/*******************************************************************************

    copyright:      Copyright (c) 2007-2008 Jascha Wetzel. All rights reserved.

    license:        BSD style: $(LICENSE)

    version:        Initial release: Jan 2008

    authors:        Jascha Wetzel

    This is a regular expression compiler and interpreter based on the Tagged NFA/DFA method.

    The Regex class is not thread safe

    See <a href="http://en.wikipedia.org/wiki/Regular_expression">Wikpedia's article on regular expressions</a>
    for details on regular expressions in general.

    The used method implies, that the expressions are <i>regular</i>, in the way language theory defines it,
    as opposed to what &quot;regular expression&quot; means in most implementations
    (e.g. PCRE or those from the standard libraries of Perl, Java or Python).
    The advantage of this method is it's performance, it's disadvantage is the inability to realize some features
    that Perl-like regular expressions have (e.g. back-references).
    See <a href="http://swtch.com/~rsc/regexp/regexp1.html">&quot;Regular Expression Matching Can Be Simple And Fast&quot;</a>
    for details on the differences.

    The time for matching a regular expression against an input string of length N is in O(M*N), where M depends on the
    number of matching brackets and the complexity of the expression. That is, M is constant wrt. the input
    and therefore matching is a linear-time process.

    The syntax of a regular expressions is as follows.
    <i>X</i> and <i>Y</i> stand for an arbitrary regular expression.

    <table border=1 cellspacing=0 cellpadding=5>
    <caption>Operators</caption>
    $(TR $(TD X|Y) $(TD alternation, i.e. X or Y) )
    $(TR $(TD (X)) $(TD matching brackets - creates a sub-match) )
    $(TR $(TD (?X)) $(TD non-matching brackets - only groups X, no sub-match is created) )
    $(TR $(TD [Z]) $(TD character class specification, Z is a string of characters or character ranges, e.g. [a-zA-Z0-9_.\-]) )
    $(TR $(TD [^Z]) $(TD negated character class specification) )
    $(TR $(TD &lt;X) $(TD lookbehind, X may be a single character or a character class) )
    $(TR $(TD &gt;X) $(TD lookahead, X may be a single character or a character class) )
    $(TR $(TD ^) $(TD start of input or start of line) )
    $(TR $(TD $) $(TD end of input or end of line) )
    $(TR $(TD \b) $(TD start or end of word, equals (?&lt;\s&gt;\S|&lt;\S&gt;\s)) )
    $(TR $(TD \B) $(TD opposite of \b, equals (?&lt;\S&gt;\S|&lt;\s&gt;\s)) )
    </table>

    <table border=1 cellspacing=0 cellpadding=5>
    <caption>Quantifiers</caption>
    $(TR $(TD X?) $(TD zero or one) )
    $(TR $(TD X*) $(TD zero or more) )
    $(TR $(TD X+) $(TD one or more) )
    $(TR $(TD X{n,m}) $(TD at least n, at most m instances of X.<br>If n is missing, it's set to 0.<br>If m is missing, it is set to infinity.) )
    $(TR $(TD X??) $(TD non-greedy version of the above operators) )
    $(TR $(TD X*?) $(TD see above) )
    $(TR $(TD X+?) $(TD see above) )
    $(TR $(TD X{n,m}?) $(TD see above) )
    </table>

    <table border=1 cellspacing=0 cellpadding=5>
    <caption>Pre-defined character classes</caption>
    $(TR $(TD .) $(TD any printable character) )
    $(TR $(TD \s) $(TD whitespace) )
    $(TR $(TD \S) $(TD non-whitespace) )
    $(TR $(TD \w) $(TD alpha-numeric characters or underscore) )
    $(TR $(TD \W) $(TD opposite of \w) )
    $(TR $(TD \d) $(TD digits) )
    $(TR $(TD \D) $(TD non-digit) )
    </table>

    Note that "alphanumeric" only applies to Latin-1.
*******************************************************************************/
module tango.text.Regex;

static import tango.core.Array;

static import tango.text.convert.Utf;

debug(TangoRegex) import tango.io.Stdout;

/* *****************************************************************************
    A simple pair
*******************************************************************************/
private struct Pair(T)
{
    static Pair opCall(T a, T b)
    {
        Pair p;
        p.a = a;
        p.b = b;
        return p;
    }

    union
    {
        struct {
            T first, second;
        }
        struct {
            T a, b;
        }
    }
}

/* *****************************************************************************
    Double linked list
*******************************************************************************/
private class List(T)
{
    class Element
    {
        T value;
        Element prev,
                next;

        this(T v)
        {
            value = v;
        }
    }

    size_t  len;
    Element head,
            tail;

    List // opCatAssign(T v)
	    opOpAssign(string op)(T v) if (op == "~")
    {
        if ( tail is null )
            head = tail = new Element(v);
        else {
            tail.next = new Element(v);
            tail.next.prev = tail;
            tail = tail.next;
        }
        ++len;
        return this;
    }

    List insertAfter(T w, T v)
    {
        foreach ( e; &this.elements )
        {
            if ( e.value is w )
                return insertAfter(e, v);
        }
        return null;
    }

    List insertAfter(Element e, T v)
    {
        auto tmp = new Element(v);
        tmp.prev = e;
        tmp.next = e.next;
        e.next.prev = tmp;
        e.next = tmp;
        if ( e is tail )
            tail = tmp;
        ++len;
        return this;
    }

    List // opCatAssign(List l)
	    opOpAssign(string op)(List l) if (op == "~")
    {
        if ( l.empty() )
            return this;
        if ( tail is null ) {
            head = l.head;
            tail = l.tail;
        }
        else {
            tail.next = l.head;
            tail.next.prev = tail;
            tail = l.tail;
        }
        len += l.len;
        return this;
    }

    List pushFront(T v)
    {
        if ( head is null )
            head = tail = new Element(v);
        else
        {
            head.prev = new Element(v);
            head.prev.next = head;
            head = head.prev;
        }
        ++len;
        return this;
    }

    List insertBefore(T w, T v)
    {
        foreach ( e; &this.elements )
        {
            if ( e.value is w )
                return insertBefore(e, v);
        }
        return null;
    }

    List insertBefore(Element e, T v)
    {
        auto tmp = new Element(v);
        tmp.prev = e.prev;
        tmp.next = e;
        e.prev.next = tmp;
        e.prev = tmp;
        if ( e is head )
            head = tmp;
        ++len;
        return this;
    }

    List pushFront(List l)
    {
        if ( l.empty() )
            return this;
        if ( head is null ) {
            head = l.head;
            tail = l.tail;
        }
        else {
            head.prev = l.tail;
            head.prev.next = head;
            head = l.head;
        }
        len += l.len;
        return this;
    }

    size_t length()
    {
        return len;
    }

    bool empty()
    {
        return head is null;
    }

    void clear()
    {
        head = null;
        tail = null;
        len = 0;
    }

    void pop()
    {
        remove(tail);
    }

    void remove(Element e)
    {
        if ( e is null )
            return;
        if ( e.prev is null )
            head = e.next;
        else
            e.prev.next = e.next;
        if ( e.next is null )
            tail = e.prev;
        else
            e.next.prev = e.prev;
        --len;
    }

    int elements(int delegate(ref Element) dg)
    {
        for ( Element e=head; e !is null; e = e.next )
        {
            int ret = dg(e);
            if ( ret )
                return ret;
        }
        return 0;
    }

    int elements_reverse(int delegate(ref Element) dg)
    {
        for ( Element e=tail; e !is null; e = e.prev )
        {
            int ret = dg(e);
            if ( ret )
                return ret;
        }
        return 0;
    }

    int opApply(int delegate(ref T) dg)
    {
        for ( Element e=head; e !is null; e = e.next )
        {
            int ret = dg(e.value);
            if ( ret )
                return ret;
        }
        return 0;
    }

    int opApplyReverse(int delegate(ref T) dg)
    {
        for ( Element e=tail; e !is null; e = e.prev )
        {
            int ret = dg(e.value);
            if ( ret )
                return ret;
        }
        return 0;
    }
}

/* *****************************************************************************
    Stack based on dynamic array
*******************************************************************************/
private struct Stack(T)
{
    size_t  _top;
    T[]     stack;

    void push(T v)
    {
        if ( _top >= stack.length )
            stack.length = stack.length*2+1;
        stack[_top] = v;
        ++_top;
    }
    // alias push opCatAssign;

    void // opCatAssign(T[] vs)
	    opOpAssign(string op)(T[] vs) if (op == "~")
    {
        size_t end = _top+vs.length;
        if ( end > stack.length )
            stack.length = end*2;
        stack[_top..end] = vs[];
        _top = end;
    }

    void pop(size_t num)
    {
        assert(_top>=num);
        _top -= num;
    }

    T pop()
    {
        assert(_top>0);
        return stack[--_top];
    }

    T top()
    {
        assert(_top>0);
        return stack[_top-1];
    }

    T* topPtr()
    {
        assert(_top>0);
        return &stack[_top-1];
    }

    bool empty()
    {
        return _top == 0;
    }

    void clear()
    {
        _top = 0;
    }

    size_t length()
    {
        return _top;
    }

    T[] array()
    {
        return stack[0.._top];
    }

    T opIndex(size_t i)
    {
        return stack[i];
    }

    Stack dup()
    {
        Stack s;
        s._top = _top;
        s.stack = stack.dup;
        return s;
    }
}

/* ************************************************************************************************
    Set container based on assoc array
**************************************************************************************************/
private struct Set(T)
{
    bool[T] data;

    static Set opCall()
    {
        Set s;
        return s;
    }

    static Set opCall(T v)
    {
        Set s;
        s ~= v;
        return s;
    }

    void opAddAssign(T v)
    {
        data[v] = true;
    }

    void opAddAssign(Set s)
    {
        foreach ( v; s.elements )
            data[v] = true;
    }
    alias opAddAssign opCatAssign;

    size_t length()
    {
        return data.length;
    }

    T[] elements()
    {
        return data.keys;
    }

    bool remove(T v)
    {
        if ( (v in data) is null )
            return false;
        data.remove(v);
        return true;
    }

    bool contains(T v)
    {
        return (v in data) !is null;
    }

    bool contains(Set s)
    {
        Set tmp = s - this;
        return tmp.empty;
    }

    bool empty()
    {
        return data.length==0;
    }

    Set opSub(Set s)
    {
        Set res = dup;
        foreach ( v; s.elements )
            res.remove(v);
        return res;
    }

    Set dup()
    {
        Set s;
        foreach ( v; data.keys )
            s.data[v] = true;
        return s;
    }
}

/* ************************************************************************************************

**************************************************************************************************/
void quickSort(T)(T[] a)
{
    quickSort(a,cast(size_t)0,a.length);
}

void quickSort(T)(T[] a, size_t l, size_t r)
{
    T t;
    auto i = r-l;
    if ( i < 3 )
    {
        if ( i < 2 )
            return;
        if ( a[l] < a[l+1] )
            return;
        t = a[l];
        a[l] = a[l+1];
        a[l+1] = t;
        return;
    }

    auto p = a[l];
    i = l;
    auto j = r;

    while ( true )
    {
        ++i;
        for ( ; i < j && a[i] < p; ++i ) {}
        --j;
        for ( ; i < j && a[j] >= p; --j ) {}
        if ( i >= j )
            break;
        t = a[i];
        a[i] = a[j];
        a[j] = t;
    }
    --i;
    a[l] = a[i];
    a[i] = p;

    quickSort(a, l, i);
    quickSort(a, i+1, r);
}
import tango.math.Math;

/* ************************************************************************************************
    A range of characters
**************************************************************************************************/
struct CharRange(char_t)
{
    char_t  l_, r_;

    static CharRange opCall(char_t c)
    {
        CharRange cr;
        cr.l_ = c;
        cr.r_ = c;
        return cr;
    }

    static CharRange opCall(char_t a, char_t b)
    {
        CharRange cr;
        cr.l_ = min(a,b);
        cr.r_ = max(a,b);
        return cr;
    }

    char_t l()
    {
        return l_;
    }

    char_t r()
    {
        return r_;
    }

    /* ********************************************************************************************
        Compares the ranges according to their beginning.
    **********************************************************************************************/
    int opCmp(const ref CharRange cr) const
    {
        if ( l_ == cr.l_ )
            return 0;
        if ( l_ < cr.l_ )
            return -1;
        return 1;
    }


    const bool opEquals(ref const CharRange cr)
    {
        if ( l_ == cr.l_ && r_ == cr.r_ )
            return true;
        return false;
    }

    bool contains(char_t c)
    {
        return c >= l_ && c <= r_;
    }

    bool contains(CharRange cr)
    {
        return l_ <= cr.l_ && r_ >= cr.r_;
    }

    bool intersects(CharRange cr)
    {
        return r_ >= cr.l_ && l_ <= cr.r_;
    }

    CharRange intersect(CharRange cr)
    {
        assert(intersects(cr));
        CharRange ir;
        ir.l_ = max(l_, cr.l_);
        ir.r_ = min(r_, cr.r_);
        if ( ir.l_ > ir.r_ )
            ir.l_ = ir.r_ = char_t.min;
        return ir;
    }

    CharRange[] subtract(CharRange cr)
    {
        CharRange[] sr;
        if ( cr.contains(this) )
            return sr;
        if ( !intersects(cr) )
            sr ~= this;
        else
        {
            CharRange d;
            if ( contains(cr) )
            {
                d.l_ = l_;
                d.r_ = cast(char_t)(cr.l_-1);
                if ( d.l_ <= d.r_ )
                    sr ~= d;
                d.l_ = cast(char_t)(cr.r_+1);
                d.r_ = r_;
                if ( d.l_ <= d.r_ )
                    sr ~= d;
            }
            else if ( cr.r_ > l_ )
            {
                d.l_ = cast(char_t)(cr.r_+1);
                d.r_ = r_;
                if ( d.l_ <= d.r_ )
                    sr ~= d;
            }
            else if ( cr.l_ < r_ )
            {
                d.l_ = l_;
                d.r_ = cast(char_t)(cr.l_-1);
                if ( d.l_ <= d.r_ )
                    sr ~= d;
            }
        }
        return sr;
    }

    string toString()
    {
        string str;

        if ( l_ == r_ )
        {
            if ( l_ > 0x20 && l_ < 0x7f )
                str = Format.convert("'{}'", l_).idup;
            else
                str = Format.convert("({:x})", cast(int)l_).idup;
        }
        else
        {
            if ( l_ > 0x20 && l_ < 0x7f )
                str = Format.convert("'{}'", l_).idup;
            else
                str = Format.convert("({:x})", cast(int)l_).idup;
            str ~= "-";
            if ( r_ > 0x20 && r_ < 0x7f )
                str ~= Format.convert("'{}'", r_).idup;
            else
                str ~= Format.convert("({:x})", cast(int)r_).idup;
        }
        return str;
    }
}

import tango.core.Compiler;

static if(DMDFE_Version == 2063)
{
    /* Workabout for http://d.puremagic.com/issues/show_bug.cgi?id=10425 */
    void _bug(CharRange!(dchar) a) {}
}

/* ************************************************************************************************
    Represents a class of characters as used in regular expressions (e.g. [0-9a-z], etc.)
**************************************************************************************************/
struct CharClass(char_t)
{
    alias CharRange!(char_t) range_t;

    //---------------------------------------------------------------------------------------------
    // pre-defined character classes
    static const CharClass!(char_t)
        line_startend = {parts: [
            {l_:0x00, r_:0x00},
            {l_:0x0a, r_:0x0a},
            {l_:0x13, r_:0x13}
        ]},
        digit = {parts: [
            {l_:0x30, r_:0x39}
        ]},
        whitespace = {parts: [
            {l_:0x09, r_:0x09},
            {l_:0x0a, r_:0x0a},
            {l_:0x0b, r_:0x0b},
            {l_:0x13, r_:0x13},
            {l_:0x14, r_:0x14},
            {l_:0x20, r_:0x20}
        ]};

    // 8bit classes
    static if ( is(char_t == char) )
    {
        static const CharClass!(char_t)
            any_char = {parts: [
                {l_:0x01, r_:0xff}
            ]},
            dot_oper = {parts: [
                {l_:0x01, r_:0xff}
            ]},
            alphanum_ = {parts: [
                {l_:0x30, r_:0x39},
                {l_:0x41, r_:0x5a},
                {l_:0x5f, r_:0x5f},
                {l_:0x61, r_:0x7a}
            ]};
    }
    // 16bit and 32bit classes
    static if ( is(char_t == wchar) || is(char_t == dchar) )
    {
        static const CharClass!(char_t)
            any_char = {parts: [
                {l_:0x0001, r_:0xffff}
            ]},
            dot_oper = {parts: [
                {l_:0x0001, r_:0xffff}
            ]},
            alphanum_ = {parts: [
                {l_:0x30, r_:0x39},
                {l_:0x41, r_:0x5a},
                {l_:0x5f, r_:0x5f},
                {l_:0x61, r_:0x7a}
            ]};
    }

    //---------------------------------------------------------------------------------------------
    range_t[] parts;

    invariant()
    {
//        foreach ( i, p; parts )
//            assert(p.l_<=p.r_, Int.toString(i)~": "~Int.toString(p.l_)~" > "~Int.toString(p.r_));
    }

    static CharClass opCall(const(CharClass) cc)
    {
        CharClass ncc;
        ncc.parts = cc.parts.dup;
        return ncc;
    }

    int opCmp(const ref CharClass cc) const
    {
        if ( parts.length < cc.parts.length )
            return -1;
        if ( parts.length > cc.parts.length )
            return 1;
        foreach ( i, p; cc.parts )
        {
            if ( p.l_ != parts[i].l_ || p.r_ != parts[i].r_ )
                return 1;
        }
        return 0;
    }

    bool empty()
    {
        return parts.length <= 0;
    }

    bool matches(char_t c)
    {
        foreach ( p; parts )
        {
            if ( p.contains(c) )
                return true;
        }
        return false;
    }

    CharClass intersect(CharClass cc)
    {
        CharClass ic;
        foreach ( p; parts )
        {
            foreach ( cp; cc.parts )
            {
                if ( p.intersects(cp) )
                    ic.parts ~= p.intersect(cp);
            }
        }
        return ic;
    }

    // requires the class to be optimized
    bool contains(range_t cr)
    {
        foreach ( p; parts )
        {
            if ( p.contains(cr) )
                return true;
        }
        return false;
    }

    // requires the class to be optimized
    bool contains(CharClass cc)
    {
        Louter: foreach ( p; cc.parts )
        {
            foreach ( p2; parts )
            {
                if ( p2.contains(p) )
                    continue Louter;
            }
            return false;
        }
        return true;
    }

    void subtract(CharClass cc)
    {
        negate();
        add(cc);
        negate();
    }

    void add(CharClass cc)
    {
        parts ~= cc.parts;
    }

    void add(range_t cr)
    {
        parts ~= cr;
    }

    void add(char_t c)
    {
        parts ~= CharRange!(char_t)(c);
    }

    /* ********************************************************************************************
        Requires the CharClass to be optimized.
    **********************************************************************************************/
    void negate()
    {
        optimize();
        char_t  start = char_t.min;

        // first part touches left boundary of value range
        if ( parts.length > 0 && parts[0].l_ == start )
        {
            start = parts[0].r_;
            if ( start < char_t.max )
                ++start;

            foreach ( i, ref cr; parts[0 .. $-1] )
            {
                cr.l_ = start;
                cr.r_ = cast(char_t)(parts[i+1].l_-1);
                start = parts[i+1].r_;
                if ( start < char_t.max )
                    ++start;
            }
            if ( start != char_t.max ) {
                parts[$-1].l_ = start;
                parts[$-1].r_ = char_t.max;
            }
            else
                parts.length = parts.length-1;
            return;
        }

        foreach ( i, ref cr; parts )
        {
            char_t tmp = cast(char_t)(cr.l_-1);
            cr.l_ = start;
            start = cr.r_;
            if ( start < char_t.max )
                ++start;
            cr.r_ = tmp;
        }

        // last part does not touch right boundary
        if ( start != char_t.max )
            parts ~= range_t(start, char_t.max);
    }

    void optimize()
    {
        if ( empty() )
            return;

        tango.core.Array.sort(parts);

        size_t i = 0;
        foreach ( p; parts[1 .. $] )
        {
            if ( p.l_ > parts[i].r_+1 ) {
                ++i;
                parts[i].l_ = p.l_;
                parts[i].r_ = p.r_;
                continue;
            }
            parts[i].r_ = max(p.r_, parts[i].r_);
            if ( parts[i].r_ >= char_t.max )
                break;
        }
        parts.length = i+1;
    }

    string toString()
    {
        string str;
        str ~= "[";
        foreach ( p; parts )
            str ~= p.toString();
        str ~= "]";
        return str;
    }
}

debug(UnitTest)
{
unittest
{
    static CharClass!(char) cc = { parts: [{l_:0,r_:10},{l_:0,r_:6},{l_:5,r_:12},{l_:12,r_:17},{l_:20,r_:100}] };
    assert(cc.toString(), "[(0)-(a)(0)-(6)(5)-(c)(c)-(11)(14)-'d']");
    cc.optimize();
    assert(cc.toString(),  "[(0)-(11)(14)-'d']");
    cc.negate();
    assert(cc.toString(),  " [(12)-(13)'e'-(ff)]");
    cc.optimize();
    assert(cc.toString(),  "[(0)-(11)(14)-'d']");
    cc.negate();
    assert(cc.toString(),  "[(12)-(13)'e'-(ff)]");

    static CharClass!(char) cc2 = { parts: [] };
    assert(cc.toString(),  "[]");
    cc2.optimize();
    assert(cc.toString(),  "[]");
    cc2.negate();
    assert(cc.toString(),  "[(0)-(ff)]");
    cc2.optimize();
    assert(cc.toString(),  "[(0)-(ff)]");
    cc2.negate();
    assert(cc.toString(),  "[]");

    static CharClass!(char) cc3 = { parts: [{l_:0,r_:100},{l_:200,r_:0xff},] };
    assert(cc3.toString(), "[(0)-'d'(c8)-(ff)]");
    cc3.negate();
    assert(cc.toString(),  "['e'-(c7)]");
    cc3.negate();
    assert(cc.toString(),  "[(0)-'d'(c8)-(ff)]");

    static CharClass!(char) cc4 = { parts: [{l_:0,r_:200},{l_:100,r_:0xff},] };
    assert(cc.toString(),  "[(0)-(c8)'d'-(ff)]");
    cc4.optimize();
    assert(cc.toString(),  "[(9)-(13)(20)-'~'(a0)-(ff)(100)-(17f)(180)-(24f)(20a3)-(20b5)]");

    static CharClass!(dchar) cc5 = { parts: [{l_:0x9,r_:0x13},{0x20,r_:'~'},{l_:0xa0,r_:0xff},{l_:0x100,r_:0x17f},{l_:0x180,r_:0x24f},{l_:0x20a3,r_:0x20b5}] };
    cc5.optimize();
    assert(cc.toString(),  "[(9)-(13)(20)-'~'(a0)-(24f)(20a3)-(20b5)]");
    cc5.negate();
    assert(cc.toString(),  "[(0)-(8)(14)-(1f)(7f)-(9f)(250)-(20a2)(20b6)-(10ffff)]");
    cc5.optimize();
    assert(cc.toString(),  "[(0)-(8)(14)-(1f)(7f)-(9f)(250)-(20a2)(20b6)-(10ffff)]");
    cc5.negate();
    assert(cc.toString(),  "[(9)-(13)(20)-'~'(a0)-(24f)(20a3)-(20b5)]");
}
}

/* ************************************************************************************************

**************************************************************************************************/
private struct Predicate(char_t)
{
    alias const(char_t)[]       string_t;
    alias CharClass!(char_t)    cc_t;
    alias CharRange!(char_t)    cr_t;

    // generic data
    enum Type {
        consume, epsilon, lookahead, lookbehind
    }

    cc_t    input;
    Type    type;

    // data for compiled predicates
    enum uint  MAX_BITMAP_LENGTH = 256,
               MAX_SEARCH_LENGTH = 256;
    enum MatchMode {
        generic, generic_l,
        single_char, bitmap, string_search,         // consume
        single_char_l, bitmap_l, string_search_l    // lookahead
    }

    MatchMode   mode;
    // data_chr had to be pulled out of the union due to
    // http://d.puremagic.com/issues/show_bug.cgi?id=2632 --- don't put it back
    // in until this is resolved!
    //
    // Keep in mind that data_str.length can't be modified directly unless the
    // new length is strictly greater than the old length. This is essentially
    // because ubyte.sizeof can be less than char_t.sizeof. If you set
    // data_str.length to anything less than or equal to data_bmp.length,
    // data_str will not be reallocated: only the length value will change but
    // nothing will realize that not that much space has actually been
    // allocated. data_str will be too small and you'll likely get segfaults or
    // such.
    //
    // In short: don't mess with data_str.length. If you have to, remove the
    // union entirely.
    //
    // -- Deewiant
    union {
        ubyte[]     data_bmp;
        string_t    data_str;
    };
    char_t      data_chr;


    void compile()
    {
        assert(input.parts.length > 0);

        // single char?
        if ( input.parts.length == 1 && input.parts[0].l_ == input.parts[0].r_ )
        {
            mode = type==Type.consume ? MatchMode.single_char : MatchMode.single_char_l;
            data_chr = input.parts[0].l_;
            return;
        }
        // check whether we can use a bitmap
        foreach ( p; input.parts )
        {
            if ( p.l_ > MAX_BITMAP_LENGTH || p.r_ > MAX_BITMAP_LENGTH )
                goto LnoBitmap;
        }

        // setup bitmap
        data_bmp.length = MAX_BITMAP_LENGTH/8;
        foreach ( p; input.parts )
        {
            for ( char_t c = p.l_; c <= p.r_; ++c )
                data_bmp[c/8] |= 1 << (c&7);
        }
        mode = type==Type.consume ? MatchMode.bitmap : MatchMode.bitmap_l;
        return;

    LnoBitmap:
/*
        // check whether the class is small enough to justify a string-search
        // TODO: consider inverse class for 8bit chars?
        uint class_size;
        foreach ( p; input.parts )
            class_size += cast(uint)p.r_+1-p.l_;
        if ( class_size > MAX_SEARCH_LENGTH )
            goto Lgeneric;
        data_str.length = class_size;
        size_t ind;
        foreach ( p; input.parts )
        {
            for ( char_t c = p.l_; c <= p.r_; ++c )
                data_str[ind++] = c;
        }
        mode = type==Type.consume ? MatchMode.string_search : MatchMode.string_search_l;
        return;
*/
    Lgeneric:
        data_str = cast(char_t[])input.parts;
        mode = type==Type.consume ? MatchMode.generic : MatchMode.generic_l;
    }

    bool matches(char_t c)
    {
        if ( type == Type.consume || type == Type.lookahead )
            return input.matches(c);
        assert(0);
    }

    Predicate intersect(Predicate p)
    {
        Predicate p2;
        if ( type != Type.epsilon && p.type != Type.epsilon )
            p2.input = input.intersect(p.input);
        return p2;
    }

    bool intersects(Predicate p)
    {
        if ( type != p.type )
            return false;
        foreach ( cr; input.parts )
        {
            foreach ( cr2; p.input.parts )
            {
                if ( cr.intersects(cr2) )
                    return true;
            }
        }
        return false;
    }

    bool exceedsMax(uint maxc)
    {
        foreach ( p; input.parts )
        {
            if ( p.l_ > maxc || p.r_ > maxc )
                return true;
        }

        return false;
    }

    @property bool empty()
    {
        return type != Type.epsilon && input.empty();
    }

    void subtract(Predicate p)
    {
        if ( type != Type.epsilon && p.type != Type.epsilon )
            input.subtract(p.input);
    }

    void negate()
    {
        assert(type != Type.epsilon);
        input.negate();
    }

    void optimize()
    {
        assert(type != Type.epsilon);
        input.optimize();
    }

    int opCmp(const ref Predicate p) const
    {
        return input.opCmp(p.input);
    }

    const bool opEquals(ref const Predicate p)
    {
        if ( type != p.type )
            return false;
        if ( input.opCmp(p.input) != 0 )
            return false;
        return true;
    }

    cc_t getInput()
    {
        return input;
    }

    void setInput(const(cc_t) cc)
    {
        input = cast(cc_t)cc;
    }

    void appendInput(cr_t cr)
    {
        input.add(cr);
    }

    void appendInput(cc_t cc)
    {
        input.add(cc);
    }

    void appendInput(Predicate p)
    {
        input.add(p.input);
    }

    string toString()
    {
        string str;
        switch ( type )
        {
            case Type.consume:      str = input.toString();       break;
            case Type.epsilon:      str = "eps";                break;
            case Type.lookahead:    str = "la:"~input.toString(); break;
            case Type.lookbehind:   str = "lb:"~input.toString(); break;
            default:
                assert(0);
        }
        return str;
    }
}
import Utf = tango.text.convert.Utf;
import tango.text.convert.Format;

/* ************************************************************************************************

**************************************************************************************************/
class RegExpException : Exception
{
    this(string msg)
    {
        super("RegExp: "~msg);
    }
}

/* ************************************************************************************************
    TNFA state
**************************************************************************************************/
private class TNFAState(char_t)
{
    bool    accept = false,
            visited = false;
    size_t  index;
    List!(TNFATransition!(char_t))  transitions;

    this()
    {
        transitions = new List!(TNFATransition!(char_t));
    }
}


/* ************************************************************************************************
    Priority classes used to linearize priorities after non-linear transition creation.
**************************************************************************************************/
private enum PriorityClass {
    greedy=0, normal=1, reluctant=2, extraReluctant=3
}

/* ********************************************************************************
    TNFA tagged transition
***********************************************************************************/
private class TNFATransition(char_t)
{
    TNFAState!(char_t)  target;
    Predicate!(char_t)  predicate;
    uint                priority,
                        tag;        /// one-based tag number, 0 = untagged
    PriorityClass       priorityClass;

    this(PriorityClass pc)
    {
        priorityClass = pc;
    }

    /******************************************************************************
        Move through states only going via epsilon transitions, and only choosing
        the one with highest priority. If the highest priority transition from a
        state isn't an epsilon transition, false is returned.
        If the accepting NFA state can be reached in this manner, true is returned.

        NOTE: This method does not look for cycles which should be kept in mind for
        later. larsivi 20090827
    *******************************************************************************/
    bool canFinish()
    {
        TNFAState!(char_t)  t = target;
        while (!t.accept) {
            TNFATransition!(char_t) highestPriTrans;
            foreach (trans; t.transitions) {
                if (!highestPriTrans || highestPriTrans.priority > trans.priority)
                    highestPriTrans = trans;
            }
            if (!(highestPriTrans.predicate.type == Predicate!(char_t).Type.epsilon))
                return false;

            t = highestPriTrans.target;
        }
        return true;
    }

}

/* ************************************************************************************************
    Fragments of TNFAs as used in the Thompson method
**************************************************************************************************/
private class TNFAFragment(char_t)
{
    alias TNFAState!(char_t)        state_t;
    alias TNFATransition!(char_t)   trans_t;

    List!(trans_t)  entries,        /// transitions to be added to the entry state
                    exits,          /// transitions to be added to the exit state
                    entry_state,    /// transitions to write the entry state to
                    exit_state;     /// transitions to write the exit state to

    bool swapMatchingBracketSyntax;

    this()
    {
        entries     = new List!(trans_t);
        exits       = new List!(trans_t);
        entry_state = new List!(trans_t);
        exit_state  = new List!(trans_t);
    }

    /* ********************************************************************************************
        Write the given state as entry state to this fragment.
    **********************************************************************************************/
    void setEntry(state_t state)
    {
        state.transitions ~= entries;
        foreach ( t; entry_state )
            t.target = state;
    }

    /* ********************************************************************************************
        Write the given state as exit state to this fragment.
    **********************************************************************************************/
    void setExit(state_t state)
    {
        state.transitions ~= exits;
        foreach ( t; exit_state )
            t.target = state;
    }
}

/* ************************************************************************************************
    Tagged NFA
**************************************************************************************************/
private final class TNFA(char_t)
{
    alias TNFATransition!(char_t)   trans_t;
    alias TNFAFragment!(char_t)     frag_t;
    alias TNFAState!(char_t)        state_t;
    alias Predicate!(char_t)        predicate_t;
    alias char_t[]                  string_t;
    alias CharRange!(char_t)        range_t;
    alias CharClass!(char_t)        cc_t;

    string_t    pattern;
    state_t[]   states;
    state_t     start;

    bool swapMatchingBracketSyntax; /// whether to make (?...) matching and (...) non-matching

    /* ********************************************************************************************
        Creates the TNFA from the given regex pattern
    **********************************************************************************************/
    this(string_t regex)
    {
        next_tag        = 1;
        transitions     = new List!(trans_t);

        pattern = regex;
    }

    /* ********************************************************************************************
        Print the TNFA (tabular representation of the delta function)
    **********************************************************************************************/
    debug(TangoRegex) void print()
    {
        foreach ( int i, s; states )
        {
            Stdout.format("{}{:d2}{}", s is start?">":" ", i, s.accept?"*":" ");

            bool first=true;
            Stdout(" {");
            foreach ( t; s.transitions )
            {
                Stdout.format("{}{}{}:{}->{}", first?"":", ", t.priority, "gnrx"[t.priorityClass], t.predicate.toString, t.target is null?-1:t.target.index);
                if ( t.tag > 0 ) {
                    Stdout.format(" t{}", t.tag);
                }
                first = false;
            }
            Stdout("}").newline;
        }
    }

    uint tagCount()
    {
        return next_tag-1;
    }

    /* ********************************************************************************************
        Constructs the TNFA using extended Thompson method.
        Uses a slightly extended version of Dijkstra's shunting yard algorithm to convert
        the regexp from infix notation.
    **********************************************************************************************/
    void parse(bool unanchored)
    {
        List!(frag_t)       frags       = new List!(frag_t);
        Stack!(Operator)    opStack;
        Stack!(uint)        tagStack;
        Stack!(Pair!(uint)) occurStack;
        opStack.push(Operator.eos);

        /* ****************************************************************************************
            Perform action on operator stack
        ******************************************************************************************/
        bool perform(Operator next_op, bool explicit_operator=true)
        {
            // calculate index in action matrix
            int index = cast(int)opStack.top()*(Operator.max+1);
            index += cast(int)next_op;

            debug(tnfa) Stdout.formatln("\t{}:{} -> {}  {} frag(s)",
                operator_names[opStack.top], operator_names[next_op], action_names[action_lookup[index]], frags.length
            );
            switch ( action_lookup[index] )
            {
                case Act.pua:
                    opStack.push(next_op);
                    if ( next_op == Operator.open_par ) {
                        tagStack.push(next_tag);
                        next_tag += 2;
                    }
                    break;
                case Act.poc:
                    switch ( opStack.top() )
                    {
                        case Operator.concat:       constructConcat(frags);                             break;
                        case Operator.altern:       constructAltern(frags);                             break;
                        case Operator.zero_one_g:   constructZeroOne(frags, PriorityClass.greedy);      break;
                        case Operator.zero_one_ng:  constructZeroOne(frags, PriorityClass.reluctant);   break;
                        case Operator.zero_one_xr:  constructZeroOne(frags, PriorityClass.extraReluctant);  break;
                        case Operator.zero_more_g:  constructZeroMore(frags, PriorityClass.greedy);     break;
                        case Operator.zero_more_ng: constructZeroMore(frags, PriorityClass.reluctant);  break;
                        case Operator.zero_more_xr: constructZeroMore(frags, PriorityClass.extraReluctant); break;
                        case Operator.one_more_g:   constructOneMore(frags, PriorityClass.greedy);      break;
                        case Operator.one_more_ng:  constructOneMore(frags, PriorityClass.reluctant);   break;
                        case Operator.one_more_xr:  constructOneMore(frags, PriorityClass.extraReluctant);  break;
                        case Operator.occur_g:
                            Pair!(uint) occur = occurStack.pop();
                            constructOccur(frags, occur.a, occur.b, PriorityClass.greedy);
                            break;
                        case Operator.occur_ng:
                            Pair!(uint) occur = occurStack.pop();
                            constructOccur(frags, occur.a, occur.b, PriorityClass.reluctant);
                            break;
                        default:
                            throw new RegExpException("cannot process operand at \""~Utf.toString(pattern[cursor..$]).idup~"\"");
                    }
                    opStack.pop();

                    perform(next_op, false);
                    break;
                case Act.poa:
                    opStack.pop();
                    break;
                case Act.pca:
                    if ( opStack.top() == Operator.open_par )
                    {
                        if ( tagStack.empty() )
                            throw new RegExpException(Format.convert("Missing opening parentheses for closing parentheses at char {} \"{}\"", cursor, Utf.toString(pattern[cursor..$])).idup);
                        constructBracket(frags, tagStack.top());
                        tagStack.pop();
                    }
                    else {
                        assert(opStack.top() == Operator.open_par_nm);
                        constructBracket(frags);
                    }
                    opStack.pop();
                    break;
                case Act.don:
                    return true;
                case Act.err:
                default:
                    throw new RegExpException(Format.convert("Unexpected operand at char {} \"{}\" in \"{}\"", cursor, Utf.toString(pattern[cursor..$]), Utf.toString(pattern)).idup);
            }

            return false;
        }

        // add implicit extra reluctant .* (with . == any_char) at the beginning for unanchored matches
        // and matching bracket for total match group
        if ( unanchored ) {
            frags ~= constructChars(cc_t.any_char, predicate_t.Type.consume);
            perform(Operator.zero_more_xr, false);
            perform(Operator.concat, false);
            perform(Operator.open_par, false);
        }

        // convert regex to postfix and create TNFA
        bool implicit_concat;
        predicate_t.Type pred_type;

        while ( !endOfPattern() )
        {
            pred_type = predicate_t.Type.consume;

            dchar c = readPattern();
            switch ( c )
            {
                case '|':
                    perform(Operator.altern);
                    implicit_concat = false;
                    break;
                case '(':
                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = false;
                    if ( peekPattern() == '?' ) {
                        readPattern();
                        perform(swapMatchingBracketSyntax?Operator.open_par:Operator.open_par_nm);
                    }
                    else
                        perform(swapMatchingBracketSyntax?Operator.open_par_nm:Operator.open_par);
                    break;
                case ')':
                    perform(Operator.close_par);
                    break;
                case '?':
                    if ( peekPattern() == '?' ) {
                        readPattern();
                        perform(Operator.zero_one_ng);
                    }
                    else
                        perform(Operator.zero_one_g);
                    break;
                case '*':
                    if ( peekPattern() == '?' ) {
                        readPattern();
                        perform(Operator.zero_more_ng);
                    }
                    else
                        perform(Operator.zero_more_g);
                    break;
                case '+':
                    if ( peekPattern() == '?' ) {
                        readPattern();
                        perform(Operator.one_more_ng);
                    }
                    else
                        perform(Operator.one_more_g);
                    break;
                case '{':
                    Pair!(uint) occur;
                    parseOccurCount(occur.a, occur.b);
                    occurStack.push(occur);
                    if ( peekPattern() == '?' ) {
                        readPattern();
                        perform(Operator.occur_ng);
                    }
                    else
                        perform(Operator.occur_g);
                    break;
                case '[':
                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = true;
                    frags ~= constructCharClass(pred_type);
                    break;
                case '.':
                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = true;
                    frags ~= constructChars(cc_t.dot_oper, pred_type);
                    break;
                case '$':
                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = true;

                    frags ~= constructChars(cc_t.line_startend, predicate_t.Type.lookahead);
                    break;
                case '^':
                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = true;

                    frags ~= constructChars(cc_t.line_startend, predicate_t.Type.lookbehind);
                    break;
                case '>':
                    c = readPattern();
                    pred_type = predicate_t.Type.lookahead;
                    if ( c == '[' )
                        goto case '[';
                    else if ( c == '\\' )
                        goto case '\\';
                    else if ( c == '.' )
                        goto case '.';
                    else
                        goto default;
                case '<':
                    c = readPattern();
                    pred_type = predicate_t.Type.lookbehind;
                    if ( c == '[' )
                        goto case '[';
                    else if ( c == '\\' )
                        goto case '\\';
                    else if ( c == '.' )
                        goto case '.';
                    else
                        goto default;
                case '\\':
                    c = readPattern();

                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = true;

                    switch ( c )
                    {
                        case 't':
                            frags ~= constructSingleChar('\t', pred_type);
                            break;
                        case 'n':
                            frags ~= constructSingleChar('\n', pred_type);
                            break;
                        case 'r':
                            frags ~= constructSingleChar('\r', pred_type);
                            break;
                        case 'w':   // alphanumeric and _
                            frags ~= constructChars(cc_t.alphanum_, pred_type);
                            break;
                        case 'W':   // non-(alphanum and _)
                            auto cc = cc_t(cc_t.alphanum_);
                            cc.negate();
                            frags ~= constructChars(cc, pred_type);
                            break;
                        case 's':   // whitespace
                            frags ~= constructChars(cc_t.whitespace, pred_type);
                            break;
                        case 'S':   // non-whitespace
                            auto cc = cc_t(cc_t.whitespace);
                            cc.negate();
                            frags ~= constructChars(cc, pred_type);
                            break;
                        case 'd':   // digit
                            frags ~= constructChars(cc_t.digit, pred_type);
                            break;
                        case 'D':   // non-digit
                            auto cc = cc_t(cc_t.digit);
                            cc.negate();
                            frags ~= constructChars(cc, pred_type);
                            break;
                        case 'b':   // either end of word
                            if ( pred_type != predicate_t.Type.consume )
                                throw new RegExpException("Escape sequence \\b not allowed in look-ahead or -behind");

                            // create (?<\S>\s|<\s>\S)
                            auto cc = cc_t(cc_t.whitespace);
                            cc.negate();

                            perform(Operator.open_par_nm);

                            frags ~= constructChars(cc, predicate_t.Type.lookbehind);
                            perform(Operator.concat, false);
                            frags ~= constructChars(cc_t.whitespace, predicate_t.Type.lookahead);
                            perform(Operator.altern, false);
                            frags ~= constructChars(cc_t.whitespace, predicate_t.Type.lookbehind);
                            perform(Operator.concat, false);
                            frags ~= constructChars(cc, predicate_t.Type.lookahead);

                            perform(Operator.close_par, false);
                            break;
                        case 'B':   // neither end of word
                            if ( pred_type != predicate_t.Type.consume )
                                throw new RegExpException("Escape sequence \\B not allowed in look-ahead or -behind");

                            // create (?<\S>\S|<\s>\s)
                            auto cc = cc_t(cc_t.whitespace);
                            cc.negate();

                            perform(Operator.open_par_nm);

                            frags ~= constructChars(cc, predicate_t.Type.lookbehind);
                            perform(Operator.concat, false);
                            frags ~= constructChars(cc, predicate_t.Type.lookahead);
                            perform(Operator.altern, false);
                            frags ~= constructChars(cc_t.whitespace, predicate_t.Type.lookbehind);
                            perform(Operator.concat, false);
                            frags ~= constructChars(cc_t.whitespace, predicate_t.Type.lookahead);

                            perform(Operator.close_par, false);
                            break;
                        case '(':
                        case ')':
                        case '[':
                        case ']':
                        case '{':
                        case '}':
                        case '*':
                        case '+':
                        case '?':
                        case '.':
                        case '\\':
                        case '^':
                        case '$':
                        case '|':
                        case '<':
                        case '>':
                        default:
                            frags ~= constructSingleChar(c, pred_type);
                            break;
//                            throw new RegExpException(Format.convert("Unknown escape sequence \\{}", c));
                    }
                    break;

                default:
                    if ( implicit_concat )
                        perform(Operator.concat, false);
                    implicit_concat = true;
                    frags ~= constructSingleChar(c, pred_type);
            }
        }

        // add implicit reluctant .* (with . == any_char) at the end for unanchored matches
        if ( unanchored )
        {
            perform(Operator.close_par, false);
            if ( implicit_concat )
                perform(Operator.concat, false);
            frags ~= constructChars(cc_t.any_char, predicate_t.Type.consume);
            perform(Operator.zero_more_ng, false);
        }

        // empty operator stack
        while ( !perform(Operator.eos) ) {}

        // set start and finish states
        start = addState();
        state_t finish = addState();
        finish.accept = true;

        foreach ( f; frags ) {
            f.setExit(finish);
            f.setEntry(start);
        }

        // set transition priorities
        List!(trans_t)[PriorityClass.max+1] trans;
        foreach ( ref t; trans )
            t = new List!(trans_t);

        Stack!(trans_t) todo;
        state_t state = start;

        while ( !todo.empty() || !state.visited )
        {
            if ( !state.visited )
            {
                state.visited = true;
                foreach_reverse ( t; state.transitions )
                    todo.push(t);
            }

            if ( todo.empty() )
                break;
            trans_t t = todo.top();
            todo.pop();
            assert(t.priorityClass<=PriorityClass.max);
            trans[t.priorityClass] ~= t;
            state = t.target;
        }

        uint nextPrio;
        foreach ( ts; trans )
        {
            foreach ( t; ts )
                t.priority = nextPrio++;
        }
    }

private:
    uint            next_tag;
    size_t          cursor,
                    next_cursor;
    List!(trans_t)  transitions;

    state_t[state_t]    clonedStates;
    trans_t[trans_t]    clonedTransitions;

    /// RegEx operators
    enum Operator {
        eos, concat, altern, open_par, close_par,
        zero_one_g, zero_more_g, one_more_g,        // greedy
        zero_one_ng, zero_more_ng, one_more_ng,     // non-greedy/reluctant
        zero_one_xr, zero_more_xr, one_more_xr,     // extra-reluctant
        open_par_nm, occur_g, occur_ng
    }
    __gshared immutable const(char)[][] operator_names = ["EOS", "concat", "|", "(", ")", "?", "*", "+", "??", "*?", "+?", "??x", "*?x", "+?x", "(?", "{x,y}", "{x,y}?"];

    /// Actions for to-postfix transformation
    enum Act {
        pua, poc, poa, pca, don, err
    }
    __gshared immutable const(char)[][] action_names = ["push+advance", "pop+copy", "pop+advance", "pop+copy+advance", "done", "error"];

    /// Action lookup for to-postfix transformation
    __gshared immutable Act[] action_lookup =
    [
    //  eos      concat   |        (        )        ?        *        +        ??       *?       +?       ??extra  *?extra  +?extra  (?       {x,y}    {x,y}?
        Act.don, Act.pua, Act.pua, Act.pua, Act.err, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua,
        Act.poc, Act.pua, Act.poc, Act.pua, Act.poc, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua,
        Act.err, Act.pua, Act.pua, Act.pua, Act.pca, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua,
        Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err, Act.err,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.err, Act.pua, Act.pua, Act.pua, Act.pca, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua, Act.pua,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc,
        Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.poc, Act.pua, Act.poc, Act.poc
    ];

    final dchar peekPattern()
    {
        auto tmp = next_cursor;
        if ( tmp < pattern.length )
            return decode(pattern, tmp);
        return 0;
    }

    final dchar readPattern()
    {
        cursor = next_cursor;
        if ( next_cursor < pattern.length )
            return decode(pattern, next_cursor);
        return 0;
    }

    final bool endOfPattern()
    {
        return next_cursor >= pattern.length;
    }

    state_t addState()
    {
        state_t s = new state_t;
        s.index = states.length;
        states ~= s;
        return s;
    }

    trans_t addTransition(PriorityClass pc = PriorityClass.normal)
    {
        trans_t trans = new trans_t(pc);
        transitions ~= trans;
        return trans;
    }

    uint parseNumber()
    {
        uint res;
        while ( !endOfPattern() )
        {
            auto c = peekPattern();
            if ( c < '0' || c > '9' )
                break;
            res = res*10+(c-'0');
            readPattern();
        }
        return res;
    }

    void parseOccurCount(out uint minOccur, out uint maxOccur)
    {
        assert(pattern[cursor] == '{');

        minOccur = parseNumber();
        if ( peekPattern() == '}' ) {
            readPattern();
            maxOccur = minOccur;
            return;
        }
        if ( peekPattern() != ',' )
            throw new RegExpException("Invalid occurence range at \""~Utf.toString(pattern[cursor..$]).idup~"\"");
        readPattern();
        maxOccur = parseNumber();
        if ( peekPattern() != '}' )
            throw new RegExpException("Invalid occurence range at \""~Utf.toString(pattern[cursor..$]).idup~"\"");
        readPattern();
        if ( maxOccur > 0 && maxOccur < minOccur )
            throw new RegExpException("Invalid occurence range (max < min) at \""~Utf.toString(pattern[cursor..$]).idup~"\"");
    }

    trans_t clone(trans_t t)
    {
        if ( t is null )
            return null;
        trans_t* tmp = t in clonedTransitions;
        if ( tmp !is null )
            return *tmp;

        trans_t t2 = new trans_t(t.priorityClass);
        clonedTransitions[t] = t2;
        t2.tag = t.tag;
        t2.priority = t.priority;
        t2.predicate = t.predicate;
        t2.target = clone(t.target);
        transitions ~= t2;
        return t2;
    }

    state_t clone(state_t s)
    {
        if ( s is null )
            return null;
        state_t* tmp = s in clonedStates;
        if ( tmp !is null )
            return *tmp;

        state_t s2 = new state_t;
        clonedStates[s] = s2;
        s2.accept = s.accept;
        s2.visited = s.visited;
        foreach ( t; s.transitions )
            s2.transitions ~= clone(t);
        s2.index = states.length;
        states ~= s2;
        return s2;
    }

    frag_t clone(frag_t f)
    {
        if ( f is null )
            return null;
        clonedStates = null;
        clonedTransitions = null;

        frag_t f2 = new frag_t;
        foreach ( t; f.entries )
            f2.entries ~= clone(t);
        foreach ( t; f.exits )
            f2.exits ~= clone(t);
        foreach ( t; f.entry_state )
            f2.entry_state ~= clone(t);
        foreach ( t; f.exit_state )
            f2.exit_state ~= clone(t);
        return f2;
    }

    //---------------------------------------------------------------------------------------------
    // Thompson constructions of NFA fragments

    frag_t constructSingleChar(char_t c, predicate_t.Type type)
    {
        debug(tnfa) Stdout.formatln("constructCharFrag {}", c);

        trans_t trans = addTransition();
        trans.predicate.appendInput(CharRange!(char_t)(c));

        trans.predicate.type = type;

        frag_t frag = new frag_t;
        frag.exit_state ~= trans;
        frag.entries    ~= trans;
        return frag;
    }

    frag_t constructChars(string_t chars, in predicate_t.Type type)
    {
        cc_t cc;
        for ( int i = 0; i < chars.length; ++i )
            cc.add(chars[i]);

        return constructChars(cc, type);
    }

    frag_t constructChars(const(cc_t) charclass, in predicate_t.Type type)
    {
        debug(tnfa) Stdout.format("constructChars type={}", type);

        trans_t trans = addTransition();
        trans.predicate.type = type;

        trans.predicate.setInput(cc_t(charclass));

        trans.predicate.optimize();
        debug(tnfa) Stdout.formatln("-> {}", trans.predicate.toString);

        frag_t frag = new frag_t;
        frag.exit_state ~= trans;
        frag.entries    ~= trans;
        return frag;
    }

    frag_t constructCharClass(predicate_t.Type type)
    {
        debug(tnfa) Stdout.format("constructCharClass type={}", type);
        auto oldCursor = cursor;

        trans_t trans = addTransition();

        bool negated=false;
        if ( peekPattern() == '^' ) {
            readPattern();
            negated = true;
        }

        char_t  last;
        bool    have_range_start,
                first_char = true;
        for ( ; !endOfPattern() && peekPattern() != ']'; )
        {
            dchar c = readPattern();
            switch ( c )
            {
                case '-':
                    if ( first_char ) {
                        trans.predicate.appendInput(range_t(c));
                        break;
                    }
                    if ( !have_range_start )
                        throw new RegExpException("Missing range start for '-' operator after \""~Utf.toString(pattern).idup~"\"");
                    else if ( endOfPattern() || peekPattern() == ']' )
                        throw new RegExpException("Missing range end for '-' operator after \""~Utf.toString(pattern).idup~"\"");
                    else {
                        c = readPattern();
                        trans.predicate.appendInput(range_t(last, c));
                        have_range_start = false;
                    }
                    break;
                case '\\':
                    if ( endOfPattern() )
                        throw new RegExpException("unexpected end of string after \""~Utf.toString(pattern).idup~"\"");
                    c = readPattern();
                    switch ( c )
                    {
                        case 't':
                            c = '\t';
                            break;
                        case 'n':
                            c = '\n';
                            break;
                        case 'r':
                            c = '\r';
                            break;
                        default:
                            break;
                    }
                    goto default;
                default:
                    if ( have_range_start )
                        trans.predicate.appendInput(range_t(last));
                    last = c;
                    have_range_start = true;
            }
            first_char = false;
        }
        if ( !endOfPattern() )
            readPattern();
        if ( last != char_t.init )
            trans.predicate.appendInput(range_t(last));
        debug(tnfa) Stdout.formatln(" {}", pattern[oldCursor..cursor]);

        if ( negated ) {
            auto tmp = cc_t(cc_t.any_char);
            tmp.subtract(trans.predicate.input);
            trans.predicate.input = tmp;
        }
        else
            trans.predicate.optimize();
        debug(tnfa) Stdout.formatln("-> {}", trans.predicate.toString);

        trans.predicate.type = type;

        frag_t frag = new frag_t;
        frag.exit_state ~= trans;
        frag.entries    ~= trans;
        return frag;
    }

    void constructBracket(List!(frag_t) frags, uint tag=0)
    {
        debug(tnfa) Stdout.formatln("constructBracket");

        state_t entry = addState(),
                exit = addState();
        frags.tail.value.setEntry(entry);
        frags.tail.value.setExit(exit);

        trans_t tag1 = addTransition(),
                tag2 = addTransition();
        tag1.predicate.type = predicate_t.Type.epsilon;
        tag2.predicate.type = predicate_t.Type.epsilon;
        if ( tag > 0 )
        {
            // make sure the tag indeces for bracket x are always
            // x*2 for the opening bracket and x*2+1 for the closing bracket
            tag1.tag = tag++;
            tag2.tag = tag;
        }
        tag1.target = entry;
        exit.transitions ~= tag2;

        frag_t frag = new frag_t;
        frag.entries ~= tag1;
        frag.exit_state ~= tag2;
        frags.pop();
        frags ~= frag;
    }

    void constructOneMore(List!(frag_t) frags, PriorityClass prioClass)
    {
        debug(tnfa) Stdout.formatln("constructOneMore");

        if ( frags.empty() )
            throw new RegExpException("too few arguments for + at \""~Utf.toString(pattern[cursor..$]).idup~"\"");

        trans_t repeat = addTransition(prioClass),
                cont = addTransition();
        repeat.predicate.type = predicate_t.Type.epsilon;
        cont.predicate.type = predicate_t.Type.epsilon;

        state_t s = addState();
        frags.tail.value.setExit(s);
        s.transitions ~= repeat;
        s.transitions ~= cont;

        frag_t frag = new frag_t;
        frag.entries ~= frags.tail.value.entries;
        frag.entry_state ~= frags.tail.value.entry_state;
        frag.entry_state ~= repeat;
        frag.exit_state ~= cont;
        frags.pop();
        frags ~= frag;
    }

    void constructZeroMore(List!(frag_t) frags, PriorityClass prioClass)
    {
        debug(tnfa) Stdout.formatln("constructZeroMore");

        if ( frags.empty() )
            throw new RegExpException("too few arguments for * at \""~Utf.toString(pattern[cursor..$]).idup~"\"");

        trans_t enter = addTransition(prioClass),
                repeat = addTransition(prioClass),
                skip = addTransition();
        skip.predicate.type = predicate_t.Type.epsilon;
        repeat.predicate.type = predicate_t.Type.epsilon;
        enter.predicate.type = predicate_t.Type.epsilon;

        state_t entry = addState(),
                exit = addState();
        frags.tail.value.setEntry(entry);
        frags.tail.value.setExit(exit);
        exit.transitions ~= repeat;
        enter.target = entry;

        frag_t frag = new frag_t;
        frag.entries ~= skip;
        frag.entries ~= enter;
        frag.exit_state ~= skip;
        frag.entry_state ~= repeat;
        frags.pop();
        frags ~= frag;
    }

    void constructZeroOne(List!(frag_t) frags, PriorityClass prioClass)
    {
        debug(tnfa) Stdout.formatln("constructZeroOne");

        if ( frags.empty() )
            throw new RegExpException("too few arguments for ? at \""~Utf.toString(pattern[cursor..$]).idup~"\"");

        trans_t use = addTransition(prioClass),
                skip = addTransition();
        use.predicate.type = predicate_t.Type.epsilon;
        skip.predicate.type = predicate_t.Type.epsilon;

        state_t s = addState();
        frags.tail.value.setEntry(s);
        use.target = s;

        frag_t frag = new frag_t;
        frag.entries ~= use;
        frag.entries ~= skip;
        frag.exits ~= frags.tail.value.exits;
        frag.exit_state ~= frags.tail.value.exit_state;
        frag.exit_state ~= skip;
        frags.pop();
        frags ~= frag;
    }

    void constructOccur(List!(frag_t) frags, uint minOccur, uint maxOccur, PriorityClass prioClass)
    {
        debug(tnfa) Stdout.formatln("constructOccur {},{}", minOccur, maxOccur);

        if ( frags.empty() )
            throw new RegExpException("too few arguments for {x,y} at \""~Utf.toString(pattern[cursor..$]).idup~"\"");

        state_t s;
        frag_t  total = new frag_t,
                prev;

        for ( int i = 0; i < minOccur; ++i )
        {
            frag_t f = clone(frags.tail.value);
            if ( prev !is null ) {
                s = addState();
                prev.setExit(s);
                f.setEntry(s);
            }
            else {
                total.entries = f.entries;
                total.entry_state = f.entry_state;
            }
            prev = f;
        }

        if ( maxOccur == 0 )
        {
            frag_t f = frags.tail.value;
            trans_t t = addTransition();
            t.predicate.type = predicate_t.Type.epsilon;
            f.entries ~= t;
            f.exit_state ~= t;

            t = addTransition();
            t.predicate.type = predicate_t.Type.epsilon;
            f.exits ~= t;
            f.entry_state ~= t;

            s = addState();
            f.setEntry(s);

            if ( prev !is null )
                prev.setExit(s);
            else {
                total.entries = f.entries;
                total.entry_state = f.entry_state;
            }

            prev = f;
        }

        for ( int i = minOccur; i < maxOccur; ++i )
        {
            frag_t f;
            if ( i < maxOccur-1 )
                f = clone(frags.tail.value);
            else
                f = frags.tail.value;
            trans_t t = addTransition();
            t.predicate.type = predicate_t.Type.epsilon;
            f.entries ~= t;
            f.exit_state ~= t;

            if ( prev !is null ) {
                s = addState();
                prev.setExit(s);
                f.setEntry(s);
            }
            else {
                total.entries = f.entries;
                total.entry_state = f.entry_state;
            }
            prev = f;
        }

        total.exits = prev.exits;
        total.exit_state = prev.exit_state;

        frags.pop();
        frags ~= total;
    }

    void constructAltern(List!(frag_t) frags)
    {
        debug(tnfa) Stdout.formatln("constructAltern");

        if ( frags.empty() || frags.head is frags.tail )
            throw new RegExpException("too few arguments for | at \""~Utf.toString(pattern[cursor..$]).idup~"\"");

        frag_t  frag = new frag_t,
                f1 = frags.tail.value,
                f2 = frags.tail.prev.value;
        frag.entry_state ~= f2.entry_state;
        frag.entry_state ~= f1.entry_state;
        frag.exit_state ~= f2.exit_state;
        frag.exit_state ~= f1.exit_state;
        frag.entries ~= f2.entries;
        frag.entries ~= f1.entries;
        frag.exits ~= f2.exits;
        frag.exits ~= f1.exits;

        frags.pop();
        frags.pop();
        frags ~= frag;
    }

    void constructConcat(List!(frag_t) frags)
    {
        debug(tnfa) Stdout.formatln("constructConcat");

        if ( frags.empty() || frags.head is frags.tail )
            throw new RegExpException("too few operands for concatenation at \""~Utf.toString(pattern[cursor..$]).idup~"\"");

        frag_t  f1 = frags.tail.value,
                f2 = frags.tail.prev.value;

        state_t state = addState();
        f2.setExit(state);
        f1.setEntry(state);

        frag_t frag = new frag_t;
        frag.entries ~= f2.entries;
        frag.exits ~= f1.exits;
        frag.entry_state ~= f2.entry_state;
        frag.exit_state ~= f1.exit_state;
        frags.pop();
        frags.pop();
        frags ~= frag;
    }
}
import tango.core.Array;

/* ************************************************************************************************
    Tagged DFA
**************************************************************************************************/
private class TDFA(char_t)
{
    alias Predicate!(char_t)    predicate_t;
    alias CharRange!(char_t)    range_t;
    alias CharClass!(char_t)    charclass_t;
    alias char_t[]              string_t;

    enum uint CURRENT_POSITION_REGISTER = ~0;

    /* ********************************************************************************************
        Tag map assignment command
    **********************************************************************************************/
    struct Command
    {
        uint        dst,    /// register index to recieve data
                    src;    /// register index or CURRENT_POSITION_REGISTER for current position

        string toString()
        {
            return Format.convert("{}<-{}", dst, src==CURRENT_POSITION_REGISTER?"p":Format.convert("{}", src)).idup;
        }

        /* ****************************************************************************************
            Order transitions by the order of their predicates.
        ******************************************************************************************/
        int opCmp(const ref Command cmd) const
        {
            if ( src == CURRENT_POSITION_REGISTER && cmd.src != CURRENT_POSITION_REGISTER )
                return 1;
            if ( src != CURRENT_POSITION_REGISTER && cmd.src == CURRENT_POSITION_REGISTER )
                return -1;
            if ( dst < cmd.dst )
                return -1;
            if ( dst == cmd.dst )
                return 0;
            return 1;
        }
    }

    struct TagIndex
    {
        uint    tag,
                index;

        const int opCmp(ref const TagIndex o)
        {
            if (tag == o.tag && index == o.index)
                return 0;
            if (tag > o.tag)
                return 1;
            if (tag < o.tag)
                return -1;
            if (index > o.index)
                return 1;
            if (index < o.index)
                return -1;
            assert(0);
        }
    }

    /* ********************************************************************************************
        TDFA state
    **********************************************************************************************/
    class State
    {
        enum Mode {
            GENERIC, MIXED, LOOKUP
        }

        enum uint  LOOKUP_LENGTH = 256,
                   INVALID_STATE = 255;

        bool            accept = false;
        bool            reluctant = false;
        size_t          index;
        Transition[]    transitions,
                        generic_transitions;
        Command[]       finishers;

        ubyte[]         lookup;
        Mode            mode;

        void optimize()
        {
            // merge transitions with equal targets (same state index and equal commands)
            size_t[] remove_indeces;
            foreach ( i, t; transitions[0 .. $-1] )
            {
                foreach ( t2; transitions[i+1 .. $] )
                {
                    if ( t.predicate.type != t2.predicate.type || !t.equalTarget(t2) )
                        continue;
                    t2.predicate.appendInput(t.predicate);
                    remove_indeces ~= i;
                    break;
                }
            }

            // remove transitions that have been merged into another
            if ( remove_indeces.length > 0 )
            {
                Transition[] tmp;
                tmp.length = transitions.length - remove_indeces.length;
                size_t next_remove, next;
                foreach ( i, t; transitions )
                {
                    if ( next_remove < remove_indeces.length && remove_indeces[next_remove] == i ) {
                        ++next_remove;
                        continue;
                    }
                    tmp[next++] = t;
                }
                transitions = tmp;

                foreach ( t; transitions )
                    t.predicate.optimize();
            }
        }

        void createLookup()
        {
            size_t count;
            foreach ( t; transitions )
            {
                if ( !t.predicate.exceedsMax(LOOKUP_LENGTH) )
                    ++count;
            }

            if ( count < 1 || transitions.length > INVALID_STATE ) {
                generic_transitions.length = transitions.length;
                generic_transitions[] = transitions[];
                return;
            }

            foreach ( t; transitions )
            {
                if ( t.predicate.exceedsMax(LOOKUP_LENGTH) )
                    generic_transitions ~= t;
            }

            // setup lookup table
            lookup.length = LOOKUP_LENGTH;
            lookup[] = INVALID_STATE;
            foreach ( i, t; transitions )
            {
                foreach ( p; t.predicate.input.parts )
                {
                    if ( p.l_ >= lookup.length )
                        continue;
                    for ( char_t c = p.l_; c <= min(p.r_, LOOKUP_LENGTH-1); ++c )
                        lookup[c] = cast(ubyte)i;
                }
            }

            mode = count < transitions.length? Mode.MIXED : Mode.LOOKUP;
        }
    }

    /* ********************************************************************************************
        TDFA transition
    **********************************************************************************************/
    class Transition
    {
        State       target;
        predicate_t predicate;
        Command[]   commands;

        /* ****************************************************************************************
            Order transitions by the order of their predicates.
        ******************************************************************************************/
        override final int opCmp(Object o)
        {
            Transition t = cast(Transition)o;
            assert(t !is null);
            return predicate.opCmp(t.predicate);
        }

        override final const bool opEquals(const Object o)
        {
            auto t = cast(const(Transition))o;
            if ( t is null )
                return false;
            if ( equalTarget(t) && t.predicate == predicate )
                return true;
            return false;
        }

        const bool equalTarget(in Transition t)
        {
            if ( t.target.index != target.index )
                return false;
            if ( commands.length != t.commands.length )
                return false;
            Louter: foreach ( cmd; commands )
            {
                foreach ( cmd2; t.commands )
                {
                    if ( cmd == cmd2 )
                        continue Louter;
                }
                return false;
            }
            return true;
        }
    }


    State[]     states;
    State       start;
    Command[]   initializer;
    uint        num_tags;

    uint[TagIndex]  registers;
    uint            next_register;

    uint num_regs()
    {
        return next_register;
    }

    /* ********************************************************************************************
        Constructs the TDFA from the given TNFA using extended power set method
    **********************************************************************************************/
    this(TNFA!(char_t) tnfa)
    {
        num_tags        = tnfa.tagCount();

        next_register   = num_tags;
        for ( int i = 1; i <= num_tags; ++i ) {
            TagIndex ti;
            ti.tag = i;
            registers[ti] = i-1;
        }

        // create epsilon closure of TNFA start state
        SubsetState subset_start    = new SubsetState;
        StateElement se             = new StateElement;
        se.nfa_state = tnfa.start;
        subset_start.elms ~= se;

        // apply lookbehind closure for string/line start
        predicate_t tmp_pred;
        tmp_pred.setInput(CharClass!(char_t).line_startend);
        subset_start = epsilonClosure(lookbehindClosure(epsilonClosure(subset_start, subset_start), tmp_pred), subset_start);

        start = addState();
        subset_start.dfa_state = start;

        // generate initializer and finisher commands for TDFA start state
        generateInitializers(subset_start);
        generateFinishers(subset_start);

        // initialize stack for state traversal
        List!(SubsetState)  subset_states   = new List!(SubsetState),
                            unmarked        = new List!(SubsetState);
        subset_states   ~= subset_start;
        unmarked        ~= subset_start;
        debug(tdfa) Stdout.formatln("\n{} = {}\n", subset_start.dfa_state.index, subset_start.toString);

        while ( !unmarked.empty() )
        {
            SubsetState state = unmarked.tail.value;
            unmarked.pop();

            // create transitions for each class, creating new states when necessary
            foreach ( pred; disjointPredicates(state) )
            {
                // find NFA state we reach with pred
                // reach will set predicate type correctly
                debug(tdfa) Stdout.format("from {} with {} reach", state.dfa_state.index, pred.toString);
                SubsetState target = reach(state, pred);
                if ( target is null ) {
                    continue;
                    debug(tdfa) Stdout.formatln(" nothing - lookbehind at beginning, skipping");
                }
                debug(tdfa) Stdout.formatln(" {}", target.toString);
                target = epsilonClosure(lookbehindClosure(epsilonClosure(target, state), pred), state);

                Transition trans = new Transition;
                state.dfa_state.transitions ~= trans;
                debug (tdfa_new_trans) Stdout.formatln("Creating with pred: {}", pred);
                trans.predicate = pred;

                // generate indeces for pos commands
                // delay creation of pos command until we have reorder-commands
                uint[uint] cmds = null;
                foreach ( e; target.elms )
                {
                    foreach ( tag, ref index; e.tags )
                    {
                        bool found=false;
                        foreach ( e2; state.elms )
                        {
                            int* i = tag in e2.tags;
                            if ( i !is null && *i == index ) {
                                found=true;
                                break;
                            }
                        }
                        if ( !found )
                        {
                            // if index is < 0 it is a temporary index
                            // used only to distinguish the state from existing ones.
                            // the previous index can be reused instead.
                            if ( index < 0 )
                                index = -index-1;
                            cmds[tag] = index;
                        }
                        else
                            assert(index>=0);
                    }
                }

                // check whether a state exists that is identical except for tag index reorder-commands
                bool exists=false;
                foreach ( equivTarget; subset_states )
                {
                    if ( reorderTagIndeces(target, equivTarget, state, trans) ) {
                        target = equivTarget;
                        exists = true;
                        break;
                    }
                }
                // else create new target state
                if ( !exists )
                {
                    State ts = addState();
                    target.dfa_state = ts;
                    subset_states   ~= target;
                    unmarked        ~= target;
                    debug(tdfa_add) {
                        Stdout.formatln("\nAdded {} = {}\n", target.dfa_state.index, target.toString);
                    }
                    generateFinishers(target);
                }

                // now generate pos commands, rewriting reorder-commands if existent
                foreach ( tag, index; cmds )
                {
                    // check whether reordering used this tag, if so, overwrite the command directly,
                    // for it's effect would be overwritten by a subsequent pos-command anyway
                    uint reg = registerFromTagIndex(tag, index);
                    bool found = false;
                    foreach ( ref cmd; trans.commands )
                    {
                        if ( cmd.src == reg ) {
                            found = true;
                            cmd.src = CURRENT_POSITION_REGISTER;
                            break;
                        }
                    }
                    if ( !found ) {
                        Command cmd;
                        cmd.dst = reg;
                        cmd.src = CURRENT_POSITION_REGISTER;
                        trans.commands ~= cmd;
                    }
                }

                trans.target = target.dfa_state;
                debug(tdfa) {
                    Stdout.formatln("=> from {} with {} reach {}", state.dfa_state.index, pred.toString, target.dfa_state.index);
                }
            }

            state.dfa_state.optimize();
        }

        // renumber registers continuously
        uint[uint]  regNums;

        for ( next_register = 0; next_register < num_tags; ++next_register )
            regNums[next_register] = next_register;

        void renumberCommand(ref Command cmd)
        {
            if ( cmd.src != CURRENT_POSITION_REGISTER && (cmd.src in regNums) is null )
                regNums[cmd.src] = next_register++;
            if ( (cmd.dst in regNums) is null )
                regNums[cmd.dst] = next_register++;
            if ( cmd.src != CURRENT_POSITION_REGISTER )
                cmd.src = regNums[cmd.src];
            cmd.dst = regNums[cmd.dst];
        }

        foreach ( state; states )
        {
            foreach ( ref cmd; state.finishers )
                renumberCommand(cmd);
            // make sure pos-commands are executed after reorder-commands and
            // reorder-commands do not overwrite each other
            tango.core.Array.sort(state.finishers);

            foreach ( trans; state.transitions )
            {
                foreach ( ref cmd; trans.commands )
                    renumberCommand(cmd);
                tango.core.Array.sort(trans.commands);
                trans.predicate.compile();
            }
        }

        debug(TangoRegex)
        {
            foreach ( ref v; registers )
            {
                if ( (v in regNums) !is null )
                    v = regNums[v];
            }
        }

        minimizeDFA();

        foreach ( state; states )
            state.createLookup();

        // TODO: optimize memory layout of TDFA

        // TODO: add lookahead for string-end somewhere
        // TODO: mark dead-end states (not leaving a non-finishing susbet)
        // TODO: mark states that can leave the finishing subset of DFA states or use a greedy transition
        //       (execution may stop in that state)
    }

    /* ********************************************************************************************
        Print the TDFA (tabular representation of the delta function)
    **********************************************************************************************/
    debug(TangoRegex) void print()
    {
        Stdout.formatln("#tags = {}", num_tags);

        auto tis = new TagIndex[registers.length];
        foreach ( k, v; registers )
            tis [v] = k;
        foreach ( r, ti; tis ) {
            Stdout.formatln("tag({},{}) in reg {}", ti.tag, ti.index, r);
        }
        Stdout.formatln("Initializer:");
        foreach ( cmd; initializer ) {
            Stdout.formatln("{}", cmd.toString);
        }
        Stdout.formatln("Delta function:");
        foreach ( int i, s; states )
        {
            Stdout.format("{}{:d2}{}", s is start?">":" ", i, s.accept?"*":" ");

            bool first=true;
            Stdout(" {");
            foreach ( t; s.transitions )
            {
                Stdout.format("{}{}->{} (", first?"":", ", t.predicate.toString, t.target.index);
                bool firstcmd=true;
                foreach ( cmd; t.commands )
                {
                    if ( firstcmd )
                        firstcmd = false;
                    else
                        Stdout(",");
                    Stdout.format("{}", cmd.toString);
                }
                Stdout(")");
                first = false;
            }
            Stdout("} (");

            bool firstcmd=true;
            foreach ( cmd; s.finishers )
            {
                if ( firstcmd )
                    firstcmd = false;
                else
                    Stdout(",");
                Stdout.format("{}", cmd.toString);
            }
            Stdout.formatln(")");
        }
    }

private:
    /* ********************************************************************************************
        A (TNFA state, tags) pair element of a subset state.
    **********************************************************************************************/
    class StateElement
    {
        TNFAState!(char_t)  nfa_state;
        int[uint]           tags;
        // use place-value priority with 2 places, value(maxPrio) > value(lastPrio)
        uint                maxPriority,
                            lastPriority;

        bool prioGreater(StateElement se)
        {
            if ( maxPriority < se.maxPriority )
                return true;
            if ( maxPriority == se.maxPriority ) {
                assert(lastPriority != se.lastPriority);
                return lastPriority < se.lastPriority;
            }
            return false;
        }

        override int opCmp(Object o)
        {
            StateElement se = cast(StateElement)o;
            assert(se !is null);
            if ( maxPriority < se.maxPriority )
                return 1;
            if ( maxPriority == se.maxPriority )
            {
                if ( lastPriority == se.lastPriority )
                    return 0;
                return lastPriority < se.lastPriority;
            }
            return -1;
        }

        override string toString()
        {
            const(char)[] str;
            str = Format.convert("{} p{}.{} {{", nfa_state.index, maxPriority, lastPriority);
            bool first = true;
            foreach ( k, v; tags ) {
                str ~= Format.convert("{}m({},{})", first?"":",", k, v);
                first = false;
            }
            str ~= "}";
            return str.idup;
        }
    }

    /* ********************************************************************************************
        Represents a state in the NFA to DFA conversion.
        Contains the set of states (StateElements) the NFA might be in at the same time and the
        corresponding DFA state that we create.
    **********************************************************************************************/
    class SubsetState
    {
        StateElement[]  elms;
        State           dfa_state;

        this(StateElement[] elms=null)
        {
            this.elms = elms;
        }

        int opApply(int delegate (ref TNFATransition!(char_t)) dg)
        {
            int res;
            foreach ( elm; elms )
            {
                foreach ( t; elm.nfa_state.transitions )
                {
                    res = dg(t);
                    if ( res )
                        return res;
                }
            }
            return res;
        }

        override string toString()
        {
            string str = "[ ";
            bool first = true;
            foreach ( s; elms ) {
                if ( !first )
                    str ~= ", ";
                str ~= s.toString();
                first = false;
            }
            return str~" ]";
        }
    }

    /* ********************************************************************************************
        Temporary structure used for disjoint predicate computation
    **********************************************************************************************/
    struct Mark
    {
        char_t  c;
        bool    end;    /// false = start of range

        int opCmp(const ref Mark m) const
        {
            if ( c < m.c )
                return -1;
            if ( c > m.c )
                return 1;
            if ( end < m.end )
                return -1;
            if ( end > m.end )
                return 1;
            return 0;
        }
    }

    /* ********************************************************************************************
        Calculates the register index for a given tag map entry. The TDFA implementation uses
        registers to save potential tag positions, the index space gets linearized here.

        Params:     tag =   tag number
                    index = tag map index
        Returns:    index of the register to use for the tag map entry
    **********************************************************************************************/
    uint registerFromTagIndex(uint tag, uint index)
    {
        if ( index > 0 )
        {
            TagIndex ti;
            ti.tag = tag;
            ti.index = index;
            uint* i = ti in registers;
            if ( i !is null )
                return *i;
            return registers[ti] = next_register++;
        }
        else
            return tag-1;
    }

    Mark[] marks_;

    /* ********************************************************************************************
        Add new TDFA state to the automaton.
    **********************************************************************************************/
    State addState()
    {
        State s = new State;
        s.index = states.length;
        states ~= s;
        return s;
    }

    /* ********************************************************************************************
        Creates disjoint predicates from all outgoing, potentially overlapping TNFA transitions.

        Params:     state = SubsetState to create the predicates from
        Returns:    List of disjoint predicates that can be used for a DFA state
    **********************************************************************************************/
    predicate_t[] disjointPredicates(SubsetState state)
    {
        alias CharRange!(char_t) range_t;
        debug(tdfa) Stdout.formatln("disjointPredicates()");

        size_t num_marks;
        foreach ( t; state )
        {
            // partitioning will consider lookbehind transitions,
            // st. lb-closure will not expand for transitions with a superset of the lb-predicate
            if ( t.predicate.type != predicate_t.Type.epsilon )
            {
                debug(tdfa) Stdout.formatln("{}", t.predicate.toString);
                if ( marks_.length < num_marks+2*t.predicate.getInput().parts.length )
                    marks_.length = num_marks+2*t.predicate.getInput().parts.length;
                foreach ( p; t.predicate.getInput().parts ) {
                    marks_[num_marks++] = Mark(p.l(), false);
                    marks_[num_marks++] = Mark(p.r(), true);
                }
            }
        }

        if ( num_marks <= 1 )
            throw new Exception("disjointPredicates: No transitions in subset state");

        debug(tdfa) Stdout("\nsorting...").newline;
        // using built-in sort somtimes gives an AV in TypeInfo.swap
        quickSort(marks_[0 .. num_marks]);
        assert(!marks_[0].end);

        debug(tdfa)
        {
            Stdout("\nsorted marks:\n");
            bool first=true;
            foreach ( m; marks_[0 .. num_marks] )
            {
                if ( first )
                    first = false;
                else
                    Stdout(",");
                if ( m.c > 0x20 && m.c < 0x7f )
                    Stdout.format("{}{}", m.end?"e":"s", m.c);
                else
                    Stdout.format("{}{:x}", m.end?"e":"s", cast(int)m.c);
            }
            Stdout.newline;
        }

        size_t  next,
                active = 1;
        char_t  start = marks_[0].c,
                end;
        range_t[]   disjoint = new range_t[num_marks/2+1];

        foreach ( m; marks_[1 .. num_marks] )
        {
            if ( m.end )
            {
                assert(active>0);
                --active;
                if ( m.c < start )
                    continue;
                end = m.c;
                // the next range cannot start at the same pos
                // because starts are sorted before endings
                if ( active > 0 )
                    ++m.c;
            }
            else
            {
                ++active;
                if ( active == 1 )
                {
                    // skip uncovered interval
                    if ( m.c > start ) {
                        start = m.c;
                        continue;
                    }
                    end = m.c;
                    ++m.c;
                }
                // skip range start if cursor already marks it
                else if ( m.c <= start )
                    continue;
                else
                    end = m.c-1;
            }

            // save range
            if ( disjoint.length <= next )
                disjoint.length = disjoint.length*2;

            disjoint[next].l_ = start;
            disjoint[next].r_ = end;
            ++next;

            // advance cursor
            start = m.c;
        }
        disjoint.length = next;

        // merge isolated ranges into sets of ranges
        // no range in a set may occur separated from the others in any predicate
        predicate_t[]   preds;
        preds.length = 1;
        Lmerge: foreach ( r; disjoint )
        {
            if ( preds[$-1].empty() )
                preds[$-1].appendInput(r);
            else
            {
                // we can merge r into the current predicate if
                // pred contains r <=> pred contains all the other ranges
                foreach ( t; state )
                {
                    if ( t.predicate.type == predicate_t.Type.epsilon )
                        continue;

                    if ( t.predicate.getInput().contains(r)
                        != t.predicate.getInput().contains(preds[$-1].getInput()) )
                    {
                        preds.length = preds.length+1;
                        break;
                    }
                }
                preds[$-1].appendInput(r);
            }
        }

        debug(tdfa)
        {
            Stdout("\ndisjoint ranges:\n");
            first=true;
            foreach ( r; disjoint )
            {
                if ( first )
                    first = false;
                else
                    Stdout(",");
                Stdout.format("{}", r);
            }
            Stdout.newline;
            Stdout("\ndisjoint predicates:\n");
            first=true;
            foreach ( ref p; preds )
            {
                if ( first )
                    first = false;
                else
                    Stdout(",");
                Stdout.format("{}", p.toString);
            }
            Stdout.newline;
        }

        debug(tdfa) Stdout.formatln("disjointPredicates() end");
        return preds;
    }

    /* ********************************************************************************************
        Finds all TNFA states that can be reached directly with the given predicate and creates
        a new SubsetState containing those target states.

        Params:     subst = SubsetState to start from
                    pred =  predicate that is matched against outgoing transitions
        Returns:    SubsetState containing the reached target states
    **********************************************************************************************/
    SubsetState reach(SubsetState subst, ref predicate_t pred)
    {
        // to handle the special case of overlapping consume and lookahead predicates,
        // we find the different intersecting predicate types
        bool    have_consume,
                have_lookahead;
        foreach ( t; subst )
        {
            if ( t.predicate.type != predicate_t.Type.consume && t.predicate.type != predicate_t.Type.lookahead )
                continue;
            auto intpred = t.predicate.intersect(pred);
            if ( !intpred.empty() )
            {
                if ( t.predicate.type == predicate_t.Type.consume )
                    have_consume = true;
                else if ( t.predicate.type == predicate_t.Type.lookahead )
                    have_lookahead = true;
            }
        }

        // if there is consume/lookahead overlap,
        // lookahead predicates are handled first
        predicate_t.Type processed_type;
        if ( have_lookahead )
            processed_type = predicate_t.Type.lookahead;
        else if ( have_consume )
            processed_type = predicate_t.Type.consume;
        else {
            debug(tdfa) Stdout.formatln("\nERROR: reach found no consume/lookahead symbol for {} in \n{}", pred.toString, subst.toString);
            return null;
        }
        pred.type = processed_type;

        // add destination states to new subsetstate
        SubsetState r = new SubsetState;
        foreach ( s; subst.elms )
        {
            foreach ( t; s.nfa_state.transitions )
            {
                if ( t.predicate.type != processed_type )
                    continue;
                auto intpred = t.predicate.intersect(pred);
                if ( !intpred.empty() ) {
                    StateElement se = new StateElement;
                    se.maxPriority = max(t.priority, s.maxPriority);
                    se.lastPriority = t.priority;
                    se.nfa_state = t.target;
                    se.tags = s.tags;
                    r.elms ~= se;
                }
            }
        }

        // if we prioritized lookaheads, the states that may consume are also added to the new subset state
        // this behaviour is somewhat similar to an epsilon closure
        if ( have_lookahead && have_consume )
        {
            foreach ( s; subst.elms )
            {
                foreach ( t; s.nfa_state.transitions )
                {
                    if ( t.predicate.type != predicate_t.Type.consume )
                        continue;
                    auto intpred = t.predicate.intersect(pred);
                    if ( !intpred.empty() ) {
                        r.elms ~= s;
                        break;
                    }
                }
            }
        }
        return r;
    }

    /* ********************************************************************************************
        Extends the given SubsetState with the states that are reached through lookbehind transitions.

        Params:     from =      SubsetState to create the lookbehind closure for
                    previous =  predicate "from" was reached with
        Returns:    SubsetState containing "from" and all states of it's lookbehind closure
    **********************************************************************************************/
    SubsetState lookbehindClosure(SubsetState from, predicate_t pred)
    {
        List!(StateElement) stack = new List!(StateElement);
        StateElement[size_t]  closure;

        foreach ( e; from.elms )
        {
            stack ~= e;
            closure[e.nfa_state.index] = e;
        }

        while ( !stack.empty() )
        {
            StateElement se = stack.tail.value;
            stack.pop();
            foreach ( t; se.nfa_state.transitions )
            {
                if ( t.predicate.type != predicate_t.Type.lookbehind )
                    continue;
                if ( t.predicate.intersect(pred).empty() )
                    continue;
                uint new_maxPri = max(t.priority, se.maxPriority);

                StateElement* tmp = t.target.index in closure;
                if ( tmp !is null )
                {
                    // if higher prio (smaller value) exists, do not use this transition
                    if ( tmp.maxPriority < new_maxPri ) {
//                         debug(tdfa) Stdout.formatln("maxPrio({}) {} beats {}, continuing", t.target.index, tmp.maxPriority, new_maxPri);
                        continue;
                    }
                    else if ( tmp.maxPriority == new_maxPri )
                    {
                        // "equal lastPrio -> first-come-first-serve"
                        // doesn't work for lexer - how to solve it properly?
                        if ( tmp.lastPriority <= t.priority ) {
//                             debug(tdfa) Stdout.formatln("lastPrio({}) {} beats {}, continuing", t.target.index, tmp.lastPriority, t.priority);
                            continue;
                        }
//                         else
//                             debug(tdfa) Stdout.formatln("lastPrio({}) {} beats {}", t.target.index, t.priority, tmp.lastPriority);
                    }
//                     else
//                         debug(tdfa) Stdout.formatln("maxPrio({}) {} beats {}", t.target.index, new_maxPri, tmp.maxPriority);
                }

                StateElement new_se = new StateElement;
                new_se.maxPriority = max(t.priority, se.maxPriority);
                new_se.lastPriority = t.priority;
                new_se.nfa_state = t.target;
                new_se.tags = se.tags;

                closure[t.target.index] = new_se;
                stack ~= new_se;
            }
        }

        SubsetState res = new SubsetState;
        res.elms = closure.values;
        return res;
    }

    /* ********************************************************************************************
        Generates the epsilon closure of the given subset state, creating tag map entries
        if tags are passed. Takes priorities into account, effectively realizing
        greediness and reluctancy.

        Params:     from =      SubsetState to create the epsilon closure for
                    previous =  SubsetState "from" was reached from
        Returns:    SubsetState containing "from" and all states of it's epsilon closure
    **********************************************************************************************/
    SubsetState epsilonClosure(SubsetState from, SubsetState previous)
    {
        int firstFreeIndex=-1;
        foreach ( e; previous.elms )
        {
            foreach ( ti; e.tags )
                firstFreeIndex = max(firstFreeIndex, cast(int)ti);
        }
        ++firstFreeIndex;

        List!(StateElement) stack = new List!(StateElement);
        StateElement[size_t]  closure;

        foreach ( e; from.elms )
        {
            stack ~= e;
            closure[e.nfa_state.index] = e;
        }

        while ( !stack.empty() )
        {
            StateElement se = stack.tail.value;
            stack.pop();
            foreach ( t; se.nfa_state.transitions )
            {
                if ( t.predicate.type != predicate_t.Type.epsilon )
                    continue;
                // this is different from Ville Laurikari's algorithm, but it's crucial
                // to take the max (instead of t.priority) to make reluctant operators work
                uint new_maxPri = max(t.priority, se.maxPriority);

                StateElement* tmp = t.target.index in closure;
                if ( tmp !is null )
                {
                    // if higher prio (smaller value) exists, do not use this transition
                    if ( tmp.maxPriority < new_maxPri ) {
//                         debug(tdfa) Stdout.formatln("maxPrio({}) {} beats {}, continuing", t.target.index, tmp.maxPriority, new_maxPri);
                        continue;
                    }
                    else if ( tmp.maxPriority == new_maxPri )
                    {
                        // "equal lastPrio -> first-come-first-serve"
                        // doesn't work for lexer - how to solve it properly?
                        if ( tmp.lastPriority <= t.priority ) {
//                             debug(tdfa) Stdout.formatln("lastPrio({}) {} beats {}, continuing", t.target.index, tmp.lastPriority, t.priority);
                            continue;
                        }
//                         else
//                             debug(tdfa) Stdout.formatln("lastPrio({}) {} beats {}", t.target.index, t.priority, tmp.lastPriority);
                    }
//                     else
//                         debug(tdfa) Stdout.formatln("maxPrio({}) {} beats {}", t.target.index, new_maxPri, tmp.maxPriority);
                }

                auto new_se = new StateElement;
                new_se.maxPriority = new_maxPri;
                new_se.lastPriority = t.priority;
                new_se.nfa_state = t.target;

                if ( t.tag > 0 )
                {
                    foreach ( k, v; se.tags )
                        new_se.tags[k] = v;
                    new_se.tags[t.tag] = firstFreeIndex;
                }
                else
                    new_se.tags = se.tags;

                closure[t.target.index] = new_se;
                stack ~= new_se;
            }
        }

        SubsetState res = new SubsetState;
        res.elms = closure.values;

        // optimize tag usage
        // all we need to do is to check whether the largest tag-index from the
        // previous state is actually used in the new state and move all tags with
        // firstFreeIndex down by one if not, but only if firstFreeIndex is not 0
        if ( firstFreeIndex > 0 )
        {
            bool seenLastUsedIndex = false;
            sluiLoop: foreach ( e; res.elms )
            {
                foreach ( i; e.tags )
                {
                    if ( i == firstFreeIndex-1 ) {
                        seenLastUsedIndex = true;
                        break sluiLoop;
                    }
                }
            }
            if ( !seenLastUsedIndex )
            {
                foreach ( e; res.elms )
                {
                    foreach ( ref i; e.tags )
                    {
                        // mark index by making it negative
                        // to signal that it can be decremented
                        // after it has been detected to be a newly used index
                        if ( i == firstFreeIndex )
                            i = -firstFreeIndex;
                    }
                }
            }
        }

        return res;
    }

    /* ********************************************************************************************
        Tries to create commands that reorder the tag map of "previous", such that "from" becomes
        tag-wise identical to "to". If successful, these commands are added to "trans". This
        is done for state re-use.

        Params:     from =      SubsetState to check for tag-wise equality to "to"
                    to =        existing SubsetState that we want to re-use
                    previous =  SubsetState we're coming from
                    trans =     Transition we went along
        Returns:    true if "from" is tag-wise identical to "to" and the necessary commands have
                    been added to "trans"
    **********************************************************************************************/
    bool reorderTagIndeces(SubsetState from, SubsetState to, SubsetState previous, Transition trans)
    {
        if ( from.elms.length != to.elms.length )
            return false;

        bool[Command]
            cmds;
        uint[TagIndex]
            reorderedIndeces;
        StateElement[TagIndex]
            reordered_elements;

        Louter: foreach ( fe; from.elms )
        {
            foreach ( te; to.elms )
            {
                if ( te.nfa_state.index != fe.nfa_state.index )
                    continue;
                if ( fe.tags.length != te.tags.length )
                    return false;
                foreach ( tag, findex; fe.tags )
                {
                    if ( (tag in te.tags) is null )
                        return false;

                    TagIndex ti;
                    ti.tag = tag;
                    ti.index = te.tags[tag];

                    // apply priority for conflicting tag indeces
                    if ( (ti in reorderedIndeces) !is null )
                    {
                        auto rse = reordered_elements[ti];
                        auto ri = reorderedIndeces[ti];
                        if ( ri != findex
                            && ( rse.maxPriority < fe.maxPriority
                                || rse.maxPriority == fe.maxPriority
                                && rse.lastPriority <= fe.lastPriority )
                        )
                            continue;
                        Command cmd;
                        cmd.src = registerFromTagIndex(tag,ri);
                        cmd.dst = registerFromTagIndex(tag,te.tags[tag]);
                        cmds.remove(cmd);
                    }
                    // if target index differs, create reordering command
                    if ( te.tags[tag] != findex )
                    {
                        Command cmd;
                        cmd.src = registerFromTagIndex(tag,findex);
                        cmd.dst = registerFromTagIndex(tag,te.tags[tag]);
                        cmds[cmd] = true;
                    }

                    reorderedIndeces[ti] = findex;
                    reordered_elements[ti] = fe;
                }
                continue Louter;
            }
            return false;
        }

        debug(tdfa) {
            Stdout.formatln("\nreorder {} to {}\n", from.toString, to.dfa_state.index);
        }

        trans.commands ~= cmds.keys;
        return true;
    }

    /* ********************************************************************************************
        Generate tag map initialization commands for start state.
    **********************************************************************************************/
    void generateInitializers(SubsetState start)
    {
        uint[uint] cmds;
        foreach ( nds; start.elms )
        {
            foreach ( k, v; nds.tags )
                cmds[k] = v;
        }

        foreach ( k, v; cmds ) {
            Command cmd;
            cmd.dst = registerFromTagIndex(k,v);
            cmd.src = CURRENT_POSITION_REGISTER;
            initializer ~= cmd;
        }
    }

    /* ********************************************************************************************
        Generates finisher commands for accepting states.
    **********************************************************************************************/
    void generateFinishers(SubsetState r)
    {
        // if at least one of the TNFA states accepts,
        // set the finishers from active tags in increasing priority
        StateElement[]  sorted_elms = r.elms.dup;
        tango.core.Array.sort(sorted_elms);
        bool reluctant = false;
        foreach ( se; sorted_elms ) {
            debug (Finishers) Stdout.formatln("Finisher: {}", se);
            if ( se.nfa_state.accept )
            {
                r.dfa_state.accept = true;

                // Knowing that we're looking at an epsilon closure with an accepting
                // state, we look at the involved transitions - if the path from the
                // nfa state in the set with the highest incoming priority (last in
                // sorted_elms list) to the accepting nfa state is via the highest
                // priority transitions, and they are all epsilon transitions, this
                // suggests we're looking at a regex ending with a reluctant pattern.
                // The NFA->DFA transformation will most likely extend the automata
                // further, but we want the matching to stop here.
                // NOTE: The grounds for choosing the last element in sorted_elms
                // are somewhat weak (empirical testing), but sofar no new
                // regressions have been discovered. larsivi 20090827
                TNFATransition!(char_t) highestPriTrans;
                if (!(sorted_elms[$-1] && sorted_elms[$-1].nfa_state &&
                            sorted_elms[$-1].nfa_state))
                    throw new Exception ("Something is NULL that is expected to
                            be non-null", __FILE__, __LINE__);

                foreach ( trans; sorted_elms[$-1].nfa_state.transitions ) {
                    if (trans.canFinish()) {
                        r.dfa_state.reluctant = true;
                        break;
                    }
                }

                bool[uint]  finished_tags;
                {
                    foreach ( t, i; se.tags )
                        if ( i > 0 && !(t in finished_tags) ) {
                            finished_tags[t] = true;
                            Command cmd;
                            cmd.dst = registerFromTagIndex(t, 0);
                            cmd.src = registerFromTagIndex(t, i);
                            r.dfa_state.finishers ~= cmd;
                        }
                }
            }
        }
    }

    /* ********************************************************************************************
        Assumes that the command-lists are sorted and transitions are optimized
    **********************************************************************************************/
    void minimizeDFA()
    {
        class DiffTable
        {
            this(size_t num) {
                diff_ = new bool[num*(num+1)/2];
            }

            ~this() { diff_.destroy; }

            bool opCall(size_t a, size_t b)
            {
                if ( a < b )
                    return diff_[b*(b+1)/2+a];
                return diff_[a*(a+1)/2+b];
            }

            void set(size_t a, size_t b)
            {
                if ( a < b )
                    diff_[b*(b+1)/2+a] = true;
                else
                    diff_[a*(a+1)/2+b] = true;
            }

            bool[]  diff_;
        }

        debug(tdfa) Stdout.formatln("Minimizing TDFA");

        scope diff = new DiffTable(states.length);
        bool new_diff = true;

        while ( new_diff )
        {
            new_diff = false;
            foreach ( i, a; states[0 .. $-1] )
            {
                Linner: foreach ( j, b; states[i+1 .. $] )
                {
                    if ( diff(i, j+i+1) )
                        continue;

                    // assume optimized transitions
                    if ( a.accept != b.accept || a.transitions.length != b.transitions.length ) {
                        diff.set(i, j+i+1);
                        new_diff = true;
                        continue;
                    }

                    if ( a.accept ) // b accepts too
                    {
                        // assume sorted finishers
                        if ( a.finishers.length != b.finishers.length ) {
                            diff.set(i, j+i+1);
                            new_diff = true;
                            continue;
                        }
                        foreach ( k, cmd; a.finishers )
                        {
                            if ( cmd != b.finishers[k] ) {
                                diff.set(i, j+i+1);
                                new_diff = true;
                                continue Linner;
                            }
                        }
                    }

                    Ltrans: foreach ( ta; a.transitions )
                    {
                        foreach ( tb; b.transitions )
                        {
                            if ( ta.predicate.intersects(tb.predicate) )
                            {
                                if ( diff(ta.target.index, tb.target.index) ) {
                                    diff.set(i, j+i+1);
                                    new_diff = true;
                                    continue Linner;
                                }
                                // assume sorted commands
                                if ( ta.commands.length != tb.commands.length ) {
                                    diff.set(i, j+i+1);
                                    new_diff = true;
                                    continue Linner;
                                }
                                foreach ( k, cmd; ta.commands )
                                {
                                    if ( cmd != tb.commands[k] ) {
                                        diff.set(i, j+i+1);
                                        new_diff = true;
                                        continue Linner;
                                    }
                                }
                                continue Ltrans;
                            }
                        }

                        diff.set(i, j+i+1);
                        new_diff = true;
                        continue Linner;
                    }

                }
            }
        }

        foreach ( i, a; states[0 .. $-1] )
        {
            foreach ( j, b; states[i+1 .. $] )
            {
                if ( !diff(i, j+i+1) )
                {
                    debug(tdfa) Stdout.formatln("State {} == {}", i, j+i+1);
                    // remap b to a
                    foreach ( k, c; states )
                    {
                        foreach ( t; c.transitions )
                        {
                            if ( t.target.index == j+i+1 )
                                t.target = a;
                        }
                    }
                }
            }
        }

    }
}
import tango.text.Util;

/**************************************************************************************************
    Regular expression compiler and interpreter.
**************************************************************************************************/
class RegExpT(char_t)
{
    alias TDFA!(dchar)      tdfa_t;
    alias TNFA!(dchar)      tnfa_t;
    alias CharClass!(dchar) charclass_t;
    alias Predicate!(dchar) predicate_t;

    /**********************************************************************************************
        Construct a RegExpT object.
        Params:
            pattern = Regular expression.
        Throws: RegExpException if there are any compilation errors.
        Example:
            Declare two variables and assign to them a Regex object:
            ---
            auto r = new Regex("pattern");
            auto s = new Regex(r"p[1-5]\s*");
            ---
    **********************************************************************************************/
    this(const(char_t)[] pattern, const(char_t)[] attributes=null)
    {
        this(pattern, false, true);
    }

    /** ditto */
    this(const(char_t)[] pattern, bool swapMBS, bool unanchored, bool printNFA=false)
    {
        pattern_ = pattern;

        debug(TangoRegex) {}
        else {
            static if ( is(char_t == dchar) ) {
                scope tnfa_t tnfa_ = new tnfa_t(pattern_);
            }
            else {
                scope tnfa_t tnfa_ = new tnfa_t(tango.text.convert.Utf.toString32(pattern_));
            }
        }
       
        tnfa_.swapMatchingBracketSyntax = swapMBS;
        tnfa_.parse(unanchored);
        if ( printNFA ) {
            debug(TangoRegex) Stdout.formatln("\nTNFA:");
            debug(TangoRegex) tnfa_.print;
        }
        tdfa_ = new tdfa_t(tnfa_);
        registers_.length = tdfa_.num_regs();
    }

    /**********************************************************************************************
        Generate instance of Regex.
        Params:
            pattern = Regular expression.
        Throws: RegExpException if there are any compilation errors.
        Example:
            Declare two variables and assign to them a Regex object:
            ---
            auto r = Regex("pattern");
            auto s = Regex(r"p[1-5]\s*");
            ---
    **********************************************************************************************/
    static RegExpT!(char_t) opCall(const(char_t)[] pattern, const(char_t)[] attributes = null)
    {
        return new RegExpT!(char_t)(pattern, attributes);
    }

    /**********************************************************************************************
        Set up for start of foreach loop.
        Returns:    Instance of RegExpT set up to search input.
        Example:
            ---
            import tango.io.Stdout;
            import tango.text.Regex;

            void main()
            {
                foreach(m; Regex("ab").search("qwerabcabcababqwer"))
                    Stdout.formatln("{}[{}]{}", m.pre, m.match(0), m.post);
            }
            // Prints:
            // qwer[ab]cabcababqwer
            // qwerabc[ab]cababqwer
            // qwerabcabc[ab]abqwer
            // qwerabcabcab[ab]qwer
            ---
    **********************************************************************************************/
    public RegExpT!(char_t) search(const(char_t)[] input)
    {
        input_ = input;
        next_start_ = 0;
        last_start_ = 0;
        return this;
    }

    /** ditto */
    public int opApply(int delegate(RegExpT!(char_t)) dg)
    {
        int result;
        while ( !result && test() )
            result = dg(this);
        return result;
    }

    /**********************************************************************************************
        Search input for match.
        Returns: false for no match, true for match
    **********************************************************************************************/
    bool test(const(char_t)[] input)
    {
        this.input_ = input;
        next_start_ = 0;
        last_start_ = 0;
        return test();
    }

    /**********************************************************************************************
        Pick up where last test(input) or test() left off, and search again.
        Returns: false for no match, true for match
    **********************************************************************************************/
    bool test()
    {
        // initialize registers
        assert(registers_.length == tdfa_.num_regs());
        registers_[0..$] = -1;
        foreach ( cmd; tdfa_.initializer ) {
            assert(cmd.src == tdfa_.CURRENT_POSITION_REGISTER);
            registers_[cmd.dst] = 0;
        }

        tdfa_t.Transition* tp, tp_end;

        // DFA execution
        auto inp = input_[next_start_ .. $];
        auto s = tdfa_.start;

        debug(TangoRegex) Stdout.formatln("{}{}: {}", s.accept?"*":" ", s.index, inp);
        LmainLoop: for ( size_t p, next_p; p < inp.length; )
        {
        Lread_char:
            dchar c = cast(dchar)inp[p];
            if ( c & 0x80 )
                c = decode(inp, next_p);
            else
                next_p = p+1;

        Lprocess_char:
            debug(TangoRegex) Stdout.formatln("{} (0x{:x})", c, cast(int)c);

            tdfa_t.Transition t = void;
            switch ( s.mode )
            {
                case s.Mode.LOOKUP:
                    if ( c < s.LOOKUP_LENGTH )
                    {
                        debug(TangoRegex) Stdout.formatln("lookup");
                        auto i = s.lookup[c];
                        if ( i == s.INVALID_STATE )
                            break LmainLoop;
                        t = s.transitions[ i ];
                        if ( t.predicate.type != t.predicate.Type.consume )
                            goto Lno_consume;
                        goto Lconsume;
                    }
                    break LmainLoop;

                case s.Mode.MIXED:
                    if ( c < s.LOOKUP_LENGTH )
                    {
                        debug(TangoRegex) Stdout.formatln("mixed");
                        auto i = s.lookup[c];
                        if ( i == s.INVALID_STATE )
                            break;
                        t = s.transitions[ i ];
                        if ( t.predicate.type != t.predicate.Type.consume )
                            goto Lno_consume;
                        goto Lconsume;
                    }
                    break;

                case s.Mode.GENERIC:
                default:
                    break;
            }

            Ltrans_loop: for ( tp = &s.generic_transitions[0], tp_end = tp+s.generic_transitions.length;
                tp < tp_end; ++tp )
            {
                t = *tp;
                switch ( t.predicate.mode )
                {
                    // single char
                    case predicate_t.MatchMode.single_char:
                        debug(TangoRegex) Stdout.formatln("single char 0x{:x} == 0x{:x}", cast(int)c, cast(int)t.predicate.data_chr);
                        if ( c != t.predicate.data_chr )
                            continue Ltrans_loop;
                        goto Lconsume;
                    case predicate_t.MatchMode.single_char_l:
                        debug(TangoRegex) Stdout.formatln("single char 0x{:x} == 0x{:x}", cast(int)c, cast(int)t.predicate.data_chr);
                        if ( c != t.predicate.data_chr )
                            continue Ltrans_loop;
                        goto Lno_consume;

                    // bitmap
                    case predicate_t.MatchMode.bitmap:
                        debug(TangoRegex) Stdout.formatln("bitmap {}\n{}", c, t.predicate.toString);
                        if ( c <= predicate_t.MAX_BITMAP_LENGTH && ( t.predicate.data_bmp[c/8] & (1 << (c&7)) ) )
                            goto Lconsume;
                        continue Ltrans_loop;
                    case predicate_t.MatchMode.bitmap_l:
                        debug(TangoRegex) Stdout.formatln("bitmap {}\n{}", c, t.predicate.toString);
                        if ( c <= predicate_t.MAX_BITMAP_LENGTH && ( t.predicate.data_bmp[c/8] & (1 << (c&7)) ) )
                            goto Lno_consume;
                        continue Ltrans_loop;

                    // string search
                    case predicate_t.MatchMode.string_search:
                        debug(TangoRegex) Stdout.formatln("string search {} in {}", c, t.predicate.data_str);
                        if ( indexOf(t.predicate.data_str.ptr, c, t.predicate.data_str.length) >= t.predicate.data_str.length )
                            continue Ltrans_loop;
                        goto Lconsume;
                    case predicate_t.MatchMode.string_search_l:
                        debug(TangoRegex) Stdout.formatln("string search {} in {}", c, t.predicate.data_str);
                        if ( indexOf(t.predicate.data_str.ptr, c, t.predicate.data_str.length) >= t.predicate.data_str.length )
                            continue Ltrans_loop;
                        goto Lno_consume;

                    // generic
                    case predicate_t.MatchMode.generic:
                        debug(TangoRegex) Stdout.formatln("generic {}\n{}", c, t.predicate.toString);
                        for ( auto cmp = t.predicate.data_str.ptr,
                            cmpend = cmp + t.predicate.data_str.length;
                            cmp < cmpend; ++cmp )
                        {
                            if ( c < *cmp ) {
                                ++cmp;
                                continue;
                            }
                            ++cmp;
                            if ( c <= *cmp )
                                goto Lconsume;
                        }
                        continue Ltrans_loop;
                    case predicate_t.MatchMode.generic_l:
                        debug(TangoRegex) Stdout.formatln("generic {}\n{}", c, t.predicate.toString);
                        for ( auto cmp = t.predicate.data_str.ptr,
                            cmpend = cmp + t.predicate.data_str.length;
                            cmp < cmpend; ++cmp )
                        {
                            if ( c < *cmp ) {
                                ++cmp;
                                continue;
                            }
                            ++cmp;
                            if ( c <= *cmp )
                                goto Lno_consume;
                        }
                        continue Ltrans_loop;

                    default:
                        assert(0);
                }

            Lconsume:
                p = next_p;
            Lno_consume:

                s = t.target;
                debug(TangoRegex) Stdout.formatln("{}{}: {}", s.accept?"*":" ", s.index, inp[p..$]);
                debug(TangoRegex) Stdout.formatln("{} commands", t.commands.length);

                foreach ( cmd; t.commands )
                {
                    if ( cmd.src == tdfa_.CURRENT_POSITION_REGISTER )
                        registers_[cmd.dst] = cast(int)p;
                    else
                        registers_[cmd.dst] = registers_[cmd.src];
                }

                if (s.accept && s.reluctant)
                    // Don't continue matching, the current find should be correct
                    goto Laccept;

                // if all input was consumed and we do not already accept, try to
                // add an explicit string/line end
                if ( p >= inp.length )
                {
                    if ( s.accept || c == 0 )
                        break;
                    c = 0;
                    goto Lprocess_char;
                }
                goto Lread_char;
            }
            // no applicable transition
            break;
        }

        if ( s.accept )
        {
        Laccept:
            foreach ( cmd; s.finishers ) {
                assert(cmd.src != tdfa_.CURRENT_POSITION_REGISTER);
                registers_[cmd.dst] = registers_[cmd.src];
            }
            if ( registers_.length > 1 && registers_[1] >= 0 ) {
                last_start_ = next_start_;
                next_start_ += registers_[1];
            }
            return true;
        }

        return false;
    }

    /**********************************************************************************************
        Return submatch with the given index.
        Params:
            index = 0 returns whole match, index > 0 returns submatch of bracket #index
        Returns:
            Slice of input for the requested submatch, or null if no such submatch exists.
    **********************************************************************************************/
    const(char_t)[] match(uint index)
    {
        if ( index > tdfa_.num_tags )
            return null;
        int start   = cast(int)last_start_+registers_[index*2],
            end     = cast(int)last_start_+registers_[index*2+1];
        if ( start >= 0 && start < end && end <= input_.length )
            return input_[start .. end];
        return null;
    }

    /** ditto */
    const(char_t)[] opIndex(uint index)
    {
        return match(index);
    }

    /**********************************************************************************************
        Return the slice of the input that precedes the matched substring.
        If no match was found, null is returned.
    **********************************************************************************************/
    const(char_t)[] pre()
    {
        auto start = registers_[0];
        if ( start < 0 )
            return null;
        return input_[0 .. last_start_+start];
    }

    /**********************************************************************************************
        Return the slice of the input that follows the matched substring.
        If no match was found, the whole slice of the input that was processed in the last test.
    **********************************************************************************************/
    const(char_t)[] post()
    {
        if ( registers_[1] >= 0 )
            return input_[next_start_ .. $];
        return input_[last_start_ .. $];
    }

    /**********************************************************************************************
        Splits the input at the matches of this regular expression into an array of slices.
        Example:
            ---
            import tango.io.Stdout;
            import tango.text.Regex;

            void main()
            {
                auto strs = Regex("ab").split("abcabcababqwer");
                foreach( s; strs )
                    Stdout.formatln("{}", s);
            }
            // Prints:
            // c
            // c
            // qwer
            ---
    **********************************************************************************************/
    inout(char_t)[][] split(inout(char_t)[] input)
    {
        auto res = new inout(char_t)[][PREALLOC];
        uint index;
        inout(char_t)[] tmp = input;

        foreach ( r; search(input) )
        {
            tmp = cast(inout(char_t)[])pre();
            res[index++] = tmp[last_start_ .. $];
            if ( index >= res.length )
                res.length = res.length*2;
            tmp = cast(inout(char_t)[])post();
        }

        res[index++] = tmp;
        res.length = index;
        return res;
    }

    /**********************************************************************************************
        Returns a copy of the input with all matches replaced by replacement.
    **********************************************************************************************/
    char_t[] replaceAll(const(char_t)[] input, const(char_t)[] replacement, char_t[] output_buffer=null)
    {
        const(char_t)[] tmp = input;
        if ( output_buffer.length <= 0 )
            output_buffer = new char_t[input.length+replacement.length];
        output_buffer.length = 0;

        foreach ( r; search(input) )
        {
            tmp = pre();
            if ( tmp.length > last_start_ )
                output_buffer ~= tmp[last_start_ .. $];
            output_buffer ~= replacement;
            tmp = post();
        }
        output_buffer ~= tmp;
        return output_buffer;
    }

    /**********************************************************************************************
        Returns a copy of the input with the last match replaced by replacement.
    **********************************************************************************************/
    char_t[] replaceLast(const(char_t)[] input, const(char_t)[] replacement, char_t[] output_buffer=null)
    {
        const(char_t)[] tmp_pre, tmp_post;
        if ( output_buffer.length <= 0 )
            output_buffer = new char_t[input.length+replacement.length];
        output_buffer.length = 0;

        foreach ( r; search(input) ) {
            tmp_pre = pre();
            tmp_post = post();
        }

        if ( tmp_pre !is null || tmp_post !is null ) {
            output_buffer ~= tmp_pre;
            output_buffer ~= replacement;
            output_buffer ~= tmp_post;
        }
        else
            output_buffer ~= input;

        return output_buffer;
    }

    /**********************************************************************************************
        Returns a copy of the input with the first match replaced by replacement.
    **********************************************************************************************/
    char_t[] replaceFirst(const(char_t)[] input, const(char_t)[] replacement, char_t[] output_buffer=null)
    {
        const(char_t)[] tmp = input;
        if ( output_buffer.length <= 0 )
            output_buffer = new char_t[input.length+replacement.length];
        output_buffer.length = 0;

        if ( test(input) )
        {
            tmp = pre();
            if ( tmp.length > last_start_ )
                output_buffer ~= tmp[last_start_ .. $];
            output_buffer ~= replacement;
            tmp = post();
        }
        output_buffer ~= tmp;
        return output_buffer;
    }

    /**********************************************************************************************
        Calls dg for each match and replaces it with dg's return value.
    **********************************************************************************************/
    char_t[] replaceAll(const(char_t)[] input, char_t[] delegate(RegExpT!(char_t)) dg, char_t[] output_buffer=null)
    {
        const(char_t)[]    tmp = input;
        if ( output_buffer.length <= 0 )
            output_buffer = new char_t[input.length];
        output_buffer.length = 0;

        foreach ( r; search(input) )
        {
            tmp = pre();
            if ( tmp.length > last_start_ )
                output_buffer ~= tmp[last_start_ .. $];
            output_buffer ~= dg(this);
            tmp = post();
        }
        output_buffer ~= tmp;
        return output_buffer;
    }

    /**********************************************************************************************
        Compiles the regular expression to D code.

        NOTE : Remember to import this module (tango.text.Regex) in the module where you put the
        generated D code.

    **********************************************************************************************/
    // TODO: input-end special case
    const(char)[] compileToD(const(char)[] func_name = "match", bool lexer=false)
    {
        const(char)[] code;
        const(char)[] str_type;
        static if ( is(char_t == char) )
            str_type = "const(char)[]";
        static if ( is(char_t == wchar) )
            str_type = "const(wchar)[]";
        static if ( is(char_t == dchar) )
            str_type = "const(dchar)[]";

        if ( lexer )
            code = Format.convert("// {}\nbool {}({} input, out uint token, out {} match", pattern_, func_name, str_type, str_type);
        else {
            code = Format.convert("// {}\nbool match({} input", pattern_, str_type);
            code ~= Format.convert(", ref {}[] groups", str_type);
        }
        code ~= Format.convert(")\n{{\n    uint s = {};", tdfa_.start.index);

        uint num_vars = tdfa_.num_regs();
        if ( num_vars > 0 )
        {
            if ( lexer )
                code ~= "\n    static int ";
            else
                code ~= "\n    int ";
            bool first = true;
            for ( int i = 0, used = 0; i < num_vars; ++i )
            {
                bool hasInit = false;
                foreach ( cmd; tdfa_.initializer )
                {
                    if ( cmd.dst == i ) {
                        hasInit = true;
                        break;
                    }
                }

                if ( first )
                    first = false;
                else
                    code ~= ", ";
                if ( used > 0 && used % 10 == 0 )
                    code ~= "\n        ";
                ++used;
                code ~= Format.convert("r{}", i);

                if ( hasInit )
                    code ~= "=0";
                else
                    code ~= "=-1";
            }
            code ~= ";";
        }

        code ~= "\n\n    for ( size_t p, next_p; p < input.length; )\n    {";
        code ~= "\n        dchar c = cast(dchar)input[p];\n        if ( c & 0x80 )\n            decode(input, next_p);";
        code ~= "\n        else\n            next_p = p+1;\n        switch ( s )\n        {";

        size_t[] finish_states;
        foreach ( s; tdfa_.states )
        {
            code ~= Format.convert("\n            case {}:", s.index);

            if ( s.accept )
            {
                finish_states ~= s.index;

                tdfa_t.State target;
                foreach ( t; s.transitions )
                {
                    if ( target is null )
                        target = t.target;
                    else if ( target !is t.target )
                    {
                        target = null;
                        break;
                    }
                }
                if ( target !is null && target is s )
                    s.transitions = null;
            }

            bool first_if=true;
            charclass_t cc, ccTest;

            tango.core.Array.sort(s.transitions);
            foreach ( t; s.transitions )
            {
                ccTest.add(t.predicate.getInput());
                ccTest.optimize();
                if ( t.predicate.getInput() < ccTest )
                    cc = t.predicate.getInput();
                else
                    cc = ccTest;

                if ( first_if ) {
                    code ~= "\n                if ( ";
                    first_if = false;
                }
                else
                    code ~= "\n                else if ( ";
                bool first_cond=true;
                foreach ( cr; cc.parts )
                {
                    if ( first_cond )
                        first_cond = false;
                    else
                        code ~= " || ";
                    if ( cr.l() == cr.r() )
                        code ~= Format.convert("c == 0x{:x}", cast(int)cr.l());
                    else
                        code ~= Format.convert("c >= 0x{:x} && c <= 0x{:x}", cast(int)cr.l(), cast(int)cr.r());
                }
                code ~= Format.convert(" ) {{\n                    s = {};", t.target.index);

                if ( t.predicate.type == typeof(t.predicate.type).consume )
                    code ~= "\n                    p = next_p;";
                foreach ( cmd; t.commands )
                    code ~= compileCommand(cmd, "                    ");
/*
                // if inp ends here and we do not already accept, try to add an explicit string/line end
                if ( p >= inp.length && !s.accept && c != 0 ) {
                    c = 0;
                    goto Lprocess_char;
                }
*/
                code ~= "\n                }";
            }

            if ( !first_if )
                code ~= Format.convert(
                    "\n                else\n                    {};\n                break;",
                    s.accept?Format.convert("goto finish{}", s.index):"return false"
                );
            else
                code ~= Format.convert("\n                {};", s.accept?Format.convert("goto finish{}", s.index):"return false");
        }

        // create finisher groups
        size_t[][size_t] finisherGroup;
        foreach ( fs; finish_states )
        {
            // check if finisher group with same commands exists
            bool haveFinisher = false;
            foreach ( fg; finisherGroup.keys )
            {
                bool equalCommands = false;
                if ( tdfa_.states[fs].finishers.length == tdfa_.states[fg].finishers.length )
                {
                    equalCommands = true;
                    foreach ( i, cmd; tdfa_.states[fs].finishers )
                    {
                        if ( cmd != tdfa_.states[fg].finishers[i] ) {
                            equalCommands = false;
                            break;
                        }
                    }
                }
                if ( equalCommands ) {
                    // use existing group for this state
                    finisherGroup[fg] ~= fs;
                    haveFinisher = true;
                    break;
                }
            }
            // create new group
            if ( !haveFinisher )
                finisherGroup[fs] ~= fs;
        }


        code ~= "\n            default:\n                assert(0);\n        }\n    }\n\n    switch ( s )\n    {";
        foreach ( group, states; finisherGroup )
        {
            foreach ( s; states )
                code ~= Format.convert("\n        case {}: finish{}:", s, s);

            foreach ( cmd; tdfa_.states[group].finishers )
            {
                if ( lexer )
                {
                    if ( tdfa_.states[group].finishers.length > 1 )
                        throw new RegExpException("Lexer error: more than one finisher in flm lexer!");
                    if ( cmd.dst % 2 == 0 || cmd.dst >= tdfa_.num_tags )
                        throw new RegExpException(Format.convert("Lexer error: unexpected dst register {} in flm lexer!", cmd.dst).idup);
                    code ~= Format.convert("\n            match = input[0 .. r{}];\n            token = {};", cmd.src, cmd.dst/2);
                }
                else
                    code ~= compileCommand(cmd, "            ");
            }

            code ~= "\n            break;";
        }
        code ~= "\n        default:\n            return false;\n    }\n";

        if ( !lexer )
        {
            code ~= Format.convert("\n    groups.length = {};", tdfa_.num_tags/2);
            for ( int i = 0; i < tdfa_.num_tags/2; ++i )
                code ~= Format.convert("\n    if ( r{} > -1 && r{} > -1 )\n        groups[{}] = input[r{} .. r{}];", 2*i, 2*i+1, i, 2*i, 2*i+1);
        }

        code ~= "\n    return true;\n}";
        return code;
    }

    /*********************************************************************************************
        Get the pattern with which this regex was constructed.
    **********************************************************************************************/
    public const(char_t)[] pattern()
    {
        return pattern_;
    }

    /*********************************************************************************************
        Get the tag count of this regex, representing the number of sub-matches.

        This value is the max valid value for match/opIndex.
    **********************************************************************************************/
    uint tagCount()
    {
        return tdfa_.num_tags;
    }

    int[]       registers_;
    size_t      next_start_,
                last_start_;

    debug(TangoRegex) tnfa_t tnfa_;
    tdfa_t      tdfa_;
private:
    enum int           PREALLOC = 16;
    const(char_t)[]    input_,
                       pattern_;

    const(char)[] compileCommand(tdfa_t.Command cmd, const(char)[] indent)
    {
        const(char)[]  code,
                dst;
        code ~= Format.convert("\n{}r{} = ", indent, cmd.dst);
        if ( cmd.src == tdfa_.CURRENT_POSITION_REGISTER )
            code ~= "p;";
        else
            code ~= Format.convert("r{};", cmd.src);
        return code;
    }
}

alias RegExpT!(char) Regex;

debug(utf) import tango.stdc.stdio;
// the following block is stolen from phobos.
// the copyright notice applies for this block only.
/*
 *  Copyright (C) 2003-2004 by Digital Mars, www.digitalmars.com
 *  Written by Walter Bright
 *
 *  This software is provided 'as-is', without any express or implied
 *  warranty. In no event will the authors be held liable for any damages
 *  arising from the use of this software.
 *
 *  Permission is granted to anyone to use this software for any purpose,
 *  including commercial applications, and to alter it and redistribute it
 *  freely, subject to the following restrictions:
 *
 *  o  The origin of this software must not be misrepresented; you must not
 *     claim that you wrote the original software. If you use this software
 *     in a product, an acknowledgment in the product documentation would be
 *     appreciated but is not required.
 *  o  Altered source versions must be plainly marked as such, and must not
 *     be misrepresented as being the original software.
 *  o  This notice may not be removed or altered from any source
 *     distribution.
 */

class UtfException : Exception
{
    size_t idx; /// index in string of where error occurred

    this(immutable(char)[] s, size_t i)
    {
        idx = i;
        super(s);
    }
}

bool isValidDchar(dchar c)
{
    /* Note: FFFE and FFFF are specifically permitted by the
     * Unicode standard for application internal use, but are not
     * allowed for interchange.
     * (thanks to Arcane Jill)
     */

    return c < 0xD800 ||
        (c > 0xDFFF && c <= 0x10FFFF /*&& c != 0xFFFE && c != 0xFFFF*/);
}

/* *************
 * Decodes and returns character starting at s[idx]. idx is advanced past the
 * decoded character. If the character is not well formed, a UtfException is
 * thrown and idx remains unchanged.
 */

dchar decode(const(char)[] s, ref size_t idx)
    {
        size_t len = s.length;
        dchar V;
        size_t i = idx;
        char u = s[i];

        if (u & 0x80)
        {   uint n;
            char u2;

            /* The following encodings are valid, except for the 5 and 6 byte
             * combinations:
             *  0xxxxxxx
             *  110xxxxx 10xxxxxx
             *  1110xxxx 10xxxxxx 10xxxxxx
             *  11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
             *  111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
             *  1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
             */
            for (n = 1; ; n++)
            {
                if (n > 4)
                    goto Lerr;          // only do the first 4 of 6 encodings
                if (((u << n) & 0x80) == 0)
                {
                    if (n == 1)
                        goto Lerr;
                    break;
                }
            }

            // Pick off (7 - n) significant bits of B from first byte of octet
            V = cast(dchar)(u & ((1 << (7 - n)) - 1));

            if (i + (n - 1) >= len)
                goto Lerr;                      // off end of string

            /* The following combinations are overlong, and illegal:
             *  1100000x (10xxxxxx)
             *  11100000 100xxxxx (10xxxxxx)
             *  11110000 1000xxxx (10xxxxxx 10xxxxxx)
             *  11111000 10000xxx (10xxxxxx 10xxxxxx 10xxxxxx)
             *  11111100 100000xx (10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx)
             */
            u2 = s[i + 1];
            if ((u & 0xFE) == 0xC0 ||
                (u == 0xE0 && (u2 & 0xE0) == 0x80) ||
                (u == 0xF0 && (u2 & 0xF0) == 0x80) ||
                (u == 0xF8 && (u2 & 0xF8) == 0x80) ||
                (u == 0xFC && (u2 & 0xFC) == 0x80))
                goto Lerr;                      // overlong combination

            for (uint j = 1; j != n; j++)
            {
                u = s[i + j];
                if ((u & 0xC0) != 0x80)
                    goto Lerr;                  // trailing bytes are 10xxxxxx
                V = (V << 6) | (u & 0x3F);
            }
            if (!isValidDchar(V))
                goto Lerr;
            i += n;
        }
        else
        {
            V = cast(dchar) u;
            i++;
        }

        idx = i;
        return V;

      Lerr:
        throw new Exception("invalid UTF-8 sequence");
    }

/*  ditto */

dchar decode(const(wchar)[] s, ref size_t idx)
    in
    {
        assert(idx >= 0 && idx < s.length);
    }
    out (result)
    {
        assert(isValidDchar(result));
    }
    body
    {
        const(char)[] msg;
        dchar V;
        size_t i = idx;
        uint u = s[i];

        if (u & ~0x7F)
        {   if (u >= 0xD800 && u <= 0xDBFF)
            {   uint u2;

                if (i + 1 == s.length)
                {   msg = "surrogate UTF-16 high value past end of string";
                    goto Lerr;
                }
                u2 = s[i + 1];
                if (u2 < 0xDC00 || u2 > 0xDFFF)
                {   msg = "surrogate UTF-16 low value out of range";
                    goto Lerr;
                }
                u = ((u - 0xD7C0) << 10) + (u2 - 0xDC00);
                i += 2;
            }
            else if (u >= 0xDC00 && u <= 0xDFFF)
            {   msg = "unpaired surrogate UTF-16 value";
                goto Lerr;
            }
            else if (u == 0xFFFE || u == 0xFFFF)
            {   msg = "illegal UTF-16 value";
                goto Lerr;
            }
            else
                i++;
        }
        else
        {
            i++;
        }

        idx = i;
        return cast(dchar)u;

      Lerr:
        throw new UtfException(msg.idup, i);
    }

/*  ditto */

dchar decode(const(dchar)[] s, ref size_t idx)
    in
    {
        assert(idx >= 0 && idx < s.length);
    }
    body
    {
        size_t i = idx;
        dchar c = s[i];

        if (!isValidDchar(c))
            goto Lerr;
        idx = i + 1;
        return c;

      Lerr:
        throw new UtfException("5invalid UTF-32 value", i);
    }



/* =================== Encode ======================= */

/* *****************************
 * Encodes character c and appends it to array s[].
 */

void encode(ref const(char)[] s, dchar c)
    in
    {
        assert(isValidDchar(c));
    }
    body
    {
        const(char)[] r = s;

        if (c <= 0x7F)
        {
            r ~= cast(char) c;
        }
        else
        {
            char[4] buf;
            uint L;

            if (c <= 0x7FF)
            {
                buf[0] = cast(char)(0xC0 | (c >> 6));
                buf[1] = cast(char)(0x80 | (c & 0x3F));
                L = 2;
            }
            else if (c <= 0xFFFF)
            {
                buf[0] = cast(char)(0xE0 | (c >> 12));
                buf[1] = cast(char)(0x80 | ((c >> 6) & 0x3F));
                buf[2] = cast(char)(0x80 | (c & 0x3F));
                L = 3;
            }
            else if (c <= 0x10FFFF)
            {
                buf[0] = cast(char)(0xF0 | (c >> 18));
                buf[1] = cast(char)(0x80 | ((c >> 12) & 0x3F));
                buf[2] = cast(char)(0x80 | ((c >> 6) & 0x3F));
                buf[3] = cast(char)(0x80 | (c & 0x3F));
                L = 4;
            }
            else
            {
                assert(0);
            }
            r ~= buf[0 .. L];
        }
        s = r;
    }

/*  ditto */

void encode(ref const(wchar)[] s, dchar c)
    in
    {
        assert(isValidDchar(c));
    }
    body
    {
        const(wchar)[] r = s;

        if (c <= 0xFFFF)
        {
            r ~= cast(wchar) c;
        }
        else
        {
            wchar[2] buf;

            buf[0] = cast(wchar) ((((c - 0x10000) >> 10) & 0x3FF) + 0xD800);
            buf[1] = cast(wchar) (((c - 0x10000) & 0x3FF) + 0xDC00);
            r ~= buf;
        }
        s = r;
    }

/*  ditto */

void encode(ref const(dchar)[] s, dchar c)
    in
    {
        assert(isValidDchar(c));
    }
    body
    {
        s ~= c;
    }

debug(UnitTest)
{
    unittest
    {
        // match a fixed string
        Regex r;
        r = new Regex("abc");
        assert(r.test("abc123"));
        assert(r.test("feabc123"));
        assert(!r.test("abe123"));

        // match a non-ASCII string
        r = new Regex("☃");
        assert(r.test("a☃c123"));

        // capture
        r = new Regex("☃(c)");
        assert(r.test("a☃c123"));
        assert(r.match(1) == "c");

        // dot
        r = new Regex("..");
        assert(r.test("a☃c123"));

        // dot capture
        r = new Regex("(..)");
        assert(r.test("a☃c123"));
        assert(r.match(1) == "a☃");



        // two captures
        r = new Regex("(.)(.)");
        assert(r.test("a☃c123"));
        assert(r.match(1) == "a");
        assert(r.match(2) == "☃");

        // multiple letters
        r = new Regex(".*(e+).*");
        assert(r.test("aeeeeec123"));
        assert(r.match(1) == "eeeee", "Expected: eeeee Got: " ~ r.match(1));
        assert(!r.test("abfua"));

        // multiple snowmen
        r = new Regex(".*(☃+).*");
        assert(r.test("a☃☃☃☃☃c123"));
        assert(r.match(1) == "☃☃☃☃☃", "Expected: ☃☃☃☃☃ Got: " ~ r.match(1));
        assert(!r.test("abeua"));

        /**
           "*" quantifier bug. In "(☃+)" pattern test ((r.match(0) == "☃☃☃☃☃")&&(r.match(1) == "☃☃☃☃☃")) has passed.
        */
        // multiple snowmen
        r = new Regex("(☃+)");
        assert(r.test("a☃☃☃☃☃c123"));
        assert(r.match(0) == "☃☃☃☃☃");
        assert(r.match(1) == "☃☃☃☃☃");
        assert(r.test("a☃beua"));
    }

    debug(RegexTestOnly)
    {
        import tango.io.Stdout;
        void main() { Stdout("All tests pass").newline(); }
    }
}