Lisp vs. Java... D?

by Lionello Lunesu, Oct. 18th 2006

Two weeks ago, Josh Stern posted an interesting link on the digitalmars.D newsgroup. The link pointed to a page describing a programming challenge that was originally used in a study in which 38 C, C++ and Java programmers were asked to write versions of a program to compare Java and C++ efficiency.

Many other programmers had taken up the challenge since. In one such comparison, Lisp turned out to be a clear winner over Java and C++ when comparing development time and lines of code, with the record standing at about 2 hours, with 45 non-comment non-blank lines. For C++, the shortest development time was 3 hours and the shortest program had 107 lines of code.

With some time to spare, I thought I should give it a try myself. Although I'm a C++ programmer myself (and have been for 10 years now), I thought it would be a great opportunity to test my skills in the D programming language.

I had not looked at the other implementations and only read the original problem statement. It took me 1 hour and 15 minutes, from reading the statement to finishing the program. The program was just 55 lines long (remember, the shortest C++ entry had 107 lines), not counting blank lines, comments, assertions, unittests, contracts and trailing curly brackets (similar to Lisp).

The program basically had to do this: read all words from a dictionary file; for each letter there was a corresponding digit (like the letters on a phone); using this mapping, read phone numbers from another text file and print all possible combinations of words for that phone number. Check the link above for more details.

What made implementing this task so easy in D? I just happened to have every construct I needed at hand. Some of the goodies used, are:

Assertions, contracts and unittests
Built-in arrays and strings, with easy slicing
Built-in associative array
Garbage collection
Type deduction and foreach
Nested functions
Returning (reference to) a array from a function

During the actual programming I did not have to stop and think about anything. The program evolved automatically, once I thought of the actual container for the dictionary. It turned out that the 'winning' Lisp version was storing the dictionary in a similar manner.

Compiling

To test the program yourself, first download the dictionary.txt and input.txt. Then, download the latest D compiler. To compile, copy-paste the D code into a newly created file phoneno.d, and compile it with:

dmd -run phoneno.d

// Created by Lionello Lunesu and placed in the public domain.
// This file has been modified from its original version.
// It has been formatted to fit your screen.
module phoneno;     // optional
import std.stdio;   // writefln     
import std.ctype;   // isdigit     
import std.stream;  // BufferedFile

// Just for readability (imagine char[][][char[]])    
alias char[] string;
alias string[] stringarray;

/// Strips non-digit characters from the string (COW)
string stripNonDigit( in string line ) 
{
    string ret;
    foreach(uint i, c; line) {
        // Error: std.ctype.isdigit at C:\dmd\src\phobos\std\ctype.d(37) 
        // conflicts with std.stream.isdigit at C:\dmd\src\phobos\std\stream.d(2924)
        if (!std.ctype.isdigit(c)) {
            if (!ret)
                ret = line[0..i];    
        }    
        else if (ret)
            ret ~= c;    
    }    
    return ret?ret:line;
}

unittest {
    assert( stripNonDigit("asdf") == ""  );
    assert( stripNonDigit("\'13-=2 4kop") ==  "1324"  );
}

/// Converts a word into a number, ignoring all non alpha characters  
string wordToNum( in string word )
{
// translation table for the task at hand
const char[256] TRANSLATE =    
    "                                "  // 0   
    "                0123456789      "  // 32     
    " 57630499617851881234762239     "  // 64   
    " 57630499617851881234762239     "
    "                                "
    "                                "
    "                                "    
    "                                ";
    string ret;
    foreach(c; cast(ubyte[])word)
        if (TRANSLATE[c] != ' ')
            ret ~= TRANSLATE[c];
    return ret;
}

unittest {
 // Test wordToNum using the table from the task description.
 assert( "01112223334455666777888999" ==
   wordToNum("E | J N Q | R W X | D S Y | F T | A M | C I V | B K U | L O P | G H Z"));
 assert( "01112223334455666777888999" == 
   wordToNum("e | j n q | r w x | d s y | f t | a m | c i v | b k u | l o p | g h z"));
 assert( "0123456789" == 
   wordToNum("0 |   1   |   2   |   3   |  4  |  5  |   6   |   7   |   8   |   9"));
}

void main( string[] args )
{
    // This associative array maps a number to an array of words.    
    stringarray[string]    num2words;

    foreach(string word; new BufferedFile("dictionary.txt" ) )
        num2words[ wordToNum(word) ] ~= word.dup;        // must dup

    /// Finds all alternatives for the given number
    /// (should have been stripped from non-digit characters)
    stringarray _FindWords( string numbers, bool digitok )
    in {
        assert(numbers.length >  0);    
    }    
    out(result) {
        foreach (a; result)
            assert( wordToNum(a) == numbers );
    }    
    body {
        stringarray ret;
        bool foundword = false;
        for (uint t=1; t<=numbers.length; ++t) {
            auto alternatives = numbers[0..t] in num2words;
            if (!alternatives)
                continue;
            foundword = true;
            if (numbers.length >  t) {
                // Combine all current alternatives with all alternatives     
                // of the rest (next piece can start with a digit)              
                foreach (a2; _FindWords( numbers[t..$], true     ) )
                    foreach(a1; *alternatives)
                       ret ~= a1 ~ " " ~ a2;
            }
            else    
                ret ~= *alternatives;    // append these alternatives
        }
        // Try to keep 1 digit, only if we're allowed and no other
        // alternatives were found
        // Testing "ret.length" makes more sense than testing "foundword",
        // but the other implementations seem to do just this.
        if (digitok && !foundword) { //ret.length == 0  
            if(numbers.length >  1) {
                // Combine 1 digit with all altenatives from the rest    
                // (next piece can not start with a digit)          
                foreach (a; _FindWords( numbers[1..$], false ) )
                    ret ~= numbers[0..1] ~ " " ~ a;
            }    
            else    
                ret ~= numbers[0..1];    // just append this digit             
        }    
        return ret;
    }

    /// (This function was inlined in the original program) 
    /// Finds all alternatives for the given phone number 
    /// Returns: array of strings 
    stringarray FindWords( string phone_number )
    {
        if (!phone_number.length)
            return null;
        // Strip the non-digit characters from the phone number, and
        // pass it to the recursive function (leading digit is allowed)
        return _FindWords( stripNonDigit(phone_number), true );    
    }    
    
    // Read the phone numbers     
    foreach(string phone; new BufferedFile("input.txt"   ) )
        foreach(alternative; FindWords( phone ) )
            writefln(phone, ": ", alternative );
}

Books

Lisp vs. Java... D?

Compiling