I made a number of Google searches of the forms below and summed the results (previous posts averaged the results):

"implemented in <language>"
  "written in <language>"

Naturally this is of very limited utility, and the numbers are only useful when comparing relatively within the same search since the number of results Google returns can vary greatly over time.

¹ combines Lisp, Scheme, Common Lisp, Arc & Clojure
² combines OCaml, (S)ML, Caml
³ summed separate searches for sml and ml

I made a number of Google searches of the forms below and averaged the results:

"implemented in <language>"
  "written in <language>"

Naturally this is of very limited utility, and the numbers are only useful when comparing relatively within one column since the number of results Google returns can vary greatly over time.

¹ combines Lisp, Scheme, Common Lisp, Arc & Clojure
² combines OCaml, (S)ML, Caml
³ summed separate searches for sml and ml

I made a number of Google searches of the forms below and averaged the results:

"implemented in <language>"
"written in <language>"

¹ combines Lisp, Scheme, Common Lisp, Arc & Clojure
² combines OCaml, (S)ML, Caml
³ summed separate searches for sml and ml

Chapter 6 – Case Analysis

Tuple types are homogeneous e.g. all values of type int*real are pairs containing an int and a real. Match failures occur at compile time. Types that have more than one form, such as int, are heterogeneous. Pattern matches fail at run time as a bind failure.

ML defines functions over heterogeneous types using clausal function expressions. For example:

fn pat1 => exp1
 | pat2 => exp2
 | ...
 | patn => expn

Each component pat => exp is called a clause, or a rule. The entire assembly of rules is a called a match. For example:

val recip : int -> int =
    fn 0 => 0 | n:int => 1 div n

The fun notation is generalized so we can be more concise:

fun recip 0 = 0
  | recip (n:int) = 1 div n

Case analysis on the values of a heterogeneous type is performed by application of a clausally-defined function. The notation:

case exp
  of pat1 => exp1
   | ...
   | patn => expn

is short for the application:

(fn pat1 => exp1
  | ...
  | patn => expn)
exp

Matches are subject to two forms of “sanity check” – exhaustiveness checking and redundancy checking.

Chapter 7 – Recursive Functions

For a function to be able to call itself, it needs a name. For example:

val rec factorial : int->int =
    fn 0 => 1 | n:int => n * factorial (n-1)

or using fun notation:

fun factorial 0 = 1
  | factorial (n:int) = n * factorial (n-1)

Iteration

If we define a helper function that accepts an accumulator we can reduce the storage needed:

fun helper (0,r:int) = r
  | helper (n:int,r:int) = helper (n-1,n*r)
fun factorial (n:int) = helper (n,1)

It’s better programming style to hide the helper function w/in a local declaration:

local
    fun helper (0,r:int) = r
      | helper (n:int,r:int) = helper (n-1,n*r)
in
    fun factorial (n:int) = helper (n,1)
end

Tail recursive functions are analogous to loops in imperative languages – they iterate the computation w/o needing auxiliary storage.

Mutual Recursion

Definitions which are mutually recursive can be joined together with the and keyword to indicate they are defined simultaneously by mutual recursion:

fun even 0 = true
  | even n = odd (n-1)
and odd 0 = false
  | odd n = even (n-1)

Chapter 8 – Type Inference and Polymorphism

Standard ML allows you to omit type information whenever it can be determined from context. Consider the following:

fn s:string => s ^ "\n"

There is no need to declare the type of s since it can be inferred from the context, so we may write:

fn s => s ^ "\n"

A type scheme is a type expression involving one or more type variables standing for an unknown, but arbitrary type expression. Type variables are written ‘a (pronounced alpha), ‘b (pronounced beta), etc. For example, the type scheme ‘a->’a has instances int->int, string->string, (int*int)->(int*int), and (’b->’b)->(’b->’b), etc. It does not have the type int->string.

We may bind an identity function to the variable I as follows:

val I : 'a->'a = fn x=>x

We may also write:

fun I(x:'a) : 'a = x

Standard ML eliminates the need to ascribe a type scheme to the variable:

val I = fn x=>x

or

fun I(x) = x

Chapter 9 – Programming with Lists

The values of type type list are the finite lists of values of type type:

1. nil is a value of type typ list.
2. if h is a value of type typ, and t is a value of type typ list,
   then h::t is a value of type typ list.
3. Nothing else is a value of type typ list.

The type expression typ list is a postfix notation for the application of the type constructor list to the type typ.

A value val of type typ list has the form:
val1 :: (val2 :: (… :: (valn :: nil) … ))

The :: operator is right-associative, so we may omit parentheses:
val1 :: val2 :: … :: valn :: nil

Or, we may use list notation:
[ val1, val2, ..., valn ]

Computing With Lists

Some examples:

fun length nil = 0
  | length (_::t) = 1 + length t

Note we do not give a name to the head of the list, instead we use a wildcard _

fun append (nil, l) = l
  | append (h::t, l) = h :: append (t, l)

The latter is built into Standard ML and is written using infix as: exp1 @ exp2

fun rev nil = nil
  | rev (h::t) = rev t @ [h]

The running time of the latter is O(n²). The following definition makes use of an accumulator and has a running time of O(n):

local
    fun helper (nil, a) = a
      | helper (h::t, a) = helper (t, h::a)
in
    fun rev' l = helper (l, nil)
end

Chapter 10 – Concrete Data Types

Non-Recursive Datatypes

Example of nullary i.e. zero argument, constructors:

datatype suit = Spades | Hearts | Diamonds | Clubs

It is conventional to capitalize the names of value constructors, but this is not required by the language.

Datatypes may be parameterized by a type:

datatype 'a option = NONE | SOME of 'a

The values are NONE or Some val, where val is a value of type typ. The option type constructor is pre-defined in Standard ML.

Option types can also be used in aggregate data structures:

type entry = { name:string, spouse string option }

An entry for an unmarried person would have a spouse field with a value of NONE.

Recursive Datatypes

datatype 'a tree =
  Empty |
  Node of 'a tree * 'a * 'a tree

1. The empty tree Empty is a binary tree.
2. If tree_1 and tree_2 are binary trees, and val is a value of type type, then Node (tree_1, val, tree_2) is a binary tree.
3. Nothing else is a binary tree.
A function to compute the height of a binary tree, and one to compute the number of nodes:

fun height Empty = 0
  | height (Node (lft, _, rht)) = 1 + max (height lft, height rht)

fun size Empty = 0
  | size (Node (lft, _, rht)) = 1 + size lft + size rht

Heterogeneous Data Structures

The tree data type above requires that the type of the data items at the nodes must be the same for every node of the tree. To represent a heterogeneous tree, the data item must be labelled with enough info to determine the type at run-time.

datatype int_or_string =
  Int of int |
  String of string

type int_or_string =
  int_or_string tree

Datatype declarations and pattern matching can be useful for manipulating the abstract syntax of a language. Consider an example representing arithmetic expressions:

datatype expr =
  Numeral of int |
  Plus of expr * expr |
  Times of expr * expr

fun eval (Numeral n) = Numeral n
  | eval (Plus (e1, e2)) =
    let
        val Numeral n1 = eval e1
        val Numeral n2 = eval e2
    in
        Numeral (n1+n2)
    end
  | eval (Times (e1, e2)) =
    let
        val Numeral n1 = eval e1
        val Numeral n2 = eval e2
    in
        Numeral (n1*n2)
    end

If we extend the expr datatype as follows:

datatype expr =
  Numeral of int |
  Plus of expr * expr |
  Times of expr * expr
  Recip of expr

The compiler will complain about eval being incompatible with the new version of expr. Recompiling eval will produce an inexhaustive match warning since eval lacks a case for Recip. This is one of the benefits of static typing provided in Standard ML.

After bumping into a number of local programmers who expressed an interest in functional programming, I thought it might be a good time to start a local group that focused on functional programming languages, so I did a couple days ago.

TriFunc.org is a group for programmers who are interested in functional programming languages and live near the Research Triangle area of North Carolina.

If you live in the area and have an interest in functional programming languages, feel free to dive in and start participating in the Google Group discussions. Once we reach a critical mass, I expect we’ll produce a meeting schedule, etc., but that will depend on where the group wants to take this.

Chapter 4 – Functions

Lambda expressions are written as:

fn var : typ => exp

For example:

fn x : real => Math.sqrt (Math.sqrt x)
(fn x : real => Math.sqrt (Math.sqrt x)) (16.0)
val fourthroot : real -> real =
  fn x : real => Math.sqrt (Math.sqrt x)
fourthroot 16.0

ML provides a special syntax for function bindings that’s more concise:

fun fourthroot (x:real):real = Math.sqrt (Math.sqrt x)

By experimenting (I expect this is covered later), I found the following is sufficient due to type inference:

fun fourthroot x = Math.sqrt (Math.sqrt x)

Local val bindings may shadow parameters and other val bindings. For example, in the following, the last occurrence of x refers to the parameter of h, but the preceding two occurrences of x refer to the local binding with a value of 2.0:

fun h(x:real):real =
  let val x:real = 2.0 in x+x end * x

Chapter 5 – Products and Records

The n-tuple is the simplest form of aggregate data structure. They are of the form:
(val₁, … ,val_n)
An n-tuple is a value of a product type of the form:
typ₁* … *typ_n
For example:

val pair : int * int = (2, 3)
val triple : int * real * string = (2, 2.0, "2")
val pair_of_pairs : (int * int) * (real * real) = ((2,3),(2.0,3.0))

A 0-tuple, also known as a null tuple, is the empty sequence of values, ( ). It is a value of type unit. 1-tuples are absent from the language.

Tuple Patterns

We can use pattern matching to extract portions of an n-tuple. For example, in the code snippet below, r will be equal to 3.14. Underscores indicate “don’t care” positions:

val foo : (int * string) * (real * char) = ((7,"hello"),(3.14,#"z"))
val ((_, _), (r:real, _)) = foo

We can give names to the first and second components of the pair – using the REPL:

- val (is:int*string,rc:real*char) = foo;
val is = (7,"hello") : int * string
val rc = (3.14,#"z") : real * char

A pattern is one of three forms:

1. A variable pattern of the form var : typ
2. A tuple pattern of the form (pat₁, …, pat_n), where each pat_i is a pattern. This includes as a special case the null-tuple pattern, ()
3. A wildcard pattern of the form _

Record Types

Tuples can become more difficult to use as the number of elements increases. Record types allow labeling each component. A record type has the form:
{ lab₁:typ₁, …, lab_n:typ_n}
A record value has the form:
{ lab₁=val₁, …, lab_n=val_n}
A record pattern has the form:
{ lab₁=pat₁, …, lab_n=pat_n}

For example, the record type hyperlink is defined as follows:

type hyperlink =
  { protocol : string,
    address  : string,
    display  : string }

The following record binding defines a variable of type hyperlink:

val mailto : hyperlink =
  { protocol = "mailto",
    address = "foo@bar.com",
    display = "Brian Adkins" }

The following record binding:

val { protocol=prot, display=disp, address=addr } = mailto

decomposes into the three variable bindings:

val prot = "mailto"
val addr = "foo@bar.com"
val disp = "Brian Adkins"

We can use wildcard to extract selected fields:

val {protocol=prot, address=_, display=_ } = mailto

However, this isn’t very helpful with many fields, so we can use ellipsis patterns:

val {protocol=prot, ... } = mailto

ML provides an abbreviated form of record pattern {lab₁,…lab_n} which stands for { lab₁=var₁, …, lab_n=var_n} where the variables have the same name as the corresponding labels. For example, the following:

val { protocol, address, display } = mailto

decomposes into these bindings:

val protocol = "mailto"
val address = "foo@bar.com"
val display = "Brian Adkins"

Multiple Arguments and Multiple Results

fun dist (x:real, y:real):real = sqrt (x*x + y*y)

Keyword parameters are supported through record patterns:

fun dist’ {x=x:real, y=y:real} = sqrt (x*x + y*y)

Invoked as follows:

dist' {x=2.0,y=3.0}

Functions with multiple results may be thought of as functions yield tuples (or records).

fun dist2 (x:real, y:real):real*real
  = (sqrt (x*x+y*y), abs(x-y))

Sharp notation allows us to conveniently access the Nth item in a tuple.

- val foo = (3,7,5,2);
val foo = (3,7,5,2) : int * int * int * int
- #3 foo;
val it = 5 : int

A similar notation is used for record field selection:

 - val foo = { name = "Brian", phone = "555-1212" };
val foo = {name="Brian",phone="555-1212"} : {name:string, phone:string}
- #name foo;
val it = "Brian" : string

However, Harper states, “Use of the sharp notation is strongly discouraged!“

• Part Two
• Part Three

The posts in this series will be notes & highlights from working through “Programming in Standard ML” by Robert Harper. See Getting Started with Standard ML for a link to the PDF. I may not always use quotes in this series, but it should be assumed that any factual information about Standard ML that I write in this series has been obtained from the PDF above – this series is, in effect, a summarization of Robert Harper’s PDF.

Overview

Attributes of Standard ML:

• Type-safe programming language
• Statically typed
• Extensible type system
• Polymorphic type inference
• Automatic storage management
• Encourages functional (effect-free) programming
• Allows imperative (effect-ful) programming where appropriate
• Pattern matching
• Extensible exception mechanism for handling error conditions
• Expressive and flexible module system
• Portable due to its precise definition
• Portable standard basis library

Chapter 1 – Programming in Standard ML

This chapter provides a brief overview of the language by considering an implementation of a regular expression package. It was slightly more familiar to me because of my brief studies of Haskell; otherwise, I think it would’ve appeared very foreign. I prefer more of a bottom up approach as much as possible instead of looking at language features that haven’t been described.

Currying

I do think that currying & partial application are cool features. Consider the following function declaration:

val match : RegExp.regexp -> string -> bool

Coming from a non-currying background, I would describe match as a function that accepts a regular expression and a string, and returns a boolean value indicating whether the string matches the regular expression, but Robert describes it this way:

It contains a function match that, when given a regular expression, returns a function that determines whether or not a given string matches that expression.

This was one of the features that made a strong impression on me when I first started looking at functional programming languages.

Chapter 2 – Types, Values, and Effects

Computation in ML is of a somewhat different nature. The emphasis in ML is on computation by evaluation of expressions, rather than execution of commands.

An expressions has three characteristics:

• It may or may not have a type
• It may or may not have a value
• It may or may not create an effect

Effects will be ignored until chapter 13, so until then, it’s assumed that all expressions are effect-free, or pure. A type is defined by specifying three things:

• a name for the type,
• the values of the type, and
• the operations that may be performed on values of the type

Notes:

• Negative numbers, as literals, are indicated with a tilde instead of a hypen e.g. ~7
• div is used for integers, / for reals

A typing assertion has the form: exp : typ For example: 3 : int 4 div 3 : int ML does not perform any implicit conversions between types. Use real(3) + 4.3, or 3 + round(4.3) Conditional expression (exp1 and exp2 must have the same type): if exp then exp1 else exp2 if not exp then exp1 else exp2

Chapter 3 – Declarations

Once a variable is bound to a value, it is bound to it for life; there is no possibility of changing the binding of a variable once it has been bound. Examples of type bindings:

type float = real
type count = int and average = real

Examples of value bindings:

val m : int = 3+2
val pi : real = 3.14 and e : real = 2.17

One thing to keep in mind is that binding is not assignment. The binding of a variable never changes; once bound to a value, it is always bound to that value (within the scope of the binding). However, we may shadow a binding by introducing a second binding for a variable within the scope of the first binding.

Limiting scope with let expressions:

let
  val m : int = 3
  val n : int = m*m
in
  m*n
end

The following expression evaluates to 54 because the binding of m is temporarily overridden during the evaluation of the let expression, then restored upon completion of this evaluation:

val m : int = 2
val r : int =
    let
        val m : int = 3
        val n : int = m*m
    in
        m*n
    end * m

Standard ML Resources

Compilers

• Standard ML of New Jersey is highly recommended and comes with a REPL.
• Moscow ML
• MLKit
• PolyML
• MLton is a Standard ML compiler with excellent performance. However, as far as I know, it doesn’t have a REPL, which makes it less than ideal for learning.

Books

Since I was unfamiliar with the Standard ML programming language, I was surprised to find there are a number of good books about the language. Following are just some of them:

Elements of ML Programming

ML for the Working Programmer

Purely Functional Data Structures

The Definition of Standard ML

The Standard ML Basis Library

Introduction to Programming using SML

Concurrent Programming in ML

Other Educational Materials

Programming in Standard ML – excellent online book by Robert Harper of Carnegie Mellon University. Since I don’t know if Standard ML will simply be a stepping stone to Haskell (which in turn may not be a primary language for me) or a language I invest a lot of time in, I’m going to restrict myself from my normal method of purchasing a book or two when learning a new language. Instead, I’ll be going through Harper’s online book initially.

Tips for Computer Scientists on Standard ML (Revised)

Hello World

There are a number of great compilers for Standard ML (listed above), but I only need one to get started, so I chose Standard ML of New Jersey despite the funky name. It’s a popular version, and it has a REPL, so it’s good enough for me for now.

I develop software on Mac OSX and deploy on Ubuntu Linux. On my Ubuntu server, installing SML/NJ was as simple as:

sudo apt-get install smlnj

On Mac OSX, there are a couple of options listed on this page. I could use a pre-built system or the generic Unix install, so naturally I chose the generic Unix install which installed easily according to the simple directions.

# Download config.tgz
tar xzf config.tgz
config/install.sh

# Wait for install to complete

~/software/smlnj$ rlwrap bin/sml
Standard ML of New Jersey v110.69 [built: Sat May  2 12:04:08 2009]
- print "hello, world\n";
hello, world
val it = () : unit
-

Great, looks like everything is working fine. Note, I use the rlwrap utility to provide a nicer REPL experience, but it’s not required.

I’ll continue with a series of posts with notes from working through “Programming in Standard ML”.

The 2008 Programming Language Plan didn’t go as well as I hoped, so I’m regrouping for another go at it. I did make progress learning some Logo and teaching it to my daughters, and I worked through seven chapters of “Programming in Haskell” which was very enjoyable, but I also spent way too much time trying to decide which language(s) to learn without actually learning them.

I have now decided which languages I want to learn this year, so I figured I’d post this blog entry. In some respects, things haven’t changed much from last years plan, but my decisions are much less tentative which is encouraging. I’m glad to switch from information gathering to actually learning a few languages.

Motivation

When I switched from C++ to Java in 1996 and noticed a large increase in productivity; however, it didn’t occur to me to consider whether switching from Java to another language might also give me a similar boost in productivity. The job demand for Java was high during the 10 years I used it, and I was getting paid for my time vs. what I produced, so the oversight wasn’t too costly.

When I switched from Java to Ruby in 2006 I experienced a comparable, if not greater, boost in productivity as I had with the switch from C++ to Java. This time I considered the effect of programming language choice on productivity more carefully and began to wonder about the optimal programming language for me. As a small business owner, my primary consideration is productivity, not popularity or the interchangeableness of programmers.

I don’t think a perfect programming language exists in general, but I think there is an optimal language for me given my particular circumstances, the problems I want to solve, the type of software I want to develop, my way of thinking, my aesthetic tastes, etc.

The Problem

There is a cost to research programming languages to determine the optimal one, and there is a cost to switch programming languages, so the benefit of a new programming language needs to exceed those costs.

Additionally, a significant catch-22 exists in that to truly determine if a language is optimal, one must reach a level of proficiency with the language. There isn’t enough time to learn all of the candidate languages, so some filtering mechanism must be used first to narrow the choices to a reasonably small set. This filtering mechanism is theoretical by nature, but I think the subtle practicalities of a language have a significant effect on productivity.

I have a long term perspective for this task, so I don’t mind investing time researching programming languages, and I also don’t mind if I need to develop/acquire supporting libraries before being ultimately more productive than I am currently with Ruby and Rails.

The Process

I briefly entertained the idea of going about this research in a more systematic, scientific way. That might be a reasonable approach, but due to the following:

• My laziness
• The high degree of subjectivity in programming language choice
• The lack of consensus among programming language researchers

I ended up with a fairly ad-hoc approach involving examining the following areas and trying to absorb enough information to formulate a plan:

Efficiency

Programming language efficiency is becoming increasingly important to me as Moore’s law loses steam, and power & cooling issues become more prominent.

On the one hand, Ruby is near the bottom of the pack with respect to efficiency, but the productivity has been outstanding, and at the volume of transactions I’ve needed to deliver thus far, performance hasn’t been a problem.

On the other hand, if programming languages exist (and I believe they do) that have similar or greater power than Ruby but are compiled instead of interpreted, that would be an advantage.

Benchmarks are notoriously controversial with respect to the degree in which they predict performance, but I don’t think they’re irrelevant. One popular benchmark site is: Programming Language Shootout. I also created some micro-benchmarks of my own to test performance.

One negative result of Ruby being inefficient is the disparity in performance of code written in Ruby vs. library code, or extensions, written in C. This creates counter-intuitive scenarios where a piece of code that appears less efficient is actually more efficient because of the use of built-in code implemented in C. I don’t like thinking in those terms; I would much prefer that user written code in the language is much closer to the efficiency of built-in library code.

Concurrency

With the flattening of CPU MHz and proliferation of CPU cores, I think concurrency may continue to grow in importance. Although I develop mainly server side web applications which can take advantage of multiple cores by virtue of having multiple independent processes, I would prefer to have better concurrency mechanisms in the language/libraries directly.

Joy

I think I first heard joy used, in the context of programming languages, in the Ruby community – probably from Matz himself. I have to agree that programming in Ruby has been more joyful than previous programming languages.

I’m not sure exactly why this is, but I expect it has something to do with the effectiveness and productivity of Ruby, but it may be helped by the incredibly friendly Ruby community.

Fundamental Nature

My preference is for a programming language to minimize arbitrariness and to maximize orthogonality. In other words, I would prefer the language to have a minimal set of core concepts/axioms/operators/etc. that are built upon systematically.

Community

Whether my optimal language is mainstream or not is irrelevant to me, so I don’t need the programming language community to be large (and in fact, being too large is a detriment), but I do think it should be active enough that there are people available to help answer questions, write libraries, maintain/improve compilers, debuggers etc.

To this end, I spent time on various usenet groups, IRC channels and blogs. Occasionally asking questions, but mostly observing the dialog and interaction among the community members.

Education

What educational materials are available?

I tend to prefer learning from books, so searching Amazon to see what books are available, how well they’re received, etc. was helpful. Although I prefer books typically, in the exploratory stages, it’s cheaper to go through online materials, so searching the web for free PDFs was useful. The existence of a few good texts is also an indication of the vitality of community.

Productivity

Ultimately, I’m after the most productive language for me; however, I think productivity is the factor that requires the most proficiency with a language to judge, so it was the most difficult factor to research prior to learning a language. I was limited to anecdotal stories, case studies, personal testimonies, etc.

Licensing

I prefer open source licensing because I think programming languages with proprietary licensing are more likely to die.

The Final Candidates

I began with Ruby as the point of reference irrespective of Rails and other libraries. Even though Rails is an important factor in my current productivity, since I’m taking a long term view, I didn’t want to exclude a fantastic language simply because it’s lacking something that Ruby was also lacking N years ago.

Due to the cost of researching, learning and switching to a new programming language, I only considered languages that had the potential of offering a significant improvement in one or more areas without losing too much in other areas.

The candidates I eventually selected fell into two groups. Both groups have strong functional capabilities, good efficiency, and long, successful track records. I believe that even the best programming language designers make serious mistakes which can only be identified with the hindsight of years of use. It’s possible that someone is about to release a brand new language that has the potential to be the most productive for me personally; however, I’m not willing to take the risk to learn it.

The Lisp Family

Lisp has a very long track record of success and adaptability. In fact, if I was forced to program in a single programming language from now on, I would probably choose a Lisp since I think it would have the greatest likelihood of being able to adapt.

The Lisp family has dynamic typing which I’ve grown to love with Ruby, but most also allow type declarations for efficiency. Lisp also appears to have a more fundamental nature than most languages. I’ve heard it said that Lisp was more discovered than invented.

I first became interested in Lisp when I learned that Ruby was influenced by it. My interest was reinforced by some of Paul Graham’s essays & books. I like the exploratory and dynamic nature of developing Lisp – this is mostly from reading comments of others, but I’ve tasted a small part of this when evaluating Lisp code in Emacs and having it be available immediately.

Logo is first on the list simply because it’s both a Lisp and a great language for teaching children how to program, so I can kill two birds with one stone.

Scheme is next because it’s a natural sequel to Logo and I still want to work through “The Structure and Interpretation of Computer Programs”. The fact that Paul Graham chose mzcheme to create his Arc language is a good “letter of recommendation”. Scheme feels the most fundamental, simple & clean thus far.

Common Lisp is after Scheme due to its eminent practicality and completeness. It has plenty of warts, but also plenty of power and functionality. It’s been called a great, big, ball of mud – but in a good way ;)

Lastly, is Clojure. It’s last in the Lisp family because I want to learn Scheme & Common Lisp first so I’ll be better equipped to judge Clojure. This will also allow more time for Clojure to mature. Clojure has a focus on functional programming & concurrency that is symbiotic with the Java platform. The latter provides a quick start and access to Java libraries, the JVM infrastructure, etc., but my preference would be to not be dependent on the JVM in the long run. I’ll withhold judgment until I’ve learned it and used it for a while.

The ML Family

I’m least familiar with the ML family and with functional programming in general. I’ve spent most of my career studying and using imperative, object-oriented programming languages.

I think the ML family is worth studying because:

• Functional programming may be beneficial with respect to concurrency.
• Prog. Lang. researchers seem to be enamored with ML family.
• There are enough anecdotal testimonies of productivity to warrant further study.

Standard ML is an important functional language that differs from Haskell in that it’s impure, i.e. it allows side effects, and it’s strict, i.e. not lazy. It shares Hindly-Milner static typing with Haskell. The community seems rather tiny, and I expect that if I go with a static typed language it will likely be Haskell, but I wanted to learn Standard ML first because of its historical importance and to be able to compare an impure/strict language with a pure/nonstrict language.

Haskell is probably the most different programming language from what I’m used to. This, in and of itself, has some advantages with respect to gaining new perspectives and ideas. In some ways, it’s a language that has taken things to extremes with respect to functional purity and laziness.

The community is very active. It offers a great compiler (GHC), software transactional memory, a decent base of libraries, etc.

At this early stage, I’m skeptical of static typing, functional purity and laziness, so becoming proficient in Haskell is a great opportunity to be able to determine how I feel about those.

The Plan

Rather than go through the candidates sequentially, I’m going to try and make progress on two tracks concurrently to allow me to compare concepts from both families:

I’m curious to find out how I feel about these languages after I’ve achieved some skill with them, but I think becoming proficient in the Lisp and ML families will be time well spent. At minimum, I’ll be better equipped to compare other languages.

"implemented in <language>"
"written in <language>"

I’m very curious to see how these stats change over time, so I’ve added a calendar item to recompute them in six months. Leave a comment if you’d like to add a programming language to the list, and I’ll update this article and it will be included in the recomputation six months from now.

¹ combines Lisp, Scheme, Common Lisp, Arc & Clojure
² combines OCaml, (S)ML, Caml
³ summed separate searches for sml and ml
Update 4/23/09 added C#, Tcl per comment requests.