pike.git/ src/ builtin_functions.c

2006-03-09

2006-03-09 20:31:26 by Henrik Grubbström (Grubba) <grubba@grubba.org>

• 3f268a98aa5e296c91a6850128c54cfa1339662f (99 lines) (+67/-32) ^{[ Show | Annotate ]}
  Branch: 7.9

Rewrote find_longest_prefix() to not recurse.
find_longest_prefix() now requires the prefix forest to be initialized before being called (this is now done by replace_many()).
Added a few FIXME's about possible optimizations.

Rev: src/builtin_functions.c:1.607

2:   || This file is part of Pike. For copyright information see COPYRIGHT.   || Pike is distributed under GPL, LGPL and MPL. See the file COPYING   || for more information. - || $Id: builtin_functions.c,v 1.606 2006/03/09 15:55:07 grubba Exp $ + || $Id: builtin_functions.c,v 1.607 2006/03/09 20:31:26 grubba Exp $   */      #include "global.h"
3277:   };      /* Magic, magic and more magic */ + /* Returns the index in v for the string that is the longest prefix of +  * str (if any). +  * +  * v is the sorted (according to generic_quick_binary_strcmp()) vector +  * of replacement strings. It also has the prefix forest identified. +  * +  * a is the lower bound. +  * b is the upper bound + 1. +  */   static int find_longest_prefix(char *str,    ptrdiff_t len,    int size_shift,
3284:    INT32 a,    INT32 b)   { -  INT32 c,match=-1; +  INT32 c,match=-1, match_len=0, d;    ptrdiff_t tmp;       check_c_stack(2048);
3301:    size_shift);    if(tmp<0)    { -  INT32 match2=find_longest_prefix(str, -  len, -  size_shift, -  v, -  c+1, -  b); -  if(match2!=-1) return match2; -  -  while(1) -  { -  if(v[c].prefix==-2) -  { -  v[c].prefix=find_longest_prefix(v[c].ind->str, -  v[c].ind->len, -  v[c].ind->size_shift, -  v, -  0 /* can this be optimized? */, -  c); -  } -  c=v[c].prefix; -  if(c<a || c<match) return match; -  -  if(!generic_quick_binary_strcmp(v[c].ind->str, -  v[c].ind->len, -  v[c].ind->size_shift, +  int d; +  /* Check if we have a partial prefix. +  */ +  /* FIXME: We could note at what point v[c].ind and str differ during +  * the comparison above, and then skip through the prefixes +  * until we get to one that is shorter. +  * See also similar FIXME in replace_many() below. +  * /grubba 2006-03-09 +  */ +  if ((v[c].ind->len > match_len) && +  (index_shared_string(v[c].ind, match_len) == +  INDEX_CHARP(str, match_len, size_shift))) { +  /* We need to look closer to see if we might have a partial prefix. */ +  int d = c; +  while ((d = v[d].prefix) >= a) { +  if (!generic_quick_binary_strcmp(v[d].ind->str, +  v[d].ind->len, +  v[d].ind->size_shift,    str, -  MINIMUM(len,v[c].ind->len), -  size_shift)) -  return c; +  MINIMUM(len, v[d].ind->len), +  size_shift)) { +  /* Found a valid prefix. */ +  match = d; +  match_len = v[d].ind->len; +  break;    }    } -  +  } +  a = c+1; +  }    else if(tmp>0)    { -  +  fprintf(stderr, "v[%d] == \"%s\" > \"%s\"\n", +  c, v[c].ind->str, str);    b=c;    }    else    { -  +  fprintf(stderr, "v[%d] == \"%s\" is prefix to \"%s\"\n", +  c, v[c].ind->str, str);    a=c+1; /* There might still be a better match... */    match=c; -  +  match_len = v[c].ind->len;    }    } -  +  fprintf(stderr, "Result: %d\n", match);    return match;   }   
3367:    struct array *from,    struct array *to)   { -  INT32 e,num; +  INT32 e, num;    ptrdiff_t s, length;    struct replace_many_context ctx;    ONERROR uwp;
3433:    x=index_shared_string(ctx.v[e].ind,0);    if((x<(INT32)NELEM(set_end)) && (x >= 0))    set_end[x]=e+1; +  x = e-1; +  while (x >= 0) { +  if (ctx.v[e].ind->len >= ctx.v[x].ind->len) { +  /* FIXME: We could speed up the loop by noting at what point +  * ctx.v[x] and ctx.v[e] differ, and then skip +  * though the prefixes until we get to one that is shorter. +  */ +  if (!generic_quick_binary_strcmp(ctx.v[e].ind->str, +  ctx.v[x].ind->len, +  ctx.v[e].ind->size_shift, +  ctx.v[x].ind->str, +  ctx.v[x].ind->len, +  ctx.v[x].ind->size_shift)) { +  /* ctx.v[x] is a valid prefix to ctx.v[e]. */ +  break;    } -  +  } +  x = ctx.v[x].prefix; +  } +  ctx.v[e].prefix = x; +  }       length=str->len;