TweetFollow Us on Twitter

May 95 Challenge
Volume Number:11
Issue Number:5
Column Tag:Programmer’s Challenge

Programmer’s Challenge

By Mike Scanlin, Mountain View, CA

Note: Source code files accompanying article are located on MacTech CD-ROM or source code disks.

Hy-phen-a-tion

Hyphenation algorithms come in two flavors: rule-based and dictionary-based. Of the two, dictionary-based is more reliable but has the downside of requiring a lot of storage space. Rule-based methods require considerably less space and have the option of using an “exceptions” dictionary to improve accuracy. This month’s Challenge is to implement a rule-based hyphenation algorithm.

The algorithm is this: Each part of the word on either side of the hyphen must include a vowel, not counting a final e, es or ed. The part of the word after the hyphen cannot begin with a vowel or a double consonant. No break is made between any two of the following letter combinations: sh, gh, ph, ch, th, wh, gr, pr, cr, tr, wr, br, fr, dr, vowel-r, vowel-n, or om. That means that if any two of those pairs occur next to each other, you can’t break them apart (i.e. ‘from’ contains ‘fr’ and ‘om’, which are both on the list; therefore you would not split it between the ‘r’ and the ‘o’). The letter ‘y’ is not a vowel.

The two routines you’ll write are:

void *
InitHyphenation(maxRAM)
ulong   maxRAM;

void
Hyphenate(privateDataPtr, inPtr, outPtr)
void  *privateDataPtr;
Str255  *inPtr;
Str255  *outPtr;

The InitHyphenation routine is an untimed routine called once that can set up whatever tables you might want to use. It must not allocate more than maxRAM bytes, which will be between 16K and 64K.

Hyphenate is the routine that does the work. PrivateDataPtr is the return value from the InitHyphenation routine. InPtr is a pointer to a read-only unhyphenated Pascal word (containing only letters ‘a’..‘z’ and ‘A’..‘Z’). OutPtr is a pointer to 256 bytes of space where you store the hyphenated word. You need to preserve the case of the input and you should insert a kHyphen byte (0x2D) everywhere the above algorithm tells you to (yes, the complete output is guaranteed to fit within a Str255).

Note that this algorithm is valid for English only. It will find about 70% of all valid hyphens and will make mistakes about 45% of the time. It is assumed that if this was part of a real application that a hyphenation exception word list would be kept. That list would be checked before calling this function. You, however, don’t need to be concerned with that.

Write to me if you have any questions.

Two Months Ago Winner

Congrats to Gustav Larsson (Mountain View, CA) for winning the Method Dispatcher Challenge. Gustav was a virtual unknown to this column a few months ago but he is rising fast in the Top 20 rankings. The 20 points he wins this month move him from 10th place to 4th place.

Here are the times and code sizes for each entry. Numbers in parens after a person’s name indicate that person’s cumulative point total for all previous Programmer Challenges, not including this one:

Name time code

Gustav Larsson (40) 261 1040

Kevin Cutts (36) 304 266

Jeff Mallett (27) 338 1192

Xan Gregg 349 158

Ernst Munter (51) 385 1466

Thomas Studer 437 572

David Howarth 632 130

Scanlin’s brute force method 420 122

Everyone who entered implemented some kind of cache. Gustav chose a 10-way set-associative cache. He splits the 16K of usable cache space into 256 cache sets, each of which holds 10 entries. He uses a hash based on the class and method numbers to map to one of the 256 caches. Once he’s got that he checks the 10 entries for a match. If he doesn’t find a match he uses an efficient binary search to look for the method within the class.

In designing any cache, the choice of hash function and amount of set-associativity (if any) is highly dependent on your data and access pattern. Gustav’s code could be tuned for a variety of efficient cache uses. Nice job, Gustav.

Poker Winner Disqualified

Turns out that I failed to do adequate testing on the winning entry for the Poker Hand Evaluator Challenge, in which Kevin Cutts was the published winner. Unfortunately, I’m going to have to retro-actively disqualify Kevin and award the 1st place prize to Gustav Larsson. My apologies to Kevin and the other MacTech readers. Rather than publish Gustav’s well-commented winning solution (it’s quite long) I’ll make it available via e-mail. If you’re interested, send me a note at scanlin@genmagic.com or at any of the Programmer Challenge electronic addresses (see p. 2).

This latent win for Gustav means that he’s 3 for 3 in the last 3 challenges, which is an unmatched winning streak. Congrats, Gustav! Also, Dave Darrah was the first person to point out the flaws in Kevin’s code and so he receives 5 extra points in the cumulative point totals for doing so. Thanks, Dave!

Top 20 Contestants of All Time

Here are the Top 20 Contestants for the 33 Programmer’s Challenges to date. The numbers below include points awarded for this months’ top 5 entrants. (Note: ties are listed alphabetically by last name -- there are 23 people listed this month because 7 people have 20 points each.)

1. Boonstra, Bob 176

2. Karsh, Bill 71

3. Stenger, Allen 65

4. Larsson, Gustav 60

5. Munter, Ernst 53

6. Riha, Stepan 51

7. Goebel, James 49

8. Cutts, Kevin 46

9. Nepsund, Ronald 40

10. Vineyard, Jeremy 40

11. Darrah, Dave 34

12. Mallet, Jeff 34

13. Landry, Larry 29

14. Elwertowski, Tom 24

15. Kasparian, Raffi 24

16. Lee, Johnny 22

17. Anderson, Troy 20

18. Burgoyne, Nick 20

19. Galway, Will 20

20. Israelson, Steve 20

21. Landweber, Greg 20

22. Noll, Bob 20

23. Pinkerton, Tom 20

There are three ways to earn points: (1) by scoring in the top 5 of any challenge, (2) by being the first person to find a bug in a published winning solution or, (3) being the first person to suggest a challenge that I use. The points you can win are:

1st place 20 points

2nd place 10 points

3rd place 7 points

4th place 4 points

5th place 2 points

finding bug 5 points

suggesting challenge 2 points

Here is Gustav’s winning solution:

Dispatch.c

Copyright ©1995 Gustav K. Larsson

#define kMethodNotFound ((void *) 0)
#define kClassNotFound  ((void *) -1)

typedef unsigned char uchar;
typedef unsigned short ushort;
typedef unsigned long ulong;

typedef ushort  ClassID;
typedef ushort  MethodNumber;
typedef void *  MethodAddress;

typedef struct {
  MethodNumber  methodNumber;
  MethodAddress methodAddress;
} MethodEntry;

typedef struct {
  ushort        inheritedCount;
  ushort        inheritedClasses[15];
  MethodNumber  largestMethodNumber;
  ushort        methodCount;
  MethodEntry   methods[];
} Class, *ClassPtr;

/* Each block in the cache is 64 bytes and holds 10 entries. The id field distinguishes class/method pairs 
that map to the same cache block. The hits field has a bit for each entry, which is set whenever there is a 
hit on the entry.  The "next" field is 0..9 and indicates where to check first when deciding which entry to replace. 
*/

#define BLOCK_SIZE 10

typedef struct {
  ushort          id [ BLOCK_SIZE ];
  ushort          hits;
  ushort          next;
  MethodAddress   methodAddress [ BLOCK_SIZE ];
} CacheBlock;

static CacheBlock Cache[256]; /* exactly 16K */
GetClassPtr
extern ClassPtr GetClassPtr( ClassID );
MethodAddress FindMethod( ClassID, MethodNumber );
static MethodAddress HandleMiss( ClassID, MethodNumber,
                                 CacheBlock *, ushort );
FindMethod
MethodAddress
FindMethod( ClassID classID, MethodNumber methodNumber )
{
  register ulong hash;
  register ushort *idPtr;
  register CacheBlock *block;
Check cache for this class & method.

/* First, generate a hash key that is unique for each class/method pair. The formula classID + 437 * method 
works since 437 is greater than the highest class ID, 400. The ideal multiplier depends heavily on the distribution 
of class/method pairs, but 437 seems to give a reasonable spread under a variety of conditions.
    The low byte of the hash key becomes the block number, and the higher bytes become the "id" (to distinguish 
class/method pairs that map to the same block). Add one to the id so it is never zero (zero indicates an unused 
cache entry). */

  hash = classID + 437L * methodNumber;
  block = &Cache[ (uchar)hash ];
  hash = (hash >> 8) + 1;   /* now hash holds just the id */

  idPtr = block->id;
  if ( *idPtr++ == (ushort)hash ) goto hit0;
  if ( *idPtr++ == (ushort)hash ) goto hit1;
  if ( *idPtr++ == (ushort)hash ) goto hit2;
  if ( *idPtr++ == (ushort)hash ) goto hit3;
  if ( *idPtr++ == (ushort)hash ) goto hit4;
  if ( *idPtr++ == (ushort)hash ) goto hit5;
  if ( *idPtr++ == (ushort)hash ) goto hit6;
  if ( *idPtr++ == (ushort)hash ) goto hit7;
  if ( *idPtr++ == (ushort)hash ) goto hit8;
  if ( *idPtr++ == (ushort)hash ) goto hit9;

  return HandleMiss( classID, methodNumber,
                     block, (ushort)hash );

/* Handle cache hit */
hit0: block->hits |= 0x001;  return block->methodAddress[0];
hit1: block->hits |= 0x002;  return block->methodAddress[1];
hit2: block->hits |= 0x004;  return block->methodAddress[2];
hit3: block->hits |= 0x008;  return block->methodAddress[3];
hit4: block->hits |= 0x010;  return block->methodAddress[4];
hit5: block->hits |= 0x020;  return block->methodAddress[5];
hit6: block->hits |= 0x040;  return block->methodAddress[6];
hit7: block->hits |= 0x080;  return block->methodAddress[7];
hit8: block->hits |= 0x100;  return block->methodAddress[8];
hit9: block->hits |= 0x200;  return block->methodAddress[9];
}
HandleMiss
static MethodAddress
HandleMiss( ClassID classID, MethodNumber methodNumber,
            CacheBlock *blockPtr, ushort id )
{
  register ClassPtr classPtr;
  MethodAddress methodAddress;
  register ushort *temp;  /* shared address register */

  /* Not in cache, so look it up the hard way */
  classPtr = GetClassPtr( classID );
  if ( classPtr != kClassNotFound ) {

    /* Look in this class. Use a binary search. */
    {
      register MethodEntry *methods = classPtr->methods;
      ushort searchSize = classPtr->methodCount;
      register MethodNumber methodReg = methodNumber;
                                /* method # in a register */

/* Unroll the binary search for the most common cases. The BINARY macro reduces the search to progressively 
more elementary cases. If classPtr->methodCount is greater than 32, we run a general binary search until 
the search is reduced to an unrolled case.  */

continueBinarySearch:
      switch ( searchSize )
      {
        bin2: case 2:
          if ( methods[1].methodNumber == methodReg ) {
            methodAddress = methods[1].methodAddress;
            goto addCache;
          }
          /* else fall through... */
        bin1: case 1:
          if ( methods[0].methodNumber == methodReg ) {
            methodAddress = methods[0].methodAddress;
            goto addCache;
          }
          else goto checkSuperclasses;

      #define BINARY(mid,mid1)                        \
        if ( methodReg < methods[mid].methodNumber )  \
          goto bin##mid;  /* like "hi = mid" */       \
        else {                                        \
          methods += mid; /* like "lo = mid" */       \
          goto bin##mid1;                             \
        }

        /* binN: case N: BINARY(N/2,(N+1)/2) */
        bin3:  case  3: BINARY(1,2)
        bin4:  case  4: BINARY(2,2)
        bin5:  case  5: BINARY(2,3)
        bin6:  case  6: BINARY(3,3)
        bin7:  case  7: BINARY(3,4)
        bin8:  case  8: BINARY(4,4)
        bin9:  case  9: BINARY(4,5)
        bin10: case 10: BINARY(5,5)
        bin11: case 11: BINARY(5,6)
        bin12: case 12: BINARY(6,6)
        bin13: case 13: BINARY(6,7)
        bin14: case 14: BINARY(7,7)
        bin15: case 15: BINARY(7,8)
        bin16: case 16: BINARY(8,8)
        bin17: case 17: BINARY(8,9)
        bin18: case 18: BINARY(9,9)
        bin19: case 19: BINARY(9,10)
        bin20: case 20: BINARY(10,10)
        bin21: case 21: BINARY(10,11)
        bin22: case 22: BINARY(11,11)
        bin23: case 23: BINARY(11,12)
        bin24: case 24: BINARY(12,12)
        bin25: case 25: BINARY(12,13)
        bin26: case 26: BINARY(13,13)
        bin27: case 27: BINARY(13,14)
        bin28: case 28: BINARY(14,14)
        bin29: case 29: BINARY(14,15)
        bin30: case 30: BINARY(15,15)
        bin31: case 31: BINARY(15,16)
        bin32: case 32: BINARY(16,16)

        #define NUM_UNROLLED 32 /* # of unrolled cases */

        default:
          if (methodReg <= classPtr->largestMethodNumber) {
            register ushort lo, mid, hi, mn;
            lo = 0;
            hi = classPtr->methodCount - 1;
            while ( lo + NUM_UNROLLED - 1 < hi ) {
              mid = (lo + hi) >> 1;
              mn = methods[mid].methodNumber;
              if ( methodReg < mn )
                hi = mid;
              else if ( methodReg > mn )
                lo = mid;
              else {
                methodAddress = methods[mid].methodAddress;
                goto addCache;
              }
            }
            /* reduced to an unrolled case */
            searchSize = hi-lo+1;
            methods += lo;
            goto continueBinarySearch;
          }
      }
    }

/* Look in superclasses. Eliminating the recursion by keeping a custom stack of critical variables doesn't 
buy us much. Recursion also makes the code clearer. */

checkSuperclasses:
    {
      register ulong i, max = classPtr->inheritedCount;
           /* shared address register points into class list */
      temp = classPtr->inheritedClasses;

      for ( i = 0; i < max; i++ ) {
        methodAddress = FindMethod(*temp++, methodNumber);
        if ( methodAddress != kMethodNotFound )
          goto addCache;
      }
    }
  }

/* There are two ways to get here: classPtr is kClassNotFound (hopefully rare), or the method was not found 
in any of the superclasses. Either way, put a "not found" entry into the cache. This helps when searching 
complicated inheritance hierarchies or repeatedly looking up the same method in several related subclasses. 
*/

  methodAddress = kMethodNotFound;

Add an entry to the cache.
addCache:

// shared address register holds block pointer
  temp = (ushort*) blockPtr;
  #define block ((CacheBlock*) temp)
  {
    register ulong hits = block->hits;
    register ulong next = block->next;
    register ulong mask = 1L << next;

/* Choose the entry to replace. Look for a cleared bit in "hits" starting at "next". If all the bits are 1 initially, 
go all the way around; we are guaranteed to stop since we clear bits as we go. */

    while ( hits & mask ) {
      hits &= ~mask;              /* no, clear the bit */
      mask <<= 1;                 /* and try next bit */
      if ( mask == (1 << BLOCK_SIZE) )
        mask = 1;
      next++;
    }
    if ( next >= BLOCK_SIZE )
      next -= BLOCK_SIZE;

    block->methodAddress[ next ] = methodAddress;
    block->id[ next ] = id;
    block->hits = hits;
    block->next = ( next < BLOCK_SIZE-1 ? next+1 : 0 );
  }

  return methodAddress;
}

 

Community Search:
MacTech Search:

Software Updates via MacUpdate

The Slaughter: Act One (Games)
The Slaughter: Act One 1.0.323 Device: iOS Universal Category: Games Price: $3.99, Version: 1.0.323 (iTunes) Description: “The game mixes realism and surrealism to create a story that can cause just as much laughter as fear. A-” -... | Read more »
NEO TURF MASTERS (Games)
NEO TURF MASTERS 1.0 Device: iOS Universal Category: Games Price: $2.99, Version: 1.0 (iTunes) Description: NEOGEO’s legendary golf game is back, in a brand-new mobile version with touch controls! NEO TURF MASTERS (also known as “BIG... | Read more »
How to send money to a friend with the F...
Facebook is already a lot of things to a lot of people. It's a social network, sure, but also a tool for catologing memories, chatting with friends, and much more. [Read more] | Read more »
Branch out into the Chinese market with...
Created a game that you’re super enthused about, only to find yourself struggling to get it the exposure that you believe it duly deserves? With the time, energy and money that goes into development, it’s only right that games should be given the... | Read more »
Pocket Arcade Story (Games)
Pocket Arcade Story 1.00 Device: iOS Universal Category: Games Price: $4.99, Version: 1.00 (iTunes) Description: Here comes a new challenger: it's the arcade simulation game you've been waiting for! Build your very own gaming... | Read more »
How to get coins faster in Rodeo Stamped...
There comes a time in a cowboy or cowgirl's life when all the riding and lassoing skills in the world aren't enough. You're going to need some cold, hard cash to keep your sky zoo expanding in Rodeo Stampede. [Read more] | Read more »
How to out-do Cam Newton in Can You Dab?
The thing about dance crazes is that you're never really sure when they've run their course. Take the Dab, for instance. Propelled by its adoption as the touchdown celebration of choice for Carolina Panthers quarterback Cam Newton, the Dab seemed... | Read more »
Artik Games releases Splashy Cats for An...
Splashy Cats had us hooked from the title alone, and when we found out the game was literally just zig-zagging one of our favourite pop-culture references, guised as a playable cat character, down a river – our appetites were whetted to say the... | Read more »
Battle Cars (Games)
Battle Cars 1.1 Device: iOS Universal Category: Games Price: $1.99, Version: 1.1 (iTunes) Description: Welcome to the world of Battle Cars. Battle Cars is a classic arcade top-down racing game with fast mini cars and funny weapons to... | Read more »
How to get started with live.ly
One could be forgiven for thinking that there are already plenty of streaming video apps out there. It's just that the App Store charts would insist that you're mistaken. [Read more] | Read more »

Price Scanner via MacPrices.net

9-inch 128GB Silver iPad Pro on sale for $50...
B&H Photo has the 9.7″ 128GB Silver Apple iPad Pro on sale for $699 including free shipping plus NY tax only. Their price is $50 off MSRP. Read more
Why Use Indie Opera And Vivaldi Instead Of Sa...
For many years my web browser workhorses were various permutations and spinoffs of the Netscape/Mozilla/Firefox Open Source platform, and the Norwegian indie browser Opera, which I took a shine to... Read more
Western Digital Launches Worlds Fastest 256GB...
At the Mobile World Congress in Shanghai Western Digital Corporation this week introduced a new suite of 256 gigabyte (GB) microSD cards, which includes the new 256GB SanDisk Extreme microSDXC UHS-I... Read more
KeyCue 8.1 Integrates With Typinator To Displ...
Ergonis Software has released KeyCue 8.1, a new version of the company’s keyboard shortcut cheat sheet. KeyCue 8 introduced a new way to define a wide variety of triggers, which can be used to... Read more
Save up to $600 with Apple refurbished Mac Pr...
Apple has Certified Refurbished Mac Pros available for up to $600 off the cost of new models. An Apple one-year warranty is included with each Mac Pro, and shipping is free. The following... Read more
21-inch 2.8GHz iMac on sale for $1199, save $...
Amazon has the 21″ 2.8GHz iMac (model #MK442LL/A) on sale for $1199.99 including free shipping. Their price is $100 off MSRP, and it’s the lowest price available for this model. Read more
13-inch 2.5GHz MacBook Pro (Apple refurbished...
Apple has Certified Refurbished 13″ 2.5GHz MacBook Pros available for $829, or $270 off the cost of new models. Apple’s one-year warranty is standard, and shipping is free: - 13″ 2.5GHz MacBook Pros... Read more
MacBook Airs on sale for up to $50-$100 off M...
B&H Photo has 13″ and 11″ MacBook Airs on sale for up to $100 off MSRP. Shipping is free, and B&H charges NY sales tax only: - 11″ 1.6GHz/128GB MacBook Air: $849 $50 off list price - 11″ 1.... Read more
Brexit Vote Result Forecast To Slash UK 2016...
Uncertainty and economic volatility can be expected to increase over the next nine months, as the Brexit and concerns over the future of the EU hit IT investment, say Canalys market analysts, with... Read more
13-inch 256GB MacBook Air on sale for $1149,...
Amazon has the 2016 13″ 1.6GHz/256GB MacBook Air (model MMGG2LL/A) on sale for $1149.99 including free shipping. Their price is also $50 off MSRP. Read more

Jobs Board

*Apple* Retail - Multiple Positions - Apple,...
Job Description: Sales Specialist - Retail Customer Service and Sales Transform Apple Store visitors into loyal Apple customers. When customers enter the store, Read more
*Apple* iPhone 6s and New Products Tester Ne...
…we therefore look forward to put out products to quality test for durability. Apple leads the digital music revolution with its iPods and iTunes online store, Read more
*Apple* iPhone 6s and New Products Tester Ne...
…we therefore look forward to put out products to quality test for durability. Apple leads the digital music revolution with its iPods and iTunes online store, Read more
*Apple* iPhone 6s and New Products Tester Ne...
…we therefore look forward to put out products to quality test for durability. Apple leads the digital music revolution with its iPods and iTunes online store, Read more
*Apple* Retail - Multiple Positions, Willow...
Job Description: Sales Specialist - Retail Customer Service and Sales Transform Apple Store visitors into loyal Apple customers. When customers enter the store, Read more
All contents are Copyright 1984-2011 by Xplain Corporation. All rights reserved. Theme designed by Icreon.