Updating Clojure namespaces

Continuing my experimentations with Emacs Lisp, I created a function that automatically updates the namespace of a Clojure buffer in conformity with its pathname. This is useful after renaming a file, since changing a file pathname requires to change its namespace, which is sometimes annoying to do by hand.
Once added to your .emacs file, you can call it with:

M-x clojure-update-ns

It assumes clojure-mode is already loaded.

Earmuffs and variables

According to the Clojure coding standards for libraries, variables names should be written with *earmuffs* only when they are intended to be rebind.

Here and there I forget this rule and use earmuffs most of the time, so I decided to create an Emacs Lisp function to add/remove earmuffs to variable:





Add it to your .emacs file and then place your cursor on a variable and call it with

M-x earmuffy

to add earmuffs, or type

C-u M-x earmuffy

to remove them.

Dear language designers, do not forget Ada

These last days I took some time to refresh my old memories of the Ada language. While Ada, like all languages, has defaults on its own I'm still very impressed by some functionalities it offers and that current mainstream languages still don't have. Most of these functionalities are there since the first standardized version in 1983 (since then the language standard was extended in 1995, 2005 and the next version is expected for 2012).

The first impressive future is about primitive types. Here are some of the types defined in Ada and their C equivalent (taken from [1]):

Ada Type	Description	C Equivalent
Character	A single character	char
Integer	An integer (32-bit) number	int
Natural	Zero or positive integer	-
Positive	A positive integer	-
Long_Integer	A big integer (same as long in Gcc)	long (same as int in Gcc)
Long_Long_Integer	A really big (64-bit) integer	long long
Short_Integer	A small (16-bit) integer	short
Short_Short_Integer	A really small (8-bit) integer	char
Float	A real number	float
Long_Float	A big real number	double
Long_Long_Float	A really big real number	long double
Short_Float	A smaller real number	?
Fixed	A fixed-point real number	-
String	An Ada fixed-length string	char array

Boolean types are also provided. New types are defined and type safety is guaranteed. Here we define a new type based on a Long_Float and a variable of this type:

type Speed is new Long_Float;

Speedy_Gonzales_Speed : Speed;

Since Speed is a new type, it is not a Long_Float and thus assignment from a Long_Float value to a Speed variable is forbidden:

X : Long_float := 300.0; 
Speedy_Gonzales_Speed := X;

The compiler effectively gives an error:

expected type "Speed" defined at line 6
found type "Standard.Long_Float"

When it makes sense, we can define a subtype of another primitive type. Assignments between types and their subtypes is allowed. Here we define the type Degree which is a subtype of Float:

subtype Degree is Long_Float;

Oven_Temperature : Degree;
Y : Long_Float := 255.0;

Oven_Temperature := Y;

Types can be constrained with a range definition (taken from [2]):

type Degrees is new Float range -273.15 .. Float'Last;

This allow a Degrees variable to range from -273.15 (absolute zero) to the last value allowed by a Float.

This strong typing provides safety and contrasts with what is seen in current mainstream programming languages. Languages like Java don't provide rich primitive types (there are for instance no unsigned integer in Java!) and require the programmer to define its own cumbersome classes if he wants new types (which poses readability problems if the types are new number types and the language does not support operator overloading). Some languages like C don't even bother so much about type conversion: a float can be assigned to an integer. This leads to solutions which are far from being satisfying (see for instance C FAQ - round).

A lot of other possibilities are offered to the Ada programmer, for instance decimal types can be defined with their precision. The precision is then guaranteed by the compiler:

type money is delta 0.01 digits 18;

Ada also supports multidimensional arrays, bit-level memory access, definition of memory pool, concurrency programming in a task-oriented way, object-oriented programming, generic packages etc.

At another level Scheme and Common Lisp supports a numeric towel. For example, 3 is an integer. Therefore 3 is also a rational, a real, and a complex (example from the Scheme standard).

;; in Scheme:
> (rational? (/ 1 5))
#t
> (rational? 3)
#t
> (real? 3)
#t

While this discussion may not make so much sense for dynamic languages such as Clojure, I think new languages being designed, which are not dynamically typed, could really benefit of having such rich primitives types. A complex type systems like the one from ML or Haskell is not needed for this purpose and primitives types is the most basic and most used feature of a programming language, so dear languages designers, next time, have a small thought for Ada.

[1] http://www.pegasoft.ca/resources/boblap/book.html
[2] http://en.wikibooks.org/wiki/Ada_Programming

Fed up with typing ns declarations?

While watching the demonstration video of the Play framework, I saw that the developer had some nice templates for its TextMate editor. A few Google searches tells us that it is also possible to have the same functionality in Emacs with the YASnippet template system.

Typing all these long namespaces declarations in Clojure is quickly boring when you create a lot of files, so why not create a template for that?

After having installed YASnippet, create a clojure-mode directory inside the yasnippet/snippets/text-mode directory and create a file named ns with this content:

(ns `(let* ((nsname '())
        (dirs (split-string (buffer-file-name) "/"))
        (aftersrc nil))
     (dolist (dir dirs)
        (if aftersrc
            (progn
                (setq nsname (cons dir nsname))
                (setq nsname (cons "." nsname)))
             (when (or (string= dir "src") (string= dir "test"))
                (setq aftersrc t))))
     (when nsname
       (replace-regexp-in-string "_" "-" (substring (apply 'concat (reverse nsname))  0 -5))))`
     (:use $1)
     (:require ))

Now, when inside a Clojure buffer type "ns" and TAB to complete ; if you are for instance in the src/mylib/utils/swing_stuff.clj buffer, this will be expanded into the following text:

(ns mylib.utils.swing-stuff
  (:use )
  (:require ))

Isn't that handy?

Note: this is my first hack with Emacs Lisp, and now that it is working I'm publishing it without any further improvements, so be indulgent regarding the implementation!

Ray tracing, once again

There are already several implementations of simple raytracers in Clojure on the Web but I couldn't resist implementing one for myself! It is based on the one of the book Ansi Common Lisp by Paul Graham.

Nothing particular, just perhaps the use of protocols for the dispatch of the normal, color and intersect functions and some code for displaying spirals.

Code is available here: GitHub Kototama

when-let maybe?

In this post we will see how to incrementally develop a macro similar to when-let but more flexible.


When-let is useful to both bind one variable and do a test on the bind value in one operation. One common usage is:

(when-let [something (get-something arg1 arg2)]
   (do-stuff something))

This is equivalent to this code:

(let [something (get-something arg1 arg2)]
   (when something
      (do-stuff something)))

but with a more concise form.

When-let only executes the code after the binding if the value assigned in the let form is logically true, that is only if the value is not nil or false. In our example, this means that if get-something returns false then do-stuff will not be executed. Sometimes this is not enough and we want to execute the code even for false values. For instance if we are getting our data from a database and false values are acceptable but nil values are not, or if our functions return nil on error.

We could define a when-nlet macro which does what when-let does but executes the body whenever the binded value is not nil. By spying the code from clojure.core we obtain:

(defmacro when-nlet [bindings & body]
  (when (not= (count bindings) 2)
    (throw (IllegalArgumentException.
            "when-nlet requires an even number of forms in binding vector")))
  (let [form (bindings 0)
        tst (bindings 1)]
    `(let [temp# ~tst]
        (when (not (nil? temp#))
          (let [~form temp#]
            ~@body)))))

This is fine. But what if we need multiple values to be bound and multiple checks on them?

We could write:

(when-nlet [val1 (get-something arg1 arg2)]
  (when-nlet [val2 (get-something2 val1)]
     (when-nlet [val3 (get-something3 val2]
        (do-stuff val1 val2 val3))))

This is not satisfying. What about writing a when-nlet* macro that does multiple binds?

We could call it like that:

(when-nlet* [val1 (get-something arg1 arg2)
             val2 (get-something2 val1)
             val3 (get-something3 val2)]
          (do-stuff val1 val2 val3))

and it would produce multiple calls to when-nlet.

Here it is:

(defmacro when-nlet* [bindings & body]
  (when (not (even? (count bindings)))
    (throw (IllegalArgumentException.
            "when-nlet* requires an even number of forms in binding vector")))
  (let [whenlets (reduce (fn [sexpr bind]
                           (let [form (first bind)
                                 tst (second bind)]
                             (conj sexpr `(when-nlet [~form ~tst]))))
                         ()
                         (partition 2 bindings))
        body (cons 'do body)]
    `(->> ~body ~@whenlets)))

After the reduce the whenlets variable is assigned this list (we can see it by playing with macroexpand, macroexpand-1 and prn at the REPL): 

((when-nlet [val3 (get-something3 val2)]) (when-nlet [val2 (get-something2 val1)]) (when-nlet [val1 (get-something arg1 arg2)]))

We then thread the body inside the when-nlets forms with the powerful ->> macro. We obtain:

(when-nlet [val1 (get-something arg1 arg2)]
  (when-nlet [val2 (get-something2 val1)]
    (when-nlet [val3 (get-something3 val2)]
      (do (do-stuff val1 val2 val3)))))

So basically what we have done is creating a macro that does multiple binds, stops after the first bind returning a nil value and executes its body if no nil value has been encountered. It is nice and it shows us how much powerful Clojure is but... if we read this tutorial on how to use monads in Clojure we quickly see that there is a simpler way to do the same thing with the maybe monad!

(domonad maybe-m [val1 (get-something "a" "b")
                  val2 (get-something2 val1)
                  val3 (get-somnil val2)]
               (do-stuff val1 val2 val3))

This code is equivalent to our usage of when-nlet*. Thus, if we still feel the need for our when-nlet* macro, we could write it simply like that (after :using clojure.contrib.monads):

(defmacro when-nlet* [binding & body]
  (let [body (cons 'do body)]
   `(domonad maybe-m ~binding ~body)))

Is that not much better?

Conclusion: we shall not write macros before learning more about monads ;-) !

How to build a GUI with NetBeans and Clojure

In this post we show how to create a simple GUI with the powerful Swing GUI builder of NetBeans, use it from Clojure and deliver it as a self-executable JAR. The application is a text generator similar to the Henley program of the chapter 8 of Ansi Common Lisp by Paul Graham: it reads some input file to generate statistics and use them to generate a random text. Throughout the design of the application we emphasize on a clean separation between the presentation logic and the logic of the data model with a MVC pattern.

Here are the tools required to create and build the application:

• Java 1.6
• Apache Maven
• NetBeans
• Leiningen

Creating the GUI

We use Leiningen as our Clojure build tool and it plays nicely with Maven, so we create a Maven project in NetBeans for the GUI. The project consists of Swing components, in our example just one JFrame. We do not write any Java code and we just use the NetBeans GUI builder to create components without any logic. The logic will be in the Clojure code. This Maven project will be a dependency of our Clojure project.

First we make a new NetBeans project by clicking on File/New Project.../Maven/Maven Project and selects Maven Quickstart Archetype which is nearly an empty Maven project. We name our project henleyuicomponents. Clojars forces you to have lower-case JAR names, so we choose a lower-case name for our project, useful if we want to publish it later. We set the Group Id to henley, the version to 1.0.0-SNAPSHOT and the package to henley.uicomponents.

We create a new JFrame called MainFrame by right-clicking on the package name in the Projects tree, and selecting "JFrame Form...". We then design the application with the GUI builder and make it looks like this:

For each Swing components (JButton, JText etc.) that we need to access to from our Clojure code, we right-click on it in the GUI builder, select "Customize code", rename it with a meaningful name and set its Access to public. When we are finished designing the GUI we can build it by right-clicking on the project name and choosing "build". This will invoke Maven with the command "mvn install". Alternatively you can go in the project directory and invokes this command with a shell. Once build, the Maven artifact is installed in your local Maven repository (in ~/.m2 on Linux for instance).

We now need to connect the Swing components to our Clojure code.

Wiring the components

We create an "henley" project with Leiningen with "lein new henley" with the following content for the project.clj:

(defproject henley "1.0.0-SNAPSHOT"
  :description "A GUI for a text generator similar to the one
 in the book Ansi Common Lisp, Paul Graham"
  :dependencies [[org.clojure/clojure "1.2.0"]
                 [org.clojure/clojure-contrib "1.2.0"]
                 [henley/henleyuicomponents "1.0.0-SNAPSHOT"]]
  :dev-dependencies [[swank-clojure "1.2.0"]
                     [lein-run "1.0.0-SNAPSHOT"]]
  :main henley.core
  )

The Maven component is listed as a dependency. The :main option specify where the main function is located. This is necessary to create a self-executable JAR.

We follow a MVC pattern where the data of the model are pushed from the controller to the view. The view consists of the Swing components in the Maven component; for our simple example it is only the JFrame. Let us see how our project is organized, here is the project tree:

• The model directory contains the logic of the application, in our case one namespace and its functions to generate some text.

• The view directory contains all the code necessary to use the Swing components. We define two protocols, View and SwingUI: they are used by the controller and allow to abstract the details of the particular graphical implementation. The View protocols is for all functions that are totally independent of Swing from the point of view of the controller. The SwingUI is for all functions that are relative to Swing, like attaching a listener. They are defined in the files view.clj and swingui.clj. The Swing components inside the Maven artifact are referenced in the uicomponents.clj file. Application.clj provides an implementation for the two protocols.

• The controller directory contains the register.clj function which uses the SwingUI interface to register Swing listeners defined in swing_listeners.clj. These listeners can use the SwingUI interface to get some information from the Swing components, for instance the number of words defined by the user in the JText field. This is indeed their purpose: extract the information and calls the function defined in handlers.clj. The functions in handlers.clj are callbacks and do not have access to the SwingUI interface, all the relevant information has been extracted by the swing_listeners and they only access the interface through the View protocol. The function in the handlers uses the model and informs the View of any changes.

Developing the application logic

In our case the logic of the application is just the text generation, a functional translation in Clojure of the Henley program available in the book ANSI Common LISP.

Building and using the application

We build and install the Swing components with the command "mvn install" launched from the henleyuicomponents directory. We then go in the henley directory and call "lein deps" to resolve all dependencies. If we want to build a self-executable we call "lein uberjar" and go outside for a walk; when we are back we should have a standalone JAR. If not, we may have more success by installing cake and do "cake uberjar". The JAR can be executed with "java -jar jarname".  
If you have a problem building the JAR you can comment the :main option in the project.clj file and type the command "lein run henley.core -main" from the henley directory to launch the application.
How to use the application? We can for instance generate a french "poem" by using the baudelaire.txt file in the test directory as an input file:

Abstraction levels

We have a lot of files and two protocols just for a simple project. What kind of abstraction do they defined?

• uicomponents.clj allows us to access the Swing components independently of the way they are defined. They could be defined with code manually, with an Eclipse project etc. We choose the NetBeans GUI builder because it is the best free Swing builder available (to the extend of our knowledge).
  
  
• the SwingUI protocol allows the controller to access the Swing components in an implementation-independent way. If for instance the Number of words JText field becomes later a JSpinner this will not affect the controller: the details of the implementation are already abstracted by the protocol.
  
  
• The callbacks defined in handlers.clj, which are the heart of the controller, are independent of the GUI. The GUI could be written in SWT: this will not affect them. At this level, the View protocol abstracts the GUI implementation.
  
  
• By pushing data from the controller to the View, the GUI is independent of the model data. It does not matter in a simple example as ours, but it will on a bigger project.

Conclusion

We have a clean and very flexible design but with one constraint: a lot of functions are defined just to do a few operations. What do you think of this design? Do you see a way to simplify it without losing flexibility?

Links

The self-executable JAR can be download here
The code is available on Git: https://github.com/kototama/henley