regexp.cpp

Go to the documentation of this file.

#ifdef HAVE_CONFIG_H
#include <../../nrnconf.h>
#endif
/*
 * Copyright (c) 1987, 1988, 1989, 1990, 1991 Stanford University
 * Copyright (c) 1991 Silicon Graphics, Inc.
 *
 * Permission to use, copy, modify, distribute, and sell this software and 
 * its documentation for any purpose is hereby granted without fee, provided
 * that (i) the above copyright notices and this permission notice appear in
 * all copies of the software and related documentation, and (ii) the names of
 * Stanford and Silicon Graphics may not be used in any advertising or
 * publicity relating to the software without the specific, prior written
 * permission of Stanford and Silicon Graphics.
 * 
 * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND, 
 * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY 
 * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.  
 *
 * IN NO EVENT SHALL STANFORD OR SILICON GRAPHICS BE LIABLE FOR
 * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
 * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
 * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF 
 * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE 
 * OF THIS SOFTWARE.
 */

/*
 * Regexp - regular expression searching
 */

#include <InterViews/regexp.h>
#include <ivstream.h>
#include <string.h>

/*
 * This version is based on the Henry Spencers public domain reimplementation
 * of the regular expression matching subroutines.  They are included as
 * static subroutines after the externally accessible routines.
 */

/*
 * Forward declarations for regcomp()'s friends.
 */
static regexp* regcomp(const char* exp);
static char* reg(int paren, int* flagp);
static char* regbranch(int* flagp);
static char* regpiece(int* flagp);
static char* regatom(int* flagp);
static char* regnode(char op);
static char* regnext(char* p);
static void regc(char b);
static void reginsert(char op, char* opnd);
static void regtail(char* p, char* val);
static void regoptail(char* p, char* val);
static void regerror(const char* s);
static int regexec(regexp* prog, char* string);
static int regtry(regexp* prog, char* string);
static int regmatch(char* prog);
static int regrepeat(char* p);


inline char *
FindNewline(char* s) {
    return strchr(s, '\n');
}

inline char *
NextLine(char* s) {
    char* newstart;

    if ((newstart = FindNewline(s)) != nil)
    newstart++;
    return newstart;
}

Regexp::Regexp (const char* pat) {
    int length = strlen(pat);
    pattern_ = new char[length+1];
    strncpy(pattern_, pat, length);
    pattern_[length] = '\0';
    c_pattern = regcomp(pattern_);
    if (!c_pattern) {
        delete [] pattern_;
        pattern_ = nil;
    }
}

Regexp::Regexp (const char* pat, int length) {
    pattern_ = new char[length+1];
    strncpy(pattern_, pat, length);
    pattern_[length] = '\0';
    c_pattern = regcomp(pattern_);
    if (!c_pattern) {
        delete [] pattern_;
        pattern_ = nil;
    }
}

Regexp::~Regexp () {
    if (pattern_) {
        delete [] pattern_;
    }
    if (c_pattern) {
        delete [] c_pattern;
    }
}

const char* Regexp::pattern() const { return pattern_; }

int Regexp::Search (const char* text, int length, int index, int range) {
    bool forwardSearch;
    bool frontAnchored;
    bool endAnchored;
    char* searchStart;
    char* searchLimit;
    char* endOfLine = nil;
    char* lastMatch = nil;
    char csave;

    /*
     * A small sanity check.  Otherwise length is unused in this function.
     * This is really what the logic embedded in the old version of this
     * routine enforced.
     */
    if (index + range > length) {
    range = length - index;
    if (range < 0)
        return -1;
    }

    if (c_pattern == nil) {
        return -1;
    }

    c_pattern->startp[0] = nil;

    if (range < 0) {
    forwardSearch = false;
    searchLimit = (char *) text + index;
    searchStart = (char *) searchLimit + range; /* range is negative */
    } else {
    forwardSearch = true;
    searchStart = (char *) text + index;
    searchLimit = (char *) searchStart + range;
    }

    /* Mark end of text string so search will stop */
    char save = *searchLimit;
    *searchLimit = '\0';

    frontAnchored = pattern_[0] == '^';
    endAnchored = pattern_[strlen(pattern_)-1] == '$';
    if (frontAnchored && (searchStart != text || searchStart[-1] == '\n')) {
    searchStart = NextLine(searchStart);
    }

    while (searchStart && searchStart < searchLimit) {
    int result;

    if (endAnchored && (endOfLine = FindNewline(searchStart)) != nil) {
        csave = *endOfLine;
        *endOfLine = '\0';
    }

    result = regexec(c_pattern, searchStart);

    if (endOfLine)
        *endOfLine = csave;

    if (result) {
        /* Found a match */
        if (forwardSearch)
        break;  /* Done */
        else {
        lastMatch = c_pattern->startp[0];
        searchStart = c_pattern->endp[0];
        if (frontAnchored)
            searchStart = NextLine(searchStart);
        continue;
        }
    }
    /* Did not find a match */
    if (frontAnchored || endAnchored)
        searchStart = NextLine(searchStart);
    else
        break;
    }

    if (!forwardSearch && lastMatch) {
    if (endAnchored && (endOfLine = FindNewline(lastMatch)) != nil) {
        csave = *endOfLine;
        *endOfLine = '\0';
    }
    (void) regexec(c_pattern, lastMatch); /* Refill startp and endp */
    if (endOfLine)
        *endOfLine = csave;
    }

    *searchLimit = save;
    c_pattern->textStart = (char *) text;

    return c_pattern->startp[0] - c_pattern->textStart;
}

int Regexp::Match (const char* text, int length, int index) {

    if (c_pattern == nil)
    return -1;

    c_pattern->startp[0] = nil;

    char save = *(text+length);
    *(char*)(text+length) = '\0';

    c_pattern->textStart = (char *) text;
    (void) regexec(c_pattern, (char *) text + index);

    *(char*)(text+length) = save;

    if (c_pattern->startp[0] != nil)
        return c_pattern->endp[0] - c_pattern->startp[0];
    else
        return -1;
}

int Regexp::BeginningOfMatch (int subexp) {
    if (subexp < 0 || subexp > NSUBEXP ||
        c_pattern == nil || c_pattern->startp[0] == nil)
    return -1;
    return c_pattern->startp[subexp] - c_pattern->textStart;
}

int Regexp::EndOfMatch (int subexp) {
    if (subexp < 0 || subexp > NSUBEXP ||
        c_pattern == nil || c_pattern->startp[0] == nil)
    return -1;
    return c_pattern->endp[subexp] - c_pattern->textStart;
}

/*
 * regcomp and regexec
 *
 *  Copyright (c) 1986 by University of Toronto.
 *  Written by Henry Spencer.  Not derived from licensed software.
 *
 *  Permission is granted to anyone to use this software for any
 *  purpose on any computer system, and to redistribute it freely,
 *  subject to the following restrictions:
 *
 *  1. The author is not responsible for the consequences of use of
 *      this software, no matter how awful, even if they arise
 *      from defects in it.
 *
 *  2. The origin of this software must not be misrepresented, either
 *      by explicit claim or by omission.
 *
 *  3. Altered versions must be plainly marked as such, and must not
 *      be misrepresented as being the original software.
 *
 * Beware that some of this code is subtly aware of the way operator
 * precedence is structured in regular expressions.  Serious changes in
 * regular-expression syntax might require a total rethink.
 */

/*
 * The "internal use only" fields in regexp.h are present to pass info from
 * compile to execute that permits the execute phase to run lots faster on
 * simple cases.  They are:
 *
 * regstart char that must begin a match; '\0' if none obvious
 * reganch  is the match anchored (at beginning-of-line only)?
 * regmust  string (pointer into program) that match must include, or nil
 * regmlen  length of regmust string
 *
 * Regstart and reganch permit very fast decisions on suitable starting points
 * for a match, cutting down the work a lot.  Regmust permits fast rejection
 * of lines that cannot possibly match.  The regmust tests are costly enough
 * that regcomp() supplies a regmust only if the r.e. contains something
 * potentially expensive (at present, the only such thing detected is * or +
 * at the start of the r.e., which can involve a lot of backup).  Regmlen is
 * supplied because the test in regexec() needs it and regcomp() is computing
 * it anyway.
 */

/*
 * Structure for regexp "program".  This is essentially a linear encoding
 * of a nondeterministic finite-state machine (aka syntax charts or
 * "railroad normal form" in parsing technology).  Each node is an opcode
 * plus a "next" pointer, possibly plus an operand.  "Next" pointers of
 * all nodes except BRANCH implement concatenation; a "next" pointer with
 * a BRANCH on both ends of it is connecting two alternatives.  (Here we
 * have one of the subtle syntax dependencies:  an individual BRANCH (as
 * opposed to a collection of them) is never concatenated with anything
 * because of operator precedence.)  The operand of some types of node is
 * a literal string; for others, it is a node leading into a sub-FSM.  In
 * particular, the operand of a BRANCH node is the first node of the branch.
 * (NB this is *not* a tree structure:  the tail of the branch connects
 * to the thing following the set of BRANCHes.)  The opcodes are:
 */

/* definition   number  opnd?   meaning */
#define END 0   /* no   End of program. */
#define BOL 1   /* no   Match "" at beginning of line. */
#define EOL 2   /* no   Match "" at end of line. */
#define ANY 3   /* no   Match any one character. */
#define ANYOF   4   /* str  Match any character in this string. */
#define ANYBUT  5   /* str  Match any character not in this string. */
#define BRANCH  6   /* node Match this alternative, or the next... */
#define BACK    7   /* no   Match "", "next" ptr points backward. */
#define EXACTLY 8   /* str  Match this string. */
#define NOTHING 9   /* no   Match empty string. */
#define STAR    10  /* node Match this (simple) thing 0 or more times. */
#define PLUS    11  /* node Match this (simple) thing 1 or more times. */
#define OPEN    20  /* no   Mark this point in input as start of #n. */
            /*  OPEN+1 is number 1, etc. */
#define CLOSE   30  /* no   Analogous to OPEN. */

/*
 * Opcode notes:
 *
 * BRANCH   The set of branches constituting a single choice are hooked
 *      together with their "next" pointers, since precedence prevents
 *      anything being concatenated to any individual branch.  The
 *      "next" pointer of the last BRANCH in a choice points to the
 *      thing following the whole choice.  This is also where the
 *      final "next" pointer of each individual branch points; each
 *      branch starts with the operand node of a BRANCH node.
 *
 * BACK     Normal "next" pointers all implicitly point forward; BACK
 *      exists to make loop structures possible.
 *
 * STAR,PLUS    '?', and complex '*' and '+', are implemented as circular
 *      BRANCH structures using BACK.  Simple cases (one character
 *      per match) are implemented with STAR and PLUS for speed
 *      and to minimize recursive plunges.
 *
 * OPEN,CLOSE   ...are numbered at compile time.
 */

/*
 * A node is one char of opcode followed by two chars of "next" pointer.
 * "Next" pointers are stored as two 8-bit pieces, high order first.  The
 * value is a positive offset from the opcode of the node containing it.
 * An operand, if any, simply follows the node.  (Note that much of the
 * code generation knows about this implicit relationship.)
 *
 * Using two bytes for the "next" pointer is vast overkill for most things,
 * but allows patterns to get big without disasters.
 */
#define OP(p)   (*(p))
#define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377))
#define OPERAND(p)  ((p) + 3)

/*
 * Utility definitions.
 */
#ifndef RE_CHARBITS
#define RE_CHARBITS 0xff
#endif

#define UCHARAT(p)  ((int)*(p)&RE_CHARBITS)

#define FAIL(m) { regerror(m); return(nil); }
#define ISMULT(c)   ((c) == '*' || (c) == '+' || (c) == '?')
#define META    "^$.[()|?+*\\"

/*
 * Flags to be passed up and down.
 */
#define HASWIDTH    01  /* Known never to match null string. */
#define SIMPLE      02  /* Simple enough to be STAR/PLUS operand. */
#define SPSTART     04  /* Starts with * or +. */
#define WORST       0   /* Worst case. */

/*
 * Global work variables for regcomp().
 */
static const char *regparse;        /* Input-scan pointer. */
static int regnpar;     /* () count. */
static char regdummy;
static char *regcode;       /* Code-emit pointer; &regdummy = don't. */
static long regsize;        /* Code size. */

/*
 - regcomp - compile a regular expression into internal code
 *
 * We can't allocate space until we know how big the compiled form will be,
 * but we can't compile it (and thus know how big it is) until we've got a
 * place to put the code.  So we cheat:  we compile it twice, once with code
 * generation turned off and size counting turned on, and once "for real".
 * This also means that we don't allocate space until we are sure that the
 * thing really will compile successfully, and we never have to move the
 * code and thus invalidate pointers into it.  (Note that it has to be in
 * one piece because free() must be able to free it all.)
 *
 * Beware that the optimization-preparation code in here knows about some
 * of the structure of the compiled regexp.
 */
static regexp *
regcomp(const char* exp) {
    regexp *r;
    char *scan;
    char *longest;
    int len;
    int flags;

    if (exp == nil)
        FAIL("nil argument");

    /* First pass: determine size, legality. */
    regparse = exp;
    regnpar = 1;
    regsize = 0L;
    regcode = &regdummy;
    regc(REGEXP_MAGIC);
    if (reg(0, &flags) == nil)
        return(nil);

    /* Small enough for pointer-storage convention? */
    if (regsize >= 32767L)      /* Probably could be 65535L. */
        FAIL("regexp too big");

    /* Allocate space. */
    r = (regexp *) new char[sizeof(regexp) + (unsigned)regsize];

    /* Second pass: emit code. */
    regparse = exp;
    regnpar = 1;
    regcode = r->program;
    regc(REGEXP_MAGIC);
    if (reg(0, &flags) == nil) {
        delete [] r;
        return(nil);
    }

    /* Dig out information for optimizations. */
    r->regstart = '\0'; /* Worst-case defaults. */
    r->reganch = 0;
    r->regmust = nil;
    r->regmlen = 0;
    scan = r->program+1;            /* First BRANCH. */
    if (OP(regnext(scan)) == END) {     /* Only one top-level choice. */
        scan = OPERAND(scan);

        /* Starting-point info. */
        if (OP(scan) == EXACTLY)
            r->regstart = *OPERAND(scan);
        else if (OP(scan) == BOL)
            r->reganch++;

        /*
         * If there's something expensive in the r.e., find the
         * longest literal string that must appear and make it the
         * regmust.  Resolve ties in favor of later strings, since
         * the regstart check works with the beginning of the r.e.
         * and avoiding duplication strengthens checking.  Not a
         * strong reason, but sufficient in the absence of others.
         */
        if (flags&SPSTART) {
            longest = nil;
            len = 0;
            for (; scan != nil; scan = regnext(scan))
                if (OP(scan) == EXACTLY && strlen(OPERAND(scan)) >= len) {
                    longest = OPERAND(scan);
                    len = strlen(OPERAND(scan));
                }
            r->regmust = longest;
            r->regmlen = len;
        }
    }

    return(r);
}

/*
 - reg - regular expression, i.e. main body or parenthesized thing
 *
 * Caller must absorb opening parenthesis.
 *
 * Combining parenthesis handling with the base level of regular expression
 * is a trifle forced, but the need to tie the tails of the branches to what
 * follows makes it hard to avoid.
 */
static char *
reg(int paren, int* flagp) {
    char *ret;
    char *br;
    char *ender;
    int parno;
    int flags;

    *flagp = HASWIDTH;  /* Tentatively. */

    /* Make an OPEN node, if parenthesized. */
    if (paren) {
        if (regnpar >= NSUBEXP)
            FAIL("too many ()");
        parno = regnpar;
        regnpar++;
        ret = regnode(OPEN+parno);
    } else
        ret = nil;

    /* Pick up the branches, linking them together. */
    br = regbranch(&flags);
    if (br == nil)
        return(nil);
    if (ret != nil)
        regtail(ret, br);   /* OPEN -> first. */
    else
        ret = br;
    if (!(flags&HASWIDTH))
        *flagp &= ~HASWIDTH;
    *flagp |= flags&SPSTART;
    while (*regparse == '|') {
        regparse++;
        br = regbranch(&flags);
        if (br == nil)
            return(nil);
        regtail(ret, br);   /* BRANCH -> BRANCH. */
        if (!(flags&HASWIDTH))
            *flagp &= ~HASWIDTH;
        *flagp |= flags&SPSTART;
    }

    /* Make a closing node, and hook it on the end. */
    ender = regnode((paren) ? CLOSE+parno : END);   
    regtail(ret, ender);

    /* Hook the tails of the branches to the closing node. */
    for (br = ret; br != nil; br = regnext(br))
        regoptail(br, ender);

    /* Check for proper termination. */
    if (paren && *regparse++ != ')') {
        FAIL("unmatched ()");
    } else if (!paren && *regparse != '\0') {
        if (*regparse == ')') {
            FAIL("unmatched ()");
        } else
            FAIL("junk on end");    /* "Can't happen". */
        /* NOTREACHED */
    }

    return(ret);
}

/*
 - regbranch - one alternative of an | operator
 *
 * Implements the concatenation operator.
 */
static char *
regbranch(int* flagp) {
    char *ret;
    char *chain;
    char *latest;
    int flags;

    *flagp = WORST;     /* Tentatively. */

    ret = regnode(BRANCH);
    chain = nil;
    while (*regparse != '\0' && *regparse != '|') {
        if (*regparse == '\\' && regparse[1] == ')') {
            regparse++;
            break;
        }
        latest = regpiece(&flags);
        if (latest == nil)
            return(nil);
        *flagp |= flags&HASWIDTH;
        if (chain == nil)   /* First piece. */
            *flagp |= flags&SPSTART;
        else
            regtail(chain, latest);
        chain = latest;
    }
    if (chain == nil)   /* Loop ran zero times. */
        (void) regnode(NOTHING);

    return(ret);
}

/*
 - regpiece - something followed by possible [*+?]
 *
 * Note that the branching code sequences used for ? and the general cases
 * of * and + are somewhat optimized:  they use the same NOTHING node as
 * both the endmarker for their branch list and the body of the last branch.
 * It might seem that this node could be dispensed with entirely, but the
 * endmarker role is not redundant.
 */
static char *
regpiece(int* flagp) {
    char *ret;
    char op;
    char *next;
    int flags;

    ret = regatom(&flags);
    if (ret == nil)
        return(nil);

    op = *regparse;
    if (!ISMULT(op)) {
        *flagp = flags;
        return(ret);
    }

    if (!(flags&HASWIDTH) && op != '?')
        FAIL("*+ operand could be empty");
    *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);

    if (op == '*' && (flags&SIMPLE))
        reginsert(STAR, ret);
    else if (op == '*') {
        /* Emit x* as (x&|), where & means "self". */
        reginsert(BRANCH, ret);         /* Either x */
        regoptail(ret, regnode(BACK));      /* and loop */
        regoptail(ret, ret);            /* back */
        regtail(ret, regnode(BRANCH));      /* or */
        regtail(ret, regnode(NOTHING));     /* null. */
    } else if (op == '+' && (flags&SIMPLE))
        reginsert(PLUS, ret);
    else if (op == '+') {
        /* Emit x+ as x(&|), where & means "self". */
        next = regnode(BRANCH);         /* Either */
        regtail(ret, next);
        regtail(regnode(BACK), ret);        /* loop back */
        regtail(next, regnode(BRANCH));     /* or */
        regtail(ret, regnode(NOTHING));     /* null. */
    } else if (op == '?') {
        /* Emit x? as (x|) */
        reginsert(BRANCH, ret);         /* Either x */
        regtail(ret, regnode(BRANCH));      /* or */
        next = regnode(NOTHING);        /* null. */
        regtail(ret, next);
        regoptail(ret, next);
    }
    regparse++;
    if (ISMULT(*regparse))
        FAIL("nested *?+");

    return(ret);
}

/*
 - regatom - the lowest level
 *
 * Optimization:  gobbles an entire sequence of ordinary characters so that
 * it can turn them into a single node, which is smaller to store and
 * faster to run.  Backslashed characters are exceptions, each becoming a
 * separate node; the code is simpler that way and it's not worth fixing.
 */
static char *
regatom(int* flagp) {
    char *ret;
    int flags;

    *flagp = WORST;     /* Tentatively. */

    switch (*regparse++) {
    case '^':
        ret = regnode(BOL);
        break;
    case '$':
        ret = regnode(EOL);
        break;
    case '.':
        ret = regnode(ANY);
        *flagp |= HASWIDTH|SIMPLE;
        break;
    case '[': {
            int classbeg;
            int classend;

            if (*regparse == '^') { /* Complement of range. */
                ret = regnode(ANYBUT);
                regparse++;
            } else
                ret = regnode(ANYOF);
            if (*regparse == ']' || *regparse == '-')
                regc(*regparse++);
            while (*regparse != '\0' && *regparse != ']') {
                if (*regparse == '-') {
                    regparse++;
                    if (*regparse == ']' || *regparse == '\0')
                        regc('-');
                    else {
                        classbeg = UCHARAT(regparse-2)+1;
                        classend = UCHARAT(regparse);
                        if (classbeg > classend+1)
                            FAIL("invalid [] range");
                        for (; classbeg <= classend; classbeg++)
                            regc(classbeg);
                        regparse++;
                    }
                } else
                    regc(*regparse++);
            }
            regc('\0');
            if (*regparse != ']')
                FAIL("unmatched []");
            regparse++;
            *flagp |= HASWIDTH|SIMPLE;
        }
        break;
    case '\0':
    case '|':
        FAIL("internal urp");   /* Supposed to be caught earlier. */
        break;
    case '?':
    case '+':
    case '*':
        FAIL("?+* follows nothing");
        break;
    case '\\':
        if (*regparse == '\0')
            FAIL("trailing \\");
        if (*regparse == '(') {
            regparse++;
            ret = reg(1, &flags);
            if (ret == nil)
                return(nil);
            *flagp |= flags&(HASWIDTH|SPSTART);
        } else {
            ret = regnode(EXACTLY);
            regc(*regparse++);
            regc('\0');
            *flagp |= HASWIDTH|SIMPLE;
        }
        break;
    default: {
            int len;
            char ender;

            regparse--;
            len = strcspn(regparse, META);
            if (len <= 0)
                FAIL("internal disaster");
            ender = *(regparse+len);
            if (len > 1 && ISMULT(ender))
                len--;      /* Back off clear of ?+* operand. */
            *flagp |= HASWIDTH;
            if (len == 1)
                *flagp |= SIMPLE;
            ret = regnode(EXACTLY);
            while (len > 0) {
                regc(*regparse++);
                len--;
            }
            regc('\0');
        }
        break;
    }

    return(ret);
}

/*
 - regnode - emit a node
 */
static char *           /* Location. */
regnode(char op) {
    char *ret;
    char *ptr;

    ret = regcode;
    if (ret == &regdummy) {
        regsize += 3;
        return(ret);
    }

    ptr = ret;
    *ptr++ = op;
    *ptr++ = '\0';      /* Null "next" pointer. */
    *ptr++ = '\0';
    regcode = ptr;

    return(ret);
}

/*
 - regc - emit (if appropriate) a byte of code
 */
static void
regc(char b) {
    if (regcode != &regdummy)
        *regcode++ = b;
    else
        regsize++;
}

/*
 - reginsert - insert an operator in front of already-emitted operand
 *
 * Means relocating the operand.
 */
static void
reginsert(char op, char* opnd) {
    char *src;
    char *dst;
    char *place;

    if (regcode == &regdummy) {
        regsize += 3;
        return;
    }

    src = regcode;
    regcode += 3;
    dst = regcode;
    while (src > opnd)
        *--dst = *--src;

    place = opnd;       /* Op node, where operand used to be. */
    *place++ = op;
    *place++ = '\0';
    *place++ = '\0';
}

/*
 - regtail - set the next-pointer at the end of a node chain
 */
static void
regtail(char* p, char* val) {
    char *scan;
    char *temp;
    int offset;

    if (p == &regdummy)
        return;

    /* Find last node. */
    scan = p;
    for (;;) {
        temp = regnext(scan);
        if (temp == nil)
            break;
        scan = temp;
    }

    if (OP(scan) == BACK)
        offset = scan - val;
    else
        offset = val - scan;
    *(scan+1) = (offset>>8)&0377;
    *(scan+2) = offset&0377;
}

/*
 - regoptail - regtail on operand of first argument; nop if operandless
 */
static void
regoptail(char* p, char* val) {
    /* "Operandless" and "op != BRANCH" are synonymous in practice. */
    if (p == nil || p == &regdummy || OP(p) != BRANCH)
        return;
    regtail(OPERAND(p), val);
}

/*
 * regexec and friends
 */

/*
 * Global work variables for regexec().
 */
static char *reginput;      /* String-input pointer. */
static char *regbol;        /* Beginning of input, for ^ check. */
static char **regstartp;    /* Pointer to startp array. */
static char **regendp;      /* Ditto for endp. */

/*
 - regexec - match a regexp against a string
 */
static int
regexec(regexp* prog, char* string) {
    char *s;

    /* Be paranoid... */
    if (prog == nil || string == nil) {
        regerror("nil parameter");
        return(0);
    }

    /* Check validity of program. */
    if (UCHARAT(prog->program) != REGEXP_MAGIC) {
        regerror("corrupted program");
        return(0);
    }

    /* If there is a "must appear" string, look for it. */
    if (prog->regmust != nil) {
        s = string;
        while ((s = strchr(s, prog->regmust[0])) != nil) {
            if (strncmp(s, prog->regmust, prog->regmlen) == 0)
                break;  /* Found it. */
            s++;
        }
        if (s == nil)   /* Not present. */
            return(0);
    }

    /* Mark beginning of line for ^ . */
    regbol = string;

    /* Simplest case:  anchored match need be tried only once. */
    if (prog->reganch)
        return(regtry(prog, string));

    /* Messy cases:  unanchored match. */
    s = string;
    if (prog->regstart != '\0')
        /* We know what char it must start with. */
        while ((s = strchr(s, prog->regstart)) != nil) {
            if (regtry(prog, s))
                return(1);
            s++;
        }
    else
        /* We don't -- general case. */
        do {
            if (regtry(prog, s))
                return(1);
        } while (*s++ != '\0');

    /* Failure. */
    return(0);
}

/*
 - regtry - try match at specific point
 */
static int          /* 0 failure, 1 success */
regtry(regexp* prog, char* string) {
    int i;
    char **sp;
    char **ep;

    reginput = string;
    regstartp = prog->startp;
    regendp = prog->endp;

    sp = prog->startp;
    ep = prog->endp;
    for (i = NSUBEXP; i > 0; i--) {
        *sp++ = nil;
        *ep++ = nil;
    }
    if (regmatch(prog->program + 1)) {
        prog->startp[0] = string;
        prog->endp[0] = reginput;
        return(1);
    } else
        return(0);
}

/*
 - regmatch - main matching routine
 *
 * Conceptually the strategy is simple:  check to see whether the current
 * node matches, call self recursively to see whether the rest matches,
 * and then act accordingly.  In practice we make some effort to avoid
 * recursion, in particular by going through "ordinary" nodes (that don't
 * need to know whether the rest of the match failed) by a loop instead of
 * by recursion.
 */
static int          /* 0 failure, 1 success */
regmatch(char* prog) {
    char *scan; /* Current node. */
    char *next;     /* Next node. */

    scan = prog;
    while (scan != nil) {
        next = regnext(scan);

        switch (OP(scan)) {
        case BOL:
            if (reginput != regbol)
                return(0);
            break;
        case EOL:
            if (*reginput != '\0')
                return(0);
            break;
        case ANY:
            if (*reginput == '\0')
                return(0);
            reginput++;
            break;
        case EXACTLY: {
                int len;
                char *opnd;

                opnd = OPERAND(scan);
                /* Inline the first character, for speed. */
                if (*opnd != *reginput)
                    return(0);
                len = strlen(opnd);
                if (len > 1 && strncmp(opnd, reginput, len) != 0)
                    return(0);
                reginput += len;
            }
            break;
        case ANYOF:
            if (*reginput == '\0')
                return(0);
            if (strchr(OPERAND(scan), *reginput) == nil)
                return(0);
            reginput++;
            break;
        case ANYBUT:
            if (*reginput == '\0')
                return(0);
            if (strchr(OPERAND(scan), *reginput) != nil)
                return(0);
            reginput++;
            break;
        case NOTHING:
            break;
        case BACK:
            break;
        case OPEN+1:
        case OPEN+2:
        case OPEN+3:
        case OPEN+4:
        case OPEN+5:
        case OPEN+6:
        case OPEN+7:
        case OPEN+8:
        case OPEN+9: {
                int no;
                char *save;

                no = OP(scan) - OPEN;
                save = reginput;

                if (regmatch(next)) {
                    /*
                     * Don't set startp if some later
                     * invocation of the same parentheses
                     * already has.
                     */
                    if (regstartp[no] == nil)
                        regstartp[no] = save;
                    return(1);
                } else
                    return(0);
            }
            break;
        case CLOSE+1:
        case CLOSE+2:
        case CLOSE+3:
        case CLOSE+4:
        case CLOSE+5:
        case CLOSE+6:
        case CLOSE+7:
        case CLOSE+8:
        case CLOSE+9: {
                int no;
                char *save;

                no = OP(scan) - CLOSE;
                save = reginput;

                if (regmatch(next)) {
                    /*
                     * Don't set endp if some later
                     * invocation of the same parentheses
                     * already has.
                     */
                    if (regendp[no] == nil)
                        regendp[no] = save;
                    return(1);
                } else
                    return(0);
            }
            break;
        case BRANCH: {
                char *save;

                if (OP(next) != BRANCH)     /* No choice. */
                    next = OPERAND(scan);   /* Avoid recursion. */
                else {
                    do {
                        save = reginput;
                        if (regmatch(OPERAND(scan)))
                            return(1);
                        reginput = save;
                        scan = regnext(scan);
                    } while (scan != nil && OP(scan) == BRANCH);
                    return(0);
                    /* NOTREACHED */
                }
            }
            break;
        case STAR:
        case PLUS: {
                char nextch;
                int no;
                char *save;
                int min;

                /*
                 * Lookahead to avoid useless match attempts
                 * when we know what character comes next.
                 */
                nextch = '\0';
                if (OP(next) == EXACTLY)
                    nextch = *OPERAND(next);
                min = (OP(scan) == STAR) ? 0 : 1;
                save = reginput;
                no = regrepeat(OPERAND(scan));
                while (no >= min) {
                    /* If it could work, try it. */
                    if (nextch == '\0' || *reginput == nextch)
                        if (regmatch(next))
                            return(1);
                    /* Couldn't or didn't -- back up. */
                    no--;
                    reginput = save + no;
                }
                return(0);
            }
            break;
        case END:
            return(1);  /* Success! */
        default:
            regerror("memory corruption");
            return(0);
        }

        scan = next;
    }

    /*
     * We get here only if there's trouble -- normally "case END" is
     * the terminating point.
     */
    regerror("corrupted pointers");
    return(0);
}

/*
 - regrepeat - repeatedly match something simple, report how many
 */
static int
regrepeat(char* p) {
    int count = 0;
    char *scan;
    char *opnd;

    scan = reginput;
    opnd = OPERAND(p);
    switch (OP(p)) {
    case ANY:
        count = strlen(scan);
        scan += count;
        break;
    case EXACTLY:
        while (*opnd == *scan) {
            count++;
            scan++;
        }
        break;
    case ANYOF:
        while (*scan != '\0' && strchr(opnd, *scan) != nil) {
            count++;
            scan++;
        }
        break;
    case ANYBUT:
        while (*scan != '\0' && strchr(opnd, *scan) == nil) {
            count++;
            scan++;
        }
        break;
    default:        /* Oh dear.  Called inappropriately. */
        regerror("internal foulup");
        count = 0;  /* Best compromise. */
        break;
    }
    reginput = scan;

    return(count);
}

/*
 - regnext - dig the "next" pointer out of a node
 */
static char *
regnext(char* p) {
    int offset;

    if (p == &regdummy)
        return(nil);

    offset = NEXT(p);
    if (offset == 0)
        return(nil);

    if (OP(p) == BACK)
        return(p-offset);
    else
        return(p+offset);
}

static void
regerror(const char* s) {
    std::cerr << "regexp: " << s << "\n";
}