Tcl Source Code

2013-03-06  Don Porter  <[email protected]>

	* generic/regc_nfa.c:	[Bugs 3604074,3606683] Rewrite of the
	* generic/regcomp.c:	fixempties() routine (and supporting
	routines) to completely eliminate the infinite loop hazard.
	Thanks to Tom Lane for the much improved solution.

2013-02-27  Jan Nijtmans  <[email protected]>

	* generic/regcomp.c:	[Bug 3606139]: missing error check allows
	* tests/regexp.test:    regexp to crash Tcl. Thanks to Tom Lane
	for providing the test-case and the patch.

2013-02-26  Jan Nijtmans  <[email protected]>

    victim->from = NULL;	/* precautions... */
    victim->to = NULL;
    victim->inchain = NULL;
    victim->outchain = NULL;
    victim->freechain = from->free;
    from->free = victim;
}

/*
 - hasnonemptyout - Does state have a non-EMPTY out arc?
 ^ static int hasnonemptyout(struct state *);
 */
static int
hasnonemptyout(
    struct state *s)
{
    struct arc *a;

    for (a = s->outs; a != NULL; a = a->outchain) {
	if (a->type != EMPTY) {
	    return 1;
	}
    }
    return 0;
}

/*
 - nonemptyouts - count non-EMPTY out arcs of a state
 ^ static int nonemptyouts(struct state *);
 */
static int
nonemptyouts(
    struct state *s)
{
    int n = 0;
    struct arc *a;

    for (a = s->outs; a != NULL; a = a->outchain) {
	if (a->type != EMPTY) {
	    n++;
	}
    }
    return n;
}

/*
 - nonemptyins - count non-EMPTY in arcs of a state
 ^ static int nonemptyins(struct state *);
 */
static int
nonemptyins(
    struct state *s)
{
    int n = 0;
    struct arc *a;

    for (a = s->ins; a != NULL; a = a->inchain) {
	if (a->type != EMPTY) {
	    n++;
	}
    }
    return n;
}

/*
 - findarc - find arc, if any, from given source with given type and color
 * If there is more than one such arc, the result is random.
 ^ static struct arc *findarc(struct state *, int, pcolor);
 */
static struct arc *

	freearc(nfa, a);
    }
    assert(oldState->nins == 0);
    assert(oldState->ins == NULL);
}

/*
 - copyins - copy in arcs of a state to another state
 * Either all arcs, or only non-empty ones as determined by all value.
 ^ static VOID copyins(struct nfa *, struct state *, struct state *, int);
 */
static void
copyins(
    struct nfa *nfa,
    struct state *oldState,
    struct state *newState,
    int all)
{
    struct arc *a;

    assert(oldState != newState);

    for (a=oldState->ins ; a!=NULL ; a=a->inchain) {
	if (all || a->type != EMPTY) {
	    cparc(nfa, a, a->from, newState);
	}
    }
}

/*
 - moveouts - move all out arcs of a state to another state
 ^ static VOID moveouts(struct nfa *, struct state *, struct state *);
 */

    while ((a = oldState->outs) != NULL) {
	cparc(nfa, a, newState, a->to);
	freearc(nfa, a);
    }
}

/*
 - copyouts - copy out arcs of a state to another state
 * Either all arcs, or only non-empty ones as determined by all value.
 ^ static VOID copyouts(struct nfa *, struct state *, struct state *, int);
 */
static void
copyouts(
    struct nfa *nfa,
    struct state *oldState,
    struct state *newState,
    int all)
{
    struct arc *a;

    assert(oldState != newState);

    for (a=oldState->outs ; a!=NULL ; a=a->outchain) {
	if (all || a->type != EMPTY) {
	    cparc(nfa, a, newState, a->to);
	}
    }
}

/*
 - cloneouts - copy out arcs of a state to another state pair, modifying type
 ^ static VOID cloneouts(struct nfa *, struct state *, struct state *,
 ^ 	struct state *, int);

     */

    if (from->nouts > 1) {
	s = newstate(nfa);
	if (NISERR()) {
	    return 0;
	}
	assert(to != from);		/* con is not an inarc */
	copyins(nfa, from, s, 1);	/* duplicate inarcs */
	cparc(nfa, con, s, to);		/* move constraint arc */
	freearc(nfa, con);
	from = s;
	con = from->outs;
    }
    assert(from->nouts == 1);

    /*

     */

    if (to->nins > 1) {
	s = newstate(nfa);
	if (NISERR()) {
	    return 0;
	}
	copyouts(nfa, to, s, 1);	/* duplicate outarcs */
	cparc(nfa, con, from, s);	/* move constraint */
	freearc(nfa, con);
	to = s;
	con = to->ins;
    }
    assert(to->nins == 1);

 */
static void
fixempties(
    struct nfa *nfa,
    FILE *f)			/* for debug output; NULL none */
{
    struct state *s;
    struct state *s2;
    struct state *nexts;
    struct arc *a;
    struct arc *nexta;


    /*
     * First, get rid of any states whose sole out-arc is an EMPTY,
     * since they're basically just aliases for their successor.  The
     * parsing algorithm creates enough of these that it's worth
     * special-casing this.
     */



    for (s = nfa->states; s != NULL && !NISERR(); s = nexts) {
	nexts = s->next;
	if (s->flag || s->nouts != 1) {
	    continue;
	}
	a = s->outs;
	assert(a != NULL && a->outchain == NULL);
	if (a->type != EMPTY) {
	    continue;
	}
	if (s != a->to) {
	    moveins(nfa, s, a->to);
	}
	dropstate(nfa, s);
    }

    /*

     * Similarly, get rid of any state with a single EMPTY in-arc, by
     * folding it into its predecessor.
     */
    for (s = nfa->states; s != NULL && !NISERR(); s = nexts) {
	nexts = s->next;
	/* Ensure tmp fields are clear for next step */
	assert(s->tmp = NULL);
	if (s->flag || s->nins != 1) {
	    continue;
	}
	a = s->ins;
	assert(a != NULL && a->inchain == NULL);
	if (a->type != EMPTY) {
	    continue;
	}
	if (s != a->from) {
	    moveouts(nfa, s, a->from);
	}
	dropstate(nfa, s);
    }

    /*





     * For each remaining NFA state, find all other states that are
     * reachable from it by a chain of one or more EMPTY arcs.  Then
     * generate new arcs that eliminate the need for each such chain.
     *



     * If we just do this straightforwardly, the algorithm gets slow in
     * complex graphs, because the same arcs get copied to all
     * intermediate states of an EMPTY chain, and then uselessly pushed
     * repeatedly to the chain's final state; we waste a lot of time in
     * newarc's duplicate checking.  To improve matters, we decree that
     * any state with only EMPTY out-arcs is "doomed" and will not be
     * part of the final NFA. That can be ensured by not adding any new

     * out-arcs to such a state. Having ensured that, we need not update
     * the state's in-arcs list either; all arcs that might have gotten
     * pushed forward to it will just get pushed directly to successor
     * states.  This eliminates most of the useless duplicate arcs.
     */
    for (s = nfa->states; s != NULL && !NISERR(); s = s->next) {
	for (s2 = emptyreachable(s, s); s2 != s && !NISERR();
		s2 = nexts) {



	    /*
	     * If s2 is doomed, we decide that (1) we will always push
	     * arcs forward to it, not pull them back to s; and (2) we
	     * can optimize away the push-forward, per comment above.
	     * So do nothing.
	     */
	    if (s2->flag || hasnonemptyout(s2)) {

		replaceempty(nfa, s, s2);
	    }


	    /* Reset the tmp fields as we walk back */
	    nexts = s2->tmp;
	    s2->tmp = NULL;
	}
	s->tmp = NULL;
    }

    /*
     * Remove all the EMPTY arcs, since we don't need them anymore.
     */
    for (s = nfa->states; s != NULL; s = s->next) {
	for (a = s->outs; a != NULL; a = nexta) {
	    nexta = a->outchain;
	    if (a->type == EMPTY) {
		freearc(nfa, a);
	    }

	}
    }

    /*
     * And remove any states that have become useless.  (This cleanup is
     * not very thorough, and would be even less so if we tried to
     * combine it with the previous step; but cleanup() will take care
     * of anything we miss.)
     */
    for (s = nfa->states; s != NULL && !NISERR(); s = nexts) {
	nexts = s->next;
	if ((s->nins == 0 || s->nouts == 0) && !s->flag) {
	    dropstate(nfa, s);
	}
    }

    if (f != NULL && !NISERR()) {
	dumpnfa(nfa, f);
    }
}

/*
 - emptyreachable - recursively find all states reachable from s by EMPTY arcs
 * The return value is the last such state found.  Its tmp field links back

 * to the next-to-last such state, and so on back to s, so that all these
 * states can be located without searching the whole NFA.
 * The maximum recursion depth here is equal to the length of the longest
 * loop-free chain of EMPTY arcs, which is surely no more than the size of
 * the NFA, and in practice will be a lot less than that.
 ^ static struct state *emptyreachable(struct state *, struct state *);
 */
static struct state *
emptyreachable(
    struct state *s,
    struct state *lastfound)
{

    struct arc *a;

    s->tmp = lastfound;
    lastfound = s;
    for (a = s->outs; a != NULL; a = a->outchain) {
	if (a->type == EMPTY && a->to->tmp == NULL) {

	    lastfound = emptyreachable(a->to, lastfound);
	}


    }
    return lastfound;
}

/*
 - replaceempty - replace an EMPTY arc chain with some non-empty arcs
 * The EMPTY arc(s) should be deleted later, but we can't do it here because
 * they may still be needed to identify other arc chains during fixempties().
 ^ static void replaceempty(struct nfa *, struct state *, struct state *);
 */
static void
replaceempty(



    struct nfa *nfa,
    struct state *from,
    struct state *to)
{

    int fromouts;
    int toins;

    assert(from != to);



    /*
     * Create replacement arcs that bypass the need for the EMPTY chain.  We
     * can do this either by pushing arcs forward (linking directly from
     * "from"'s predecessors to "to") or by pulling them back (linking
     * directly from "from" to "to"'s successors).  In general, we choose
     * whichever way creates greater fan-out or fan-in, so as to improve the
     * odds of reducing the other state to zero in-arcs or out-arcs and
     * thereby being able to delete it.  However, if "from" is doomed (has no
     * non-EMPTY out-arcs), we must keep it so, so always push forward in that
     * case.
     *
     * The fan-out/fan-in comparison should count only non-EMPTY arcs.  If
     * "from" is doomed, we can skip counting "to"'s arcs, since we want to
     * force taking the copynonemptyins path in that case.
     */
    fromouts = nonemptyouts(from);
    toins = (fromouts == 0) ? 1 : nonemptyins(to);

    if (fromouts > toins) {
	copyouts(nfa, to, from, 0);
	return;
    }
    if (fromouts < toins) {
	copyins(nfa, from, to, 0);
	return;
    }

    /*
     * fromouts == toins.  Decide on secondary issue: copy fewest arcs.

     *
     * Doesn't seem to be worth the trouble to exclude empties from these
     * comparisons; that takes extra time and doesn't seem to improve the
     * resulting graph much.
     */
    if (from->nins > to->nouts) {
	copyouts(nfa, to, from, 0);
	return;
    }

    copyins(nfa, from, to, 0);

}

/*
 - cleanup - clean up NFA after optimizations
 ^ static VOID cleanup(struct nfa *);
 */
static void

static struct state *newfstate(struct nfa *, int flag);
static void dropstate(struct nfa *, struct state *);
static void freestate(struct nfa *, struct state *);
static void destroystate(struct nfa *, struct state *);
static void newarc(struct nfa *, int, pcolor, struct state *, struct state *);
static struct arc *allocarc(struct nfa *, struct state *);
static void freearc(struct nfa *, struct arc *);
static int hasnonemptyout(struct state *);
static int nonemptyouts(struct state *);
static int nonemptyins(struct state *);
static struct arc *findarc(struct state *, int, pcolor);
static void cparc(struct nfa *, struct arc *, struct state *, struct state *);
static void moveins(struct nfa *, struct state *, struct state *);
static void copyins(struct nfa *, struct state *, struct state *, int);
static void moveouts(struct nfa *, struct state *, struct state *);
static void copyouts(struct nfa *, struct state *, struct state *, int);
static void cloneouts(struct nfa *, struct state *, struct state *, struct state *, int);
static void delsub(struct nfa *, struct state *, struct state *);
static void deltraverse(struct nfa *, struct state *, struct state *);
static void dupnfa(struct nfa *, struct state *, struct state *, struct state *, struct state *);
static void duptraverse(struct nfa *, struct state *, struct state *);
static void cleartraverse(struct nfa *, struct state *);
static void specialcolors(struct nfa *);
static long optimize(struct nfa *, FILE *);
static void pullback(struct nfa *, FILE *);
static int pull(struct nfa *, struct arc *);
static void pushfwd(struct nfa *, FILE *);
static int push(struct nfa *, struct arc *);
#define	INCOMPATIBLE	1	/* destroys arc */
#define	SATISFIED	2	/* constraint satisfied */
#define	COMPATIBLE	3	/* compatible but not satisfied yet */
static int combine(struct arc *, struct arc *);
static void fixempties(struct nfa *, FILE *);
static struct state *emptyreachable(struct state *, struct state *);
static void replaceempty(struct nfa *, struct state *, struct state *);
static void cleanup(struct nfa *);
static void markreachable(struct nfa *, struct state *, struct state *, struct state *);
static void markcanreach(struct nfa *, struct state *, struct state *, struct state *);
static long analyze(struct nfa *);
static void compact(struct nfa *, struct cnfa *);
static void carcsort(struct carc *, struct carc *);
static void freecnfa(struct cnfa *);

    /*
     * Do the splits.
     */

    for (s=slist ; s!=NULL ; s=s2) {
	s2 = newstate(nfa);

	copyouts(nfa, s, s2, 1);
	for (a=s->ins ; a!=NULL ; a=b) {
	    b = a->inchain;

	    if (a->from != pre) {
		cparc(nfa, a, a->from, s2);
		freearc(nfa, a);
	    }