., on GitHub), and details of the license.
  • src is where your Clojure source code lives. Clojure itself places very few restrictions on the structure of this (beyond assuming that the file structure matches the namespace structure within the Clojure source files, much the same way Java’s file structure matches its package structure). If you have a Clojure namespace with - in its name, the corresponding folder will have _ in its name.
  • test is where your Clojure test code lives. It’s a common convention that if your source file is foo/bar/baz.clj which will have a namespace of foo.bar.baz, then the tests for that code will live in foo/bar/baz_test.clj with a namespace of foo.bar.baz-test. We have hello/controllers/main_test.clj which contains tests for the source code in hello/controllers/main.clj.

If you run lein test in the 6helloclojure folder, you’ll see:

lein test hello.controllers.main-test

lein test hello.services.greeter-test

Ran 2 tests containing 3 assertions.
0 failures, 0 errors.

We’ll take a look at that in a minute. You can also run boot test (if you have Boot installed) and you will see similar output.

Let’s start with the CFML files and look at Application.cfc first, then the views folder. As usual index.cfm is an empty file.


This is based on /framework/Application.cfc and allows us to use FW/1 without a global mapping. It creates a per-application mapping for /framework via a relative path using expandPath( '../../../framework' ), and then it creates an instance of MyApplication.cfc, passing in the FW/1 configuration structure.

There are a few items of note here, and the first is that we specify diComponent as "framework.ioclj" to tell FW/1 that we want to use ioclj.cfc rather than the default ioc.cfc for the Dependency Injection component (the bean factory). ioclj.cfc extends DI/1 and provides the Clojure-specific magic. Note that we leave diEngine as the default (which is "di1") because ioclj.cfc is just an extension to DI/1.

Next we specify diLocations as the full filesystem path of the current folder. In a CFML / Clojure application, you potentially need to tell the bean factory about three things:

  • Where to look for your CFML beans (if any).
  • Where to find your project.clj or build.boot file.
  • Where to find Clojure src folders to look for Clojure source code to be “managed” (and autowired).

Locations are specified as either a comma-separated list of file paths, or an array of paths, and one of them must specify the exact directory path to where project.clj or build.boot lives. By default, each path is searched in order, for each of the three types of files. That means you can store both your Clojure code and your CFML beans in the same tree structure if you wish, or you can store them separately and provide multiple file paths in diLocations. File paths may optionally be prefixed with clj: or cfml: which will restrict searching to just that type of file.

If you want to run 6helloclojure with Boot, just add diConfig : { boot : "boot" } into the framework configuration and reload the application.


This is based on /framework/MyApplication.cfc and allows us to override parts of FW/1. It extends framework.one (using the per-application mapping created in Application.cfc), and then overrides any FW/1 we want to customize.

The code of note is that setupRequest() reloads the Clojure code whenever the framework itself is reloaded. Normally, in a FW/1 app, you can specify an application reload and your bean factory is recreated. Because of the way Clojure code is compiled and loaded into the JVM, reloading your bean factory is not sufficient to force a reload of those parts of the JVM, so you need to do this programmatically somehow. Note that this will only reload the namespaces that follow the FW/1 conventions to be discovered. See below for more on this.


This is a regular FW/1 views folder, containing a subfolder for each section of the app (just main in this case) and a file for each item (there are three views here). As expected we have a main.default view and a main.error view which are basic defaults for FW/1 applications. We also have a main.stopped view. We’ll see how each of these is used when we look at the main controller.

The default view references rc.greeting which we’ll see being set up in the main.clj controller below, and it also gets a greeterService from the bean factory and calls a function in that. We’ll see where greeterService comes from below as well.


In this folder we have a single Clojure file, main.clj. As you might guess from the file path, this is our application controller. Inside you’ll see a namespace declaration (the ns expression) and four functions which represent our handlers.

As with CFML controller functions, each function is passed an argument called rc which is the request context. Unlike CFML controllers which might modify elements of the rc struct directly, Clojure controllers return an updated version of the rc data structure to the framework.

In the default handler, we get the :name element of the rc and we return rc with an additional element called :greeting. In Clojure, (:foo bar) is roughly equivalent to CFML’s bar.foo, and (:foo bar "baz") is similar to:

structKeyExists( bar, "foo" ) ? bar.foo : "baz"

So default passes rc.name (or "anonymous" if name isn’t present in rc) to the greet/hello function and then stores the result in the greeting element of rc. Note that the assoc function (pronounced assosh like the word associate) returns a new struct with the key added – it does not modify the original struct. This seems very strange at first but you’ll get used to it and it’s very powerful (and very safe) since rc is immutable. We’ll look at where greet/hello comes from in a minute.

Next we have the do-redirect handler. Yes, Clojure functions can have - in their names. The standard naming convention in Clojure is words-like-this rather than wordsLikeThis or WordsLikeThis. It’s a long-standing Lisp tradition. Yes, it’s strange but you’ll get used to it and you’ll soon find it more readable than CamelCase. Anyway, this handler tells FW/1 to redirect to main.default with the specified query string. Adding a key called :redirect to rc is the equivalent of calling variables.fw.redirect() in a pure CFML FW/1 app. All four arguments to the redirect() function can be specified as keys in the :redirect struct (:append, :preserve, :queryString, :statusCode).

Then we have the stop-it handler. By the way, the URL actions that correspond to do-redirect and stop-it are do_redirect and stop_it respectively, following the filesystem convention of swapping _ for - in Clojure. This handler tells FW/1 it wants to abort the controller cycle and also set the view to the main.stopped action. Adding a key called :abort with a value of :controller is equivalent to calling variables.fw.abortController(). Adding a key called :view is equivalent to calling variables.fw.setView(). You’ll note that assoc can take any number of key/value pairs and add them all into the given struct.

Finally we have the json handler. By this point, it won’t surprise you to learn that this tells FW/1 to render the specified data as JSON. The Clojure struct {:a 1 :b "two" :c [3 4 5]} is equivalent to the CFML struct {a : 1, b : "two", c : [3, 4, 5]}. The Clojure Primer below explains basic symbols and data structures, when you’re ready to read more about this.


In this folder we have a single Clojure file, greeter.clj. As with the controller convention, the file path tells FW/1 that this is a service (and it would be autowired into any CFML code that declared property greeterService; as a dependency). This can also be pulled from the bean factory as "greeterService", as seen in the main.default view file.

There’s a single function hello in here that takes a string and returns it wrapped with "Hello " and "!".


Finally, we’ll look at the tests, first for the main controller, then the greeter service.

In controllers/main_test.clj, we have one test function default-item-test which contains two related tests. The arrow syntax means “take this thing and pass it through these functions” so:

(-> {} default :greeting)

takes an empty struct and passes it as the first argument to the default function (our handler method being tested) and then pass the result to the :greeting function – remember that (:greeting my-struct) is like my_struct.greeting in CFML. So this tests that if the rc is empty and you run main.default you get a new key called :greeting whose value is "Hello anonymous!".

The second test checks that if rc contains a :name element, you get back the appropriate greeting based on the value of that.

The (testing "label" ...) expression can have multiple is tests in it, even tho’ ours do not.

In services/greeter_test.clj, we again have one test function hello-test that verifies the hello function does what we expect.

FW/1 Conventions with Clojure

The naming and file path conventions for how you structure your Clojure code for use with FW/1 should be familiar to you if you’ve already used DI/1 in the past: plural folder names, containing “components”, become beans named for the component, with a suffix that is the singular of the folder name:

  • src/hello/controllers/main.clj -> mainController
  • src/hello/services/greeter.clj -> greeterService

In a typical Clojure application, you have src/ and test/ folders containing your source code and your test code. Then you have the top-level name of your application – hello in the paths above – and then below that you have all your Clojure “components”.

Therefore, FW/1 requires that there are at least three “segments” in a file path and name for this convention:

  • src/myapp/controllers/main.clj has myapp.controllers.main as the namespace – three segments – becomes mainController
  • src/hello/admin/services/user.clj has hello.admin.services.user as the namespace – four segments – becomes userService

That means that just src/controllers/main.clj would not match the convention as its namespace would be controllers.main (only two segments – no top-level application name).

FW/1 looks for the first plural segment that follows the top-level application name, and if there are multiple segments after that first plural segment, they are concatenated to form the bean name:

  • src/myapp/common/libraries/security/checker.clj has myapp.common.libraries.security.checker as the namespace and, assuming liberal : true is specified in the configuration, libraries will be seen as the first plural and mapped to library, and then the bean name will be securityChecker, so the full bean name will be securityCheckerLibrary.
  • src/myapp/jobs/mail/alerts.clj has myapp.jobs.mail.alerts as the namespace and the first plural is jobs, so the full bean name will be mailAlertsJob.

Note: without liberal : true in the diConfig structure, libraries would still be treated as plural but the singular would be assumed to be librarie!

Uniqueness of Namespace to Bean Mappings

One additional restriction for FW/1 to automatically manage injection of Clojure namespaces into your CFCs is that the filename + suffix must be unique across your whole application within the Clojure code (so also having src/hello/public/services/user.clj would conflict with src/hello/admin/services/user.clj, but not src/hello/services/admin/user.clj which would be mapped to adminUserService).

Additional Clojure Code

Aside: You can have additional Clojure code that doesn’t follow this convention, but the bean factory reload() function only attempts to reload Clojure files that it “knows” about via this convention. You can explicitly reload others – if you allow for a URL variable that can be passed to the reload() method of the bean factory, identifying a single namespace – but there are some subtleties there which are beyond the scope of this documentation (if you want to learn more, read the clojure.core/require docstring and know that reload("all") does (require ... :reload) on each namespace covered by the convention but reload("some.namespace") does (require 'some.namespace :reload-all) for an explicitly provided namespace).

Configuring framework.ioclj

Since ioclj.cfc is a extension of ioc.cfc (DI/1), all of the same configuration options apply as well as the following:

  • boot - string - default "". If you specify a non-empty value here, FW/1 will assume you want to use Boot instead of Leiningen and it wlil look for build.boot instead of project.clj. When a non-empty value is used here, the value of lein (below) is ignored.
  • debug - boolean - default false. If true, the CFML/Clojure bridge will write a lot of additional information to the server console (e.g., catalina.out), which will help you check where FW/1 is looking for your Clojure files and which ones it may be ignoring (due to naming conventions not matching). Another useful form of debugging is <cfdump var=#getBeanFactory().getBeanInfo()#"/> in a view so you can see all of the beans / namespaces that were found.
  • lein - string - default "lein". Use this to specify the full file system path to your Leiningen shell/batch script if just the lein command on its own won’t be found on the path used by your CFML server. At World Singles, we kept the Leiningen shell script under version control so we could manage which version was used our applications, and we specified the path to that script, which was part of our application deployment, using the lein configuration setting.
  • server - boolean - default false. By default, a new instance of the CFML/Clojure bridge CFC (cfmljure.cfc) is created every time the bean factory (ioclj.cfc) is created. However, since this bridge modifies the classpath of your server on the fly – not just your application (because JVM settings are per server instance, not per application instance) – it is a reasonable optimization to cache the bridge instance in server scope (since running lein classpath or boot show -C can be slow). Specifying server : true will tell framework.ioclj to perform that optimization for you. This will create server.__cfmljure, which will be reused each time the bean factory is recreated, until the server is restarted, or you explicitly delete that variable. If you change the dependencies in project.clj or build.boot, you will need to force the bridge instance to be recreated in order for those new dependencies to be picked up.
  • timeout - numeric - default 300. This is the number of seconds that FW/1 will wait for the lein classpath or boot show -C command to execute at startup. This should be sufficient even if you have a large number of dependencies in your project.clj or build.boot.

In addition, the folder paths you provide are searched as follows:

  • No prefix: path is searched (recursively) for CFCs, path is checked for project.clj or build.boot (according to whether the boot configuration is omitted or present), path is checked for a src folder which is then searched (recursively) for Clojure files to manage and autowire into CFCs.
  • clj: prefix: path is checked for project.clj or build.boot (according to whether the boot configuration is omitted or present), path is checked for a src folder which is then searched (recursively) for Clojure files to manage and autowire into CFCs.
  • cfml: prefix: path is searched (recursively) for CFCs.

If your Clojure build file is alongside your src folder, and all your CFCs are also in (subfolders of) that same location, you can specify just one folder path with no prefix. Otherwise you will need to specify the path(s) to your CFCs, the path to your Clojure build file, and the path to your Clojure src folders separately.

Example 1:

        ... CFCs ...
        ... CFCs ...
        ... Clojure source files ...
        ... Clojure test files ...

In this project, just supplying the path to approot is all you need: ioclj will find project.clj there as well as the src folder (which it will recursively search for .clj files) and DI/1 will recursively search the whole thing for CFCs.

Example 2:

        ... CFCs ...
            ... CFCs ...
            ... Clojure source files ...
            ... Clojure test files ...

In this project, you can supply the path to approot and it will find build.boot (assuming you also specify boot in your configuration), and it will recursively find all your CFCs, but if you want any Clojure code to be managed and injected, you’ll need to supply the path to model as well (where src is) and because you don’t want DI/1 to search both locations for CFCs (because that path overlaps with part of the approot tree), you will want to specify that is Clojure only:

diLocations : [ "/path/to/approot", "clj:/path/to/approot/model" ]

It would be better to be explicit about which paths are to be searched for what types of files:

diLocations : [
    "clj:/path/to/approot",             // build.boot location
    "clj:/path/to/approot/model",       // Clojure src location
    "cfml:/path/to/approot/model/cfcs", // Model CFCs
    "cfml:/path/to/approot/controllers" // Controller CFCs

As you can see, you can organize you CFML and Clojure source code – and your Clojure build file – however you like (modulo the src folder requirement for Clojure) and still tell FW/1 exactly what you want it to do.

What is ns all about?

Probably the most complex and confusing aspect when you are first learning Clojure is the ns expression. ns serves two main purposes:

  • It declares the namespace in which the code in a given file lives.
  • It declares what additional namespaces you depend on and how you’re going to access functions in them.

Think of a namespace as a “package” or “module”. The default convention in Clojure is that the namespace matches the filesystem path below your src or test folder, except that - in a namespace identifier matches _ in the file path. Technically, a namespace can be implemented across multiple files but it isn’t very common to see that, so don’t worry about it.

In the 6helloclojure example, you’ll see hello.controllers.main for src/hello/controllers/main.clj and hello.controllers.main-test for test/hello/controllers/main_test.clj. If you follow this basic convention, you won’t go wrong. If your ns declaration doesn’t match the filesystem path, you can get strange errors when you attempt to access it. Think of it much like the dotted-path used to access CFCs in CFML.

The second important part of ns is the list of namespaces your code :requires. There are two basic forms here:

  • [some.namespace :as alias] which lets you use alias/name as a way to reference any functions defined in some.namespace.
  • [some.namespace :refer [fn1 fn2]] which lets you use fn1 and fn2 directly from some.namespace without a prefix. The shorthand :refer :all brings in every function from that namespace, and is generally only used in test code.

Prefer the first form – which you see in hello.controllers.main where hello.services.greeter is given an alias of greet so that the controller can call greet/hello.

You’ll use the :require expression to bring in Clojure standard libraries, 3rd party Clojure libraries, and parts of your own code. For most 3rd party libraries, you’ll also need to add an entry to :dependencies in your project.clj file in order to tell Leiningen that you need that library downloaded (or in your build.boot file for Boot). We’ll see this when we learn about database access below.

I recommend writing your required namespaces in alphabetical order – most production Clojure code relies on quite a long list of other namespaces so organization is important.

You might think this sounds a bit like import statements in Java and you’d be right, except that :require is purely for Clojure code. There is another form for importing Java classes, called :import, but Java interop is a whole other subject and you should consult a Clojure book or the online documentation for more details about that.

Building Your Own CFML / Clojure Application

Since we want to focus on the Clojure aspect of a combined CFML / Clojure project, we’ll start out by creating a Clojure project, and adding the necessary FW/1 CFML files to it in order to create a web application.

We will build a very simple task manager, backed by a database. We will initially create a CFML controller and write a Clojure service to interact with the database, then we’ll replace the CFML controller with a Clojure controller to wrap things up.

We’ll assume you’re using Leiningen for this exercise. I may update this section with instructions for Boot once project templates are avaialble for that tool.

Creating the Clojure project

Our Clojure code doesn’t need to be in our CFML webroot so just pick a folder and we’ll create a new Clojure application, in which we’ll write our task manager code:

lein new app taskmanager

This will create a folder called taskmanager containing a bare bones Clojure project. It’ll have a -main function so we can use it to run specific operations from the command line, and it’ll have tests we can run to validate our code.

Accessing a Database from Clojure

In order to access a database (via JDBC) from Clojure, you will first need to update your project.clj file to include dependencies on Clojure’s JDBC library and what your choice of JDBC driver is. For now we’ll use Apache Derby but if you look at the java.jdbc page on GitHub, you’ll see links to several other common database drivers (if you use SQL Server, you’ll find it easiest to use the jTDS driver to get started!).

Edit project.clj and update the :dependencies section to include:

[cfml-interop "0.2.0"]
[org.apache.derby/derby ""]
[org.clojure/java.jdbc "0.4.1"]

You’ll now have a vector with four vectors inside it like this:

:dependencies [[cfml-interop "0.2.0"]
               [org.apache.derby/derby ""]
               [org.clojure/java.jdbc "0.4.1"]
               [org.clojure/clojure "1.8.0"]]

Note: if the last vector has org.clojure/clojure "1.6.0", you must update it to use "1.8.0" instead!

The first entry is the open source version of the CFML / Clojure interop library we use at World Singles. The second entry is the JDBC driver for Apache Derby. The third entry is Clojure’s core JDBC library. The fourth entry is the Clojure language and core functions.

Now run lein repl and we can try this out. As the REPL starts up, it will download the new libraries and then you’ll get the prompt. Let’s create test database and write and read some data with it:

taskmanager.core> (require '[clojure.java.jdbc :as sql])
nil ;; now we have the JDBC library loaded with an alias
taskmanager.core> (def db {:dbtype "derby" :dbname "/path/to/clojure/taskmanager/cfmltest" :create true})
#'user/db ;; this is our database spec -- adjust the path accordingly!
user=> (sql/execute! db [(str "CREATE TABLE task ("
  "task VARCHAR(32),"
  "done BOOLEAN DEFAULT false"
[0] ;; success! we created the task table
taskmanager.core> (sql/insert! db :task {:task "Test database"})
({:1 1M}) ;; the sequence of inserted keys:
;; there is just one key, labeled :1, with the value 1
;; the M indicates a BigDecimal value
taskmanager.core> (sql/insert! db :task {:task "Read some data"})
({:1 2M}) ;; generated key is 2 this time
taskmanager.core> (sql/query db ["SELECT * FROM task WHERE NOT done"])
({:done false, :task "Test database", :id 1} {:done false, :task "Read some data", :id 2})
;; our two records came back, let's update one
taskmanager.core> (sql/update! db :task {:done true} ["id = ?" 1])
(1) ;; one row was updated
taskmanager.core> (sql/query db ["SELECT * FROM task WHERE NOT done"])
({:done false, :task "Read some data", :id 2})
;; yup, that's our only task not done now
taskmanager.core> (sql/update! db :task {:done true} ["id = ?" 2])
(1) ;; one row was updated
;; let's delete our table to clean up
taskmanager.core> (sql/execute! db [ "DROP TABLE task"])
[0] ;; success! we dropped the table

Press control-d to exit the REPL.

Some notes on the syntax:

  • Functions that modify the database end in ! to indicate they have side effects
  • SQL and any parameters are specified together in a vector:
    • ["a SQL string" param1 param2 param3]
    • Use ? in the SQL string to indicate a parameter placeholder
    • A PreparedStatement is used for all operations (which, together with using ?, protects you from SQL injection attacks)
  • execute! can be used for arbitrary SQL or DDL
  • query is intended for SELECT only and returns a result set
  • insert! inserts a new row from a hash map (struct)
  • update! sets the column values shown in the hash map for all rows that match the WHERE clause provided in the vector, so it is shorthand for ["UPDATE table SET col1 = ?, col2 = ?, col3 = ? WHERE clause" param1 param2 param3]

Adding the initial CFML files

We’re going to put our CFML files in a folder somewhere in our webroot, e.g., a taskmanager folder. We’ll need Application.cfc and an empty index.cfm. Here’s what needs to be in our Application.cfc file:

component extends=framework.one {
    variables.framework = {
        diComponent : "framework.ioclj",
        diLocations : "/path/to/clojure/taskmanager"
    function setupRequest() {
        // reload Clojure when FW/1 is reloaded:
        if ( isFrameworkReloadRequest() ) {
            getBeanFactory().reload( "all" );

We need to tell FW/1 to use the ioclj extension to DI/1, and we need to tell it the full path to the Clojure code (the folder where our project.clj lives). We also add the code to programmatically reload our Clojure files, in setupRequest().

Then we need a default main.default view so we can test the app:

<!--- views/main/default.cfm --->
Hello, World!

If this loads without error, FW/1 has successfully located your Clojure project and loaded it.

Writing a Clojure Service

We’ll start by writing a simple service containing just a single function so we can check that FW/1 finds and loads it correctly and that we can access it from our view. In your taskmanager Clojure folder, create a services subfolder under src/taskmanager and then create this greeting.clj file:

;; src/taskmanager/services/greeting.clj
(ns taskmanager.services.greeting)

(defn hello [name] (str "Hello, " name "!"))

Now we’ll update our main.default view to look like this:

  #getBeanFactory().getBean("greetingService").hello( "Clojure" )#

If you just try to hit the taskmanager app in your browser, you’ll get an error that greetingService does not exist. You need to reload FW/1 (via adding ?reload=true to the URL) so that it will reload the bean factory and find our newly created Clojure service. If you do that, you should see Hello, Clojure! in your browser.

Now we can get to work writing our simple task manager service!

We already wrote most of the code we need, as part of our REPL session above, and if you look in that taskmanager folder, you’ll find a (hidden) file called .lein-repl-history which contains the code you typed in. We’re going to use that as the basis of our service. Let’s create task.clj in our services folder like this:

;; src/taskmanager/services/task.clj
(ns taskmanager.services.task
  (:require [clojure.java.jdbc :as sql]))

;; for now we'll just hard-code one database spec but
;; we could pass it in from our controller as needed
(def db {:dbtype "derby" :dbname "cfmltest" :create true})

(defn create-task-table []
  (sql/execute! db [(str "CREATE TABLE task ("
                         "id INT GENERATED ALWAYS AS IDENTITY,"
                         "task VARCHAR(32),"
                         "done BOOLEAN DEFAULT false"

(defn add-task
  "Given a task name, add it to our database and return the new row's ID."
  (-> (sql/insert! db :task {:task task}) first :1))

(defn complete-task
  "Given a task ID, mark it as done and return the number of rows updated."
  (-> (sql/update! db :task {:done true} ["id = ?" id]) first))

(defn task-list
  "Return the tasks.
  If all is true, return all tasks, else just the incomplete ones."
  (sql/query db [(str "SELECT * FROM task"
                      (when-not all " WHERE NOT done"))]))

You can try out each function in the REPL or you can write unit tests, but they’re all fairly simple and we “tested” them when we were exploring how to work with a database in the REPL earlier.

Fleshing out the Views

Next we’ll create the views we need for our simple task manager. The default view will show a list of tasks with options to:

  • List all / outstanding tasks
  • Add a new task
  • Next to each task, the status, and the ability to mark a task as complete

We’ll reuse main.default as our list view and add main.newtask for adding new tasks. We’ll additionally have actions for main.addtask and main.completetask which will both redirect to main.default. We’ll have an optional boolean argument to main.default which will specify showing all tasks or just outstanding tasks.

Here’s our main.default view:

<!--- views/main/default.cfm --->
<cfparam name="rc.all" default="false"/>
        <cfif rc.all>
            <a href="#buildURL( 'main.default?all=false' )#">List To-Do</a>
            <a href="#buildURL( 'main.default?all=true' )#">List All</a>
        <a href="#buildURL( 'main.newtask' )#">Add New Task</a>
    <cfif arrayLen( rc.tasks )>
            <cfloop index="task" array="#rc.tasks#">
                    <form action="#buildURL( 'main.completetask' )#" method="post">
                        <cfif !task.done>
                            <input type="hidden" name="id" value="#task.id#"/>
                            <input type="submit" value="Done!"/>
        <p>You have nothing to do!</p>

Here’s our main.newtask view:

<!--- views/main/newtask.cfm --->
    <form action="#buildURL( 'main.addtask' )#" method="post">
            <input name="task" type="text"/>
            <input type="submit" value="Add!"/>
    <p><a href="#buildURL( 'main.default' )#">Cancel</a></p>

Testing with a CFML Controller

We’ll add a dummy controller so we can test this:

// controllers/main.cfc
component accessors=true {
    property framework;
    function default( rc ) {
        rc.tasks = [ ];
    function addtask( rc ) {
        framework.redirect( 'main.default' );
    function completetask( rc ) {
        framework.redirect( 'main.default' );

If we hit the app in our browser with ?reload=true in the URL, you should see:

List All | Add New Task

You have nothing to do!

If you click Add New Task and fill in the form and click Add!, you’ll end up back on this page with no tasks listed.

Using the Clojure Service

Now we’ll wire in our Clojure service that we wrote above and make something useful happen!

We’ll change our controller to look like this:

// controllers/main.cfc
component accessors=true {
    property framework;
    property taskService; // add this
    function default( rc ) {
        // default rc.all and pass boolean to Clojure:
        param name="rc.all" default="false";
        rc.tasks = taskService.task_list( rc.all ? true : false );
    function addtask( rc ) {
        // call Clojure
        taskService.add_task( rc.task );
        framework.redirect( 'main.default' );
    function completetask( rc ) {
        // call Clojure
        taskService.complete_task( rc.id );
        framework.redirect( 'main.default' );

Note: because URL and form values come into CFML as strings that it will automatically coerce to numeric, boolean, etc, we need to explicitly pass either true or false to taskService.task_list() which is why we write rc.all ? true : false. Most languages do not have CFML’s flexibility when interpreting data, but they also provide a lot more type safety! You might wonder why we don’t convert rc.id to numeric in the call to taskService.complete_task()? We can get away with it there because it’s passed directly to the JDBC driver and that knows that the id field is numeric and it will parse the string for you. In general, you probably shouldn’t rely on that and should explicitly convert your inbound string data from rc to the type your Clojure code is expecting.

If we run this (with ?reload=true in the URL), we’ll probably get a SQL exception because our table doesn’t exist. We dropped the table at the end of our REPL session (unless you recreated it again while testing the task.clj service?).

Let’s make our controller automatically create the table if the task list fails:

    function default( rc ) {
        // default rc.all and pass boolean to Clojure:
        param name="rc.all" default="false";
        try {
            rc.tasks = taskService.task_list( rc.all ? true : false );
        } catch ( any e ) {
            rc.tasks = taskService.task_list( rc.all ? true : false );

Run it again (with ?reload=true) and you should see an empty list of tasks. Add a new task and… oh dear! We got an exception:

Key [TASK] doesn't exist in Map (clojure.lang.PersistentArrayMap)

The reason for this is that raw Clojure structs (hashmaps) are not quite compatible with CFML structs (because they use keywords as keys, whereas CFML uses case-insensitive strings). The Clojure integration provides a way to convert back and forth, so we’ll need our controller to depend on cfmljure, and then we’ll use toCFML() to convert the Clojure data structure to a CFML-compatible data structure so we can display it:

// controllers/main.cfc
component accessors=true {
    property framework;
    property taskService;
    property cfmljure; // add this
    function default( rc ) {
        param name="rc.all" default="false";
        try {
            rc.tasks = taskService.task_list( rc.all ? true : false );
        } catch ( any e ) {
            rc.tasks = taskService.task_list( rc.all ? true : false );
        rc.tasks = cfmljure.toCFML( rc.tasks ); // make it CFML-compatible

Note: if you don’t add the cfml-interop dependency to a project, the toCFML() and toClojure() functions available in cfmljure will still “work” but the struct keys will be case-sensitive and lowercase so you would need to use myStruct['key'] notation to access them instead of just myStruct.key.

Writing a Clojure Controller

In this final section of Building Your Own CFML / Clojure Application, we’re going to replace our CFML Controller with an equivalent Clojure Controller. In your Clojure taskmanager project, create a src/taskmanager/controllers/ folder and inside it create main.clj containing:

;; src/taskmanager/controllers/main.clj
(ns taskmanager.controllers.main
  (:require [cfml.coerce :refer [->boolean]
            [taskmanager.services.task :as task]))

(defn default [rc]
  (let [all? (->boolean (:all rc))
        tasks (try
                (task/task-list all?)
                (catch Exception _
                  (task/task-list all?)))]
    (assoc rc :tasks tasks)))

(defn addtask [rc]
  (task/add-task (:task rc))
  (assoc rc :redirect {:action "main.default"}))

(defn completetask [rc]
  (task/complete-task (:id rc))
  (assoc rc :redirect {:action "main.default"}))

This is the equivalent of our CFML controller above so let’s walk through each piece of it:

  1. (ns ...) specifies taskmanager.controllers.main as our namespace and makes our service available with the task alias.
  2. default does the following:
    1. binds all? to the boolean value of rc.all (which comes in as the string "true" or "false"!).
    2. binds tasks to the result of calling task-list in our service (including catching any exception, attempting to create the task table, and calling task-list again – try/catch is an expression in Clojure that returns a value, just like any other expression).
    3. returns rc with the task list added (as rc.tasks).
  3. addtask calls add-task in our service and returns rc with a redirect added to it.
  4. completetask calls complete-task in our service and returns rc with a redirect added to it.

Now run this with ?reload=true in the URL and it should work just as it did before, except now it’s using the Clojure controller instead of the CFML controller! How do you know it’s using the Clojure version? Because framework.ioclj adds the Clojure namespaces it finds after any CFCs it finds, overwriting any beans with the same alias. If you want to convince yourself, remove controllers/main.cfc from your taskmanager folder in the webroot, and reload the application again.

So why would we write our controllers in Clojure instead of CFML?

  • As simple functions (that take rc as input and produce an updated rc as output), they’re easy to write unit tests for.
  • You don’t need to convert rc data structures back and forth between CFML and Clojure since that’s taken care of automatically in FW/1 (technically in cljcontroller.cfc).
  • You can work in the REPL building and testing your entire application’s functionality (and then work on the views with a browser).
  • We get the full power of Clojure’s data abstractions, concurrency, and immutability working for us.
  • With your services and controllers in Clojure, you’re one step away from building all-Clojure web applications using FW/1 for Clojure.

RC Value Conversion

Since all values come into the rc (from URL and form scope) as strings, you’ll usually need to convert them to the appropriate numeric, boolean, or other data type in order to use them in your Clojure code. We saw above that you can get away with some things, but in the pure Clojure world, you need to be much more specific about types.

The cfml-interop library, containing the cfml.interop namespace, contains some useful functions for converting from string to various types - the ->type naming is just a convention in Clojure for conversion functions:

  • ->long – Accepts any value, string or numeric, and tries to convert it to a Long value. You can provide a second argument which specifies the default to return if conversion fails (otherwise it will return 0).
  • ->double – Accepts any value, string or numeric, and tries to convert it to a Double value. You can provide a second argument which specifies the default to return if conversion fails (otherwise it will return 0.0).
  • ->boolean – Accepts any value, boolean, string or numeric, and tries to convert it to a Boolean value. You can provide a second argument which specifies the default to return if conversion fails (otherwise it will return false). This treats the strings "true" and "yes" as true (not case sensitive), as well as treating non-zero numeric values as true (and zero as false), just like CFML.

Augmenting RC With Clojure Expressions

Normally rc is automatically converted back and forth between a CFML “struct” and a Clojure hash map so that inside Clojure controllers, the keys are lowercase keywords (as expected in Clojure). This automatic conversion handles “regular” data types just fine but there may be times when you want to pass more complex data into Clojure controllers, such as a Component providing the application’s configuration or state. The cljcontroller.cfc file provides an extension point for this – callClojure() – to allow you to add Clojure expressions to the request context (after conversion), before calling the actual controller (and then removing them again before returning to CFML). Write a CFC that extends framework.cljcontroller and overrides callClojure(). Make sure to call super.callClojure(). Then tell DI/1 that cljcontroller is your CFC instead:

// myapp/cljcontroller.cfc:
component extends=framework.cljcontroller {

    // add/remove application.clojure_component to the request context
    // this is a Clojure Component (Stuart Sierra's "reloaded" workflow)
    // and is not convertible to/from CFML -- note that rcClj is a Clojure
    // hash map at this point so you must use Clojure functions to manipulate it
    function callClojure( string qualifiedName, any rcClj ) {
        var core = variables.cfmljure.clojure.core;
        var appComponent = core.keyword( "application-component" );
        rcClj = core.assoc( rcClj, appComponent, application.clojure_component );
        var rawResult = super.callClojure( qualifiedName, rcClj );
        return core.dissoc( rawResult, appComponent );


// framework configuration:
            diConfig : {
                cljcontroller : "myapp.cljcontroller",

A Clojure Primer

To learn about Clojure in any depth, I’d recommend you go through the More Stuff to Read section at the end of this page, but I’m going to give you a quick run through of some useful basics that should get you up and running more quickly.

As I claimed earlier, Clojure is a very simple language with only a few pieces of syntax:

  • A semicolon introduces a comment. Typically a single semicolon ; is used for an end of line comment and a double semicolon ;; is used for a whole line comment.
  • (func arg1 arg2 arg3) represents a function call (with three arguments). You can use commas if you want but they are just whitespace: (func arg1, arg2, arg3). Most Clojure developers omit commas in function calls. Almost everything in Clojure is a function call. A few things that look like function calls are actually “special forms” but you can pretend they’re really function calls until you get a bit more proficient (a function call evaluates all its arguments before the call, a special form may not evaluate its arguments until it needs them).
  • a is a symbol – a variable or function name – that evaluates to whatever it has been bound to. See def, defn and let below.
  • :a is a keyword – it evaluates to itself – that is a bit like a string except it is cached (so two :as in different parts of your code are the exact same thing and can be compared for identity based on their address, not their value). Keywords are commonly used as the keys in a hash map (see below).
  • [1 2 3 4] represents a vector (array) with the specified elements. You can use commas if you want, but they are just whitespace: [1, 2, 3, 4]. Most Clojure developers omit commas in vectors.
  • {:a 1, :b 2, :c 3} represents a hash map (struct) with keys :a, :b, :c and values 1, 2, 3 respectively. The commas are just whitespace but some Clojure developers use them anyway to clearly delineate pairs.
  • #{1 2 3} represents a set of values. If you add 4 to that set you get #{1 2 3 4} but if you add 1, 2, or 3 to it, it remains unchanged because those values are already in the set. CFML doesn’t have a set data type, but you might simulate it with a struct whose values are true (or some other arbitrary value) and the presence or absence of a key says whether it’s in the set or not.

That’s about it. When you learn about macros (in one of the Clojure books), you’ll encounter a few new pieces of syntax but you don’t need that to get going.

Some Basic Clojure Functions

Functions all the way down. Some functions / special forms are very important so I’m going to go through some of those here.

def, fn, and defn

The def special form creates a global (top-level) binding of a name to a value within the current namespace:

(def a 42)
;; binds a to the value 42 -- technically a is a Var
;; produces 42

The fn special form creates an anonymous function:

(fn [a b c] (* a b c))
;; creates a function with three arguments a, b, and c
;; that multiplies those three values together

Anonymous functions are often used as arguments to other functions:

(map (fn [x] (* x 2)) [1 2 3 4 5])
;; produces (2 4 6 8 10)

Strictly speaking, a function can contain more than one expression: they are all evaluated in order, but only the value of the last expression is returned. The others are thrown away. So why have multiple expressions? You might have some operations that cause side effects, such as logging to file or writing to a database, and you want to evaluate those for their effects but not necessarily their result.

You can bind a name to an anonymous function, to create a named function:

(def twice (fn [x] (* x 2)))

This is so common that defn exists as a shorthand for it:

(defn twice [x] (* x 2))

Documentation is built into Clojure and you can (and should) provide a “docstring” for all your functions that say what they do and possibly provide example usage:

(defn twice "Doubles its argument." [x] (* x 2))

You’ll often see this written on multiple lines like this:

(defn sum-of-squares
  "Given two values, return the sum of their squares."
  [a b]
  (+ (* a a) (* b b)))

It’s common for functions to have short argument names because each function should be very short and simple, and the function names should be descriptive.

The docstring – and the source code – of functions is available in the REPL (and in any Clojure editor that supports live evaluation) through the doc – and source – functions (in the clojure.repl namespace). This makes it very easy to experiment with new libraries since all of the documentation and source code is right there at your fingertips!

if, when, and do

In CFML we have an if statement. If the condition is true, the first group of statements is executed, else the second group of statements is executed. In functional languages like Clojure, if is an expression (a special form) and it takes a condition and two expressions. If the condition is true, the first expression is evaluated, else the second expression is evaluated. It’s more like the ternary operator ? : in CFML.

Also, it’s important to note that Clojure has strong views on what is true and false. You’ll hear of “truthy” and “falsey” in the Clojure world because Clojure treats everything as true except for false and nil. In particular, that means that 0 is true, unlike in CFML. The reason for this is an idiom called “nil punning”: it’s common for Clojure functions to return nil for “no such value” instead of throwing an exception, and allowing nil to mean false makes it easy to test for such things:

(if (:name rc) (str "Hello " (:name rc) "!") "Who?")

In CFML, if you did rc.name and it wasn’t present, you’d get an exception. In Clojure, (:name rc) simple returns nil if there’s no such key and you just test for that. You can prevent repetition of (:name rc) with if-let – see below.

If you don’t have an “else” expression (and want to return nil instead), use when instead of if:

(when (:name rc) (str "Hello " (:name rc) "!"))

If the condition is “falsey”, you get nil back. Be aware that when allows multiple expressions, just like fn above and let below, and it evaluates all of them but only returns the value of the last expression.

If you want to evaluate multiple expressions in an if, you need do:

(if some-condition

do can be used where any single expression is accepted and behaves like the body of fn.

It might not surprise you to learn that when is really defined in terms of if and do something like this:

(when condition expr1 expr2 expr3)
;; is treated like
(if condition (do expr1 expr2 expr3) nil)

let and its cousins

Global bindings are fine for functions that you want exposed to the world, but you often want local bindings inside a function. That’s what the let special form is for:

(defn sum-of-squares
  "Given two values, return the sum of their squares."
  [a b]
  (let [a-squared (* a a)
        b-squared (* b b)]
    (+ a-squared b-squared)))

let is followed by a vector of bindings – pairs of symbol and expression – and then one or more expressions. It creates a local binding for each pair by evaluating the expression and binding it to the name (symbol), and then it evaluates the expressions in its body and returns the last expression’s value (in the same way function bodies are evaluated).

There are several variants of let that are useful. The most common is probably if-let:

(if-let [name (:name rc)]
  (str "Hello " name "!")

Unlike let, if-let only allows one binding pair followed by two expressions. If the binding is “truthy”, the first expression will be evaluated, else the second expression will be evaluated. Note that the bound symbol is only available in the first expression!

loop and recur

Since I often tell people that functional programming means no mutable variables and therefore no loops, it might surprise you to learn that Clojure has a loop construct. It isn’t quite what it seems!

First, let’s look at a recursive function:

(defn fact [n] (if (zero? n) 1 (* n (fact (dec n)))))
;; (fact 1) => 1
;; (fact 3) => 6
;; (fact 5) => 120

Because it is recursive, it creates a new stack frame for every call to itself. You can still call it with some big numbers but eventually you will get a stack overflow:

(fact 1000N) => 402387260077093773543702433923003985719374...
(fact 10000N) => StackOverflowError   java.math.BigInteger.valueOf (BigInteger.java:1098)

There’s a technique called tail recursion that allows some languages to evaluate recursive functions without needing stack frames. You have to rewrite the function to ensure the recursive call is in the tail position (the fact call above is nested inside multiplcation). Often, you need to create a helper function:

(defn fact-helper [n prod] (if (zero? n) prod (fact-helper (dec n) (* n prod))))
(defn fact [n] (fact-helper n 1))

This produces the same results as fact above, but now the recursive call to fact-helper is in the tail position. Clojure doesn’t optimize this directly but provides recur for you to tell the compiler you want it optimized – and to verify you really are using tail recursion:

(defn fact [n] (if (zero? n) 1 (* n (recur (dec n)))))
;; CompilerException: Can only recur from tail position


(defn fact-helper [n prod] (if (zero? n) prod (recur (dec n) (* n prod))))
(defn fact [n] (fact-helper n 1))
(fact 10000N) => 28462596809170545189064132121198688901480514...

You can see that recur replaces the recursive call, but what it’s really doing behind the scenes is rebinding the function arguments and “jumping” back to the beginning of the function, to avoid a function call altogether. Since this is really a loop-with-rebinding, Clojure provides a loop construct that lets you do this directly:

(defn fact [n] (loop [n n, prod 1] (if (zero? n) prod (recur (dec n) (* n prod)))))

This single function is identical in behavior to the fact / fact-helper pair shown immediately above. You can see how the loop binding behaves just like the initial call to fact-helper, binding facts argument n to the local symbol n and 1 to the local symbol prod.

Useful core Functions

Since Clojure is all about data structures, there is a rich selection of functions that operate on them. Some of the valuable ones to know are:

  • (first [1 2 3 4]) returns the first element of a sequence: 1
  • (rest [1 2 3 4]) returns the rest of a sequence: (2 3 4)

The rest of a single element sequence is an empty sequence – () – and the rest of an empty sequence is also an empty sequence, as is rest of nil!

  • (seq some-collection) returns a sequence of some-collections elements if it is non-empty, else returns nil

It is common to test for empty sequences with if (seq some-collection).

  • (cons 1 [2 3 4]) returns a new sequence with that element added: (1 2 3 4)
  • (assoc {:a 1} :b 2) returns a new hash map with the key associated to the value: {:a 1, :b 2}
  • (dissoc {:a 1, :b 2} :a) returns a new hash map with the key removed (dissociated): {:b 2}
  • (map inc [1 2 3 4]) returns a new sequence with the function applied to each element: (2 3 4 5]
  • (filter even? [1 2 3 4]) returns a new sequence containing just the elements for which the predicate is “truthy”: (2 4)
  • (reduce + 0 [1 2 3 4]) returns the value computed by repeatedly applying the function to the initial value and each element of the sequence: 10

If the initial value is omitted, the first element of the sequence is used so:

(reduce func some-collection)
;; is the same as
(reduce func (first some-collection) (rest some-collection))

The project.clj File

The piece of project.clj you’ll touch most often is the :dependencies entry. This is a list of all the libraries your program needs and the versions of each you want to use:

:dependencies [[clj-time "0.9.0"]
               [org.clojure/clojure "1.8.0"]]

Libraries come from two locations by default: Maven Central and Clojars.

Most Clojure libraries tell you what to put in project.clj to pull them in so you mostly won’t need to care which location they actually come from but it’s instructive to at least know how to read this stuff and how to search for libraries yourself.

Each entry is called a “coordinate” and contains a “group ID” and an “artifact ID”. A single name just means that the group ID and artifact ID are the same thing (so clj-time is shorthand for clj-time/clj-time).

To search Maven Central for org.clojure/clojure you would use the query g:"org.clojure" AND a:"clojure" which asks for group ID org.clojure and artifact ID clojure. Right now there are over 90 versions of that library on Maven Central and the latest is 1.9.0-alpha3 but if you click the All (92) link, you’ll see the most recent non-prerelease version is 1.8.0 which is what Leiningen puts in project.clj by default (in Leiningen 2.6.1 and later).

On the other hand, clj-time comes from Clojars because it is a community project. If you search for clj-time you’ll get a lot of results but most of them are not canonical versions. The most recent canonical version is https://clojars.org/clj-time but there are other, earlier canonical versions, such as https://clojars.org/backtype/clj-time so you need to be a bit careful. If in doubt, get on IRC, Slack, or the mailing list and ask!

For an overview of all the possible settings in project.clj, take a look at the Sample project.clj File on GitHub.

The build.boot File

As with Leiningen’s project.clj file, the piece you’ll touch most often in Boot’s build.boot file will be the :dependencies entry (in a (set-env! ..) call). The only additional wrinkle with Boot is that you’ll need a boot.properties file as well if the version of Clojure you want to use in your project is not the same as the default version that your installation of Boot already uses. For example, at World Singles we’re using Clojure 1.8.0 RC 4 so our boot.properties file contains:


and then in our build.boot file we specify [org.clojure/clojure "1.8.0-RC4"] as a dependency. This is because Boot uses the same JVM for itself as for your project’s code (whereas Leiningen uses two separate JVMs).

Boot’s build file is “just Clojure” so you can do pretty much anything you want there. Boot provides a number of important functions for setting up and managing projects but the main one you’ll use is set-env! to specify the libraries you need (:dependencies), where your resources – source files etc – are (:resource-paths), and where your tests are (:source-paths). For more detail, consult the Boot web site.

About Functional Programming

Functional programming isn’t new. It’s origins lie in Lisp which was created in the 1950’s and is the second-oldest computer language (second only to FORTRAN). Throughout the 70’s and 80’s a lot of functional languages were created, mostly in academia, to study the benefits of the functional style, as well look at levels of expressiveness in programming languages. Classic functional languages include Standard ML, Miranda, and Haskell. Haskell was a result of the proliferation of similar functional languages being created by each university in England (and elsewhere). It was decided that a single, committee-designed functional language should exist that included the best ideas of all of the diverse variants out there. Haskell is probably the most widely used language today from that era. It has an extremely powerful type system and a very strong view of purity – lack of side effects – but it has been used extensively over the last 25 years in industry as well as academia.

The recent resurgence of functional programming has shown itself in languages like F# from Microsoft, Scala, and Clojure, even Rust, as well as some compile-to-JS languages like Elm and PureScript. The reason behind this resurgence is that immutable data structures and pure functions offer the ability to write concurrent code a lot more easily and lot more safely than the mainstream OOP approach. And we need concurrency in order to take advantage of multi-core machines, now that we’re no longer seeing continued speed increases in individual cores like we had for the previous several decades.

While it may seem obvious that functional programming leans heavily on functions as building blocks, the real core values of functional programming are avoiding mutable data and avoiding side effects in functions. The more that you can push side effects to the edges of your program, the more of your code becomes pure functions that can be easily reasoned about, easily tested, and often easily reused. Functional programming focuses on small, pure functions that can be composed to create larger pieces of functionality. If you have a function inc that adds one to its argument and a function twice that doubles its argument, then (comp twice inc) is a function that adds one to its argument and then doubles it: functions are like Legos that you can easily assemble to build products.

Immutable or Persistent Data Structures

In the context of functional programming, you’ll hear a lot of talk of immutable data structures and persistent data structures. In OOP languages, you typically perform operations on a data structure to modify it in place. That means you can’t safely share it with other pieces of code, especially across multiple threads. By contrast, in a functional language, when you perform an operation on a data structure, you get back a new data structure that shares as much structure as possible with the original data structure – and yet leaves the original data structure unchanged.

While they are designed for efficiency, it is usually at scale, rather than for small examples. In Clojure, many data structures are “chunked” internally into groups of 32 elements. A vector of 100 elements is going to be four chunks and is optimized for adding elements to the end of the vector. A list is optimized for adding elements at the start of it. A hash map is also chunked and optimized for adding elements in random locations. They are also optimized for different patterns of access: a list is optimized for purely sequential access, a vector for indexed (random) access, and a hash map for keyed (random) access.

Clojure and Scala share a lot of heavily optimized implementation details in their persistent data structures.

Despite all this efficiency, you still need to think about how to use data structures, and there are going to be some algorithms where bashing a data structure in place is just going to be faster. It won’t be as safe, just faster. Clojure is fine with the idea of localized mutation and has versions of vectors and hash maps that are optimized for that purpose (known as transients).

The most important aspect of these data structures in Clojure is the set of abstractions over them, including “sequence” and “associative”. These are uniform ways to think about data structures as just sequences of data (accessed in order, or randomly by keys), no matter what their actual implementation is, so that you can apply all of Clojure’s core functions to arbitrary data structures in standardized ways. The lack of unique types for each data structure you use means that you don’t need unique functions to operate on them all. A function that operates on a sequence can accept any data structure that supports the sequence abstraction. A function that operates on an associative collection can accept any data structure that supports the associative abstraction.

For example, a result set from a database query is simply a sequence of associative collections. It can be mapped, filtered, reduced using standard functions and its rows transformed using any of the standard associative functions. This allows abstraction over different data stores as well since, from Clojure’s point of view, MySQL and MongoDB look very similar.

Functions as Building Blocks

I’ve already emphasized that functional programming favors small, simple, pure functions but you can do that sort of functional decomposition in most languages. Many modern languages allow you to write anonymous functions and pass them around as arguments, as well as return them from functions. Even with those features available, it takes a while to shift from an imperative style with some functions being passed around to a functional style where functions are your primary abstraction.

You’ll hear the term “higher order functions” a lot but all this means is a function that accepts a function as one (or more) of its arguments or a function that returns a function as a result. You’ll see the former in CFML with the new map, filter, and reduce member functions – or perhaps you’ve seen it in JavaScript.

Consider these two functions (in CFML):

function saveStuff( data ) {
    if ( validStuff( data ) ) {
        dbSave( "stuffTable", data );
    } else {
        writeLog( "Invalid stuff, not saved" );

function saveThing( data ) {
    if ( thingIsValid( "all", data ) ) {
        dbSave( "thingTable", data );
    } else {
        writeLog( "Invalid thing, not saved" );

Now, consider this function:

function saveValidData( validator, type, data ) {
    if ( validator( data ) ) {
        dbSave( type & "Table", data );
    } else {
        writeLog( "Invalid #type#, not saved" );

This is probably not a transform you would do in CFML but it’s a natural one in a functional language, because you would see the commonality and want to remove the duplication:

var saveStuff = function( data ) { return saveValidData( validStuff, "stuff", data ); };
var saveThing = function( data ) {
    return saveValidData(
        function( data ) { return thingIsValid( "all", data ); },

Now imagine you had a higher order function called partial that accepted a function and one or more of its arguments and returned a new function that accepted the rest of its arguments and then called it, it would be very natural to make this transform and then write:

var saveStuff = partial( saveValidData, validStuff, "stuff" );
var saveThing = partial( saveValidData, partial( thingIsValid, "all" ), "thing" );

This is much cleaner, especially if you had quite a few of these “save data if valid” variants.

Another way to write this without partial would be:

function thingIsAllValid( data ) { return thingIsValid( "all", data ); }
function saveValidData( validator, type ) {
    // accept validator and type, return a function that accepts data
    return function( data ) {
        if ( validator( data ) ) {
            dbSave( type & "Table", data );
        } else {
            writeLog( "Invalid #type#, not saved" );

var saveStuff = saveValidData( validStuff, "stuff" ); // returns a function
var saveThing = saveValidData( thingIsAllValid, "thing" );

You could also write “thing” validation like this:

function thingValidator( scope ) {
    return function( data ) {
        return thingIsValid( scope, data );

var saveThing = saveValidData( thingValidator( "all" ), "thing" );

By making our functions more flexible in how they accept arguments, we make it easier to reuse them. This is functional thinking!

In Clojure we can define a function with multiple argument lists so this becomes even easier:

(defn thing-is-valid
  ([scope] (fn [data] (thing-is-valid scope data)))
  ([scope data] ... return true or false ...))

(defn save-valid-data
  ([validator type] (fn [data] (save-valid-data validator type data)))
  ([validator type data]
   (if (validator data)
     (db-save (str type "Table") data)
     (write-log (str "Invalid " type ", not saved")))))

(def save-stuff (save-valid-data valid-stuff "stuff"))
(def save-thing (save-valid-data (thing-is-valid "all") "thing")

No need for partial (although Clojure has that built-in), no need for helper functions.

All You Know About OO Programming is Wrong

I learned old-fashioned imperative procedural programming first. I learned BASIC, assembly language, Pascal and later COBOL and FORTRAN. Although OOP has its roots back in the 50’s and 60’s (ironically, with Lisp, just like FP has its roots in Lisp), it didn’t really go mainstream until the mid-to-late 80’s with the arrival of Eiffel and C++. I learned a lot of FP during the 80’s but since it wasn’t going mainstream and OOP was, I switched horses to stay employable. By the time Java appeared, OOP had become the default “standard” way to build software, even though C++ and Java were not at all what Alan Kay had in mind when he coined the phrase “object-oriented”.

What most developers know as modern OOP focuses on polymorphism, inheritance, encapsulation and objects that have both state and behavior bundled together. You construct an object with its initial state. You run a bunch of methods on it, interacting with other objects, to modify its state, and then you query the object to get that state back out.

A common idiom in the OOP world for collections is an iterator. In CFML this shows up in query objects. As you loop over the iterator, it changes its state to refer to successive elements of the underlying collection. A CFML query refers to successive rows of the result set as you loop over it and the currentrow element is updated at each iteration. Instead of a result set looking like a sequence of rows, we’re used to looking at a snapshot of the “current row” – and that’s what iterators do to us for collections as well.

In addition, instead of processing collections holistically to produce either new collections or specific results, we’re using to iterating through the elements and either modifying them in place or performing side effecting operations along the way.

What makes this problematic is that you can’t then easily run this code concurrently to take advantage of multiple cores: code that mutates collections in place or generates side effects is rarely thread safe.

In other words, mutable state is bad.

What about OOP’s other basic tenets?

Why do you encapsulate data? The primary reason is so that you can control mutation of that data. If your data is immutable, encapsulation is no longer needed for that. The other argument for encapsulation is so that you can change the representation of the state without affecting your clients. In a functional world, if you change your representation, you can always provide a function that transforms it to the original structure and then simply compose that transformation with any client function that needs to access it. That composed client + transform can be refactored away over time as the client transitions to the new API. In other words, your data is your API and there’s no need to encapsulate it (at least, not for the traditional OOP reasons).

Inheritance? Inheritance exists in OOP because the notion of data types is inherently tied to classes and objects. Inheritance represents a strong coupling between an implementation and an interface or between one implementation and another related implementation. It exists because there’s no way to separate out the notion of data types and their relationships from the class implementation relationships. Needless to say, in a functional world, you can choose to have relationships between data types as you need them, without being forced to create relationships between data representations. In Clojure, in particular, you can create hierarchies of types independent of any data and use those to guide function call dispatching.

Which brings us nicely to polymorphism! Polymorphism is great. It’s very useful. Unfortunately, in OOP it is tied to inheritance which, as we’ve just seen, is all about coupling when you’re dealing with classes. If you only ever use interfaces and pure implementations, you can free yourself from some of the problems of coupling forced on you by OOP, but you still get stuck if you have a class (implementation) that isn’t declared to implement an interface that you need to use, even though it has the right methods. Nor can you easily take an arbitrary existing class and make it implement your interface (you can extend the class and implement your interface, only if the class is not final in Java, for example). In addition to all that, polymorphism in OOP is only effective on the first argument – the object type itself – which means that when faced with more complex problems, you have to resort to design patterns like Visitor and implement double dispatch. But then you are forced to modify the “visited” class every time you want to visit it with a new class. What you really need there is polymorphism based on multiple arguments. Fortunately, you can have that in functional programming.

So what have we learned?

Polymorphism is very limited in OOP but can be very powerful in FP once we remove the restriction of single dispatch and the coupling to a static inheritance hierarchy: this gives you ad hoc or a la carte polymorphism.

Inheritance as seen in OOP is essentially an implementation detail. You can have ad hoc inheritance in FP, which provides expressiveness where you want it and avoids boilerplate and coupling where you don’t.

Encapsulation is required in OOP when you have mutatable state and is also needed in order to allow changes in implementation. In other words, encapsulation is also essentially an implementation detail. FP solves this by making data immutable and making functions easily composable.

Mutable state is just bad. It prevents refactoring to leverage concurrency, it leads to hard to find bugs in complex programs, and it also erases any notion of time in your program (because you have only the current version of the state, rather than the series of values that were transformed to get to that point). FP solves that by removing mutability.

If FP is so great, why aren’t we all using it already?

Good question. The OOP industry is vast. Design Patterns, training and consulting, higher education based on teaching OOP (ironically after supplanting a lot of courses that taught FP!), testing, tooling, IDEs. The momentum behind OOP is huge and the inertia of industry to keep doing things the way they know is almost overwhelming. Yet we see functional features in nearly every new language being designed, we see functional features being added to nearly every existing language over time, we see the functional style of programming being advocated even in traditional languages – with less reliance on mutable state. Most of that pressure is coming from the need to do more concurrency and to avoid the bugs that arise from side-effecting code. In other words, a lot of people already know that FP is a better way to solve a lot of problems.

Remember that modern OOP – as enshrined in Java and C# particularly – is not what the originators of OOP had in mind. They imagined objects as proxies for real world elements such as displays and control devices, that objects would be coarse-grained and communicate by sending messages between themselves. In other words, they would be more like “actors”… which you have in both Clojure (as agents) and Scala (as actors)!

More Stuff to Read

Once you’ve got a taste for Clojure, there are lots of online resources and a host of great books you can read. Here’s a small sample, roughly in order of approachability:

  • The online tutorial for learning Clojure, no installation required: http://www.tryclj.com
  • The “4Clojure” puzzles online: https://www.4clojure.com
    • These quickly get hard enough that you’ll want a REPL open locally to play with!
  • The Clojure Koans: http://clojurekoans.com
    • You need at least Java, Leiningen, and Git installed for these.
  • Some great books to read:
    • Clojure for the Brave and True http://www.braveclojure.com (introductory level)
    • Living Clojure http://shop.oreilly.com/product/0636920034292.do (introduces Clojure, but assumes you’re an experienced developer in other languages)
    • Clojure Programming http://www.clojurebook.com (deep technical coverage)
    • The Joy Of Clojure http://www.joyofclojure.com (focused on the “why” of Clojure)

Digging Into Reloading

More to come on this, eventually!