"Linux Gazette...making Linux just a little more fun!"

A Glimpse of Icon

By Clinton Jeffery and Shamim Mohamed

Motivation

Many languages introduce special capabilities for specific kinds of applications, but few present us with more powerful control structures or programming paradigms. You may be comfortable sticking with a language you already know, but if you are challenged to write complex programs and are short on time, you need the best language for the job. Icon is a high-level programming language that looks like many other programming languages but offers many advanced features that add up to big gains in productivity. Before we get to all that, let us write the canonical first program:

procedure main()
   write("Hello, world!")
end

icont hello -x

We are introducing many concepts, so don't expect to understand everything the first time through -- the only way to learn a language is to write programs in it; so get Icon, and take it for a test drive.

Genealogy

Icon was developed over several years at the University of Arizona by a team led by Ralph Griswold. Today, it runs on many platforms and is used by researchers in algorithms, compilers, and linguistics as well as system administrators and hobbyists. The implementation and source code are in the public domain.

Variables, Expressions, and Type

Icon's expression syntax starts out much as do most languages. For example, i+j represents the arithmetic addition of the values stored in the variables i and j, f(x) is a call to f with argument x, variables may be global or local to a procedure, and so on.

Variable declarations are not required, and variables can hold any type of value. However, Icon is a strongly typed language; it knows the type of each value and it does not allow you to mix invalid types in expressions. The basic scalar types are integers, real numbers, strings, and sets of characters (csets). Integers and strings can be arbitrarily large; and strings can contain any characters. There are also structured types: lists, associative arrays, sets and records. Icon performs automatic storage management.

Goal-directed Expression Evaluation

Success and Failure

Control structures such as if check for success, so

   if i > 3 then ...

Since functions that fail do not need to return an error code separately from the results, detecting cases such as end-of-file is simpler, as in:

   if line := read() then write(process(line))

   write(process(read())

Generators

   find("or", "horror")

   find("or", "horror") > 3 

The most obvious generator is the alternation operator |. The expression

   expr

| expr

   x = (3 | 5)

   (x | y) = (3 | 5)

   (x == 3) || (x == 5) || (y == 3) || (y == 5)

In addition to | , Icon has a generate operator ! that generates elements of data structures, and a generator to that produces ranges of integers. For example, !L generates the elements of list L, and 1 to 10 generates the first ten positive integers. Besides these operators that generate results, most generators in Icon take the form of calls to built-in and user-defined procedures. Procedures are discussed below.

Iteration

   every i := 1 to 10 do ...

   every f(1 | 1 | 2 | 3 | 5 | 8)

Procedures

Procedures are a basic building block in most languages, including Icon. Like C, an Icon program is organized as a collection of procedures and execution starts from a procedure named main(). Here is an example of an ordinary procedure. This one generates and sums the elements of a list L, whose elements had better be numbers (or convertible to numbers).

procedure sum(L)
   total := 0
   every total +:= !L
   return total
end

A user can write her own generator by including a

   suspend expr

procedure nonzero(L)
   every i := !L do
      if i ~= 0 then suspend i
end

String Processing

Besides expression evaluation, Icon offers compelling features to reduce the effort required to write complex programs. From Icon's ancestor SNOBOL4, the granddaddy of all string processing languages, Icon inherits some of the most flexible and readable built-in data structures found in any language.

Strings

   s[7] := " linux "
   s[14:19] := "journal"

String Scanning

   s ? expr

   text ? {
      while move(1) do
         write(move(1))
   }

procedure getword()
   while line := read() do 
      line ? while tab(upto(wchar)) do {
         word := tab(many(wchar))
         suspend word
      }
end

Regular Expressions

The Icon Program Library (included with the distribution) provides regular expression matching functions. To use it, include the line link regexp at the top of the program. Here is an example of `search-and-replace':

procedure re_sub(str, re, repl)
   result := ""
   str ? {
      while j := ReFind(re) do {
         result ||:= tab(j) || repl
         tab(ReMatch(re))
      }
      result ||:= tab(0)
   }
   return result
end

Structures

Icon has several structured (or non-scalar) types as well that help organize and store collections of arbitrary (and possibly mixed) types of values. A table is an associative array, where values are stored indexed by keys which may be of arbitrary type; a list is a group of values accessed by integer indices as well as stack and queue operations; a set is an unordered group of values, etc.

Tables

   wordcount := table(0)
   every word := getword() do
      wordcount[word] +:= 1
   every word := key(wordcount) do
      write(word, "\t", wordcount[word])

Lists

   L := ["linux", 2.0, "unix"]

   write(L[3])

Records and Sets

record complex(re, im)

   i := complex(0, 0)
   j := complex(1, -1)
   if a.re = b.re then ...

A set is an unordered collection of values with the uniqueness property i.e. an element can only be present in a set once.

   S := set(["rock lobster", 'B', 52])

Graphs

Since there is no restriction on the types of values in a list, they can be other lists too. Here's an example of a how a graph or tree may be implemented with lists:

record node(label, links)
   ...
   barney := node("Barney", list())
   betty := node("Betty", list())
   bambam := node("Bam-Bam", list())
   put(bambam.links, barney, betty)

An Example

global wchar
procedure main(args)
   wchar := &ucase ++ &lcase
   (*args = 1) | stop("Need a file!")
   f := open(args[1]) | stop("Couldn't open ", args[1])
   wordlist := table()
   lineno := 0
   while line := read(f) do {
      lineno +:= 1
      every word := getword(line) do 
         if *word > 3 then {
            # if word isn't in the table, set entry to empty list
            /wordlist[word] := list()
            put(wordlist[word], lineno)
         }
   }
   L := sort(wordlist)
   every l := !L do {
      writes(l[1], "\t")
      linelist := ""
      # Collect line numbers into a string
      every linelist ||:= (!l[2] || ", ")
      write(linelist[1:-2])
   }
end

procedure getword(s)
   s ? while tab(upto(wchar)) do {
      word := tab(many(wchar))
      suspend word
   }
end

Sing, Mother, sing.
Can Mother sing?
Mother can sing.
Sing, Mother, sing!

Mother  1, 2, 3, 4
Sing    1, 4
sing    1, 2, 3, 4

Co-expressions

Another novel control facility in Icon is the co-expression, which is an expression encapsulated in a thread-like execution context where its results can be picked apart one at a time. Co-expressions are are more portable and more fine-grained than comparable facilities found in most languages. Co-expressions let you `capture' generators and then use their results from multiple places in your code. Co-expressions are created by

   create expr

As a small example, suppose you have a procedure prime() that generates an infinite sequence of prime numbers, and want to number each prime as you print them out, one per line. Icon's seq() function will generate the numbers to precede the primes, but there is no way to generate elements from the two generators in tandem; no way except using co-expressions, as in the following:

   numbers := create seq()
   primes := create prime()
   every write(@numbers, ": ", @primes)

Graphics

Icon features high-level graphics facilities that are portable across platforms. The most robust implementations are X Window and Microsoft Windows; Presentation Manager, Macintosh, and Amiga ports are in various stages of progress. The most important characteristics of the graphics facilities are:

• simplicity, ease of learning
• windows are integrated with Icon's existing I/O functions
• straightforward input event model

procedure main()
   &window := open("LJ example","g")
   repeat case e := Event() of {
      &lpress | &ldrag : DrawPoint(&x,&y)
      &rpress : GotoXY(&x,&y)
      "\e"    : break
      default : if type(e)=="string" then writes(&window, e)
      }
end

POSIX Functions

An Icon program that uses the POSIX functions should include the header file posixdef.icn. On error, the POSIX functions fail and set the keyword &errno; the corresponding printable error string is obtained by calling sys_errstr().

Unix functions that return a C struct (or a list, in Perl) return records in Icon. The fields in the return values have names similar to the Unix counterparts: stat() returns a record with fields ino, nlink, mode etc.

A complete description of the POSIX interfaces is included in the distribution; an HTML version is available on the web, at http://www.drones.com/unicon/. We look at a few small examples here.

An Implementation of ls

Let us look at how a simple version of the Unix ls command may be written. What we need to do is to read the directory, and perform a stat call on each name we find. In Icon, opening a directory is exactly the same as opening a file for reading; every read returns one filename.

     f := open(dir) | stop(name, ":", sys_errstr(&errno))
     names := list()
     while name := read(f) do
          push(names, name)
     every name := !sort(names) do
          write(format(lstat(name), name, dir))

The function to format the information is simple; it also checks to see if the name is a symbolic link, in which case it prints the value of the link also.

link printf
procedure format(p, name, dir)
   s := sprintf("%7s %4s %s %3s %8s %8s %8s %s %s",
	   p.ino, p.blocks, p.mode, p.nlink,
	   p.uid, p.gid, p.size, ctime(p.mtime)[5:17], name)

   if p.mode[1] == "l" then
      s ||:= " -> " || readlink(dir || "/" || name)

   return s
end

Polymorphism and other pleasant things

It's not just stat that uses human-readable values -- chmod can accept an integer that represents a bit pattern to set the file mode to, but it also takes a string just like the shell command:

     chmod(f, "a+r")

     owner := getpw("ickenham")
     root := getpw(0)
     while u := getpw() do ...

Using select

The select() function waits for input to become available on a set of files. Here is an example of the usage -- this program waits for typed input or for a window event, with a timeout of 1000 milliseconds:

   repeat {
      while *(L := select([&input, &window], 1000)) = 0 do
         ... handle timeout
      if &errno ~= 0 then
         stop("Select failed: ", sys_errstr(&errno))

      every f := !L do 
         case f of {
            &input  : handle_input()
            &window : handle_evt()
      }
   }

Networking

Icon provides a much simpler interface to BSD-style sockets. Instead of the four different system calls that are required to start a TCP/IP server using Perl, only one is needed in Icon--the open function opens network connections as well as files. The first argument to open is the network address to connect to -- host:port for Internet domain connections, and a filename for Unix domain sockets. The second argument specifies the type of connection.

Here is an Internet domain TCP server listening on port 1888:

procedure main()
     while f := open(":1888", "na") do
          if fork() = 0 then {
               service_request(f)
               exit()
          } else
               close(f)
     stop("Open failed: ", sys_errstr(&errno))
end

procedure finger(name, host)
     static fserv
     initial fserv := getserv("finger") |
          stop("Couldn't get service: ", sys_errstr(&errno))

     f := open(host || ":" || fserv.port, "n") | fail
     write(f, name) | fail
     while line := read(f) do
          write(line)
end

UDP

   s := getserv("daytime", "udp")
   
   f := open(host || ":" || s.port, "nu") |
      stop("Open failed: ", sys_errstr(&errno))

   writes(f, " ")

   if *select([f], 5000) = 0 then
      stop("Connection timed out.")

   r := receive(f)
   write("Time on ", host, " is ", r.msg)

Icon and other languages

The popular languages Perl and Java have been covered in LJ, and we think it is worth discussing how Icon stacks up against these dreadnaughts.

Perl and Icon

But when it comes to language design, Perl and Icon are not at all alike. Perl has been designed with very little structure -- or, as Larry Wall puts it, it's more like a natural language than a programming language. Perl looks strange but underneath the loose syntax its semantics are those of a conventional imperative language. Icon, on the other hand, looks more like a conventional imperative language but has richer semantics.

Advantages of Perl

Some misfeatures of Perl

Namespace confusion: it is a bad idea to allow scalar variables, vector variables and functions to have the same name. This seems like a useful thing to do, but it leads to write-only code. We think this is primarily why it's hard to maintain Perl programs. A couple of things are beyond belief -- $foo and %foo are different things, but the expression $foo{bar} actually refers to an element of %foo!

Parameter passing is a mess. Passing arrays by name is just too confusing! Even after careful study and substantial practice, we still are not absolutely certain about how to use *foo in Perl. As if to make up for the difficulty of passing arrays by reference, all scalars are passed by reference! That's very unaesthetic.

Why are there no formal parameters? Instead, one has to resort to something that looks like a function call to declare local variables and assign @_ to it. Allowing the parentheses to be left off subroutine calls is also unfortunate; it is another `easy to write, hard to read' construct. And the distinction between built-in functions and user-defined subroutines is ugly.

Variables like $` are a bad idea. We think of special characters as punctuation, we don't expect them to be (borrowing Wall's terminology) nouns. And the mnemonics that are required are evidence that these variables place an additional burden of memorization upon the programmer. (Quick, if you write Perl programs: What's `$('?)

The distinction between array and scalar contexts also leads to obfuscated code. Certainly after you've been writing Perl for a while, you get used to it (and might even like it), but again, this is just confusing. All the warnings in the Perl documentation about being certain you are evaluating in the right context is evidence of this.

Java and Icon

The important differences between Java and Icon are differences of philosophy. The Java philosophy is that everything is an object, nothing is built-in to the language, and programmers should learn class libraries for all non-trivial structures and algorithms. Java's lack of operator overloading means that its object-oriented notation allows no "shorthand" as does C++. Java's simplicity is a welcome relief after C++, but its expressive power is so weak compared to Icon (and several other very high level languages) that it is hard to argue that Java can supplant these languages. Most of Java's market share is being carved out of the C and C++ industries.

How to Improve Java

Getting Icon

Users who become serious about the language will want a copy of `The Icon Programming Language', by Ralph and Madge Griswold, Peer-to-Peer Communications 1997, ISBN 1-57398-001-3.

Lots of documentation for Icon is available from the University of Arizona, at http://www.cs.arizona.edu/icon/ There is also a newsgroup on Usenet: comp.lang.icon.

The Icon source distribution is at:
ftp://ftp.cs.arizona.edu/icon/packages/unix/unix.tgz
The POSIX functions are in the following patch that you need to apply if you wish to build from sources:
ftp://ftp.drones.com/unicon-patches.tar.gz

Linux binaries (kernel 2.0 ELF, libgdbm 2.0.0, libX11 6.0, libdl 1.7.14, libm 5.0.0 and libc 5.2.18) for Icon (with X11 and POSIX support) are available at

ftp://ftp.drones.com/icon-9.3-3.i386.rpm	Icon ftp://ftp.drones.com/icon-ipl-9.3-3.i386.rpm	Icon Program Library
ftp://ftp.drones.com/icon-idol-9.3-3.i386.rpm	Idol: Object-oriented Icon
ftp://ftp.drones.com/icon-vib-9.3-3.i386.rpm	VIB: The Visual Interface Builder
ftp://ftp.drones.com/icon-docs-9.3-3.i386.rpm	Documentation