isomorfeus-ferret 0.12.0

data/ext/isomorfeus_ferret_ext/frb_utils.c

@@ -0,0 +1,1127 @@
+#include "frt_bitvector.h"
+#include "frt_multimapper.h"
+#include "isomorfeus_ferret.h"
+#include <ruby/st.h>
+
+/*****************
+ *** BitVector ***
+ *****************/
+static VALUE cBitVector;
+
+static void
+frb_bv_free(void *p)
+{
+    object_del(p);
+    frt_bv_destroy((FrtBitVector *)p);
+}
+
+static VALUE
+frb_bv_alloc(VALUE klass)
+{
+    FrtBitVector *bv = frt_bv_new();
+    VALUE rbv = Data_Wrap_Struct(klass, NULL, &frb_bv_free, bv);
+    object_add(bv, rbv);
+    return rbv;
+}
+
+#define GET_BV(bv, self) Data_Get_Struct(self, FrtBitVector, bv)
+
+VALUE
+frb_get_bv(FrtBitVector *bv)
+{
+    VALUE rbv;
+    if ((rbv = object_get(bv)) == Qnil) {
+        rbv = Data_Wrap_Struct(cBitVector, NULL, &frb_bv_free, bv);
+        FRT_REF(bv);
+        object_add(bv, rbv);
+    }
+    return rbv;
+}
+
+/*
+ *  call-seq:
+ *     BitVector.new() -> new_bit_vector
+ *
+ *  Returns a new empty bit vector object
+ */
+static VALUE
+frb_bv_init(VALUE self)
+{
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv[i] = bool  -> bool
+ *
+ *  Set the bit and _i_ to *val* (+true+ or
+ *  +false+).
+ */
+VALUE
+frb_bv_set(VALUE self, VALUE rindex, VALUE rstate)
+{
+    FrtBitVector *bv;
+    int index = FIX2INT(rindex);
+    GET_BV(bv, self);
+    if (index < 0) {
+        rb_raise(rb_eIndexError, "%d < 0", index);
+    }
+    if (RTEST(rstate)) {
+        frt_bv_set(bv, index);
+    }
+    else {
+        frt_bv_unset(bv, index);
+    }
+
+    return rstate;
+}
+
+/*
+ *  call-seq:
+ *     bv.set(i) -> self
+ *
+ *  Set the bit at _i_ to *on* (+true+)
+ */
+VALUE
+frb_bv_set_on(VALUE self, VALUE rindex)
+{
+    frb_bv_set(self, rindex, Qtrue);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv.unset(i) -> self
+ *
+ *  Set the bit at _i_ to *off* (+false+)
+ */
+VALUE
+frb_bv_set_off(VALUE self, VALUE rindex)
+{
+    frb_bv_set(self, rindex, Qfalse);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv.get(i) -> bool
+ *     bv[i]     -> bool
+ *
+ *  Get the bit value at _i_
+ */
+VALUE
+frb_bv_get(VALUE self, VALUE rindex)
+{
+    FrtBitVector *bv;
+    int index = FIX2INT(rindex);
+    GET_BV(bv, self);
+    if (index < 0) {
+        rb_raise(rb_eIndexError, "%d < 0", index);
+    }
+
+    return frt_bv_get(bv, index) ? Qtrue : Qfalse;
+}
+
+/*
+ *  call-seq:
+ *     bv.count -> bit_count
+ *
+ *  Count the number of bits set in the bit vector. If the bit vector has been
+ *  negated using +#not+ then count the number of unset bits
+ *  instead.
+ */
+VALUE
+frb_bv_count(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    return INT2FIX(bv->count);
+}
+
+/*
+ *  call-seq:
+ *     bv.clear -> self
+ *
+ *  Clears all set bits in the bit vector. Negated bit vectors will still have
+ *  all bits set to *off*.
+ */
+VALUE
+frb_bv_clear(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    frt_bv_clear(bv);
+    frt_bv_scan_reset(bv);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv1 == bv2    -> bool
+ *     bv1 != bv2    -> bool
+ *     bv1.eql(bv2)  -> bool
+ *
+ *  Compares two bit vectors and returns true if both bit vectors have the same
+ *  bits set.
+ */
+VALUE
+frb_bv_eql(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    return frt_bv_eq(bv1, bv2) ? Qtrue : Qfalse;
+}
+
+/*
+ *  call-seq:
+ *     bv.hash -> int
+ *
+ *  Used to store bit vectors in Hashes. Especially useful if you want to
+ *  cache them.
+ */
+VALUE
+frb_bv_hash(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    return ULONG2NUM(frt_bv_hash(bv));
+}
+
+/*
+ *  call-seq:
+ *     bv1 & bv2    -> anded_bv
+ *     bv1.and(bv2) -> anded_bv
+ *
+ *  Perform a boolean _and_ operation on +bv1+ and
+ *  +bv2+
+ */
+VALUE
+frb_bv_and(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    return Data_Wrap_Struct(cBitVector, NULL, &frt_bv_destroy, frt_bv_and(bv1, bv2));
+}
+
+/*
+ *  call-seq:
+ *     bv1.and!(bv2) -> self
+ *
+ *  Perform a boolean _and_ operation on +bv1+ and
+ *  +bv2+ in place on +bv1+
+ */
+VALUE
+frb_bv_and_x(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    frt_bv_and_x(bv1, bv2);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv1 | bv2   -> ored_bv
+ *     bv1.or(bv2) -> ored_bv
+ *
+ *  Perform a boolean _or_ operation on +bv1+ and
+ *  +bv2+
+ */
+VALUE
+frb_bv_or(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    return Data_Wrap_Struct(cBitVector, NULL, &frt_bv_destroy, frt_bv_or(bv1, bv2));
+}
+
+/*
+ *  call-seq:
+ *     bv1.or!(bv2) -> self
+ *
+ *  Perform a boolean _or_ operation on +bv1+ and
+ *  +bv2+ in place on +bv1+
+ */
+VALUE
+frb_bv_or_x(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    frt_bv_or_x(bv1, bv2);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv1 ^ bv2    -> xored_bv
+ *     bv1.xor(bv2) -> xored_bv
+ *
+ *  Perform a boolean _xor_ operation on +bv1+ and
+ *  +bv2+
+ */
+VALUE
+frb_bv_xor(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    return Data_Wrap_Struct(cBitVector, NULL, &frt_bv_destroy, frt_bv_xor(bv1, bv2));
+}
+
+/*
+ *  call-seq:
+ *     bv1.xor!(bv2) -> self
+ *
+ *  Perform a boolean _xor_ operation on +bv1+ and
+ *  +bv2+ in place on +bv1+
+ */
+VALUE
+frb_bv_xor_x(VALUE self, VALUE other)
+{
+    FrtBitVector *bv1, *bv2;
+    GET_BV(bv1, self);
+    GET_BV(bv2, other);
+    frt_bv_xor_x(bv1, bv2);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     ~bv -> bv
+ *     bv.not -> bv
+ *
+ *  Perform a boolean _not_ operation on +bv+
+ *  */
+VALUE
+frb_bv_not(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    return Data_Wrap_Struct(cBitVector, NULL, &frt_bv_destroy, frt_bv_not(bv));
+}
+
+/*
+ *  call-seq:
+ *     bv.not! -> self
+ *
+ *  Perform a boolean _not_ operation on +bv+ in-place
+ */
+VALUE
+frb_bv_not_x(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    frt_bv_not_x(bv);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv.reset_scan -> self
+ *
+ *  Resets the BitVector ready for scanning. You should call this method
+ *  before calling +#next+ or +#next_unset+. It isn't
+ *  necessary for the other scan methods or for the +#each+ method.
+ */
+VALUE
+frb_bv_reset_scan(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    frt_bv_scan_reset(bv);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv.next -> bit_num
+ *
+ *  Returns the next set bit in the bit vector scanning from low order to high
+ *  order. You should call +#reset_scan+ before calling this method
+ *  if you want to scan from the beginning. It is automatically reset when you
+ *  first create the bit vector.
+ */
+VALUE
+frb_bv_next(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    return INT2FIX(frt_bv_scan_next(bv));
+}
+
+/*
+ *  call-seq:
+ *     bv.next_unset -> bit_num
+ *
+ *  Returns the next unset bit in the bit vector scanning from low order to
+ *  high order. This method should only be called on bit vectors which have
+ *  been flipped (negated). You should call +#reset_scan+ before
+ *  calling this method if you want to scan from the beginning. It is
+ *  automatically reset when you first create the bit vector.
+ */
+VALUE
+frb_bv_next_unset(VALUE self)
+{
+    FrtBitVector *bv;
+    GET_BV(bv, self);
+    return INT2FIX(frt_bv_scan_next_unset(bv));
+}
+
+/*
+ *  call-seq:
+ *     bv.next_from(from) -> bit_num
+ *
+ *  Returns the next set bit in the bit vector scanning from low order to
+ *  high order and starting at +from+. The scan is inclusive so if
+ *  +from+ is equal to 10 and +bv[10]+ is set it will
+ *  return the number 10. If the bit vector has been negated than you should
+ *  use the +#next_unset_from+ method.
+ */
+VALUE
+frb_bv_next_from(VALUE self, VALUE rfrom)
+{
+    FrtBitVector *bv;
+    int from = FIX2INT(rfrom);
+    GET_BV(bv, self);
+    if (from < 0) {
+        from = 0;
+    }
+    return INT2FIX(frt_bv_scan_next_from(bv, from));
+}
+
+/*
+ *  call-seq:
+ *     bv.next_unset_from(from) -> bit_num
+ *
+ *  Returns the next unset bit in the bit vector scanning from low order to
+ *  high order and starting at +from+. The scan is inclusive so if
+ *  +from+ is equal to 10 and +bv[10]+ is unset it will
+ *  return the number 10. If the bit vector has not been negated than you
+ *  should use the +#next_from+ method.
+ */
+VALUE
+frb_bv_next_unset_from(VALUE self, VALUE rfrom)
+{
+    FrtBitVector *bv;
+    int from = FIX2INT(rfrom);
+    GET_BV(bv, self);
+    if (from < 0) {
+        from = 0;
+    }
+    return INT2FIX(frt_bv_scan_next_unset_from(bv, from));
+}
+
+/*
+ *  call-seq:
+ *     bv.each { |bit_num| }
+ *
+ *  Iterate through all the set bits in the bit vector yielding each one in
+ *  order
+ */
+VALUE
+frb_bv_each(VALUE self)
+{
+    FrtBitVector *bv;
+    int bit;
+    GET_BV(bv, self);
+    frt_bv_scan_reset(bv);
+    if (bv->extends_as_ones) {
+        while ((bit = frt_bv_scan_next_unset(bv)) >= 0) {
+            rb_yield(INT2FIX(bit));
+        }
+    }
+    else {
+        while ((bit = frt_bv_scan_next(bv)) >= 0) {
+            rb_yield(INT2FIX(bit));
+        }
+    }
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     bv.to_a
+ *
+ *  Iterate through all the set bits in the bit vector adding the index of
+ *  each set bit to an array. This is useful if you want to perform array
+ *  methods on the bit vector. If you want to convert an array to a bit_vector
+ *  simply do this;
+ *
+ *    bv = [1, 12, 45, 367, 455].inject(FrtBitVector.new) {|bv, i| bv.set(i)}
+ */
+VALUE
+frb_bv_to_a(VALUE self)
+{
+    FrtBitVector *bv;
+    int bit;
+    VALUE ary;
+    GET_BV(bv, self);
+    ary = rb_ary_new();
+    frt_bv_scan_reset(bv);
+    if (bv->extends_as_ones) {
+        while ((bit = frt_bv_scan_next_unset(bv)) >= 0) {
+            rb_ary_push(ary, INT2FIX(bit));
+        }
+    }
+    else {
+        while ((bit = frt_bv_scan_next(bv)) >= 0) {
+            rb_ary_push(ary, INT2FIX(bit));
+        }
+    }
+    return ary;
+}
+
+static VALUE mUtils;
+
+/*
+ * Document-class: Ferret::Utils::BitVector
+ *
+ * == Summary
+ *
+ * A BitVector is pretty easy to implement in Ruby using Ruby's BigNum class.
+ * This BitVector however allows you to count the set bits with the
+ * +#count+ method (or unset bits of flipped bit vectors) and also
+ * to quickly scan the set bits.
+ *
+ * == Boolean Operations
+ *
+ * BitVector handles four boolean operations;
+ *
+ * * +&+
+ * * +|+
+ * * +^+
+ * * +~+
+ *
+ *    bv1 = BitVector.new
+ *    bv2 = BitVector.new
+ *    bv3 = BitVector.new
+ *
+ *    bv4 = (bv1 & bv2) | ~bv3
+ *
+ * You can also do the operations in-place;
+ *
+ * * +and!+
+ * * +or!+
+ * * +xor!+
+ * * +not!+
+ *
+ *    bv4.and!(bv5).not!
+ *
+ * == Set Bit Scanning
+ *
+ * Perhaps the most useful functionality in BitVector is the ability to
+ * quickly scan for set bits. To print all set bits;
+ *
+ *    bv.each {|bit| puts bit }
+ *
+ * Alternatively you could use the lower level +next+ or
+ * +next_unset+ methods. Note that the +each+ method will
+ * automatically scan unset bits if the BitVector has been flipped (using
+ * +not+).
+ */
+static void
+Init_BitVector(void)
+{
+    /* BitVector */
+    cBitVector = rb_define_class_under(mUtils, "BitVector", rb_cObject);
+    rb_define_alloc_func(cBitVector, frb_bv_alloc);
+
+    rb_define_method(cBitVector, "initialize", frb_bv_init, 0);
+    rb_define_method(cBitVector, "set", frb_bv_set_on, 1);
+    rb_define_method(cBitVector, "unset", frb_bv_set_off, 1);
+    rb_define_method(cBitVector, "[]=", frb_bv_set, 2);
+    rb_define_method(cBitVector, "get", frb_bv_get, 1);
+    rb_define_method(cBitVector, "[]", frb_bv_get, 1);
+    rb_define_method(cBitVector, "count", frb_bv_count, 0);
+    rb_define_method(cBitVector, "clear", frb_bv_clear, 0);
+    rb_define_method(cBitVector, "eql?", frb_bv_eql, 1);
+    rb_define_method(cBitVector, "==", frb_bv_eql, 1);
+    rb_define_method(cBitVector, "hash", frb_bv_hash, 0);
+    rb_define_method(cBitVector, "and!", frb_bv_and_x, 1);
+    rb_define_method(cBitVector, "and", frb_bv_and, 1);
+    rb_define_method(cBitVector, "&", frb_bv_and, 1);
+    rb_define_method(cBitVector, "or!", frb_bv_or_x, 1);
+    rb_define_method(cBitVector, "or", frb_bv_or, 1);
+    rb_define_method(cBitVector, "|", frb_bv_or, 1);
+    rb_define_method(cBitVector, "xor!", frb_bv_xor_x, 1);
+    rb_define_method(cBitVector, "xor", frb_bv_xor, 1);
+    rb_define_method(cBitVector, "^", frb_bv_xor, 1);
+    rb_define_method(cBitVector, "not!", frb_bv_not_x, 0);
+    rb_define_method(cBitVector, "not", frb_bv_not, 0);
+    rb_define_method(cBitVector, "~", frb_bv_not, 0);
+    rb_define_method(cBitVector, "reset_scan", frb_bv_reset_scan, 0);
+    rb_define_method(cBitVector, "next", frb_bv_next, 0);
+    rb_define_method(cBitVector, "next_unset", frb_bv_next_unset, 0);
+    rb_define_method(cBitVector, "next_from", frb_bv_next_from, 1);
+    rb_define_method(cBitVector, "next_unset_from", frb_bv_next_unset_from, 1);
+    rb_define_method(cBitVector, "each", frb_bv_each, 0);
+    rb_define_method(cBitVector, "to_a", frb_bv_to_a, 0);
+}
+
+/*******************
+ *** MultiMapper ***
+ *******************/
+static VALUE cMultiMapper;
+
+static void
+frb_mulmap_free(void *p)
+{
+    object_del(p);
+    frt_mulmap_destroy((FrtMultiMapper *)p);
+}
+
+static VALUE
+frb_mulmap_alloc(VALUE klass)
+{
+    FrtMultiMapper *mulmap = frt_mulmap_new();
+    VALUE rmulmap = Data_Wrap_Struct(klass, NULL, &frb_mulmap_free, mulmap);
+    object_add(mulmap, rmulmap);
+    return rmulmap;
+}
+
+/* XXX: Duplication from frb_add_mapping_i in r_analysis.c */
+static void frb_mulmap_add_mapping_i(FrtMultiMapper *mulmap, VALUE from,
+                                            const char *to)
+{
+    switch (TYPE(from)) {
+        case T_STRING:
+            frt_mulmap_add_mapping(mulmap, rs2s(from), to);
+            break;
+        case T_SYMBOL:
+            frt_mulmap_add_mapping(mulmap, rb_id2name(SYM2ID(from)), to);
+            break;
+        default:
+            rb_raise(rb_eArgError,
+                     "cannot map from %s with MappingFilter",
+                     rs2s(rb_obj_as_string(from)));
+            break;
+    }
+}
+
+/* XXX: Duplication from frb_add_mappings_i in r_analysis.c */
+static int frb_mulmap_add_mappings_i(VALUE key, VALUE value, VALUE arg)
+{
+    if (key == Qundef) {
+        return ST_CONTINUE;
+    } else {
+        FrtMultiMapper *mulmap = (FrtMultiMapper *)arg;
+        const char *to;
+        switch (TYPE(value)) {
+            case T_STRING:
+                to = rs2s(value);
+                break;
+            case T_SYMBOL:
+                to = rb_id2name(SYM2ID(value));
+                break;
+            default:
+                rb_raise(rb_eArgError,
+                         "cannot map to %s with MultiMapper",
+                         rs2s(rb_obj_as_string(key)));
+                break;
+        }
+        if (TYPE(key) == T_ARRAY) {
+            int i;
+            for (i = RARRAY_LEN(key) - 1; i >= 0; i--) {
+                frb_mulmap_add_mapping_i(mulmap, RARRAY_PTR(key)[i], to);
+            }
+        }
+        else {
+            frb_mulmap_add_mapping_i(mulmap, key, to);
+        }
+    }
+    return ST_CONTINUE;
+}
+
+/*
+ *  call-seq:
+ *     MultiMapper.new() -> new_multi_mapper
+ *
+ *  Returns a new multi-mapper object and compiles it for optimization.
+ *
+ *  Note that MultiMapper is immutable.
+ */
+static VALUE
+frb_mulmap_init(VALUE self, VALUE rmappings)
+{
+    FrtMultiMapper *mulmap = DATA_PTR(self);
+    rb_hash_foreach(rmappings, frb_mulmap_add_mappings_i, (VALUE)mulmap);
+    frt_mulmap_compile(mulmap);
+
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     multi_mapper.map(string) -> mapped_string
+ *
+ *  Performs all the mappings on the string.
+ */
+VALUE
+frb_mulmap_map(VALUE self, VALUE rstring)
+{
+    FrtMultiMapper *mulmap = DATA_PTR(self);
+    char *string = rs2s(rb_obj_as_string(rstring));
+    char *mapped_string = frt_mulmap_dynamic_map(mulmap, string);
+    VALUE rmapped_string = rb_str_new2(mapped_string);
+    free(mapped_string);
+    return rmapped_string;
+}
+
+/*
+ *  Document-class: Ferret::Utils::MultiMapper
+ *
+ *  == Summary
+ *
+ *  A MultiMapper performs a list of mappings from one string to another. You
+ *  could of course just use gsub to do this but when you are just mapping
+ *  strings, this is much faster.
+ *
+ *  Note that MultiMapper is immutable.
+ *
+ *  == Example
+ *
+ *     mapping = {
+ *       ['à','á','â','ã','ä','å','ā','ă']         => 'a',
+ *       'æ'                                       => 'ae',
+ *       ['ď','đ']                                 => 'd',
+ *       ['ç','ć','č','ĉ','ċ']                     => 'c',
+ *       ['è','é','ê','ë','ē','ę','ě','ĕ','ė',]    => 'e',
+ *       ['ƒ']                                     => 'f',
+ *       ['ĝ','ğ','ġ','ģ']                         => 'g',
+ *       ['ĥ','ħ']                                 => 'h',
+ *       ['ì','ì','í','î','ï','ī','ĩ','ĭ']         => 'i',
+ *       ['į','ı','ij','ĵ']                         => 'j',
+ *       ['ķ','ĸ']                                 => 'k',
+ *       ['ł','ľ','ĺ','ļ','ŀ']                     => 'l',
+ *       ['ñ','ń','ň','ņ','ʼn','ŋ']                 => 'n',
+ *       ['ò','ó','ô','õ','ö','ø','ō','ő','ŏ','ŏ'] => 'o',
+ *       ['œ']                                     => 'oek',
+ *       ['ą']                                     => 'q',
+ *       ['ŕ','ř','ŗ']                             => 'r',
+ *       ['ś','š','ş','ŝ','ș']                     => 's',
+ *       ['ť','ţ','ŧ','ț']                         => 't',
+ *       ['ù','ú','û','ü','ū','ů','ű','ŭ','ũ','ų'] => 'u',
+ *       ['ŵ']                                     => 'w',
+ *       ['ý','ÿ','ŷ']                             => 'y',
+ *       ['ž','ż','ź']                             => 'z'
+ *     mapper = MultiMapper.new(mapping)
+ *     mapped_string = mapper.map(string)
+ */
+static void
+Init_MultiMapper(void)
+{
+    /* MultiMapper */
+    cMultiMapper = rb_define_class_under(mUtils, "MultiMapper", rb_cObject);
+    rb_define_alloc_func(cMultiMapper, frb_mulmap_alloc);
+
+    rb_define_method(cMultiMapper, "initialize", frb_mulmap_init, 1);
+    rb_define_method(cMultiMapper, "map", frb_mulmap_map, 1);
+}
+
+/*********************
+ *** PriorityQueue ***
+ *********************/
+typedef struct PriQ
+{
+    int size;
+    int capa;
+    int mem_capa;
+    VALUE *heap;
+    VALUE proc;
+} PriQ;
+
+#define PQ_START_CAPA 32
+
+static bool frb_pq_lt(VALUE proc, VALUE v1, VALUE v2)
+{
+    if (proc == Qnil) {
+        return RTEST(rb_funcall(v1, id_lt, 1, v2));
+    }
+    else {
+        return RTEST(rb_funcall(proc, id_call, 2, v1, v2));
+    }
+}
+
+static void frb_pq_up(PriQ *pq)
+{
+    VALUE *heap = pq->heap;
+    VALUE node;
+    int i = pq->size;
+    int j = i >> 1;
+
+    node = heap[i];
+
+    while ((j > 0) && frb_pq_lt(pq->proc, node, heap[j])) {
+        heap[i] = heap[j];
+        i = j;
+        j = j >> 1;
+    }
+    heap[i] = node;
+}
+
+static void frb_pq_down(PriQ *pq)
+{
+    register int i = 1;
+    register int j = 2;         /* i << 1; */
+    register int k = 3;         /* j + 1;  */
+    register int size = pq->size;
+    VALUE *heap = pq->heap;
+    VALUE node = heap[i];       /* save top node */
+
+    if ((k <= size) && (frb_pq_lt(pq->proc, heap[k], heap[j]))) {
+        j = k;
+    }
+
+    while ((j <= size) && frb_pq_lt(pq->proc, heap[j], node)) {
+        heap[i] = heap[j];      /* shift up child */
+        i = j;
+        j = i << 1;
+        k = j + 1;
+        if ((k <= size) && frb_pq_lt(pq->proc, heap[k], heap[j])) {
+            j = k;
+        }
+    }
+    heap[i] = node;
+}
+
+static void frb_pq_push(PriQ *pq, VALUE elem)
+{
+    pq->size++;
+    if (pq->size >= pq->mem_capa) {
+        pq->mem_capa <<= 1;
+        FRT_REALLOC_N(pq->heap, VALUE, pq->mem_capa);
+    }
+    pq->heap[pq->size] = elem;
+    frb_pq_up(pq);
+}
+
+static VALUE cPriorityQueue;
+
+static void
+frb_pq_mark(void *p)
+{
+    PriQ *pq = (PriQ *)p;
+    int i;
+    for (i = pq->size; i > 0; i--) {
+        rb_gc_mark_maybe(pq->heap[i]);
+    }
+}
+
+static void frb_pq_free(PriQ *pq)
+{
+    free(pq->heap);
+    free(pq);
+}
+
+static VALUE
+frb_pq_alloc(VALUE klass)
+{
+    PriQ *pq = FRT_ALLOC_AND_ZERO(PriQ);
+    pq->capa = PQ_START_CAPA;
+    pq->mem_capa = PQ_START_CAPA;
+    pq->heap = FRT_ALLOC_N(VALUE, PQ_START_CAPA);
+    pq->proc = Qnil;
+    return Data_Wrap_Struct(klass, &frb_pq_mark, &frb_pq_free, pq);
+}
+
+#define GET_PQ(pq, self) Data_Get_Struct(self, PriQ, pq)
+/*
+ *  call-seq:
+ *     PriorityQueue.new(capacity = 32) -> new_pq
+ *     PriorityQueue.new({:capacity => 32,
+ *                        :less_than_proc => lambda{|a, b| a < b}) -> new_pq
+ *     PriorityQueue.new({:capacity => 32}) {|a, b| a < b} -> new_pq
+ *
+ *  Returns a new empty priority queue object with an optional capacity.
+ *  Once the capacity is filled, the lowest valued elements will be
+ *  automatically popped off the top of the queue as more elements are
+ *  inserted into the queue.
+ */
+static VALUE
+frb_pq_init(int argc, VALUE *argv, VALUE self)
+{
+    if (argc >= 1) {
+        PriQ *pq;
+        VALUE options = argv[0];
+        VALUE param;
+        int capa = PQ_START_CAPA;
+        GET_PQ(pq, self);
+        switch (TYPE(options)) {
+            case T_FIXNUM:
+                capa = FIX2INT(options);
+                break;
+            case T_HASH:
+                if (!NIL_P(param = rb_hash_aref(options,
+                                                ID2SYM(id_capacity)))) {
+                    capa = FIX2INT(param);
+                }
+                if (!NIL_P(param = rb_hash_aref(options,
+                                                ID2SYM(id_less_than)))) {
+                    pq->proc = param;
+                }
+                break;
+            default:
+                rb_raise(rb_eArgError,
+                         "PriorityQueue#initialize only takes a Hash or "
+                         "an integer");
+
+                break;
+        }
+        if (capa < 0) {
+            rb_raise(rb_eIndexError,
+                     "PriorityQueue must have a capacity > 0. %d < 0",
+                     capa);
+        }
+        pq->capa = capa;
+        if (rb_block_given_p()) {
+            pq->proc = rb_block_proc();
+        }
+        if (argc > 1) {
+            rb_raise(rb_eArgError,
+                     "PriorityQueue#initialize only takes one parameter");
+        }
+    }
+
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     pq.clone -> frt_pq_clone
+ *
+ *  Returns a shallow clone of the priority queue. That is only the priority
+ *  queue is cloned, its contents are not cloned.
+ */
+static VALUE
+frb_pq_clone(VALUE self)
+{
+    PriQ *pq, *new_pq = ALLOC(PriQ);
+    GET_PQ(pq, self);
+    memcpy(new_pq, pq, sizeof(PriQ));
+    new_pq->heap = FRT_ALLOC_N(VALUE, new_pq->mem_capa);
+    memcpy(new_pq->heap, pq->heap, sizeof(VALUE) * (new_pq->size + 1));
+
+    return Data_Wrap_Struct(cPriorityQueue, &frb_pq_mark, &frb_pq_free, new_pq);
+}
+
+/*
+ *  call-seq:
+ *     pq.clear -> self
+ *
+ *  Clears all elements from the priority queue. The size will be reset to 0.
+ */
+static VALUE
+frb_pq_clear(VALUE self)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    pq->size = 0;
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     pq.insert(elem) -> self
+ *     pq << elem -> self
+ *
+ *  Insert an element into a queue. It will be inserted into the correct
+ *  position in the queue according to its priority.
+ */
+static VALUE
+frb_pq_insert(VALUE self, VALUE elem)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    if (pq->size < pq->capa) {
+        frb_pq_push(pq, elem);
+    }
+    else if (pq->size > 0 && frb_pq_lt(pq->proc, pq->heap[1], elem)) {
+        pq->heap[1] = elem;
+        frb_pq_down(pq);
+    }
+    /* else ignore the element */
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     pq.adjust -> self
+ *
+ *  Sometimes you modify the top element in the priority queue so that its
+ *  priority changes. When you do this you need to reorder the queue and you
+ *  do this by calling the adjust method.
+ */
+static VALUE
+frb_pq_adjust(VALUE self)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    frb_pq_down(pq);
+    return self;
+}
+
+/*
+ *  call-seq:
+ *     pq.top -> elem
+ *
+ *  Returns the top element in the queue but does not remove it from the
+ *  queue.
+ */
+static VALUE
+frb_pq_top(VALUE self)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    return (pq->size > 0) ? pq->heap[1] : Qnil;
+}
+
+/*
+ *  call-seq:
+ *     pq.pop -> elem
+ *
+ *  Returns the top element in the queue removing it from the queue.
+ */
+static VALUE
+frb_pq_pop(VALUE self)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    if (pq->size > 0) {
+        VALUE result = pq->heap[1];       /* save first value */
+        pq->heap[1] = pq->heap[pq->size]; /* move last to first */
+        pq->heap[pq->size] = Qnil;
+        pq->size--;
+        frb_pq_down(pq);                      /* adjust heap */
+        return result;
+    }
+    else {
+        return Qnil;
+    }
+}
+
+/*
+ *  call-seq:
+ *     pq.size -> integer
+ *
+ *  Returns the size of the queue, ie. the number of elements currently stored
+ *  in the queue. The _size_ of a PriorityQueue can never be greater than
+ *  its _capacity_
+ */
+static VALUE
+frb_pq_size(VALUE self)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    return INT2FIX(pq->size);
+}
+
+/*
+ *  call-seq:
+ *     pq.capacity -> integer
+ *
+ *  Returns the capacity of the queue, ie. the number of elements that can be
+ *  stored in a Priority queue before they start to drop off the end.  The
+ *  _size_ of a PriorityQueue can never be greater than its
+ *  _capacity_
+ */
+static VALUE
+frb_pq_capa(VALUE self)
+{
+    PriQ *pq;
+    GET_PQ(pq, self);
+    return INT2FIX(pq->capa);
+}
+
+/*
+ *  Document-class: Ferret::Utils::PriorityQueue
+ *
+ *  == Summary
+ *
+ *  A PriorityQueue is a very useful data structure and one that needs a fast
+ *  implementation. Hence this priority queue is implemented in C. It is
+ *  pretty easy to use; basically you just insert elements into the queue and
+ *  pop them off.
+ *
+ *  The elements are sorted with the lowest valued elements on the top of
+ *  the heap, ie the first to be popped off. Elements are ordered using the
+ *  less_than '<' method. To change the order of the queue you can either
+ *  reimplement the '<' method pass a block when you initialize the queue.
+ *
+ *  You can also set the capacity of the PriorityQueue. Once you hit the
+ *  capacity, the lowest values elements are automatically popped of the top
+ *  of the queue as more elements are added.
+ *
+ *  == Example
+ *
+ *  Here is a toy example that sorts strings by their length and has a capacity
+ *  of 5;
+ *
+ *    q = PriorityQueue.new(5) {|a, b| a.size < b.size}
+ *    q << "x"
+ *    q << "xxxxx"
+ *    q << "xxx"
+ *    q << "xxxx"
+ *    q << "xxxxxx"
+ *    q << "xx" # hit capacity so "x" will be popped off the top
+ *
+ *    puts q.size     #=> 5
+ *    word = q.pop    #=> "xx"
+ *    q.top << "yyyy" # "xxxyyyy" will still be at the top of the queue
+ *    q.adjust        # move "xxxyyyy" to its correct location in queue
+ *    word = q.pop    #=> "xxxx"
+ *    word = q.pop    #=> "xxxxx"
+ *    word = q.pop    #=> "xxxxxx"
+ *    word = q.pop    #=> "xxxyyyy"
+ *    word = q.pop    #=> nil
+ */
+static void
+Init_PriorityQueue(void)
+{
+    /* PriorityQueue */
+    cPriorityQueue = rb_define_class_under(mUtils, "PriorityQueue", rb_cObject);
+    rb_define_alloc_func(cPriorityQueue, frb_pq_alloc);
+
+    rb_define_method(cPriorityQueue, "initialize", frb_pq_init, -1);
+    rb_define_method(cPriorityQueue, "clone", frb_pq_clone, 0);
+    rb_define_method(cPriorityQueue, "clear", frb_pq_clear, 0);
+    rb_define_method(cPriorityQueue, "insert", frb_pq_insert, 1);
+    rb_define_method(cPriorityQueue, "<<", frb_pq_insert, 1);
+    rb_define_method(cPriorityQueue, "top", frb_pq_top, 0);
+    rb_define_method(cPriorityQueue, "pop", frb_pq_pop, 0);
+    rb_define_method(cPriorityQueue, "size", frb_pq_size, 0);
+    rb_define_method(cPriorityQueue, "capacity", frb_pq_capa, 0);
+    rb_define_method(cPriorityQueue, "adjust", frb_pq_adjust, 0);
+}
+
+/* rdoc hack
+extern VALUE mFerret = rb_define_module("Ferret");
+*/
+
+/*
+ *  Document-module: Ferret::Utils
+ *
+ *  The Utils module contains a number of helper classes and modules that are
+ *  useful when indexing with Ferret. They are;
+ *
+ *  * BitVector
+ *  * MultiMapper
+ *  * PriorityQueue
+ *  * => more to come
+ *
+ *  These helper classes could also be quite useful outside of Ferret and may
+ *  one day find themselves in their own separate library.
+ */
+void
+Init_Utils(void)
+{
+    mUtils = rb_define_module_under(mFerret, "Utils");
+
+    Init_BitVector();
+    Init_MultiMapper();
+    Init_PriorityQueue();
+}