% This file is part of the Stanford GraphBase (c) Stanford University 1992
\def\title{LADDERS}
@i boilerplate.w %<< legal stuff: PLEASE READ IT BEFORE MAKING ANY CHANGES!

\prerequisites{GB\_WORDS}{GB\_\thinspace DIJK}
@* Introduction. This demonstration program uses graphs
constructed by the |gb_words| module to produce
an interactive program called \.{ladders}, which finds shortest paths
between two given five-letter words of English.

The program assumes that \UNIX\ conventions are being used. Some code in
sections listed under `\UNIX\ dependencies' in the index may need to change
if this program is ported to other operating systems.

\def\<#1>{$\langle${\rm#1}$\rangle$}
To run the program under \UNIX, say `\.{ladders} \<options>', where \<options>
consists of zero or more of the following specifications in any order:

{\narrower
\def\\#1 {\smallskip\noindent
  \hbox to 6em{\tt#1\hfill}\hangindent 8em\hangafter1 }
\\-v Verbosely print all words encountered during the shortest-path computation,
     showing also their distances from the goal word.
\\-a Use alphabetic distance instead of considering adjacent words to be one
     unit apart; for example, the alphabetic distance from `\.{words}' to
     `\.{woods}' is~3, because `\.r' is three places from `\.o' in the
     alphabet.
\\-f Use distance based on frequency (see below), instead of considering
     adjacent words to be one unit apart. This option is ignored if \.{-a}
     has been specified or if \.{-r} has been specified.
\\-h Use a lower-bound heuristic to shorten the search (see below). This option
     is ignored if option \.{-f} has been selected.
\\-e Echo the input to the output (useful if input comes from a file instead
     of from the terminal).
\\-n\<number> Limit the graph to the |n| most common English words, where |n| is
     the given \<number>.
\\-r\<number> Limit the graph to \<number> randomly selected words. This option
     is incompatible with~\.{-n}.
\\-s\<number> Use \<number> instead of 0 as the seed for random numbers, to get
     different random samples or to explore words of equal frequency in
     a different order.
\smallskip}
\noindent Option \.{-f} assigns a cost of 0 to the most common words and a
cost of 16 to the least common words; a cost between 0 and~16 is assigned to
words of intermediate frequency. The word ladders found will then have
minimum total cost by this criterion.
\smallskip
Option \.{-h} attempts to focus the search by giving priority to words that
are near the goal. (More precisely, it modifies distances between adjacent
words by using a heuristic function $\\{hh}(v)$, which would be the shortest
possible distance between |v| and the goal if every five-letter combination
happened to be an English word.) The |gb_dijk| module explains more about
such heuristics; this option is most interesting to watch when used in
conjunction with \.{-v}.

@ The program will prompt you for a starting word. If you simply type \<return>,
it exits; otherwise you should enter a five-letter word (with no uppercase
letters) before typing \<return>.

Then the program will prompt you for a goal word. If you simply type
\<return> at this point, it will go back and ask for a new starting word;
otherwise you should specify another five-letter word.

Then the program will find and display an optimal word ladder from the start
to the goal, if there is a path from one to the other
that changes only one letter at a time.

And then you have a chance to start all over again, with another starting word.

The start and goal words need not be present in the program's graph of
``known'' words. They are temporarily added to that graph, but removed
again whenever new start and goal words are given. If the \.{-f} option is
being used, the cost of the goal word will be 20 when it is not in the
program's dictionary.

@ We use the data types \&{Vertex}, \&{Arc}, and \&{Graph}
defined in |gb_graph|.

@f Vertex int
@f Arc int
@f Graph int

@ Here is the general layout of this program, as seen by the \Cee\ compiler:
@^UNIX dependencies@>

@p
#include <ctype.h> /* system file for character types */
#include "gb_graph.h" /* the standard GraphBase data structures */
#include "gb_words.h" /* routines for five-letter word graphs */
#include "gb_dijk.h" /* routines for shortest paths */
@#
@<Global variables@>@;
@<Subroutines@>@;
main(argc,argv)
  int argc; /* the number of command-line arguments */
  char *argv[]; /* an array of strings containing those arguments */
{
  @<Scan the command line options@>;
  @<Set up the graph of words@>;
  while(1) {
    @<Prompt for starting word and goal word; |break| if none given@>;
    @<Find a minimal ladder from |start| to |goal|, if one exists,
        and print it@>;
  }
}

@* Parsing the options. Let's get the \UNIX\ command-line junk out of the
way first, so that we can concentrate on meatier stuff. Our job in this part
of the program is to see if the default value zero of external variable
|verbose| should change, and/or if the default values of any of the following
internal variables should change:

@<Global variables@>=
char alph=0; /* nonzero if the alphabetic distance option is selected */
char freq=0; /* nonzero if the frequency-based distance option is selected */
char heur=0; /* nonzero if the heuristic search option is selected */
char echo=0; /* nonzero if the input-echo option is selected */
unsigned n=0; /* maximum number of words in the graph (0 means infinity) */
char rand=0; /* nonzero if we will ignore the weight of words */
long seed=0; /* seed for random number generator */

@ @<Scan the command line options@>=
while (--argc) {
@^UNIX dependencies@>
  if (strcmp(argv[argc],"-v")==0) verbose=1;
  else if (strcmp(argv[argc],"-a")==0) alph=1;
  else if (strcmp(argv[argc],"-f")==0) freq=1;
  else if (strcmp(argv[argc],"-h")==0) heur=1;
  else if (strcmp(argv[argc],"-e")==0) echo=1;
  else if (sscanf(argv[argc],"-n%u",&n)==1) rand=0;
  else if (sscanf(argv[argc],"-r%u",&n)==1) rand=1;
  else if (sscanf(argv[argc],"-s%ld",&seed)==1) ;
  else {
    fprintf(stderr,"Usage: %s [-v][-a][-f][-h][-e][-nN][-rN][-sN]\n",argv[0]);
    return -2;
  }
}
if (alph || rand) freq=0;
if (freq) heur=0;

@*Creating the graph. The GraphBase |words| procedure will produce the
five-letter words we want, organized in a graph structure.

@d quit_if(x,c)
  if (x) {
    fprintf(stderr,
      "Sorry, I couldn't build a dictionary (trouble code %d)!\n",c);
    return c;
  }

@<Set up the graph of words@>=
g=words(n,(rand? zero_vector: NULL), 0,seed);
quit_if(g==NULL,panic_code);
@<Confirm the options selected@>;
@<Modify the edge lengths, if the |alph| or |freq| option was selected@>;
@<Modify the priority queue algorithm, if unequal edge lengths are possible@>;

@ @<Glob...@>=
Graph *g; /* graph created by |words| */
int zero_vector[9]; /* weights to use when ignoring all frequency information */

@ The actual number of words may be decreased to the size of the GraphBase
dictionary, so we wait until the graph is generated before confirming
the user-selected options.

@<Confirm the options selected@>=
if (verbose) {
  if (alph) printf("(alphabetic distance selected)\n");
  if (freq) printf("(frequency-based distances selected)\n");
  if (heur) printf("(lowerbound heuristic will be used to focus the search)\n");
  if (rand) printf("(random selection of %d words with seed %d)\n",g->n,seed);
  else printf("(the graph has %d words)\n",g->n);
}

@ The edges in a |words| graph normally have length 1, so we must change them
if the user has selected |alph| or |freq|. The character position in which
adjacent words differ is recorded in the |loc| field of each arc. The
frequency of a word is stored in the |weight| field of its vertex.

@d a_dist(k) (*(p+k)<*(q+k)? *(q+k)-*(p+k): *(p+k)-*(q+k))

@<Modify the edge lengths, if the |alph| or |freq| option was selected@>=
if (alph) {@+register Vertex *u;
  for (u=g->vertices+g->n-1; u>=g->vertices; u--) {@+register Arc *a;
    register char *p=u->name;
    for (a=u->arcs; a; a=a->next) {register char *q=a->tip->name;
      a->len = a_dist(a->loc);
    }
  }
} else if (freq) {@+register Vertex *u;
  for (u=g->vertices+g->n-1; u>=g->vertices; u--) {@+register Arc *a;
    for (a=u->arcs; a; a=a->next)
      a->len = freq_cost(a->tip);
  }
}

@ The default priority queue algorithm of |dijkstra| is quite efficient
when all edge lengths are~1. Otherwise we will change it to the
alternative method that works best for edge lengths less than~128.

@<Modify the priority queue algorithm...@>=
if (alph || freq || heur) {
  init_queue=init_128;
  delete_min=delete_from_128;
  enqueue=enqueue_128;
  requeue=requeue_128;
}

@ The frequency has been computed with the default weights explained in the
documentation of |words|; it is usually less than $2^{16}$.
A word whose frequency is 0 costs~16; a word whose frequency is 1 costs~15;
a word whose frequency is 2 or 3 costs~14; and the costs keeps decreasing
by~1 as the frequency doubles, until we get down to a cost of~0.

@<Sub...@>=
int freq_cost(v)
  Vertex *v;
{@+register long acc=v->weight; /* the frequency, to be shifted right */
  register k=16;
  while (acc) k--, acc>>=1;
  return (k<0? 0: k);
}

@* Minimal ladders. The guts of this program is a routine to compute shortest
paths between two given words, |start| and |goal|.

The |dijkstra| procedure does this, in any graph with nonnegative arc lengths.
The only complication we need to deal with here is that |start| and |goal|
might not themselves be present in the graph. In that case we want to add
them, albeit temporarily.

The conventions of |gb_graph| allow us to do the desired augmentation
by creating a new graph |gg| whose vertices are borrowed from~|g|. The
graph~|g| has space for two more vertices (actually for four), and any
new memory blocks allocated for the additional arcs present in~|gg| will
be freed later by the operation |gb_recycle(gg)| without confusion.

@<Glob...@>=
Graph *gg; /* clone of |g| with possible additional words */
char start[6], goal[6];
   /* \.{words} dear to the user's \.{heart}, plus |'\0'| */
Vertex *uu, *vv; /* start and goal vertices in |gg| */

@ @<Find a minimal ladder from |start| to |goal|...@>=
@<Build the amplified graph |gg|@>;
@<Let |dijkstra| do the hard work@>;
@<Print the answer@>;
@<Remove all traces of |gg|@>;

@ @<Build the amplified graph |gg|@>=
gg=gb_new_graph(0);
quit_if(gg==NULL,20);  /* out of memory */
gg->vertices = g->vertices;
gg->n = g->n;
@<Put the |start| word into |gg|, and let |uu| point to it@>;
@<Put the |goal| word into |gg|, and let |vv| point to it@>;
if (gg->n==g->n+2) @<Check if |start| is adjacent to |goal|@>;
quit_if(gb_alloc_trouble,21); /* out of memory */

@ The |find_word| procedure returns |NULL| if it can't find the given word
in the graph just constructed by |words|. In that case it has applied its
second argument to every adjacent word. Hence the program logic here
does everything needed to add a new vertex to~|gg| when necessary.

@<Put the |start| word into |gg|, and let |uu| point to it@>=
(gg->vertices+gg->n)->name = start; /* a tentative new vertex */
uu=find_word(start,plant_new_edge);
if (!uu)
  uu = gg->vertices + gg->n++; /* recognize the new vertex and refer to it */

@ @<Put the |goal|...@>=
if (strncmp(start,goal,5)==0) vv=uu; /* avoid inserting a word twice */
else {
  (gg->vertices+gg->n)->name = goal; /* a tentative new vertex */
  vv=find_word(goal,plant_new_edge);
  if (!vv)
    vv = gg->vertices + gg->n++; /* recognize the new vertex and refer to it */
}

@ @<Sub...@>=
void plant_new_edge(v)
  Vertex *v;
{@+Vertex *u=gg->vertices+gg->n; /* the new edge runs from |u| to |v| */
  gb_new_edge(u,v,1);
  if (alph)
    u->arcs->len=(u->arcs-1)->len=alph_dist(u->name,v->name);
  else if (freq) {
    u->arcs->len=20; /* adjust the arc length from |v| to |u| */
    (u->arcs-1)->len=freq_cost(v); /* adjust the arc length from |u| to |v| */
  }
}

@ The |alph_dist| subroutine calculates the alphabetic distance between
arbitrary five-letter words, whether they are adjacent or not.

@<Sub...@>=
int alph_dist(p,q)
  register char *p, *q;
{
  return a_dist(0)+a_dist(1)+a_dist(2)+a_dist(3)+a_dist(4);
}

@ There's a bug in the above logic that could be embarrassing,
although it will come up only when a user is trying to be clever: The
|find_word| routine knows only the words of~|g|, so it will fail to
make any direct connection between |start| and |goal| if they happen
to be adjacent to each other yet not in the original graph. We had
better fix this, or else the ladder program will look stupid.

@<Check if |start|...@>=
if (hamm_dist(start,goal)==1) {
  gg->n--; /* temporarily pretend |vv| hasn't been added yet */
  plant_new_edge(uu); /* make |vv| adjacent to |uu| */
  gg->n++; /* and recognize it again */
}

@ The Hamming distance between words is the number of character positions
in which they differ.

@d h_dist(k) (*(p+k)==*(q+k)? 0: 1)

@<Sub...@>=
int hamm_dist(p,q)
  register char *p, *q;
{
  return h_dist(0)+h_dist(1)+h_dist(2)+h_dist(3)+h_dist(4);
}

@ OK, now we've got a graph in which |dijkstra| can operate.

@<Let |dijkstra| do the hard work@>=
if (!heur) min_dist=dijkstra(uu,vv,gg,NULL);
else if (alph) min_dist=dijkstra(uu,vv,gg,alph_heur);
else min_dist=dijkstra(uu,vv,gg,hamm_heur);

@ @<Sub...@>=
long alph_heur(v)
  Vertex *v;
{@+return alph_dist(v->name,goal);@+}
@#
long hamm_heur(v)
  Vertex *v;
{@+return hamm_dist(v->name,goal);@+}

@ @<Glob...@>=
long min_dist; /* length of the shortest ladder */

@ @<Print the answer@>=
if (min_dist<0) printf("Sorry, there's no ladder from %s to %s.\n",start,goal);
else print_dijkstra_result(vv);

@ Finally, we have to clean up our tracks. It's easy to remove all arcs
from the new vertices of~|gg| to the old vertices of~|g|; it's a bit
more tricky to remove the arcs from old to new. The loop here will also
remove arcs properly between start and goal vertices, if they both
belong to |gg| not~|g|.

@<Remove all traces of |gg|@>=
for (uu=g->vertices+gg->n-1; uu>=g->vertices+g->n; uu--) {@+register Arc *a;
  for (a=uu->arcs; a; a=a->next) {
    vv=a->tip; /* now |vv->arcs==a-1|, since arcs for edges come in pairs */
    vv->arcs=vv->arcs->next;
  }
  uu->arcs=NULL; /* we needn't clear |uu->name| */
}
gb_recycle(gg); /* the |gg->data| blocks disappear, but |g->data| remains */

@* Terminal interaction. We've finished doing all the interesting things;
only one minor part of the program still remains to be written.

@<Prompt for...@>=
putchar('\n'); /* make a blank line for visual punctuation */
restart: /* if we try to avoid this label,
              the |break| command will be broken */
if (prompt_for_five("Starting",start)!=0) break;
if (prompt_for_five("    Goal",goal)!=0) goto restart;

@ @<Sub...@>=
int prompt_for_five(s,p)
  char *s; /* string used in prompt message */
  register char *p; /* where to put a string typed by the user */
{@+register char *q; /* current position to store characters */
  register int c; /* current character of input */
  while (1) {
    printf("%s word: ",s);
    fflush(stdout); /* make sure the user sees the prompt */
    q=p;
    while (1) {
      c=getchar();
      if (c==EOF) return -1; /* end-of-file */
      if (echo) putchar(c);
      if (c=='\n') break;
      if (!islower(c)) q=p+5;
      else if (q<p+5) *q=c;
      q++;
    }
    if (q==p+5) return 0; /* got a good five-letter word */
    if (q==p) return 1; /* got just \<return> */
    printf("(Please type five lowercase letters and RETURN.)\n");
  }
}

@* Index. Finally, here's a list that shows where the identifiers of this
program are defined and used.