PostgreSQL Source Code git master
All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Pages
regc_nfa.c File Reference
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros

#define NISERR()   VISERR(nfa->v)
 
#define NERR(e)   VERR(nfa->v, (e))
 
#define BULK_ARC_OP_USE_SORT(nsrcarcs, ndestarcs)    ((nsrcarcs) < 4 ? 0 : ((nsrcarcs) > 32 || (ndestarcs) > 32))
 
#define CA(ct, at)   (((ct)<<CHAR_BIT) | (at))
 

Functions

static struct nfanewnfa (struct vars *v, struct colormap *cm, struct nfa *parent)
 
static void freenfa (struct nfa *nfa)
 
static struct statenewstate (struct nfa *nfa)
 
static struct statenewfstate (struct nfa *nfa, int flag)
 
static void dropstate (struct nfa *nfa, struct state *s)
 
static void freestate (struct nfa *nfa, struct state *s)
 
static void newarc (struct nfa *nfa, int t, color co, struct state *from, struct state *to)
 
static void createarc (struct nfa *nfa, int t, color co, struct state *from, struct state *to)
 
static struct arcallocarc (struct nfa *nfa)
 
static void freearc (struct nfa *nfa, struct arc *victim)
 
static void changearcsource (struct arc *a, struct state *newfrom)
 
static void changearctarget (struct arc *a, struct state *newto)
 
static int hasnonemptyout (struct state *s)
 
static struct arcfindarc (struct state *s, int type, color co)
 
static void cparc (struct nfa *nfa, struct arc *oa, struct state *from, struct state *to)
 
static void sortins (struct nfa *nfa, struct state *s)
 
static int sortins_cmp (const void *a, const void *b)
 
static void sortouts (struct nfa *nfa, struct state *s)
 
static int sortouts_cmp (const void *a, const void *b)
 
static void moveins (struct nfa *nfa, struct state *oldState, struct state *newState)
 
static void copyins (struct nfa *nfa, struct state *oldState, struct state *newState)
 
static void mergeins (struct nfa *nfa, struct state *s, struct arc **arcarray, int arccount)
 
static void moveouts (struct nfa *nfa, struct state *oldState, struct state *newState)
 
static void copyouts (struct nfa *nfa, struct state *oldState, struct state *newState)
 
static void cloneouts (struct nfa *nfa, struct state *old, struct state *from, struct state *to, int type)
 
static void delsub (struct nfa *nfa, struct state *lp, struct state *rp)
 
static void deltraverse (struct nfa *nfa, struct state *leftend, struct state *s)
 
static void dupnfa (struct nfa *nfa, struct state *start, struct state *stop, struct state *from, struct state *to)
 
static void duptraverse (struct nfa *nfa, struct state *s, struct state *stmp)
 
static void removeconstraints (struct nfa *nfa, struct state *start, struct state *stop)
 
static void removetraverse (struct nfa *nfa, struct state *s)
 
static void cleartraverse (struct nfa *nfa, struct state *s)
 
static struct statesingle_color_transition (struct state *s1, struct state *s2)
 
static void specialcolors (struct nfa *nfa)
 
static long optimize (struct nfa *nfa, FILE *f)
 
static void pullback (struct nfa *nfa, FILE *f)
 
static int pull (struct nfa *nfa, struct arc *con, struct state **intermediates)
 
static void pushfwd (struct nfa *nfa, FILE *f)
 
static int push (struct nfa *nfa, struct arc *con, struct state **intermediates)
 
static int combine (struct nfa *nfa, struct arc *con, struct arc *a)
 
static void fixempties (struct nfa *nfa, FILE *f)
 
static struct stateemptyreachable (struct nfa *nfa, struct state *s, struct state *lastfound, struct arc **inarcsorig)
 
static int isconstraintarc (struct arc *a)
 
static int hasconstraintout (struct state *s)
 
static void fixconstraintloops (struct nfa *nfa, FILE *f)
 
static int findconstraintloop (struct nfa *nfa, struct state *s)
 
static void breakconstraintloop (struct nfa *nfa, struct state *sinitial)
 
static void clonesuccessorstates (struct nfa *nfa, struct state *ssource, struct state *sclone, struct state *spredecessor, struct arc *refarc, char *curdonemap, char *outerdonemap, int nstates)
 
static void removecantmatch (struct nfa *nfa)
 
static void cleanup (struct nfa *nfa)
 
static void markreachable (struct nfa *nfa, struct state *s, struct state *okay, struct state *mark)
 
static void markcanreach (struct nfa *nfa, struct state *s, struct state *okay, struct state *mark)
 
static long analyze (struct nfa *nfa)
 
static void checkmatchall (struct nfa *nfa)
 
static bool checkmatchall_recurse (struct nfa *nfa, struct state *s, bool **haspaths)
 
static bool check_out_colors_match (struct state *s, color co1, color co2)
 
static bool check_in_colors_match (struct state *s, color co1, color co2)
 
static void compact (struct nfa *nfa, struct cnfa *cnfa)
 
static void carcsort (struct carc *first, size_t n)
 
static int carc_cmp (const void *a, const void *b)
 
static void freecnfa (struct cnfa *cnfa)
 
static void dumpnfa (struct nfa *nfa, FILE *f)
 

Macro Definition Documentation

◆ BULK_ARC_OP_USE_SORT

#define BULK_ARC_OP_USE_SORT (   nsrcarcs,
  ndestarcs 
)     ((nsrcarcs) < 4 ? 0 : ((nsrcarcs) > 32 || (ndestarcs) > 32))

Definition at line 758 of file regc_nfa.c.

◆ CA

#define CA (   ct,
  at 
)    (((ct)<<CHAR_BIT) | (at))

◆ NERR

#define NERR (   e)    VERR(nfa->v, (e))

Definition at line 40 of file regc_nfa.c.

◆ NISERR

#define NISERR ( )    VISERR(nfa->v)

Definition at line 39 of file regc_nfa.c.

Function Documentation

◆ allocarc()

static struct arc * allocarc ( struct nfa nfa)
static

Definition at line 368 of file regc_nfa.c.

369{
370 struct arc *a;
371
372 /* first, recycle anything that's on the freelist */
373 if (nfa->freearcs != NULL)
374 {
375 a = nfa->freearcs;
376 nfa->freearcs = a->freechain;
377 }
378 /* otherwise, is there anything left in the last arcbatch? */
379 else if (nfa->lastab != NULL && nfa->lastabused < nfa->lastab->narcs)
380 {
381 a = &nfa->lastab->a[nfa->lastabused++];
382 }
383 /* otherwise, need to allocate a new arcbatch */
384 else
385 {
386 struct arcbatch *newAb;
387 size_t narcs;
388
390 {
392 return NULL;
393 }
394 narcs = (nfa->lastab != NULL) ? nfa->lastab->narcs * 2 : FIRSTABSIZE;
395 if (narcs > MAXABSIZE)
397 newAb = (struct arcbatch *) MALLOC(ARCBATCHSIZE(narcs));
398 if (newAb == NULL)
399 {
401 return NULL;
402 }
404 newAb->narcs = narcs;
405 newAb->next = nfa->lastab;
406 nfa->lastab = newAb;
407 nfa->lastabused = 1;
408 a = &newAb->a[0];
409 }
410
411 return a;
412}
int a
Definition: isn.c:68
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:76
#define NERR(e)
Definition: regc_nfa.c:40
#define MALLOC(n)
Definition: regcustom.h:52
#define REG_ETOOBIG
Definition: regex.h:233
#define REG_ESPACE
Definition: regex.h:227
#define ARCBATCHSIZE(n)
Definition: regguts.h:317
#define MAXABSIZE
Definition: regguts.h:320
#define FIRSTABSIZE
Definition: regguts.h:319
#define REG_MAX_COMPILE_SPACE
Definition: regguts.h:451
Definition: regguts.h:296
struct arcbatch * next
Definition: regguts.h:313
size_t narcs
Definition: regguts.h:314
struct arc a[FLEXIBLE_ARRAY_MEMBER]
Definition: regguts.h:315
Definition: regguts.h:349
struct vars * v
Definition: regguts.h:369
size_t lastabused
Definition: regguts.h:362
struct arc * freearcs
Definition: regguts.h:358
struct arcbatch * lastab
Definition: regguts.h:360
size_t spaceused
Definition: regcomp.c:309

References a, arcbatch::a, ARCBATCHSIZE, FIRSTABSIZE, nfa::freearcs, if(), nfa::lastab, nfa::lastabused, MALLOC, MAXABSIZE, arcbatch::narcs, NERR, arcbatch::next, REG_ESPACE, REG_ETOOBIG, REG_MAX_COMPILE_SPACE, vars::spaceused, and nfa::v.

Referenced by createarc().

◆ analyze()

static long analyze ( struct nfa nfa)
static

Definition at line 3051 of file regc_nfa.c.

3052{
3053 struct arc *a;
3054 struct arc *aa;
3055
3056 if (NISERR())
3057 return 0;
3058
3059 /* Detect whether NFA can't match anything */
3060 if (nfa->pre->outs == NULL)
3061 return REG_UIMPOSSIBLE;
3062
3063 /* Detect whether NFA matches all strings (possibly with length bounds) */
3065
3066 /* Detect whether NFA can possibly match a zero-length string */
3067 for (a = nfa->pre->outs; a != NULL; a = a->outchain)
3068 for (aa = a->to->outs; aa != NULL; aa = aa->outchain)
3069 if (aa->to == nfa->post)
3070 return REG_UEMPTYMATCH;
3071 return 0;
3072}
#define NISERR()
Definition: regc_nfa.c:39
static void checkmatchall(struct nfa *nfa)
Definition: regc_nfa.c:3097
#define REG_UEMPTYMATCH
Definition: regex.h:149
#define REG_UIMPOSSIBLE
Definition: regex.h:150
struct arc * outchain
Definition: regguts.h:301
struct state * to
Definition: regguts.h:300
struct state * post
Definition: regguts.h:353
struct state * pre
Definition: regguts.h:350
struct arc * outs
Definition: regguts.h:330

References a, checkmatchall(), NISERR, arc::outchain, state::outs, nfa::post, nfa::pre, REG_UEMPTYMATCH, REG_UIMPOSSIBLE, and arc::to.

Referenced by ExecVacuum(), GetCommandLogLevel(), and optimize().

◆ breakconstraintloop()

static void breakconstraintloop ( struct nfa nfa,
struct state sinitial 
)
static

Definition at line 2558 of file regc_nfa.c.

2559{
2560 struct state *s;
2561 struct state *shead;
2562 struct state *stail;
2563 struct state *sclone;
2564 struct state *nexts;
2565 struct arc *refarc;
2566 struct arc *a;
2567 struct arc *nexta;
2568
2569 /*
2570 * Start by identifying which loop step we want to break at.
2571 * Preferentially this is one with only one constraint arc. (XXX are
2572 * there any other secondary heuristics we want to use here?) Set refarc
2573 * to point to the selected lone constraint arc, if there is one.
2574 */
2575 refarc = NULL;
2576 s = sinitial;
2577 do
2578 {
2579 nexts = s->tmp;
2580 assert(nexts != s); /* should not see any one-element loops */
2581 if (refarc == NULL)
2582 {
2583 int narcs = 0;
2584
2585 for (a = s->outs; a != NULL; a = a->outchain)
2586 {
2587 if (a->to == nexts && isconstraintarc(a))
2588 {
2589 refarc = a;
2590 narcs++;
2591 }
2592 }
2593 assert(narcs > 0);
2594 if (narcs > 1)
2595 refarc = NULL; /* multiple constraint arcs here, no good */
2596 }
2597 s = nexts;
2598 } while (s != sinitial);
2599
2600 if (refarc)
2601 {
2602 /* break at the refarc */
2603 shead = refarc->from;
2604 stail = refarc->to;
2605 assert(stail == shead->tmp);
2606 }
2607 else
2608 {
2609 /* for lack of a better idea, break after sinitial */
2610 shead = sinitial;
2611 stail = sinitial->tmp;
2612 }
2613
2614 /*
2615 * Reset the tmp fields so that we can use them for local storage in
2616 * clonesuccessorstates. (findconstraintloop won't mind, since it's just
2617 * going to abandon its search anyway.)
2618 */
2619 for (s = nfa->states; s != NULL; s = s->next)
2620 s->tmp = NULL;
2621
2622 /*
2623 * Recursively build clone state(s) as needed.
2624 */
2625 sclone = newstate(nfa);
2626 if (sclone == NULL)
2627 {
2628 assert(NISERR());
2629 return;
2630 }
2631
2632 clonesuccessorstates(nfa, stail, sclone, shead, refarc,
2633 NULL, NULL, nfa->nstates);
2634
2635 if (NISERR())
2636 return;
2637
2638 /*
2639 * It's possible that sclone has no outarcs at all, in which case it's
2640 * useless. (We don't try extremely hard to get rid of useless states
2641 * here, but this is an easy and fairly common case.)
2642 */
2643 if (sclone->nouts == 0)
2644 {
2645 freestate(nfa, sclone);
2646 sclone = NULL;
2647 }
2648
2649 /*
2650 * Move shead's constraint-loop arcs to point to sclone, or just drop them
2651 * if we discovered we don't need sclone.
2652 */
2653 for (a = shead->outs; a != NULL; a = nexta)
2654 {
2655 nexta = a->outchain;
2656 if (a->to == stail && isconstraintarc(a))
2657 {
2658 if (sclone)
2659 cparc(nfa, a, shead, sclone);
2660 freearc(nfa, a);
2661 if (NISERR())
2662 break;
2663 }
2664 }
2665}
static void cparc(struct nfa *nfa, struct arc *oa, struct state *from, struct state *to)
Definition: regc_nfa.c:608
static void clonesuccessorstates(struct nfa *nfa, struct state *ssource, struct state *sclone, struct state *spredecessor, struct arc *refarc, char *curdonemap, char *outerdonemap, int nstates)
Definition: regc_nfa.c:2704
static int isconstraintarc(struct arc *a)
Definition: regc_nfa.c:2331
static void freearc(struct nfa *nfa, struct arc *victim)
Definition: regc_nfa.c:418
static struct state * newstate(struct nfa *nfa)
Definition: regc_nfa.c:137
static void freestate(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:242
#define assert(x)
Definition: regcustom.h:56
struct state * from
Definition: regguts.h:299
int nstates
Definition: regguts.h:354
struct state * states
Definition: regguts.h:355
Definition: regguts.h:323
int nouts
Definition: regguts.h:328
struct state * tmp
Definition: regguts.h:331
struct state * next
Definition: regguts.h:332

References a, assert, clonesuccessorstates(), cparc(), freearc(), freestate(), arc::from, isconstraintarc(), newstate(), state::next, NISERR, state::nouts, nfa::nstates, state::outs, nfa::states, state::tmp, and arc::to.

Referenced by findconstraintloop().

◆ carc_cmp()

static int carc_cmp ( const void *  a,
const void *  b 
)
static

Definition at line 3612 of file regc_nfa.c.

3613{
3614 const struct carc *aa = (const struct carc *) a;
3615 const struct carc *bb = (const struct carc *) b;
3616
3617 if (aa->co < bb->co)
3618 return -1;
3619 if (aa->co > bb->co)
3620 return +1;
3621 if (aa->to < bb->to)
3622 return -1;
3623 if (aa->to > bb->to)
3624 return +1;
3625 /* This is unreached, since there should be no duplicate arcs now: */
3626 return 0;
3627}
int b
Definition: isn.c:69
Definition: regguts.h:401
int to
Definition: regguts.h:403
color co
Definition: regguts.h:402

References a, b, carc::co, and carc::to.

Referenced by carcsort().

◆ carcsort()

static void carcsort ( struct carc first,
size_t  n 
)
static

Definition at line 3605 of file regc_nfa.c.

3606{
3607 if (n > 1)
3608 qsort(first, n, sizeof(struct carc), carc_cmp);
3609}
#define qsort(a, b, c, d)
Definition: port.h:474
static int carc_cmp(const void *a, const void *b)
Definition: regc_nfa.c:3612

References carc_cmp(), and qsort.

Referenced by compact().

◆ changearcsource()

static void changearcsource ( struct arc a,
struct state newfrom 
)
static

Definition at line 489 of file regc_nfa.c.

490{
491 struct state *oldfrom = a->from;
492 struct arc *predecessor;
493
494 assert(oldfrom != newfrom);
495
496 /* take it off old source's out-chain */
497 assert(oldfrom != NULL);
498 predecessor = a->outchainRev;
499 if (predecessor == NULL)
500 {
501 assert(oldfrom->outs == a);
502 oldfrom->outs = a->outchain;
503 }
504 else
505 {
506 assert(predecessor->outchain == a);
507 predecessor->outchain = a->outchain;
508 }
509 if (a->outchain != NULL)
510 {
511 assert(a->outchain->outchainRev == a);
512 a->outchain->outchainRev = predecessor;
513 }
514 oldfrom->nouts--;
515
516 a->from = newfrom;
517
518 /* prepend it to new source's out-chain */
519 a->outchain = newfrom->outs;
520 a->outchainRev = NULL;
521 if (newfrom->outs)
522 newfrom->outs->outchainRev = a;
523 newfrom->outs = a;
524 newfrom->nouts++;
525}
struct arc * outchainRev
Definition: regguts.h:302

References a, assert, state::nouts, arc::outchain, arc::outchainRev, and state::outs.

Referenced by moveouts().

◆ changearctarget()

static void changearctarget ( struct arc a,
struct state newto 
)
static

Definition at line 533 of file regc_nfa.c.

534{
535 struct state *oldto = a->to;
536 struct arc *predecessor;
537
538 assert(oldto != newto);
539
540 /* take it off old target's in-chain */
541 assert(oldto != NULL);
542 predecessor = a->inchainRev;
543 if (predecessor == NULL)
544 {
545 assert(oldto->ins == a);
546 oldto->ins = a->inchain;
547 }
548 else
549 {
550 assert(predecessor->inchain == a);
551 predecessor->inchain = a->inchain;
552 }
553 if (a->inchain != NULL)
554 {
555 assert(a->inchain->inchainRev == a);
556 a->inchain->inchainRev = predecessor;
557 }
558 oldto->nins--;
559
560 a->to = newto;
561
562 /* prepend it to new target's in-chain */
563 a->inchain = newto->ins;
564 a->inchainRev = NULL;
565 if (newto->ins)
566 newto->ins->inchainRev = a;
567 newto->ins = a;
568 newto->nins++;
569}
struct arc * inchainRev
Definition: regguts.h:305
struct arc * inchain
Definition: regguts.h:304
int nins
Definition: regguts.h:327
struct arc * ins
Definition: regguts.h:329

References a, assert, arc::inchain, arc::inchainRev, state::ins, and state::nins.

Referenced by moveins().

◆ check_in_colors_match()

static bool check_in_colors_match ( struct state s,
color  co1,
color  co2 
)
static

Definition at line 3469 of file regc_nfa.c.

3470{
3471 bool result = true;
3472 struct arc *a;
3473
3474 /*
3475 * Identical algorithm to check_out_colors_match, except examine the
3476 * from-states of s' inarcs.
3477 */
3478 for (a = s->ins; a != NULL; a = a->inchain)
3479 {
3480 if (a->co == co1)
3481 {
3482 assert(a->from->tmp == NULL);
3483 a->from->tmp = a->from;
3484 }
3485 }
3486 for (a = s->ins; a != NULL; a = a->inchain)
3487 {
3488 if (a->co == co2)
3489 {
3490 if (a->from->tmp != NULL)
3491 a->from->tmp = NULL;
3492 else
3493 result = false; /* unmatched co2 arc */
3494 }
3495 }
3496 for (a = s->ins; a != NULL; a = a->inchain)
3497 {
3498 if (a->co == co1)
3499 {
3500 if (a->from->tmp != NULL)
3501 {
3502 result = false; /* unmatched co1 arc */
3503 a->from->tmp = NULL;
3504 }
3505 }
3506 }
3507 return result;
3508}

References a, assert, and state::ins.

Referenced by checkmatchall().

◆ check_out_colors_match()

static bool check_out_colors_match ( struct state s,
color  co1,
color  co2 
)
static

Definition at line 3415 of file regc_nfa.c.

3416{
3417 bool result = true;
3418 struct arc *a;
3419
3420 /*
3421 * To do this in linear time, we assume that the NFA contains no duplicate
3422 * arcs. Run through the out-arcs, marking states reachable by arcs of
3423 * color co1. Run through again, un-marking states reachable by arcs of
3424 * color co2; if we see a not-marked state, we know this co2 arc is
3425 * unmatched. Then run through again, checking for still-marked states,
3426 * and in any case leaving all the tmp fields reset to NULL.
3427 */
3428 for (a = s->outs; a != NULL; a = a->outchain)
3429 {
3430 if (a->co == co1)
3431 {
3432 assert(a->to->tmp == NULL);
3433 a->to->tmp = a->to;
3434 }
3435 }
3436 for (a = s->outs; a != NULL; a = a->outchain)
3437 {
3438 if (a->co == co2)
3439 {
3440 if (a->to->tmp != NULL)
3441 a->to->tmp = NULL;
3442 else
3443 result = false; /* unmatched co2 arc */
3444 }
3445 }
3446 for (a = s->outs; a != NULL; a = a->outchain)
3447 {
3448 if (a->co == co1)
3449 {
3450 if (a->to->tmp != NULL)
3451 {
3452 result = false; /* unmatched co1 arc */
3453 a->to->tmp = NULL;
3454 }
3455 }
3456 }
3457 return result;
3458}

References a, assert, and state::outs.

Referenced by checkmatchall().

◆ checkmatchall()

static void checkmatchall ( struct nfa nfa)
static

Definition at line 3097 of file regc_nfa.c.

3098{
3099 bool **haspaths;
3100 struct state *s;
3101 int i;
3102
3103 /*
3104 * If there are too many states, don't bother trying to detect matchall.
3105 * This limit serves to bound the time and memory we could consume below.
3106 * Note that even if the graph is all-RAINBOW, if there are significantly
3107 * more than DUPINF states then it's likely that there are paths of length
3108 * more than DUPINF, which would force us to fail anyhow. In practice,
3109 * plausible ways of writing a matchall regex with maximum finite path
3110 * length K tend not to have very many more than K states.
3111 */
3112 if (nfa->nstates > DUPINF * 2)
3113 return;
3114
3115 /*
3116 * First, scan all the states to verify that only RAINBOW arcs appear,
3117 * plus pseudocolor arcs adjacent to the pre and post states. This lets
3118 * us quickly eliminate most cases that aren't matchall NFAs.
3119 */
3120 for (s = nfa->states; s != NULL; s = s->next)
3121 {
3122 struct arc *a;
3123
3124 for (a = s->outs; a != NULL; a = a->outchain)
3125 {
3126 if (a->type != PLAIN)
3127 return; /* any LACONs make it non-matchall */
3128 if (a->co != RAINBOW)
3129 {
3130 if (nfa->cm->cd[a->co].flags & PSEUDO)
3131 {
3132 /*
3133 * Pseudocolor arc: verify it's in a valid place (this
3134 * seems quite unlikely to fail, but let's be sure).
3135 */
3136 if (s == nfa->pre &&
3137 (a->co == nfa->bos[0] || a->co == nfa->bos[1]))
3138 /* okay BOS/BOL arc */ ;
3139 else if (a->to == nfa->post &&
3140 (a->co == nfa->eos[0] || a->co == nfa->eos[1]))
3141 /* okay EOS/EOL arc */ ;
3142 else
3143 return; /* unexpected pseudocolor arc */
3144 /* We'll check these arcs some more below. */
3145 }
3146 else
3147 return; /* any other color makes it non-matchall */
3148 }
3149 }
3150 /* Also, assert that the tmp fields are available for use. */
3151 assert(s->tmp == NULL);
3152 }
3153
3154 /*
3155 * The next cheapest check we can make is to verify that the BOS/BOL
3156 * outarcs of the pre state reach the same states as its RAINBOW outarcs.
3157 * If they don't, the NFA expresses some constraints on the character
3158 * before the matched string, making it non-matchall. Likewise, the
3159 * EOS/EOL inarcs of the post state must match its RAINBOW inarcs.
3160 */
3165 return;
3166
3167 /*
3168 * Initialize an array of path-length arrays, in which
3169 * checkmatchall_recurse will return per-state results. This lets us
3170 * memo-ize the recursive search and avoid exponential time consumption.
3171 */
3172 haspaths = (bool **) MALLOC(nfa->nstates * sizeof(bool *));
3173 if (haspaths == NULL)
3174 return; /* fail quietly */
3175 memset(haspaths, 0, nfa->nstates * sizeof(bool *));
3176
3177 /*
3178 * Recursively search the graph for all-RAINBOW paths to the "post" state,
3179 * starting at the "pre" state, and computing the lengths of the paths.
3180 * (Given the preceding checks, there should be at least one such path.
3181 * However we could get back a false result anyway, in case there are
3182 * multi-state loops, paths exceeding DUPINF+1 length, or non-algorithmic
3183 * failures such as ENOMEM.)
3184 */
3185 if (checkmatchall_recurse(nfa, nfa->pre, haspaths))
3186 {
3187 /* The useful result is the path length array for the pre state */
3188 bool *haspath = haspaths[nfa->pre->no];
3189 int minmatch,
3190 maxmatch,
3191 morematch;
3192
3193 assert(haspath != NULL);
3194
3195 /*
3196 * haspath[] now represents the set of possible path lengths; but we
3197 * want to reduce that to a min and max value, because it doesn't seem
3198 * worth complicating regexec.c to deal with nonconsecutive possible
3199 * match lengths. Find min and max of first run of lengths, then
3200 * verify there are no nonconsecutive lengths.
3201 */
3202 for (minmatch = 0; minmatch <= DUPINF + 1; minmatch++)
3203 {
3204 if (haspath[minmatch])
3205 break;
3206 }
3207 assert(minmatch <= DUPINF + 1); /* else checkmatchall_recurse lied */
3208 for (maxmatch = minmatch; maxmatch < DUPINF + 1; maxmatch++)
3209 {
3210 if (!haspath[maxmatch + 1])
3211 break;
3212 }
3213 for (morematch = maxmatch + 1; morematch <= DUPINF + 1; morematch++)
3214 {
3215 if (haspath[morematch])
3216 {
3217 haspath = NULL; /* fail, there are nonconsecutive lengths */
3218 break;
3219 }
3220 }
3221
3222 if (haspath != NULL)
3223 {
3224 /*
3225 * Success, so record the info. Here we have a fine point: the
3226 * path length from the pre state includes the pre-to-initial
3227 * transition, so it's one more than the actually matched string
3228 * length. (We avoided counting the final-to-post transition
3229 * within checkmatchall_recurse, but not this one.) This is why
3230 * checkmatchall_recurse allows one more level of path length than
3231 * might seem necessary. This decrement also takes care of
3232 * converting checkmatchall_recurse's definition of "infinity" as
3233 * "DUPINF+1" to our normal representation as "DUPINF".
3234 */
3235 assert(minmatch > 0); /* else pre and post states were adjacent */
3236 nfa->minmatchall = minmatch - 1;
3237 nfa->maxmatchall = maxmatch - 1;
3238 nfa->flags |= MATCHALL;
3239 }
3240 }
3241
3242 /* Clean up */
3243 for (i = 0; i < nfa->nstates; i++)
3244 {
3245 if (haspaths[i] != NULL)
3246 FREE(haspaths[i]);
3247 }
3248 FREE(haspaths);
3249}
int i
Definition: isn.c:72
static bool checkmatchall_recurse(struct nfa *nfa, struct state *s, bool **haspaths)
Definition: regc_nfa.c:3277
static bool check_in_colors_match(struct state *s, color co1, color co2)
Definition: regc_nfa.c:3469
static bool check_out_colors_match(struct state *s, color co1, color co2)
Definition: regc_nfa.c:3415
#define PLAIN
Definition: regcomp.c:331
#define PSEUDO
Definition: regguts.h:186
#define RAINBOW
Definition: regguts.h:159
#define DUPINF
Definition: regguts.h:99
#define MATCHALL
Definition: regguts.h:412
int flags
Definition: regguts.h:184
struct colordesc * cd
Definition: regguts.h:236
int minmatchall
Definition: regguts.h:367
int maxmatchall
Definition: regguts.h:368
color bos[2]
Definition: regguts.h:364
color eos[2]
Definition: regguts.h:365
int flags
Definition: regguts.h:366
struct colormap * cm
Definition: regguts.h:363
int no
Definition: regguts.h:324
@ FREE
Definition: task.c:94

References a, assert, nfa::bos, colormap::cd, check_in_colors_match(), check_out_colors_match(), checkmatchall_recurse(), nfa::cm, DUPINF, nfa::eos, colordesc::flags, nfa::flags, FREE, i, MALLOC, MATCHALL, nfa::maxmatchall, nfa::minmatchall, state::next, state::no, nfa::nstates, state::outs, PLAIN, nfa::post, nfa::pre, PSEUDO, RAINBOW, nfa::states, and state::tmp.

Referenced by analyze().

◆ checkmatchall_recurse()

static bool checkmatchall_recurse ( struct nfa nfa,
struct state s,
bool **  haspaths 
)
static

Definition at line 3277 of file regc_nfa.c.

3278{
3279 bool result = false;
3280 bool foundloop = false;
3281 bool *haspath;
3282 struct arc *a;
3283
3284 /*
3285 * Since this is recursive, it could be driven to stack overflow. But we
3286 * need not treat that as a hard failure; just deem the NFA non-matchall.
3287 */
3288 if (STACK_TOO_DEEP(nfa->v->re))
3289 return false;
3290
3291 /* In case the search takes a long time, check for cancel */
3292 INTERRUPT(nfa->v->re);
3293
3294 /* Create a haspath array for this state */
3295 haspath = (bool *) MALLOC((DUPINF + 2) * sizeof(bool));
3296 if (haspath == NULL)
3297 return false; /* again, treat as non-matchall */
3298 memset(haspath, 0, (DUPINF + 2) * sizeof(bool));
3299
3300 /* Mark this state as being visited */
3301 assert(s->tmp == NULL);
3302 s->tmp = s;
3303
3304 for (a = s->outs; a != NULL; a = a->outchain)
3305 {
3306 if (a->co != RAINBOW)
3307 continue; /* ignore pseudocolor arcs */
3308 if (a->to == nfa->post)
3309 {
3310 /* We found an all-RAINBOW path to the post state */
3311 result = true;
3312
3313 /*
3314 * Mark this state as being zero steps away from the string end
3315 * (the transition to the post state isn't counted).
3316 */
3317 haspath[0] = true;
3318 }
3319 else if (a->to == s)
3320 {
3321 /* We found a cycle of length 1, which we'll deal with below. */
3322 foundloop = true;
3323 }
3324 else if (a->to->tmp != NULL)
3325 {
3326 /* It's busy, so we found a cycle of length > 1, so fail. */
3327 result = false;
3328 break;
3329 }
3330 else
3331 {
3332 /* Consider paths forward through this to-state. */
3333 bool *nexthaspath;
3334 int i;
3335
3336 /* If to-state was not already visited, recurse */
3337 if (haspaths[a->to->no] == NULL)
3338 {
3339 result = checkmatchall_recurse(nfa, a->to, haspaths);
3340 /* Fail if any recursive path fails */
3341 if (!result)
3342 break;
3343 }
3344 else
3345 {
3346 /* The previous visit must have found path(s) to the end */
3347 result = true;
3348 }
3349 assert(a->to->tmp == NULL);
3350 nexthaspath = haspaths[a->to->no];
3351 assert(nexthaspath != NULL);
3352
3353 /*
3354 * Now, for every path of length i from a->to to the string end,
3355 * there is a path of length i + 1 from s to the string end.
3356 */
3357 if (nexthaspath[DUPINF] != nexthaspath[DUPINF + 1])
3358 {
3359 /*
3360 * a->to has a path of length exactly DUPINF, but not longer;
3361 * or it has paths of all lengths > DUPINF but not one of
3362 * exactly that length. In either case, we cannot represent
3363 * the possible path lengths from s correctly, so fail.
3364 */
3365 result = false;
3366 break;
3367 }
3368 /* Merge knowledge of these path lengths into what we have */
3369 for (i = 0; i < DUPINF; i++)
3370 haspath[i + 1] |= nexthaspath[i];
3371 /* Infinity + 1 is still infinity */
3372 haspath[DUPINF + 1] |= nexthaspath[DUPINF + 1];
3373 }
3374 }
3375
3376 if (result && foundloop)
3377 {
3378 /*
3379 * If there is a length-1 loop at this state, then find the shortest
3380 * known path length to the end. The loop means that every larger
3381 * path length is possible, too. (It doesn't matter whether any of
3382 * the longer lengths were already known possible.)
3383 */
3384 int i;
3385
3386 for (i = 0; i <= DUPINF; i++)
3387 {
3388 if (haspath[i])
3389 break;
3390 }
3391 for (i++; i <= DUPINF + 1; i++)
3392 haspath[i] = true;
3393 }
3394
3395 /* Report out the completed path length map */
3396 assert(s->no < nfa->nstates);
3397 assert(haspaths[s->no] == NULL);
3398 haspaths[s->no] = haspath;
3399
3400 /* Mark state no longer busy */
3401 s->tmp = NULL;
3402
3403 return result;
3404}
#define INTERRUPT(re)
Definition: regcustom.h:55
#define STACK_TOO_DEEP(re)
Definition: regguts.h:523
regex_t * re
Definition: regcomp.c:283

References a, assert, checkmatchall_recurse(), DUPINF, i, INTERRUPT, MALLOC, state::no, nfa::nstates, state::outs, nfa::post, RAINBOW, vars::re, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by checkmatchall(), and checkmatchall_recurse().

◆ cleanup()

static void cleanup ( struct nfa nfa)
static

Definition at line 2964 of file regc_nfa.c.

2965{
2966 struct state *s;
2967 struct state *nexts;
2968 int n;
2969
2970 if (NISERR())
2971 return;
2972
2973 /* clear out unreachable or dead-end states */
2974 /* use pre to mark reachable, then post to mark can-reach-post */
2975 markreachable(nfa, nfa->pre, (struct state *) NULL, nfa->pre);
2977 for (s = nfa->states; s != NULL && !NISERR(); s = nexts)
2978 {
2979 nexts = s->next;
2980 if (s->tmp != nfa->post && !s->flag)
2981 dropstate(nfa, s);
2982 }
2983 assert(NISERR() || nfa->post->nins == 0 || nfa->post->tmp == nfa->post);
2985 assert(NISERR() || nfa->post->nins == 0 || nfa->post->tmp == NULL);
2986 /* the nins==0 (final unreachable) case will be caught later */
2987
2988 /* renumber surviving states */
2989 n = 0;
2990 for (s = nfa->states; s != NULL; s = s->next)
2991 s->no = n++;
2992 nfa->nstates = n;
2993}
static void cleartraverse(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:1488
static void dropstate(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:226
static void markcanreach(struct nfa *nfa, struct state *s, struct state *okay, struct state *mark)
Definition: regc_nfa.c:3025
static void markreachable(struct nfa *nfa, struct state *s, struct state *okay, struct state *mark)
Definition: regc_nfa.c:2999
char flag
Definition: regguts.h:326

References assert, cleartraverse(), dropstate(), state::flag, markcanreach(), markreachable(), state::next, state::nins, NISERR, state::no, nfa::nstates, nfa::post, nfa::pre, nfa::states, and state::tmp.

Referenced by optimize().

◆ cleartraverse()

static void cleartraverse ( struct nfa nfa,
struct state s 
)
static

Definition at line 1488 of file regc_nfa.c.

1490{
1491 struct arc *a;
1492
1493 /* Since this is recursive, it could be driven to stack overflow */
1494 if (STACK_TOO_DEEP(nfa->v->re))
1495 {
1497 return;
1498 }
1499
1500 if (s->tmp == NULL)
1501 return;
1502 s->tmp = NULL;
1503
1504 for (a = s->outs; a != NULL; a = a->outchain)
1505 cleartraverse(nfa, a->to);
1506}

References a, cleartraverse(), NERR, state::outs, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by cleanup(), cleartraverse(), dupnfa(), and removeconstraints().

◆ cloneouts()

static void cloneouts ( struct nfa nfa,
struct state old,
struct state from,
struct state to,
int  type 
)
static

Definition at line 1256 of file regc_nfa.c.

1261{
1262 struct arc *a;
1263
1264 assert(old != from);
1265 assert(type == AHEAD || type == BEHIND);
1266
1267 for (a = old->outs; a != NULL; a = a->outchain)
1268 {
1269 assert(a->type == PLAIN);
1270 newarc(nfa, type, a->co, from, to);
1271 }
1272}
static void newarc(struct nfa *nfa, int t, color co, struct state *from, struct state *to)
Definition: regc_nfa.c:281
#define BEHIND
Definition: regcomp.c:343
#define AHEAD
Definition: regcomp.c:342
const char * type

References a, AHEAD, assert, BEHIND, arc::from, newarc(), state::outs, PLAIN, arc::to, and type.

◆ clonesuccessorstates()

static void clonesuccessorstates ( struct nfa nfa,
struct state ssource,
struct state sclone,
struct state spredecessor,
struct arc refarc,
char *  curdonemap,
char *  outerdonemap,
int  nstates 
)
static

Definition at line 2704 of file regc_nfa.c.

2712{
2713 char *donemap;
2714 struct arc *a;
2715
2716 /* Since this is recursive, it could be driven to stack overflow */
2717 if (STACK_TOO_DEEP(nfa->v->re))
2718 {
2720 return;
2721 }
2722
2723 /* If this state hasn't already got a donemap, create one */
2724 donemap = curdonemap;
2725 if (donemap == NULL)
2726 {
2727 donemap = (char *) MALLOC(nstates * sizeof(char));
2728 if (donemap == NULL)
2729 {
2731 return;
2732 }
2733
2734 if (outerdonemap != NULL)
2735 {
2736 /*
2737 * Not at outermost recursion level, so copy the outer level's
2738 * donemap; this ensures that we see states in process of being
2739 * visited at outer levels, or already merged into predecessor
2740 * states, as ones we shouldn't traverse back to.
2741 */
2742 memcpy(donemap, outerdonemap, nstates * sizeof(char));
2743 }
2744 else
2745 {
2746 /* At outermost level, only spredecessor is off-limits */
2747 memset(donemap, 0, nstates * sizeof(char));
2748 assert(spredecessor->no < nstates);
2749 donemap[spredecessor->no] = 1;
2750 }
2751 }
2752
2753 /* Mark ssource as visited in the donemap */
2754 assert(ssource->no < nstates);
2755 assert(donemap[ssource->no] == 0);
2756 donemap[ssource->no] = 1;
2757
2758 /*
2759 * We proceed by first cloning all of ssource's outarcs, creating new
2760 * clone states as needed but not doing more with them than that. Then in
2761 * a second pass, recurse to process the child clone states. This allows
2762 * us to have only one child clone state per reachable source state, even
2763 * when there are multiple outarcs leading to the same state. Also, when
2764 * we do visit a child state, its set of inarcs is known exactly, which
2765 * makes it safe to apply the constraint-is-already-checked optimization.
2766 * Also, this ensures that we've merged all the states we can into the
2767 * current clone before we recurse to any children, thus possibly saving
2768 * them from making extra images of those states.
2769 *
2770 * While this function runs, child clone states of the current state are
2771 * marked by setting their tmp fields to point to the original state they
2772 * were cloned from. This makes it possible to detect multiple outarcs
2773 * leading to the same state, and also makes it easy to distinguish clone
2774 * states from original states (which will have tmp == NULL).
2775 */
2776 for (a = ssource->outs; a != NULL && !NISERR(); a = a->outchain)
2777 {
2778 struct state *sto = a->to;
2779
2780 /*
2781 * We do not consider cloning successor states that have no constraint
2782 * outarcs; just link to them as-is. They cannot be part of a
2783 * constraint loop so there is no need to make copies. In particular,
2784 * this rule keeps us from trying to clone the post state, which would
2785 * be a bad idea.
2786 */
2787 if (isconstraintarc(a) && hasconstraintout(sto))
2788 {
2789 struct state *prevclone;
2790 int canmerge;
2791 struct arc *a2;
2792
2793 /*
2794 * Back-link constraint arcs must not be followed. Nor is there a
2795 * need to revisit states previously merged into this clone.
2796 */
2797 assert(sto->no < nstates);
2798 if (donemap[sto->no] != 0)
2799 continue;
2800
2801 /*
2802 * Check whether we already have a child clone state for this
2803 * source state.
2804 */
2805 prevclone = NULL;
2806 for (a2 = sclone->outs; a2 != NULL; a2 = a2->outchain)
2807 {
2808 if (a2->to->tmp == sto)
2809 {
2810 prevclone = a2->to;
2811 break;
2812 }
2813 }
2814
2815 /*
2816 * If this arc is labeled the same as refarc, or the same as any
2817 * arc we must have traversed to get to sclone, then no additional
2818 * constraints need to be met to get to sto, so we should just
2819 * merge its outarcs into sclone.
2820 */
2821 if (refarc && a->type == refarc->type && a->co == refarc->co)
2822 canmerge = 1;
2823 else
2824 {
2825 struct state *s;
2826
2827 canmerge = 0;
2828 for (s = sclone; s->ins; s = s->ins->from)
2829 {
2830 if (s->nins == 1 &&
2831 a->type == s->ins->type && a->co == s->ins->co)
2832 {
2833 canmerge = 1;
2834 break;
2835 }
2836 }
2837 }
2838
2839 if (canmerge)
2840 {
2841 /*
2842 * We can merge into sclone. If we previously made a child
2843 * clone state, drop it; there's no need to visit it. (This
2844 * can happen if ssource has multiple pathways to sto, and we
2845 * only just now found one that is provably a no-op.)
2846 */
2847 if (prevclone)
2848 dropstate(nfa, prevclone); /* kills our outarc, too */
2849
2850 /* Recurse to merge sto's outarcs into sclone */
2852 sto,
2853 sclone,
2854 spredecessor,
2855 refarc,
2856 donemap,
2857 outerdonemap,
2858 nstates);
2859 /* sto should now be marked as previously visited */
2860 assert(NISERR() || donemap[sto->no] == 1);
2861 }
2862 else if (prevclone)
2863 {
2864 /*
2865 * We already have a clone state for this successor, so just
2866 * make another arc to it.
2867 */
2868 cparc(nfa, a, sclone, prevclone);
2869 }
2870 else
2871 {
2872 /*
2873 * We need to create a new successor clone state.
2874 */
2875 struct state *stoclone;
2876
2877 stoclone = newstate(nfa);
2878 if (stoclone == NULL)
2879 {
2880 assert(NISERR());
2881 break;
2882 }
2883 /* Mark it as to what it's a clone of */
2884 stoclone->tmp = sto;
2885 /* ... and add the outarc leading to it */
2886 cparc(nfa, a, sclone, stoclone);
2887 }
2888 }
2889 else
2890 {
2891 /*
2892 * Non-constraint outarcs just get copied to sclone, as do outarcs
2893 * leading to states with no constraint outarc.
2894 */
2895 cparc(nfa, a, sclone, sto);
2896 }
2897 }
2898
2899 /*
2900 * If we are at outer level for this clone state, recurse to all its child
2901 * clone states, clearing their tmp fields as we go. (If we're not
2902 * outermost for sclone, leave this to be done by the outer call level.)
2903 * Note that if we have multiple outarcs leading to the same clone state,
2904 * it will only be recursed-to once.
2905 */
2906 if (curdonemap == NULL)
2907 {
2908 for (a = sclone->outs; a != NULL && !NISERR(); a = a->outchain)
2909 {
2910 struct state *stoclone = a->to;
2911 struct state *sto = stoclone->tmp;
2912
2913 if (sto != NULL)
2914 {
2915 stoclone->tmp = NULL;
2917 sto,
2918 stoclone,
2919 spredecessor,
2920 refarc,
2921 NULL,
2922 donemap,
2923 nstates);
2924 }
2925 }
2926
2927 /* Don't forget to free sclone's donemap when done with it */
2928 FREE(donemap);
2929 }
2930}
static const FormData_pg_attribute a2
Definition: heap.c:156
static int hasconstraintout(struct state *s)
Definition: regc_nfa.c:2349
int type
Definition: regguts.h:297
color co
Definition: regguts.h:298

References a, a2, assert, clonesuccessorstates(), arc::co, cparc(), dropstate(), FREE, arc::from, hasconstraintout(), state::ins, isconstraintarc(), MALLOC, NERR, newstate(), state::nins, NISERR, state::no, state::outs, vars::re, REG_ESPACE, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, arc::type, and nfa::v.

Referenced by breakconstraintloop(), and clonesuccessorstates().

◆ combine()

static int combine ( struct nfa nfa,
struct arc con,
struct arc a 
)
static

Definition at line 1987 of file regc_nfa.c.

1990{
1991#define CA(ct,at) (((ct)<<CHAR_BIT) | (at))
1992
1993 switch (CA(con->type, a->type))
1994 {
1995 case CA('^', PLAIN): /* newlines are handled separately */
1996 case CA('$', PLAIN):
1997 return INCOMPATIBLE;
1998 break;
1999 case CA(AHEAD, PLAIN): /* color constraints meet colors */
2000 case CA(BEHIND, PLAIN):
2001 if (con->co == a->co)
2002 return SATISFIED;
2003 if (con->co == RAINBOW)
2004 {
2005 /* con is satisfied unless arc's color is a pseudocolor */
2006 if (!(nfa->cm->cd[a->co].flags & PSEUDO))
2007 return SATISFIED;
2008 }
2009 else if (a->co == RAINBOW)
2010 {
2011 /* con is incompatible if it's for a pseudocolor */
2012 /* (this is hypothetical; we make no such constraints today) */
2013 if (nfa->cm->cd[con->co].flags & PSEUDO)
2014 return INCOMPATIBLE;
2015 /* otherwise, constraint constrains arc to be only its color */
2016 return REPLACEARC;
2017 }
2018 return INCOMPATIBLE;
2019 break;
2020 case CA('^', '^'): /* collision, similar constraints */
2021 case CA('$', '$'):
2022 if (con->co == a->co) /* true duplication */
2023 return SATISFIED;
2024 return INCOMPATIBLE;
2025 break;
2026 case CA(AHEAD, AHEAD): /* collision, similar constraints */
2027 case CA(BEHIND, BEHIND):
2028 if (con->co == a->co) /* true duplication */
2029 return SATISFIED;
2030 if (con->co == RAINBOW)
2031 {
2032 /* con is satisfied unless arc's color is a pseudocolor */
2033 if (!(nfa->cm->cd[a->co].flags & PSEUDO))
2034 return SATISFIED;
2035 }
2036 else if (a->co == RAINBOW)
2037 {
2038 /* con is incompatible if it's for a pseudocolor */
2039 /* (this is hypothetical; we make no such constraints today) */
2040 if (nfa->cm->cd[con->co].flags & PSEUDO)
2041 return INCOMPATIBLE;
2042 /* otherwise, constraint constrains arc to be only its color */
2043 return REPLACEARC;
2044 }
2045 return INCOMPATIBLE;
2046 break;
2047 case CA('^', BEHIND): /* collision, dissimilar constraints */
2048 case CA(BEHIND, '^'):
2049 case CA('$', AHEAD):
2050 case CA(AHEAD, '$'):
2051 return INCOMPATIBLE;
2052 break;
2053 case CA('^', '$'): /* constraints passing each other */
2054 case CA('^', AHEAD):
2055 case CA(BEHIND, '$'):
2056 case CA(BEHIND, AHEAD):
2057 case CA('$', '^'):
2058 case CA('$', BEHIND):
2059 case CA(AHEAD, '^'):
2060 case CA(AHEAD, BEHIND):
2061 case CA('^', LACON):
2062 case CA(BEHIND, LACON):
2063 case CA('$', LACON):
2064 case CA(AHEAD, LACON):
2065 return COMPATIBLE;
2066 break;
2067 }
2069 return INCOMPATIBLE; /* for benefit of blind compilers */
2070}
#define CA(ct, at)
#define SATISFIED
Definition: regcomp.c:200
#define INCOMPATIBLE
Definition: regcomp.c:199
#define COMPATIBLE
Definition: regcomp.c:201
#define REPLACEARC
Definition: regcomp.c:202
#define LACON
Definition: regcomp.c:341
#define NOTREACHED
Definition: regguts.h:96

References a, AHEAD, assert, BEHIND, CA, colormap::cd, nfa::cm, arc::co, COMPATIBLE, colordesc::flags, INCOMPATIBLE, LACON, NOTREACHED, PLAIN, PSEUDO, RAINBOW, REPLACEARC, SATISFIED, and arc::type.

Referenced by pull(), and push().

◆ compact()

static void compact ( struct nfa nfa,
struct cnfa cnfa 
)
static

Definition at line 3514 of file regc_nfa.c.

3516{
3517 struct state *s;
3518 struct arc *a;
3519 size_t nstates;
3520 size_t narcs;
3521 struct carc *ca;
3522 struct carc *first;
3523
3524 assert(!NISERR());
3525
3526 nstates = 0;
3527 narcs = 0;
3528 for (s = nfa->states; s != NULL; s = s->next)
3529 {
3530 nstates++;
3531 narcs += s->nouts + 1; /* need one extra for endmarker */
3532 }
3533
3534 cnfa->stflags = (char *) MALLOC(nstates * sizeof(char));
3535 cnfa->states = (struct carc **) MALLOC(nstates * sizeof(struct carc *));
3536 cnfa->arcs = (struct carc *) MALLOC(narcs * sizeof(struct carc));
3537 if (cnfa->stflags == NULL || cnfa->states == NULL || cnfa->arcs == NULL)
3538 {
3539 if (cnfa->stflags != NULL)
3540 FREE(cnfa->stflags);
3541 if (cnfa->states != NULL)
3542 FREE(cnfa->states);
3543 if (cnfa->arcs != NULL)
3544 FREE(cnfa->arcs);
3546 return;
3547 }
3548 cnfa->nstates = nstates;
3549 cnfa->pre = nfa->pre->no;
3550 cnfa->post = nfa->post->no;
3551 cnfa->bos[0] = nfa->bos[0];
3552 cnfa->bos[1] = nfa->bos[1];
3553 cnfa->eos[0] = nfa->eos[0];
3554 cnfa->eos[1] = nfa->eos[1];
3555 cnfa->ncolors = maxcolor(nfa->cm) + 1;
3556 cnfa->flags = nfa->flags;
3559
3560 ca = cnfa->arcs;
3561 for (s = nfa->states; s != NULL; s = s->next)
3562 {
3563 assert((size_t) s->no < nstates);
3564 cnfa->stflags[s->no] = 0;
3565 cnfa->states[s->no] = ca;
3566 first = ca;
3567 for (a = s->outs; a != NULL; a = a->outchain)
3568 switch (a->type)
3569 {
3570 case PLAIN:
3571 ca->co = a->co;
3572 ca->to = a->to->no;
3573 ca++;
3574 break;
3575 case LACON:
3576 assert(s->no != cnfa->pre);
3577 assert(a->co >= 0);
3578 ca->co = (color) (cnfa->ncolors + a->co);
3579 ca->to = a->to->no;
3580 ca++;
3581 cnfa->flags |= HASLACONS;
3582 break;
3583 default:
3585 return;
3586 }
3587 carcsort(first, ca - first);
3588 ca->co = COLORLESS;
3589 ca->to = 0;
3590 ca++;
3591 }
3592 assert(ca == &cnfa->arcs[narcs]);
3593 assert(cnfa->nstates != 0);
3594
3595 /* mark no-progress states */
3596 for (a = nfa->pre->outs; a != NULL; a = a->outchain)
3597 cnfa->stflags[a->to->no] = CNFA_NOPROGRESS;
3599}
static color maxcolor(struct colormap *cm)
Definition: regc_color.c:172
static void carcsort(struct carc *first, size_t n)
Definition: regc_nfa.c:3605
#define REG_ASSERT
Definition: regex.h:229
short color
Definition: regguts.h:155
#define CNFA_NOPROGRESS
Definition: regguts.h:420
#define COLORLESS
Definition: regguts.h:158
#define HASLACONS
Definition: regguts.h:411
Definition: regguts.h:407
int pre
Definition: regguts.h:415
color eos[2]
Definition: regguts.h:418
struct carc ** states
Definition: regguts.h:421
int ncolors
Definition: regguts.h:409
int maxmatchall
Definition: regguts.h:426
int flags
Definition: regguts.h:410
int post
Definition: regguts.h:416
int minmatchall
Definition: regguts.h:425
struct carc * arcs
Definition: regguts.h:423
color bos[2]
Definition: regguts.h:417
int nstates
Definition: regguts.h:408
char * stflags
Definition: regguts.h:419

References a, cnfa::arcs, assert, nfa::bos, cnfa::bos, carcsort(), nfa::cm, CNFA_NOPROGRESS, carc::co, COLORLESS, nfa::eos, cnfa::eos, nfa::flags, cnfa::flags, FREE, HASLACONS, LACON, MALLOC, maxcolor(), nfa::maxmatchall, cnfa::maxmatchall, nfa::minmatchall, cnfa::minmatchall, cnfa::ncolors, NERR, state::next, NISERR, state::no, state::nouts, cnfa::nstates, state::outs, PLAIN, nfa::post, cnfa::post, nfa::pre, cnfa::pre, REG_ASSERT, REG_ESPACE, nfa::states, cnfa::states, cnfa::stflags, and carc::to.

◆ copyins()

static void copyins ( struct nfa nfa,
struct state oldState,
struct state newState 
)
static

Definition at line 882 of file regc_nfa.c.

885{
886 assert(oldState != newState);
887 assert(newState->nins == 0); /* see comment above */
888
889 if (newState->nins == 0)
890 {
891 /* No need for de-duplication */
892 struct arc *a;
893
894 for (a = oldState->ins; a != NULL; a = a->inchain)
895 createarc(nfa, a->type, a->co, a->from, newState);
896 }
897#ifdef NOT_USED /* see comment above */
898 else if (!BULK_ARC_OP_USE_SORT(oldState->nins, newState->nins))
899 {
900 /* With not too many arcs, just do them one at a time */
901 struct arc *a;
902
903 for (a = oldState->ins; a != NULL; a = a->inchain)
904 cparc(nfa, a, a->from, newState);
905 }
906 else
907 {
908 /*
909 * With many arcs, use a sort-merge approach. Note that createarc()
910 * will put new arcs onto the front of newState's chain, so it does
911 * not break our walk through the sorted part of the chain.
912 */
913 struct arc *oa;
914 struct arc *na;
915
916 /*
917 * Because we bypass newarc() in this code path, we'd better include a
918 * cancel check.
919 */
920 INTERRUPT(nfa->v->re);
921
922 sortins(nfa, oldState);
923 sortins(nfa, newState);
924 if (NISERR())
925 return; /* might have failed to sort */
926 oa = oldState->ins;
927 na = newState->ins;
928 while (oa != NULL && na != NULL)
929 {
930 struct arc *a = oa;
931
932 switch (sortins_cmp(&oa, &na))
933 {
934 case -1:
935 /* newState does not have anything matching oa */
936 oa = oa->inchain;
937 createarc(nfa, a->type, a->co, a->from, newState);
938 break;
939 case 0:
940 /* match, advance in both lists */
941 oa = oa->inchain;
942 na = na->inchain;
943 break;
944 case +1:
945 /* advance only na; oa might have a match later */
946 na = na->inchain;
947 break;
948 default:
950 }
951 }
952 while (oa != NULL)
953 {
954 /* newState does not have anything matching oa */
955 struct arc *a = oa;
956
957 oa = oa->inchain;
958 createarc(nfa, a->type, a->co, a->from, newState);
959 }
960 }
961#endif /* NOT_USED */
962}
static int sortins_cmp(const void *a, const void *b)
Definition: regc_nfa.c:662
#define BULK_ARC_OP_USE_SORT(nsrcarcs, ndestarcs)
Definition: regc_nfa.c:758
static void createarc(struct nfa *nfa, int t, color co, struct state *from, struct state *to)
Definition: regc_nfa.c:323
static void sortins(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:620

References a, assert, BULK_ARC_OP_USE_SORT, cparc(), createarc(), arc::inchain, state::ins, INTERRUPT, state::nins, NISERR, NOTREACHED, vars::re, sortins(), sortins_cmp(), and nfa::v.

Referenced by pull().

◆ copyouts()

static void copyouts ( struct nfa nfa,
struct state oldState,
struct state newState 
)
static

Definition at line 1167 of file regc_nfa.c.

1170{
1171 assert(oldState != newState);
1172 assert(newState->nouts == 0); /* see comment above */
1173
1174 if (newState->nouts == 0)
1175 {
1176 /* No need for de-duplication */
1177 struct arc *a;
1178
1179 for (a = oldState->outs; a != NULL; a = a->outchain)
1180 createarc(nfa, a->type, a->co, newState, a->to);
1181 }
1182#ifdef NOT_USED /* see comment above */
1183 else if (!BULK_ARC_OP_USE_SORT(oldState->nouts, newState->nouts))
1184 {
1185 /* With not too many arcs, just do them one at a time */
1186 struct arc *a;
1187
1188 for (a = oldState->outs; a != NULL; a = a->outchain)
1189 cparc(nfa, a, newState, a->to);
1190 }
1191 else
1192 {
1193 /*
1194 * With many arcs, use a sort-merge approach. Note that createarc()
1195 * will put new arcs onto the front of newState's chain, so it does
1196 * not break our walk through the sorted part of the chain.
1197 */
1198 struct arc *oa;
1199 struct arc *na;
1200
1201 /*
1202 * Because we bypass newarc() in this code path, we'd better include a
1203 * cancel check.
1204 */
1205 INTERRUPT(nfa->v->re);
1206
1207 sortouts(nfa, oldState);
1208 sortouts(nfa, newState);
1209 if (NISERR())
1210 return; /* might have failed to sort */
1211 oa = oldState->outs;
1212 na = newState->outs;
1213 while (oa != NULL && na != NULL)
1214 {
1215 struct arc *a = oa;
1216
1217 switch (sortouts_cmp(&oa, &na))
1218 {
1219 case -1:
1220 /* newState does not have anything matching oa */
1221 oa = oa->outchain;
1222 createarc(nfa, a->type, a->co, newState, a->to);
1223 break;
1224 case 0:
1225 /* match, advance in both lists */
1226 oa = oa->outchain;
1227 na = na->outchain;
1228 break;
1229 case +1:
1230 /* advance only na; oa might have a match later */
1231 na = na->outchain;
1232 break;
1233 default:
1235 }
1236 }
1237 while (oa != NULL)
1238 {
1239 /* newState does not have anything matching oa */
1240 struct arc *a = oa;
1241
1242 oa = oa->outchain;
1243 createarc(nfa, a->type, a->co, newState, a->to);
1244 }
1245 }
1246#endif /* NOT_USED */
1247}
static void sortouts(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:687
static int sortouts_cmp(const void *a, const void *b)
Definition: regc_nfa.c:729

References a, assert, BULK_ARC_OP_USE_SORT, cparc(), createarc(), INTERRUPT, NISERR, NOTREACHED, state::nouts, arc::outchain, state::outs, vars::re, sortouts(), sortouts_cmp(), and nfa::v.

Referenced by push().

◆ cparc()

static void cparc ( struct nfa nfa,
struct arc oa,
struct state from,
struct state to 
)
static

Definition at line 608 of file regc_nfa.c.

612{
613 newarc(nfa, oa->type, oa->co, from, to);
614}

References arc::co, arc::from, newarc(), arc::to, and arc::type.

Referenced by breakconstraintloop(), clonesuccessorstates(), copyins(), copyouts(), duptraverse(), moveins(), moveouts(), pull(), and push().

◆ createarc()

static void createarc ( struct nfa nfa,
int  t,
color  co,
struct state from,
struct state to 
)
static

Definition at line 323 of file regc_nfa.c.

328{
329 struct arc *a;
330
331 a = allocarc(nfa);
332 if (NISERR())
333 return;
334 assert(a != NULL);
335
336 a->type = t;
337 a->co = co;
338 a->to = to;
339 a->from = from;
340
341 /*
342 * Put the new arc on the beginning, not the end, of the chains; it's
343 * simpler here, and freearc() is the same cost either way. See also the
344 * logic in moveins() and its cohorts, as well as fixempties().
345 */
346 a->inchain = to->ins;
347 a->inchainRev = NULL;
348 if (to->ins)
349 to->ins->inchainRev = a;
350 to->ins = a;
351 a->outchain = from->outs;
352 a->outchainRev = NULL;
353 if (from->outs)
355 from->outs = a;
356
357 from->nouts++;
358 to->nins++;
359
360 if (COLORED(a) && nfa->parent == NULL)
361 colorchain(nfa->cm, a);
362}
static void colorchain(struct colormap *cm, struct arc *a)
Definition: regc_color.c:984
static struct arc * allocarc(struct nfa *nfa)
Definition: regc_nfa.c:368
#define COLORED(a)
Definition: regcomp.c:352
struct nfa * parent
Definition: regguts.h:370

References a, allocarc(), assert, nfa::cm, arc::co, colorchain(), COLORED, arc::from, arc::inchainRev, state::ins, state::nins, NISERR, state::nouts, arc::outchainRev, state::outs, nfa::parent, and arc::to.

Referenced by copyins(), copyouts(), mergeins(), moveins(), moveouts(), and newarc().

◆ delsub()

static void delsub ( struct nfa nfa,
struct state lp,
struct state rp 
)
static

Definition at line 1281 of file regc_nfa.c.

1284{
1285 assert(lp != rp);
1286
1287 rp->tmp = rp; /* mark end */
1288
1289 deltraverse(nfa, lp, lp);
1290 if (NISERR())
1291 return; /* asserts might not hold after failure */
1292 assert(lp->nouts == 0 && rp->nins == 0); /* did the job */
1293 assert(lp->no != FREESTATE && rp->no != FREESTATE); /* no more */
1294
1295 rp->tmp = NULL; /* unmark end */
1296 lp->tmp = NULL; /* and begin, marked by deltraverse */
1297}
static void deltraverse(struct nfa *nfa, struct state *leftend, struct state *s)
Definition: regc_nfa.c:1304
#define FREESTATE(state)
Definition: tuplesort.c:399

References assert, deltraverse(), FREESTATE, state::nins, NISERR, state::no, state::nouts, and state::tmp.

◆ deltraverse()

static void deltraverse ( struct nfa nfa,
struct state leftend,
struct state s 
)
static

Definition at line 1304 of file regc_nfa.c.

1307{
1308 struct arc *a;
1309 struct state *to;
1310
1311 /* Since this is recursive, it could be driven to stack overflow */
1312 if (STACK_TOO_DEEP(nfa->v->re))
1313 {
1315 return;
1316 }
1317
1318 if (s->nouts == 0)
1319 return; /* nothing to do */
1320 if (s->tmp != NULL)
1321 return; /* already in progress */
1322
1323 s->tmp = s; /* mark as in progress */
1324
1325 while ((a = s->outs) != NULL)
1326 {
1327 to = a->to;
1328 deltraverse(nfa, leftend, to);
1329 if (NISERR())
1330 return; /* asserts might not hold after failure */
1331 assert(to->nouts == 0 || to->tmp != NULL);
1332 freearc(nfa, a);
1333 if (to->nins == 0 && to->tmp == NULL)
1334 {
1335 assert(to->nouts == 0);
1336 freestate(nfa, to);
1337 }
1338 }
1339
1340 assert(s->no != FREESTATE); /* we're still here */
1341 assert(s == leftend || s->nins != 0); /* and still reachable */
1342 assert(s->nouts == 0); /* but have no outarcs */
1343
1344 s->tmp = NULL; /* we're done here */
1345}

References a, assert, deltraverse(), freearc(), FREESTATE, freestate(), NERR, state::nins, NISERR, state::no, state::nouts, state::outs, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by delsub(), and deltraverse().

◆ dropstate()

static void dropstate ( struct nfa nfa,
struct state s 
)
static

Definition at line 226 of file regc_nfa.c.

228{
229 struct arc *a;
230
231 while ((a = s->ins) != NULL)
232 freearc(nfa, a);
233 while ((a = s->outs) != NULL)
234 freearc(nfa, a);
235 freestate(nfa, s);
236}

References a, freearc(), freestate(), state::ins, and state::outs.

Referenced by cleanup(), clonesuccessorstates(), fixconstraintloops(), fixempties(), pullback(), and pushfwd().

◆ dumpnfa()

static void dumpnfa ( struct nfa nfa,
FILE *  f 
)
static

Definition at line 3646 of file regc_nfa.c.

3648{
3649#ifdef REG_DEBUG
3650 struct state *s;
3651 int nstates = 0;
3652 int narcs = 0;
3653
3654 fprintf(f, "pre %d, post %d", nfa->pre->no, nfa->post->no);
3655 if (nfa->bos[0] != COLORLESS)
3656 fprintf(f, ", bos [%ld]", (long) nfa->bos[0]);
3657 if (nfa->bos[1] != COLORLESS)
3658 fprintf(f, ", bol [%ld]", (long) nfa->bos[1]);
3659 if (nfa->eos[0] != COLORLESS)
3660 fprintf(f, ", eos [%ld]", (long) nfa->eos[0]);
3661 if (nfa->eos[1] != COLORLESS)
3662 fprintf(f, ", eol [%ld]", (long) nfa->eos[1]);
3663 if (nfa->flags & HASLACONS)
3664 fprintf(f, ", haslacons");
3665 if (nfa->flags & HASCANTMATCH)
3666 fprintf(f, ", hascantmatch");
3667 if (nfa->flags & MATCHALL)
3668 {
3669 fprintf(f, ", minmatchall %d", nfa->minmatchall);
3670 if (nfa->maxmatchall == DUPINF)
3671 fprintf(f, ", maxmatchall inf");
3672 else
3673 fprintf(f, ", maxmatchall %d", nfa->maxmatchall);
3674 }
3675 fprintf(f, "\n");
3676 for (s = nfa->states; s != NULL; s = s->next)
3677 {
3678 dumpstate(s, f);
3679 nstates++;
3680 narcs += s->nouts;
3681 }
3682 fprintf(f, "total of %d states, %d arcs\n", nstates, narcs);
3683 if (nfa->parent == NULL)
3684 dumpcolors(nfa->cm, f);
3685 fflush(f);
3686#endif
3687}
#define fprintf(file, fmt, msg)
Definition: cubescan.l:21
static void const char fflush(stdout)
#define HASCANTMATCH
Definition: regguts.h:413

References nfa::bos, nfa::cm, COLORLESS, DUPINF, nfa::eos, fflush(), nfa::flags, fprintf, HASCANTMATCH, HASLACONS, MATCHALL, nfa::maxmatchall, nfa::minmatchall, state::next, state::no, state::nouts, nfa::parent, nfa::post, nfa::pre, and nfa::states.

Referenced by fixconstraintloops(), fixempties(), optimize(), pullback(), and pushfwd().

◆ dupnfa()

static void dupnfa ( struct nfa nfa,
struct state start,
struct state stop,
struct state from,
struct state to 
)
static

Definition at line 1355 of file regc_nfa.c.

1360{
1361 if (start == stop)
1362 {
1363 newarc(nfa, EMPTY, 0, from, to);
1364 return;
1365 }
1366
1367 stop->tmp = to;
1368 duptraverse(nfa, start, from);
1369 /* done, except for clearing out the tmp pointers */
1370
1371 stop->tmp = NULL;
1373}
return str start
static void duptraverse(struct nfa *nfa, struct state *s, struct state *stmp)
Definition: regc_nfa.c:1379
#define EMPTY
Definition: regcomp.c:329

References cleartraverse(), duptraverse(), EMPTY, newarc(), start, and state::tmp.

◆ duptraverse()

static void duptraverse ( struct nfa nfa,
struct state s,
struct state stmp 
)
static

Definition at line 1379 of file regc_nfa.c.

1382{
1383 struct arc *a;
1384
1385 /* Since this is recursive, it could be driven to stack overflow */
1386 if (STACK_TOO_DEEP(nfa->v->re))
1387 {
1389 return;
1390 }
1391
1392 if (s->tmp != NULL)
1393 return; /* already done */
1394
1395 s->tmp = (stmp == NULL) ? newstate(nfa) : stmp;
1396 if (s->tmp == NULL)
1397 {
1398 assert(NISERR());
1399 return;
1400 }
1401
1402 for (a = s->outs; a != NULL && !NISERR(); a = a->outchain)
1403 {
1404 duptraverse(nfa, a->to, (struct state *) NULL);
1405 if (NISERR())
1406 break;
1407 assert(a->to->tmp != NULL);
1408 cparc(nfa, a, s->tmp, a->to->tmp);
1409 }
1410}

References a, assert, cparc(), duptraverse(), NERR, newstate(), NISERR, state::outs, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by dupnfa(), and duptraverse().

◆ emptyreachable()

static struct state * emptyreachable ( struct nfa nfa,
struct state s,
struct state lastfound,
struct arc **  inarcsorig 
)
static

Definition at line 2303 of file regc_nfa.c.

2307{
2308 struct arc *a;
2309
2310 /* Since this is recursive, it could be driven to stack overflow */
2311 if (STACK_TOO_DEEP(nfa->v->re))
2312 {
2314 return lastfound;
2315 }
2316
2317 s->tmp = lastfound;
2318 lastfound = s;
2319 for (a = inarcsorig[s->no]; a != NULL; a = a->inchain)
2320 {
2321 if (a->type == EMPTY && a->from->tmp == NULL)
2322 lastfound = emptyreachable(nfa, a->from, lastfound, inarcsorig);
2323 }
2324 return lastfound;
2325}
static struct state * emptyreachable(struct nfa *nfa, struct state *s, struct state *lastfound, struct arc **inarcsorig)
Definition: regc_nfa.c:2303

References a, EMPTY, emptyreachable(), NERR, state::no, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by emptyreachable(), and fixempties().

◆ findarc()

static struct arc * findarc ( struct state s,
int  type,
color  co 
)
static

Definition at line 592 of file regc_nfa.c.

595{
596 struct arc *a;
597
598 for (a = s->outs; a != NULL; a = a->outchain)
599 if (a->type == type && a->co == co)
600 return a;
601 return NULL;
602}

References a, arc::co, state::outs, and type.

Referenced by colorcomplement().

◆ findconstraintloop()

static int findconstraintloop ( struct nfa nfa,
struct state s 
)
static

Definition at line 2469 of file regc_nfa.c.

2470{
2471 struct arc *a;
2472
2473 /* Since this is recursive, it could be driven to stack overflow */
2474 if (STACK_TOO_DEEP(nfa->v->re))
2475 {
2477 return 1; /* to exit as quickly as possible */
2478 }
2479
2480 if (s->tmp != NULL)
2481 {
2482 /* Already proven uninteresting? */
2483 if (s->tmp == s)
2484 return 0;
2485 /* Found a loop involving s */
2487 /* The tmp fields have been cleaned up by breakconstraintloop */
2488 return 1;
2489 }
2490 for (a = s->outs; a != NULL; a = a->outchain)
2491 {
2492 if (isconstraintarc(a))
2493 {
2494 struct state *sto = a->to;
2495
2496 assert(sto != s);
2497 s->tmp = sto;
2498 if (findconstraintloop(nfa, sto))
2499 return 1;
2500 }
2501 }
2502
2503 /*
2504 * If we get here, no constraint loop exists leading out from s. Mark it
2505 * with s->tmp == s so we need not rediscover that fact again later.
2506 */
2507 s->tmp = s;
2508 return 0;
2509}
static int findconstraintloop(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:2469
static void breakconstraintloop(struct nfa *nfa, struct state *sinitial)
Definition: regc_nfa.c:2558

References a, assert, breakconstraintloop(), findconstraintloop(), isconstraintarc(), NERR, state::outs, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by findconstraintloop(), and fixconstraintloops().

◆ fixconstraintloops()

static void fixconstraintloops ( struct nfa nfa,
FILE *  f 
)
static

Definition at line 2370 of file regc_nfa.c.

2372{
2373 struct state *s;
2374 struct state *nexts;
2375 struct arc *a;
2376 struct arc *nexta;
2377 int hasconstraints;
2378
2379 /*
2380 * In the trivial case of a state that loops to itself, we can just drop
2381 * the constraint arc altogether. This is worth special-casing because
2382 * such loops are far more common than loops containing multiple states.
2383 * While we're at it, note whether any constraint arcs survive.
2384 */
2385 hasconstraints = 0;
2386 for (s = nfa->states; s != NULL && !NISERR(); s = nexts)
2387 {
2388 nexts = s->next;
2389 /* while we're at it, ensure tmp fields are clear for next step */
2390 assert(s->tmp == NULL);
2391 for (a = s->outs; a != NULL && !NISERR(); a = nexta)
2392 {
2393 nexta = a->outchain;
2394 if (isconstraintarc(a))
2395 {
2396 if (a->to == s)
2397 freearc(nfa, a);
2398 else
2399 hasconstraints = 1;
2400 }
2401 }
2402 /* If we removed all the outarcs, the state is useless. */
2403 if (s->nouts == 0 && !s->flag)
2404 dropstate(nfa, s);
2405 }
2406
2407 /* Nothing to do if no remaining constraint arcs */
2408 if (NISERR() || !hasconstraints)
2409 return;
2410
2411 /*
2412 * Starting from each remaining NFA state, search outwards for a
2413 * constraint loop. If we find a loop, break the loop, then start the
2414 * search over. (We could possibly retain some state from the first scan,
2415 * but it would complicate things greatly, and multi-state constraint
2416 * loops are rare enough that it's not worth optimizing the case.)
2417 */
2418restart:
2419 for (s = nfa->states; s != NULL && !NISERR(); s = s->next)
2420 {
2421 if (findconstraintloop(nfa, s))
2422 goto restart;
2423 }
2424
2425 if (NISERR())
2426 return;
2427
2428 /*
2429 * Now remove any states that have become useless. (This cleanup is not
2430 * very thorough, and would be even less so if we tried to combine it with
2431 * the previous step; but cleanup() will take care of anything we miss.)
2432 *
2433 * Because findconstraintloop intentionally doesn't reset all tmp fields,
2434 * we have to clear them after it's done. This is a convenient place to
2435 * do that, too.
2436 */
2437 for (s = nfa->states; s != NULL; s = nexts)
2438 {
2439 nexts = s->next;
2440 s->tmp = NULL;
2441 if ((s->nins == 0 || s->nouts == 0) && !s->flag)
2442 dropstate(nfa, s);
2443 }
2444
2445 if (f != NULL)
2446 dumpnfa(nfa, f);
2447}
static void dumpnfa(struct nfa *nfa, FILE *f)
Definition: regc_nfa.c:3646

References a, assert, dropstate(), dumpnfa(), findconstraintloop(), state::flag, freearc(), isconstraintarc(), state::next, state::nins, NISERR, state::nouts, state::outs, nfa::states, and state::tmp.

Referenced by optimize().

◆ fixempties()

static void fixempties ( struct nfa nfa,
FILE *  f 
)
static

Definition at line 2076 of file regc_nfa.c.

2078{
2079 struct state *s;
2080 struct state *s2;
2081 struct state *nexts;
2082 struct arc *a;
2083 struct arc *nexta;
2084 int totalinarcs;
2085 struct arc **inarcsorig;
2086 struct arc **arcarray;
2087 int arccount;
2088 int prevnins;
2089 int nskip;
2090
2091 /*
2092 * First, get rid of any states whose sole out-arc is an EMPTY, since
2093 * they're basically just aliases for their successor. The parsing
2094 * algorithm creates enough of these that it's worth special-casing this.
2095 */
2096 for (s = nfa->states; s != NULL && !NISERR(); s = nexts)
2097 {
2098 nexts = s->next;
2099 if (s->flag || s->nouts != 1)
2100 continue;
2101 a = s->outs;
2102 assert(a != NULL && a->outchain == NULL);
2103 if (a->type != EMPTY)
2104 continue;
2105 if (s != a->to)
2106 moveins(nfa, s, a->to);
2107 dropstate(nfa, s);
2108 }
2109
2110 /*
2111 * Similarly, get rid of any state with a single EMPTY in-arc, by folding
2112 * it into its predecessor.
2113 */
2114 for (s = nfa->states; s != NULL && !NISERR(); s = nexts)
2115 {
2116 nexts = s->next;
2117 /* while we're at it, ensure tmp fields are clear for next step */
2118 assert(s->tmp == NULL);
2119 if (s->flag || s->nins != 1)
2120 continue;
2121 a = s->ins;
2122 assert(a != NULL && a->inchain == NULL);
2123 if (a->type != EMPTY)
2124 continue;
2125 if (s != a->from)
2126 moveouts(nfa, s, a->from);
2127 dropstate(nfa, s);
2128 }
2129
2130 if (NISERR())
2131 return;
2132
2133 /*
2134 * For each remaining NFA state, find all other states from which it is
2135 * reachable by a chain of one or more EMPTY arcs. Then generate new arcs
2136 * that eliminate the need for each such chain.
2137 *
2138 * We could replace a chain of EMPTY arcs that leads from a "from" state
2139 * to a "to" state either by pushing non-EMPTY arcs forward (linking
2140 * directly from "from"'s predecessors to "to") or by pulling them back
2141 * (linking directly from "from" to "to"'s successors). We choose to
2142 * always do the former; this choice is somewhat arbitrary, but the
2143 * approach below requires that we uniformly do one or the other.
2144 *
2145 * Suppose we have a chain of N successive EMPTY arcs (where N can easily
2146 * approach the size of the NFA). All of the intermediate states must
2147 * have additional inarcs and outarcs, else they'd have been removed by
2148 * the steps above. Assuming their inarcs are mostly not empties, we will
2149 * add O(N^2) arcs to the NFA, since a non-EMPTY inarc leading to any one
2150 * state in the chain must be duplicated to lead to all its successor
2151 * states as well. So there is no hope of doing less than O(N^2) work;
2152 * however, we should endeavor to keep the big-O cost from being even
2153 * worse than that, which it can easily become without care. In
2154 * particular, suppose we were to copy all S1's inarcs forward to S2, and
2155 * then also to S3, and then later we consider pushing S2's inarcs forward
2156 * to S3. If we include the arcs already copied from S1 in that, we'd be
2157 * doing O(N^3) work. (The duplicate-arc elimination built into newarc()
2158 * and its cohorts would get rid of the extra arcs, but not without cost.)
2159 *
2160 * We can avoid this cost by treating only arcs that existed at the start
2161 * of this phase as candidates to be pushed forward. To identify those,
2162 * we remember the first inarc each state had to start with. We rely on
2163 * the fact that newarc() and friends put new arcs on the front of their
2164 * to-states' inchains, and that this phase never deletes arcs, so that
2165 * the original arcs must be the last arcs in their to-states' inchains.
2166 *
2167 * So the process here is that, for each state in the NFA, we gather up
2168 * all non-EMPTY inarcs of states that can reach the target state via
2169 * EMPTY arcs. We then sort, de-duplicate, and merge these arcs into the
2170 * target state's inchain. (We can safely use sort-merge for this as long
2171 * as we update each state's original-arcs pointer after we add arcs to
2172 * it; the sort step of mergeins probably changed the order of the old
2173 * arcs.)
2174 *
2175 * Another refinement worth making is that, because we only add non-EMPTY
2176 * arcs during this phase, and all added arcs have the same from-state as
2177 * the non-EMPTY arc they were cloned from, we know ahead of time that any
2178 * states having only EMPTY outarcs will be useless for lack of outarcs
2179 * after we drop the EMPTY arcs. (They cannot gain non-EMPTY outarcs if
2180 * they had none to start with.) So we need not bother to update the
2181 * inchains of such states at all.
2182 */
2183
2184 /* Remember the states' first original inarcs */
2185 /* ... and while at it, count how many old inarcs there are altogether */
2186 inarcsorig = (struct arc **) MALLOC(nfa->nstates * sizeof(struct arc *));
2187 if (inarcsorig == NULL)
2188 {
2190 return;
2191 }
2192 totalinarcs = 0;
2193 for (s = nfa->states; s != NULL; s = s->next)
2194 {
2195 inarcsorig[s->no] = s->ins;
2196 totalinarcs += s->nins;
2197 }
2198
2199 /*
2200 * Create a workspace for accumulating the inarcs to be added to the
2201 * current target state. totalinarcs is probably a considerable
2202 * overestimate of the space needed, but the NFA is unlikely to be large
2203 * enough at this point to make it worth being smarter.
2204 */
2205 arcarray = (struct arc **) MALLOC(totalinarcs * sizeof(struct arc *));
2206 if (arcarray == NULL)
2207 {
2209 FREE(inarcsorig);
2210 return;
2211 }
2212
2213 /* And iterate over the target states */
2214 for (s = nfa->states; s != NULL && !NISERR(); s = s->next)
2215 {
2216 /* Ignore target states without non-EMPTY outarcs, per note above */
2217 if (!s->flag && !hasnonemptyout(s))
2218 continue;
2219
2220 /* Find predecessor states and accumulate their original inarcs */
2221 arccount = 0;
2222 for (s2 = emptyreachable(nfa, s, s, inarcsorig); s2 != s; s2 = nexts)
2223 {
2224 /* Add s2's original inarcs to arcarray[], but ignore empties */
2225 for (a = inarcsorig[s2->no]; a != NULL; a = a->inchain)
2226 {
2227 if (a->type != EMPTY)
2228 arcarray[arccount++] = a;
2229 }
2230
2231 /* Reset the tmp fields as we walk back */
2232 nexts = s2->tmp;
2233 s2->tmp = NULL;
2234 }
2235 s->tmp = NULL;
2236 assert(arccount <= totalinarcs);
2237
2238 /* Remember how many original inarcs this state has */
2239 prevnins = s->nins;
2240
2241 /* Add non-duplicate inarcs to target state */
2242 mergeins(nfa, s, arcarray, arccount);
2243
2244 /* Now we must update the state's inarcsorig pointer */
2245 nskip = s->nins - prevnins;
2246 a = s->ins;
2247 while (nskip-- > 0)
2248 a = a->inchain;
2249 inarcsorig[s->no] = a;
2250 }
2251
2252 FREE(arcarray);
2253 FREE(inarcsorig);
2254
2255 if (NISERR())
2256 return;
2257
2258 /*
2259 * Now remove all the EMPTY arcs, since we don't need them anymore.
2260 */
2261 for (s = nfa->states; s != NULL; s = s->next)
2262 {
2263 for (a = s->outs; a != NULL; a = nexta)
2264 {
2265 nexta = a->outchain;
2266 if (a->type == EMPTY)
2267 freearc(nfa, a);
2268 }
2269 }
2270
2271 /*
2272 * And remove any states that have become useless. (This cleanup is not
2273 * very thorough, and would be even less so if we tried to combine it with
2274 * the previous step; but cleanup() will take care of anything we miss.)
2275 */
2276 for (s = nfa->states; s != NULL; s = nexts)
2277 {
2278 nexts = s->next;
2279 if ((s->nins == 0 || s->nouts == 0) && !s->flag)
2280 dropstate(nfa, s);
2281 }
2282
2283 if (f != NULL)
2284 dumpnfa(nfa, f);
2285}
char * s2
static void mergeins(struct nfa *nfa, struct state *s, struct arc **arcarray, int arccount)
Definition: regc_nfa.c:971
static void moveouts(struct nfa *nfa, struct state *oldState, struct state *newState)
Definition: regc_nfa.c:1066
static int hasnonemptyout(struct state *s)
Definition: regc_nfa.c:575
static void moveins(struct nfa *nfa, struct state *oldState, struct state *newState)
Definition: regc_nfa.c:778

References a, assert, dropstate(), dumpnfa(), EMPTY, emptyreachable(), state::flag, FREE, freearc(), hasnonemptyout(), state::ins, MALLOC, mergeins(), moveins(), moveouts(), NERR, state::next, state::nins, NISERR, state::no, state::nouts, nfa::nstates, state::outs, REG_ESPACE, s2, nfa::states, and state::tmp.

Referenced by optimize().

◆ freearc()

static void freearc ( struct nfa nfa,
struct arc victim 
)
static

Definition at line 418 of file regc_nfa.c.

420{
421 struct state *from = victim->from;
422 struct state *to = victim->to;
423 struct arc *predecessor;
424
425 assert(victim->type != 0);
426
427 /* take it off color chain if necessary */
428 if (COLORED(victim) && nfa->parent == NULL)
429 uncolorchain(nfa->cm, victim);
430
431 /* take it off source's out-chain */
432 assert(from != NULL);
433 predecessor = victim->outchainRev;
434 if (predecessor == NULL)
435 {
436 assert(from->outs == victim);
437 from->outs = victim->outchain;
438 }
439 else
440 {
441 assert(predecessor->outchain == victim);
442 predecessor->outchain = victim->outchain;
443 }
444 if (victim->outchain != NULL)
445 {
446 assert(victim->outchain->outchainRev == victim);
447 victim->outchain->outchainRev = predecessor;
448 }
449 from->nouts--;
450
451 /* take it off target's in-chain */
452 assert(to != NULL);
453 predecessor = victim->inchainRev;
454 if (predecessor == NULL)
455 {
456 assert(to->ins == victim);
457 to->ins = victim->inchain;
458 }
459 else
460 {
461 assert(predecessor->inchain == victim);
462 predecessor->inchain = victim->inchain;
463 }
464 if (victim->inchain != NULL)
465 {
466 assert(victim->inchain->inchainRev == victim);
467 victim->inchain->inchainRev = predecessor;
468 }
469 to->nins--;
470
471 /* clean up and place on NFA's free list */
472 victim->type = 0;
473 victim->from = NULL; /* precautions... */
474 victim->to = NULL;
475 victim->inchain = NULL;
476 victim->inchainRev = NULL;
477 victim->outchain = NULL;
478 victim->outchainRev = NULL;
479 victim->freechain = nfa->freearcs;
480 nfa->freearcs = victim;
481}
static void uncolorchain(struct colormap *cm, struct arc *a)
Definition: regc_color.c:1001

References assert, nfa::cm, COLORED, nfa::freearcs, arc::from, arc::inchain, arc::inchainRev, state::ins, state::nins, state::nouts, arc::outchain, arc::outchainRev, state::outs, nfa::parent, arc::to, arc::type, and uncolorchain().

Referenced by breakconstraintloop(), deltraverse(), dropstate(), fixconstraintloops(), fixempties(), moveins(), moveouts(), pull(), pullback(), push(), pushfwd(), removecantmatch(), and removetraverse().

◆ freecnfa()

static void freecnfa ( struct cnfa cnfa)
static

Definition at line 3633 of file regc_nfa.c.

3634{
3635 assert(!NULLCNFA(*cnfa)); /* not empty already */
3636 FREE(cnfa->stflags);
3637 FREE(cnfa->states);
3638 FREE(cnfa->arcs);
3639 ZAPCNFA(*cnfa);
3640}
#define ZAPCNFA(cnfa)
Definition: regguts.h:436
#define NULLCNFA(cnfa)
Definition: regguts.h:438

References cnfa::arcs, assert, FREE, NULLCNFA, cnfa::states, cnfa::stflags, and ZAPCNFA.

◆ freenfa()

static void freenfa ( struct nfa nfa)
static

Definition at line 107 of file regc_nfa.c.

108{
109 struct statebatch *sb;
110 struct statebatch *sbnext;
111 struct arcbatch *ab;
112 struct arcbatch *abnext;
113
114 for (sb = nfa->lastsb; sb != NULL; sb = sbnext)
115 {
116 sbnext = sb->next;
118 FREE(sb);
119 }
120 nfa->lastsb = NULL;
121 for (ab = nfa->lastab; ab != NULL; ab = abnext)
122 {
123 abnext = ab->next;
124 nfa->v->spaceused -= ARCBATCHSIZE(ab->narcs);
125 FREE(ab);
126 }
127 nfa->lastab = NULL;
128
129 nfa->nstates = -1;
130 FREE(nfa);
131}
#define STATEBATCHSIZE(n)
Definition: regguts.h:343
struct statebatch * lastsb
Definition: regguts.h:359
struct statebatch * next
Definition: regguts.h:339
size_t nstates
Definition: regguts.h:340

References ARCBATCHSIZE, FREE, nfa::lastab, nfa::lastsb, arcbatch::narcs, arcbatch::next, statebatch::next, statebatch::nstates, nfa::nstates, vars::spaceused, STATEBATCHSIZE, and nfa::v.

Referenced by newnfa().

◆ freestate()

static void freestate ( struct nfa nfa,
struct state s 
)
static

Definition at line 242 of file regc_nfa.c.

244{
245 assert(s != NULL);
246 assert(s->nins == 0 && s->nouts == 0);
247
248 s->no = FREESTATE;
249 s->flag = 0;
250 if (s->next != NULL)
251 s->next->prev = s->prev;
252 else
253 {
254 assert(s == nfa->slast);
255 nfa->slast = s->prev;
256 }
257 if (s->prev != NULL)
258 s->prev->next = s->next;
259 else
260 {
261 assert(s == nfa->states);
262 nfa->states = s->next;
263 }
264 s->prev = NULL;
265 s->next = nfa->freestates; /* don't delete it, put it on the free list */
266 nfa->freestates = s;
267}
struct state * slast
Definition: regguts.h:356
struct state * freestates
Definition: regguts.h:357
struct state * prev
Definition: regguts.h:334

References assert, state::flag, FREESTATE, nfa::freestates, state::next, state::nins, state::no, state::nouts, state::prev, nfa::slast, and nfa::states.

Referenced by breakconstraintloop(), deltraverse(), and dropstate().

◆ hasconstraintout()

static int hasconstraintout ( struct state s)
static

Definition at line 2349 of file regc_nfa.c.

2350{
2351 struct arc *a;
2352
2353 for (a = s->outs; a != NULL; a = a->outchain)
2354 {
2355 if (isconstraintarc(a))
2356 return 1;
2357 }
2358 return 0;
2359}

References a, isconstraintarc(), and state::outs.

Referenced by clonesuccessorstates().

◆ hasnonemptyout()

static int hasnonemptyout ( struct state s)
static

Definition at line 575 of file regc_nfa.c.

576{
577 struct arc *a;
578
579 for (a = s->outs; a != NULL; a = a->outchain)
580 {
581 if (a->type != EMPTY)
582 return 1;
583 }
584 return 0;
585}

References a, EMPTY, and state::outs.

Referenced by fixempties().

◆ isconstraintarc()

static int isconstraintarc ( struct arc a)
inlinestatic

Definition at line 2331 of file regc_nfa.c.

2332{
2333 switch (a->type)
2334 {
2335 case '^':
2336 case '$':
2337 case BEHIND:
2338 case AHEAD:
2339 case LACON:
2340 return 1;
2341 }
2342 return 0;
2343}

References a, AHEAD, BEHIND, and LACON.

Referenced by breakconstraintloop(), clonesuccessorstates(), findconstraintloop(), fixconstraintloops(), and hasconstraintout().

◆ markcanreach()

static void markcanreach ( struct nfa nfa,
struct state s,
struct state okay,
struct state mark 
)
static

Definition at line 3025 of file regc_nfa.c.

3029{
3030 struct arc *a;
3031
3032 /* Since this is recursive, it could be driven to stack overflow */
3033 if (STACK_TOO_DEEP(nfa->v->re))
3034 {
3036 return;
3037 }
3038
3039 if (s->tmp != okay)
3040 return;
3041 s->tmp = mark;
3042
3043 for (a = s->ins; a != NULL; a = a->inchain)
3044 markcanreach(nfa, a->from, okay, mark);
3045}

References a, state::ins, markcanreach(), NERR, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by cleanup(), and markcanreach().

◆ markreachable()

static void markreachable ( struct nfa nfa,
struct state s,
struct state okay,
struct state mark 
)
static

Definition at line 2999 of file regc_nfa.c.

3003{
3004 struct arc *a;
3005
3006 /* Since this is recursive, it could be driven to stack overflow */
3007 if (STACK_TOO_DEEP(nfa->v->re))
3008 {
3010 return;
3011 }
3012
3013 if (s->tmp != okay)
3014 return;
3015 s->tmp = mark;
3016
3017 for (a = s->outs; a != NULL; a = a->outchain)
3018 markreachable(nfa, a->to, okay, mark);
3019}

References a, markreachable(), NERR, state::outs, vars::re, REG_ETOOBIG, STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by cleanup(), and markreachable().

◆ mergeins()

static void mergeins ( struct nfa nfa,
struct state s,
struct arc **  arcarray,
int  arccount 
)
static

Definition at line 971 of file regc_nfa.c.

975{
976 struct arc *na;
977 int i;
978 int j;
979
980 if (arccount <= 0)
981 return;
982
983 /*
984 * Because we bypass newarc() in this code path, we'd better include a
985 * cancel check.
986 */
987 INTERRUPT(nfa->v->re);
988
989 /* Sort existing inarcs as well as proposed new ones */
990 sortins(nfa, s);
991 if (NISERR())
992 return; /* might have failed to sort */
993
994 qsort(arcarray, arccount, sizeof(struct arc *), sortins_cmp);
995
996 /*
997 * arcarray very likely includes dups, so we must eliminate them. (This
998 * could be folded into the next loop, but it's not worth the trouble.)
999 */
1000 j = 0;
1001 for (i = 1; i < arccount; i++)
1002 {
1003 switch (sortins_cmp(&arcarray[j], &arcarray[i]))
1004 {
1005 case -1:
1006 /* non-dup */
1007 arcarray[++j] = arcarray[i];
1008 break;
1009 case 0:
1010 /* dup */
1011 break;
1012 default:
1013 /* trouble */
1015 }
1016 }
1017 arccount = j + 1;
1018
1019 /*
1020 * Now merge into s' inchain. Note that createarc() will put new arcs
1021 * onto the front of s's chain, so it does not break our walk through the
1022 * sorted part of the chain.
1023 */
1024 i = 0;
1025 na = s->ins;
1026 while (i < arccount && na != NULL)
1027 {
1028 struct arc *a = arcarray[i];
1029
1030 switch (sortins_cmp(&a, &na))
1031 {
1032 case -1:
1033 /* s does not have anything matching a */
1034 createarc(nfa, a->type, a->co, a->from, s);
1035 i++;
1036 break;
1037 case 0:
1038 /* match, advance in both lists */
1039 i++;
1040 na = na->inchain;
1041 break;
1042 case +1:
1043 /* advance only na; array might have a match later */
1044 na = na->inchain;
1045 break;
1046 default:
1048 }
1049 }
1050 while (i < arccount)
1051 {
1052 /* s does not have anything matching a */
1053 struct arc *a = arcarray[i];
1054
1055 createarc(nfa, a->type, a->co, a->from, s);
1056 i++;
1057 }
1058}
int j
Definition: isn.c:73

References a, assert, createarc(), i, arc::inchain, state::ins, INTERRUPT, j, NISERR, NOTREACHED, qsort, vars::re, sortins(), sortins_cmp(), and nfa::v.

Referenced by fixempties().

◆ moveins()

static void moveins ( struct nfa nfa,
struct state oldState,
struct state newState 
)
static

Definition at line 778 of file regc_nfa.c.

781{
782 assert(oldState != newState);
783
784 if (newState->nins == 0)
785 {
786 /* No need for de-duplication */
787 struct arc *a;
788
789 while ((a = oldState->ins) != NULL)
790 {
791 createarc(nfa, a->type, a->co, a->from, newState);
792 freearc(nfa, a);
793 }
794 }
795 else if (!BULK_ARC_OP_USE_SORT(oldState->nins, newState->nins))
796 {
797 /* With not too many arcs, just do them one at a time */
798 struct arc *a;
799
800 while ((a = oldState->ins) != NULL)
801 {
802 cparc(nfa, a, a->from, newState);
803 freearc(nfa, a);
804 }
805 }
806 else
807 {
808 /*
809 * With many arcs, use a sort-merge approach. Note changearctarget()
810 * will put the arc onto the front of newState's chain, so it does not
811 * break our walk through the sorted part of the chain.
812 */
813 struct arc *oa;
814 struct arc *na;
815
816 /*
817 * Because we bypass newarc() in this code path, we'd better include a
818 * cancel check.
819 */
820 INTERRUPT(nfa->v->re);
821
822 sortins(nfa, oldState);
823 sortins(nfa, newState);
824 if (NISERR())
825 return; /* might have failed to sort */
826 oa = oldState->ins;
827 na = newState->ins;
828 while (oa != NULL && na != NULL)
829 {
830 struct arc *a = oa;
831
832 switch (sortins_cmp(&oa, &na))
833 {
834 case -1:
835 /* newState does not have anything matching oa */
836 oa = oa->inchain;
837
838 /*
839 * Rather than doing createarc+freearc, we can just unlink
840 * and relink the existing arc struct.
841 */
842 changearctarget(a, newState);
843 break;
844 case 0:
845 /* match, advance in both lists */
846 oa = oa->inchain;
847 na = na->inchain;
848 /* ... and drop duplicate arc from oldState */
849 freearc(nfa, a);
850 break;
851 case +1:
852 /* advance only na; oa might have a match later */
853 na = na->inchain;
854 break;
855 default:
857 }
858 }
859 while (oa != NULL)
860 {
861 /* newState does not have anything matching oa */
862 struct arc *a = oa;
863
864 oa = oa->inchain;
865 changearctarget(a, newState);
866 }
867 }
868
869 assert(oldState->nins == 0);
870 assert(oldState->ins == NULL);
871}
static void changearctarget(struct arc *a, struct state *newto)
Definition: regc_nfa.c:533

References a, assert, BULK_ARC_OP_USE_SORT, changearctarget(), cparc(), createarc(), freearc(), arc::inchain, state::ins, INTERRUPT, state::nins, NISERR, NOTREACHED, vars::re, sortins(), sortins_cmp(), and nfa::v.

Referenced by fixempties(), and pull().

◆ moveouts()

static void moveouts ( struct nfa nfa,
struct state oldState,
struct state newState 
)
static

Definition at line 1066 of file regc_nfa.c.

1069{
1070 assert(oldState != newState);
1071
1072 if (newState->nouts == 0)
1073 {
1074 /* No need for de-duplication */
1075 struct arc *a;
1076
1077 while ((a = oldState->outs) != NULL)
1078 {
1079 createarc(nfa, a->type, a->co, newState, a->to);
1080 freearc(nfa, a);
1081 }
1082 }
1083 else if (!BULK_ARC_OP_USE_SORT(oldState->nouts, newState->nouts))
1084 {
1085 /* With not too many arcs, just do them one at a time */
1086 struct arc *a;
1087
1088 while ((a = oldState->outs) != NULL)
1089 {
1090 cparc(nfa, a, newState, a->to);
1091 freearc(nfa, a);
1092 }
1093 }
1094 else
1095 {
1096 /*
1097 * With many arcs, use a sort-merge approach. Note changearcsource()
1098 * will put the arc onto the front of newState's chain, so it does not
1099 * break our walk through the sorted part of the chain.
1100 */
1101 struct arc *oa;
1102 struct arc *na;
1103
1104 /*
1105 * Because we bypass newarc() in this code path, we'd better include a
1106 * cancel check.
1107 */
1108 INTERRUPT(nfa->v->re);
1109
1110 sortouts(nfa, oldState);
1111 sortouts(nfa, newState);
1112 if (NISERR())
1113 return; /* might have failed to sort */
1114 oa = oldState->outs;
1115 na = newState->outs;
1116 while (oa != NULL && na != NULL)
1117 {
1118 struct arc *a = oa;
1119
1120 switch (sortouts_cmp(&oa, &na))
1121 {
1122 case -1:
1123 /* newState does not have anything matching oa */
1124 oa = oa->outchain;
1125
1126 /*
1127 * Rather than doing createarc+freearc, we can just unlink
1128 * and relink the existing arc struct.
1129 */
1130 changearcsource(a, newState);
1131 break;
1132 case 0:
1133 /* match, advance in both lists */
1134 oa = oa->outchain;
1135 na = na->outchain;
1136 /* ... and drop duplicate arc from oldState */
1137 freearc(nfa, a);
1138 break;
1139 case +1:
1140 /* advance only na; oa might have a match later */
1141 na = na->outchain;
1142 break;
1143 default:
1145 }
1146 }
1147 while (oa != NULL)
1148 {
1149 /* newState does not have anything matching oa */
1150 struct arc *a = oa;
1151
1152 oa = oa->outchain;
1153 changearcsource(a, newState);
1154 }
1155 }
1156
1157 assert(oldState->nouts == 0);
1158 assert(oldState->outs == NULL);
1159}
static void changearcsource(struct arc *a, struct state *newfrom)
Definition: regc_nfa.c:489

References a, assert, BULK_ARC_OP_USE_SORT, changearcsource(), cparc(), createarc(), freearc(), INTERRUPT, NISERR, NOTREACHED, state::nouts, arc::outchain, state::outs, vars::re, sortouts(), sortouts_cmp(), and nfa::v.

Referenced by fixempties(), and push().

◆ newarc()

static void newarc ( struct nfa nfa,
int  t,
color  co,
struct state from,
struct state to 
)
static

Definition at line 281 of file regc_nfa.c.

286{
287 struct arc *a;
288
289 assert(from != NULL && to != NULL);
290
291 /*
292 * This is a handy place to check for operation cancel during regex
293 * compilation, since no code path will go very long without making a new
294 * state or arc.
295 */
296 INTERRUPT(nfa->v->re);
297
298 /* check for duplicate arc, using whichever chain is shorter */
299 if (from->nouts <= to->nins)
300 {
301 for (a = from->outs; a != NULL; a = a->outchain)
302 if (a->to == to && a->co == co && a->type == t)
303 return;
304 }
305 else
306 {
307 for (a = to->ins; a != NULL; a = a->inchain)
308 if (a->from == from && a->co == co && a->type == t)
309 return;
310 }
311
312 /* no dup, so create the arc */
313 createarc(nfa, t, co, from, to);
314}

References a, assert, arc::co, createarc(), arc::from, state::ins, INTERRUPT, state::nins, state::nouts, state::outs, vars::re, arc::to, and nfa::v.

Referenced by cloneouts(), colorcomplement(), cparc(), dupnfa(), newnfa(), okcolors(), pull(), pullback(), push(), pushfwd(), rainbow(), removetraverse(), subcolorcvec(), subcoloronechr(), and subcoloronerow().

◆ newfstate()

static struct state * newfstate ( struct nfa nfa,
int  flag 
)
static

Definition at line 212 of file regc_nfa.c.

213{
214 struct state *s;
215
216 s = newstate(nfa);
217 if (s != NULL)
218 s->flag = (char) flag;
219 return s;
220}
char * flag(int b)
Definition: test-ctype.c:33

References state::flag, flag(), and newstate().

Referenced by newnfa().

◆ newnfa()

static struct nfa * newnfa ( struct vars v,
struct colormap cm,
struct nfa parent 
)
static

Definition at line 47 of file regc_nfa.c.

50{
51 struct nfa *nfa;
52
53 nfa = (struct nfa *) MALLOC(sizeof(struct nfa));
54 if (nfa == NULL)
55 {
57 return NULL;
58 }
59
60 /* Make the NFA minimally valid, so freenfa() will behave sanely */
61 nfa->states = NULL;
62 nfa->slast = NULL;
63 nfa->freestates = NULL;
64 nfa->freearcs = NULL;
65 nfa->lastsb = NULL;
66 nfa->lastab = NULL;
67 nfa->lastsbused = 0;
68 nfa->lastabused = 0;
69 nfa->nstates = 0;
70 nfa->cm = cm;
71 nfa->v = v;
72 nfa->bos[0] = nfa->bos[1] = COLORLESS;
73 nfa->eos[0] = nfa->eos[1] = COLORLESS;
74 nfa->flags = 0;
76 nfa->parent = parent; /* Precedes newfstate so parent is valid. */
77
78 /* Create required infrastructure */
79 nfa->post = newfstate(nfa, '@'); /* number 0 */
80 nfa->pre = newfstate(nfa, '>'); /* number 1 */
81 nfa->init = newstate(nfa); /* may become invalid later */
83 if (ISERR())
84 {
85 freenfa(nfa);
86 return NULL;
87 }
89 newarc(nfa, '^', 1, nfa->pre, nfa->init);
90 newarc(nfa, '^', 0, nfa->pre, nfa->init);
92 newarc(nfa, '$', 1, nfa->final, nfa->post);
93 newarc(nfa, '$', 0, nfa->final, nfa->post);
94
95 if (ISERR())
96 {
97 freenfa(nfa);
98 return NULL;
99 }
100 return nfa;
101}
#define ERR
Definition: _int.h:161
static void rainbow(struct nfa *nfa, struct colormap *cm, int type, color but, struct state *from, struct state *to)
Definition: regc_color.c:1031
static void freenfa(struct nfa *nfa)
Definition: regc_nfa.c:107
static struct state * newfstate(struct nfa *nfa, int flag)
Definition: regc_nfa.c:212
#define ISERR()
Definition: regcomp.c:317
struct state * final
Definition: regguts.h:352
size_t lastsbused
Definition: regguts.h:361
struct state * init
Definition: regguts.h:351

References nfa::bos, nfa::cm, COLORLESS, nfa::eos, ERR, nfa::final, nfa::flags, nfa::freearcs, freenfa(), nfa::freestates, nfa::init, ISERR, nfa::lastab, nfa::lastabused, nfa::lastsb, nfa::lastsbused, MALLOC, nfa::maxmatchall, nfa::minmatchall, newarc(), newfstate(), newstate(), nfa::nstates, nfa::parent, PLAIN, nfa::post, nfa::pre, rainbow(), REG_ESPACE, nfa::slast, nfa::states, and nfa::v.

◆ newstate()

static struct state * newstate ( struct nfa nfa)
static

Definition at line 137 of file regc_nfa.c.

138{
139 struct state *s;
140
141 /*
142 * This is a handy place to check for operation cancel during regex
143 * compilation, since no code path will go very long without making a new
144 * state or arc.
145 */
146 INTERRUPT(nfa->v->re);
147
148 /* first, recycle anything that's on the freelist */
149 if (nfa->freestates != NULL)
150 {
151 s = nfa->freestates;
152 nfa->freestates = s->next;
153 }
154 /* otherwise, is there anything left in the last statebatch? */
155 else if (nfa->lastsb != NULL && nfa->lastsbused < nfa->lastsb->nstates)
156 {
157 s = &nfa->lastsb->s[nfa->lastsbused++];
158 }
159 /* otherwise, need to allocate a new statebatch */
160 else
161 {
162 struct statebatch *newSb;
163 size_t nstates;
164
166 {
168 return NULL;
169 }
170 nstates = (nfa->lastsb != NULL) ? nfa->lastsb->nstates * 2 : FIRSTSBSIZE;
173 newSb = (struct statebatch *) MALLOC(STATEBATCHSIZE(nstates));
174 if (newSb == NULL)
175 {
177 return NULL;
178 }
180 newSb->nstates = nstates;
181 newSb->next = nfa->lastsb;
182 nfa->lastsb = newSb;
183 nfa->lastsbused = 1;
184 s = &newSb->s[0];
185 }
186
187 assert(nfa->nstates >= 0);
188 s->no = nfa->nstates++;
189 s->flag = 0;
190 if (nfa->states == NULL)
191 nfa->states = s;
192 s->nins = 0;
193 s->ins = NULL;
194 s->nouts = 0;
195 s->outs = NULL;
196 s->tmp = NULL;
197 s->next = NULL;
198 if (nfa->slast != NULL)
199 {
200 assert(nfa->slast->next == NULL);
201 nfa->slast->next = s;
202 }
203 s->prev = nfa->slast;
204 nfa->slast = s;
205 return s;
206}
#define MAXSBSIZE
Definition: regguts.h:346
#define FIRSTSBSIZE
Definition: regguts.h:345
struct state s[FLEXIBLE_ARRAY_MEMBER]
Definition: regguts.h:341

References assert, FIRSTSBSIZE, state::flag, nfa::freestates, if(), state::ins, INTERRUPT, nfa::lastsb, nfa::lastsbused, MALLOC, MAXSBSIZE, NERR, state::next, statebatch::next, state::nins, state::no, state::nouts, statebatch::nstates, nfa::nstates, state::outs, state::prev, vars::re, REG_ESPACE, REG_ETOOBIG, REG_MAX_COMPILE_SPACE, statebatch::s, nfa::slast, vars::spaceused, STATEBATCHSIZE, nfa::states, state::tmp, and nfa::v.

Referenced by breakconstraintloop(), clonesuccessorstates(), duptraverse(), newfstate(), newnfa(), pull(), push(), and SetMatViewPopulatedState().

◆ optimize()

static long optimize ( struct nfa nfa,
FILE *  f 
)
static

Definition at line 1594 of file regc_nfa.c.

1596{
1597#ifdef REG_DEBUG
1598 int verbose = (f != NULL) ? 1 : 0;
1599
1600 if (verbose)
1601 fprintf(f, "\ninitial cleanup:\n");
1602#endif
1603 /* If we have any CANTMATCH arcs, drop them; but this is uncommon */
1604 if (nfa->flags & HASCANTMATCH)
1605 {
1607 nfa->flags &= ~HASCANTMATCH;
1608 }
1609 cleanup(nfa); /* may simplify situation */
1610#ifdef REG_DEBUG
1611 if (verbose)
1612 dumpnfa(nfa, f);
1613 if (verbose)
1614 fprintf(f, "\nempties:\n");
1615#endif
1616 fixempties(nfa, f); /* get rid of EMPTY arcs */
1617#ifdef REG_DEBUG
1618 if (verbose)
1619 fprintf(f, "\nconstraints:\n");
1620#endif
1621 fixconstraintloops(nfa, f); /* get rid of constraint loops */
1622 pullback(nfa, f); /* pull back constraints backward */
1623 pushfwd(nfa, f); /* push fwd constraints forward */
1624#ifdef REG_DEBUG
1625 if (verbose)
1626 fprintf(f, "\nfinal cleanup:\n");
1627#endif
1628 cleanup(nfa); /* final tidying */
1629#ifdef REG_DEBUG
1630 if (verbose)
1631 dumpnfa(nfa, f);
1632#endif
1633 return analyze(nfa); /* and analysis */
1634}
int verbose
static void cleanup(struct nfa *nfa)
Definition: regc_nfa.c:2964
static long analyze(struct nfa *nfa)
Definition: regc_nfa.c:3051
static void fixempties(struct nfa *nfa, FILE *f)
Definition: regc_nfa.c:2076
static void removecantmatch(struct nfa *nfa)
Definition: regc_nfa.c:2938
static void pushfwd(struct nfa *nfa, FILE *f)
Definition: regc_nfa.c:1811
static void pullback(struct nfa *nfa, FILE *f)
Definition: regc_nfa.c:1640
static void fixconstraintloops(struct nfa *nfa, FILE *f)
Definition: regc_nfa.c:2370

References analyze(), cleanup(), dumpnfa(), fixconstraintloops(), fixempties(), nfa::flags, fprintf, HASCANTMATCH, pullback(), pushfwd(), removecantmatch(), and verbose.

◆ pull()

static int pull ( struct nfa nfa,
struct arc con,
struct state **  intermediates 
)
static

Definition at line 1720 of file regc_nfa.c.

1723{
1724 struct state *from = con->from;
1725 struct state *to = con->to;
1726 struct arc *a;
1727 struct arc *nexta;
1728 struct state *s;
1729
1730 assert(from != to); /* should have gotten rid of this earlier */
1731 if (from->flag) /* can't pull back beyond start */
1732 return 0;
1733 if (from->nins == 0)
1734 { /* unreachable */
1735 freearc(nfa, con);
1736 return 1;
1737 }
1738
1739 /*
1740 * First, clone from state if necessary to avoid other outarcs. This may
1741 * seem wasteful, but it simplifies the logic, and we'll get rid of the
1742 * clone state again at the bottom.
1743 */
1744 if (from->nouts > 1)
1745 {
1746 s = newstate(nfa);
1747 if (NISERR())
1748 return 0;
1749 copyins(nfa, from, s); /* duplicate inarcs */
1750 cparc(nfa, con, s, to); /* move constraint arc */
1751 freearc(nfa, con);
1752 if (NISERR())
1753 return 0;
1754 from = s;
1755 con = from->outs;
1756 }
1757 assert(from->nouts == 1);
1758
1759 /* propagate the constraint into the from state's inarcs */
1760 for (a = from->ins; a != NULL && !NISERR(); a = nexta)
1761 {
1762 nexta = a->inchain;
1763 switch (combine(nfa, con, a))
1764 {
1765 case INCOMPATIBLE: /* destroy the arc */
1766 freearc(nfa, a);
1767 break;
1768 case SATISFIED: /* no action needed */
1769 break;
1770 case COMPATIBLE: /* swap the two arcs, more or less */
1771 /* need an intermediate state, but might have one already */
1772 for (s = *intermediates; s != NULL; s = s->tmp)
1773 {
1774 assert(s->nins > 0 && s->nouts > 0);
1775 if (s->ins->from == a->from && s->outs->to == to)
1776 break;
1777 }
1778 if (s == NULL)
1779 {
1780 s = newstate(nfa);
1781 if (NISERR())
1782 return 0;
1783 s->tmp = *intermediates;
1784 *intermediates = s;
1785 }
1786 cparc(nfa, con, a->from, s);
1787 cparc(nfa, a, s, to);
1788 freearc(nfa, a);
1789 break;
1790 case REPLACEARC: /* replace arc's color */
1791 newarc(nfa, a->type, con->co, a->from, to);
1792 freearc(nfa, a);
1793 break;
1794 default:
1796 break;
1797 }
1798 }
1799
1800 /* remaining inarcs, if any, incorporate the constraint */
1801 moveins(nfa, from, to);
1802 freearc(nfa, con);
1803 /* from state is now useless, but we leave it to pullback() to clean up */
1804 return 1;
1805}
static int combine(struct nfa *nfa, struct arc *con, struct arc *a)
Definition: regc_nfa.c:1987
static void copyins(struct nfa *nfa, struct state *oldState, struct state *newState)
Definition: regc_nfa.c:882

References a, assert, arc::co, combine(), COMPATIBLE, copyins(), cparc(), state::flag, freearc(), arc::from, INCOMPATIBLE, state::ins, moveins(), newarc(), newstate(), state::nins, NISERR, NOTREACHED, state::nouts, state::outs, REPLACEARC, SATISFIED, state::tmp, and arc::to.

Referenced by pullback().

◆ pullback()

static void pullback ( struct nfa nfa,
FILE *  f 
)
static

Definition at line 1640 of file regc_nfa.c.

1642{
1643 struct state *s;
1644 struct state *nexts;
1645 struct arc *a;
1646 struct arc *nexta;
1647 struct state *intermediates;
1648 int progress;
1649
1650 /* find and pull until there are no more */
1651 do
1652 {
1653 progress = 0;
1654 for (s = nfa->states; s != NULL && !NISERR(); s = nexts)
1655 {
1656 nexts = s->next;
1657 intermediates = NULL;
1658 for (a = s->outs; a != NULL && !NISERR(); a = nexta)
1659 {
1660 nexta = a->outchain;
1661 if (a->type == '^' || a->type == BEHIND)
1662 if (pull(nfa, a, &intermediates))
1663 progress = 1;
1664 }
1665 /* clear tmp fields of intermediate states created here */
1666 while (intermediates != NULL)
1667 {
1668 struct state *ns = intermediates->tmp;
1669
1670 intermediates->tmp = NULL;
1671 intermediates = ns;
1672 }
1673 /* if s is now useless, get rid of it */
1674 if ((s->nins == 0 || s->nouts == 0) && !s->flag)
1675 dropstate(nfa, s);
1676 }
1677 if (progress && f != NULL)
1678 dumpnfa(nfa, f);
1679 } while (progress && !NISERR());
1680 if (NISERR())
1681 return;
1682
1683 /*
1684 * Any ^ constraints we were able to pull to the start state can now be
1685 * replaced by PLAIN arcs referencing the BOS or BOL colors. There should
1686 * be no other ^ or BEHIND arcs left in the NFA, though we do not check
1687 * that here (compact() will fail if so).
1688 */
1689 for (a = nfa->pre->outs; a != NULL; a = nexta)
1690 {
1691 nexta = a->outchain;
1692 if (a->type == '^')
1693 {
1694 assert(a->co == 0 || a->co == 1);
1695 newarc(nfa, PLAIN, nfa->bos[a->co], a->from, a->to);
1696 freearc(nfa, a);
1697 }
1698 }
1699}
static int progress
Definition: pgbench.c:261
static int pull(struct nfa *nfa, struct arc *con, struct state **intermediates)
Definition: regc_nfa.c:1720

References a, assert, BEHIND, nfa::bos, dropstate(), dumpnfa(), state::flag, freearc(), newarc(), state::next, state::nins, NISERR, state::nouts, state::outs, PLAIN, nfa::pre, progress, pull(), nfa::states, and state::tmp.

Referenced by optimize().

◆ push()

static int push ( struct nfa nfa,
struct arc con,
struct state **  intermediates 
)
static

Definition at line 1891 of file regc_nfa.c.

1894{
1895 struct state *from = con->from;
1896 struct state *to = con->to;
1897 struct arc *a;
1898 struct arc *nexta;
1899 struct state *s;
1900
1901 assert(to != from); /* should have gotten rid of this earlier */
1902 if (to->flag) /* can't push forward beyond end */
1903 return 0;
1904 if (to->nouts == 0)
1905 { /* dead end */
1906 freearc(nfa, con);
1907 return 1;
1908 }
1909
1910 /*
1911 * First, clone to state if necessary to avoid other inarcs. This may
1912 * seem wasteful, but it simplifies the logic, and we'll get rid of the
1913 * clone state again at the bottom.
1914 */
1915 if (to->nins > 1)
1916 {
1917 s = newstate(nfa);
1918 if (NISERR())
1919 return 0;
1920 copyouts(nfa, to, s); /* duplicate outarcs */
1921 cparc(nfa, con, from, s); /* move constraint arc */
1922 freearc(nfa, con);
1923 if (NISERR())
1924 return 0;
1925 to = s;
1926 con = to->ins;
1927 }
1928 assert(to->nins == 1);
1929
1930 /* propagate the constraint into the to state's outarcs */
1931 for (a = to->outs; a != NULL && !NISERR(); a = nexta)
1932 {
1933 nexta = a->outchain;
1934 switch (combine(nfa, con, a))
1935 {
1936 case INCOMPATIBLE: /* destroy the arc */
1937 freearc(nfa, a);
1938 break;
1939 case SATISFIED: /* no action needed */
1940 break;
1941 case COMPATIBLE: /* swap the two arcs, more or less */
1942 /* need an intermediate state, but might have one already */
1943 for (s = *intermediates; s != NULL; s = s->tmp)
1944 {
1945 assert(s->nins > 0 && s->nouts > 0);
1946 if (s->ins->from == from && s->outs->to == a->to)
1947 break;
1948 }
1949 if (s == NULL)
1950 {
1951 s = newstate(nfa);
1952 if (NISERR())
1953 return 0;
1954 s->tmp = *intermediates;
1955 *intermediates = s;
1956 }
1957 cparc(nfa, con, s, a->to);
1958 cparc(nfa, a, from, s);
1959 freearc(nfa, a);
1960 break;
1961 case REPLACEARC: /* replace arc's color */
1962 newarc(nfa, a->type, con->co, from, a->to);
1963 freearc(nfa, a);
1964 break;
1965 default:
1967 break;
1968 }
1969 }
1970
1971 /* remaining outarcs, if any, incorporate the constraint */
1972 moveouts(nfa, to, from);
1973 freearc(nfa, con);
1974 /* to state is now useless, but we leave it to pushfwd() to clean up */
1975 return 1;
1976}
static void copyouts(struct nfa *nfa, struct state *oldState, struct state *newState)
Definition: regc_nfa.c:1167

References a, assert, arc::co, combine(), COMPATIBLE, copyouts(), cparc(), state::flag, freearc(), arc::from, INCOMPATIBLE, state::ins, moveouts(), newarc(), newstate(), state::nins, NISERR, NOTREACHED, state::nouts, state::outs, REPLACEARC, SATISFIED, state::tmp, and arc::to.

Referenced by pushfwd().

◆ pushfwd()

static void pushfwd ( struct nfa nfa,
FILE *  f 
)
static

Definition at line 1811 of file regc_nfa.c.

1813{
1814 struct state *s;
1815 struct state *nexts;
1816 struct arc *a;
1817 struct arc *nexta;
1818 struct state *intermediates;
1819 int progress;
1820
1821 /* find and push until there are no more */
1822 do
1823 {
1824 progress = 0;
1825 for (s = nfa->states; s != NULL && !NISERR(); s = nexts)
1826 {
1827 nexts = s->next;
1828 intermediates = NULL;
1829 for (a = s->ins; a != NULL && !NISERR(); a = nexta)
1830 {
1831 nexta = a->inchain;
1832 if (a->type == '$' || a->type == AHEAD)
1833 if (push(nfa, a, &intermediates))
1834 progress = 1;
1835 }
1836 /* clear tmp fields of intermediate states created here */
1837 while (intermediates != NULL)
1838 {
1839 struct state *ns = intermediates->tmp;
1840
1841 intermediates->tmp = NULL;
1842 intermediates = ns;
1843 }
1844 /* if s is now useless, get rid of it */
1845 if ((s->nins == 0 || s->nouts == 0) && !s->flag)
1846 dropstate(nfa, s);
1847 }
1848 if (progress && f != NULL)
1849 dumpnfa(nfa, f);
1850 } while (progress && !NISERR());
1851 if (NISERR())
1852 return;
1853
1854 /*
1855 * Any $ constraints we were able to push to the post state can now be
1856 * replaced by PLAIN arcs referencing the EOS or EOL colors. There should
1857 * be no other $ or AHEAD arcs left in the NFA, though we do not check
1858 * that here (compact() will fail if so).
1859 */
1860 for (a = nfa->post->ins; a != NULL; a = nexta)
1861 {
1862 nexta = a->inchain;
1863 if (a->type == '$')
1864 {
1865 assert(a->co == 0 || a->co == 1);
1866 newarc(nfa, PLAIN, nfa->eos[a->co], a->from, a->to);
1867 freearc(nfa, a);
1868 }
1869 }
1870}
static int push(struct nfa *nfa, struct arc *con, struct state **intermediates)
Definition: regc_nfa.c:1891

References a, AHEAD, assert, dropstate(), dumpnfa(), nfa::eos, state::flag, freearc(), state::ins, newarc(), state::next, state::nins, NISERR, state::nouts, PLAIN, nfa::post, progress, push(), nfa::states, and state::tmp.

Referenced by optimize().

◆ removecantmatch()

static void removecantmatch ( struct nfa nfa)
static

Definition at line 2938 of file regc_nfa.c.

2939{
2940 struct state *s;
2941
2942 for (s = nfa->states; s != NULL; s = s->next)
2943 {
2944 struct arc *a;
2945 struct arc *nexta;
2946
2947 for (a = s->outs; a != NULL; a = nexta)
2948 {
2949 nexta = a->outchain;
2950 if (a->type == CANTMATCH)
2951 {
2952 freearc(nfa, a);
2953 if (NISERR())
2954 return;
2955 }
2956 }
2957 }
2958}
#define CANTMATCH
Definition: regcomp.c:346

References a, CANTMATCH, freearc(), state::next, NISERR, state::outs, and nfa::states.

Referenced by optimize().

◆ removeconstraints()

static void removeconstraints ( struct nfa nfa,
struct state start,
struct state stop 
)
static

Definition at line 1419 of file regc_nfa.c.

1422{
1423 if (start == stop)
1424 return;
1425
1426 stop->tmp = stop;
1428 /* done, except for clearing out the tmp pointers */
1429
1430 stop->tmp = NULL;
1432}
static void removetraverse(struct nfa *nfa, struct state *s)
Definition: regc_nfa.c:1438

References cleartraverse(), removetraverse(), start, and state::tmp.

◆ removetraverse()

static void removetraverse ( struct nfa nfa,
struct state s 
)
static

Definition at line 1438 of file regc_nfa.c.

1440{
1441 struct arc *a;
1442 struct arc *oa;
1443
1444 /* Since this is recursive, it could be driven to stack overflow */
1445 if (STACK_TOO_DEEP(nfa->v->re))
1446 {
1448 return;
1449 }
1450
1451 if (s->tmp != NULL)
1452 return; /* already done */
1453
1454 s->tmp = s;
1455 for (a = s->outs; a != NULL && !NISERR(); a = oa)
1456 {
1457 removetraverse(nfa, a->to);
1458 if (NISERR())
1459 break;
1460 oa = a->outchain;
1461 switch (a->type)
1462 {
1463 case PLAIN:
1464 case EMPTY:
1465 case CANTMATCH:
1466 /* nothing to do */
1467 break;
1468 case AHEAD:
1469 case BEHIND:
1470 case '^':
1471 case '$':
1472 case LACON:
1473 /* replace it */
1474 newarc(nfa, EMPTY, 0, s, a->to);
1475 freearc(nfa, a);
1476 break;
1477 default:
1479 break;
1480 }
1481 }
1482}

References a, AHEAD, BEHIND, CANTMATCH, EMPTY, freearc(), LACON, NERR, newarc(), NISERR, state::outs, PLAIN, vars::re, REG_ASSERT, REG_ETOOBIG, removetraverse(), STACK_TOO_DEEP, state::tmp, and nfa::v.

Referenced by removeconstraints(), and removetraverse().

◆ single_color_transition()

static struct state * single_color_transition ( struct state s1,
struct state s2 
)
static

Definition at line 1525 of file regc_nfa.c.

1526{
1527 struct arc *a;
1528
1529 /* Ignore leading EMPTY arc, if any */
1530 if (s1->nouts == 1 && s1->outs->type == EMPTY)
1531 s1 = s1->outs->to;
1532 /* Likewise for any trailing EMPTY arc */
1533 if (s2->nins == 1 && s2->ins->type == EMPTY)
1534 s2 = s2->ins->from;
1535 /* Perhaps we could have a single-state loop in between, if so reject */
1536 if (s1 == s2)
1537 return NULL;
1538 /* s1 must have at least one outarc... */
1539 if (s1->outs == NULL)
1540 return NULL;
1541 /* ... and they must all be PLAIN arcs to s2 */
1542 for (a = s1->outs; a != NULL; a = a->outchain)
1543 {
1544 if (a->type != PLAIN || a->to != s2)
1545 return NULL;
1546 }
1547 /* OK, return s1 as the possessor of the relevant outarcs */
1548 return s1;
1549}
char * s1

References a, EMPTY, PLAIN, s1, and s2.

◆ sortins()

static void sortins ( struct nfa nfa,
struct state s 
)
static

Definition at line 620 of file regc_nfa.c.

622{
623 struct arc **sortarray;
624 struct arc *a;
625 int n = s->nins;
626 int i;
627
628 if (n <= 1)
629 return; /* nothing to do */
630 /* make an array of arc pointers ... */
631 sortarray = (struct arc **) MALLOC(n * sizeof(struct arc *));
632 if (sortarray == NULL)
633 {
635 return;
636 }
637 i = 0;
638 for (a = s->ins; a != NULL; a = a->inchain)
639 sortarray[i++] = a;
640 assert(i == n);
641 /* ... sort the array */
642 qsort(sortarray, n, sizeof(struct arc *), sortins_cmp);
643 /* ... and rebuild arc list in order */
644 /* it seems worth special-casing first and last items to simplify loop */
645 a = sortarray[0];
646 s->ins = a;
647 a->inchain = sortarray[1];
648 a->inchainRev = NULL;
649 for (i = 1; i < n - 1; i++)
650 {
651 a = sortarray[i];
652 a->inchain = sortarray[i + 1];
653 a->inchainRev = sortarray[i - 1];
654 }
655 a = sortarray[i];
656 a->inchain = NULL;
657 a->inchainRev = sortarray[i - 1];
658 FREE(sortarray);
659}

References a, assert, FREE, i, state::ins, MALLOC, NERR, state::nins, qsort, REG_ESPACE, and sortins_cmp().

Referenced by copyins(), mergeins(), and moveins().

◆ sortins_cmp()

static int sortins_cmp ( const void *  a,
const void *  b 
)
static

Definition at line 662 of file regc_nfa.c.

663{
664 const struct arc *aa = *((const struct arc *const *) a);
665 const struct arc *bb = *((const struct arc *const *) b);
666
667 /* we check the fields in the order they are most likely to be different */
668 if (aa->from->no < bb->from->no)
669 return -1;
670 if (aa->from->no > bb->from->no)
671 return 1;
672 if (aa->co < bb->co)
673 return -1;
674 if (aa->co > bb->co)
675 return 1;
676 if (aa->type < bb->type)
677 return -1;
678 if (aa->type > bb->type)
679 return 1;
680 return 0;
681}

References a, b, arc::co, arc::from, state::no, and arc::type.

Referenced by copyins(), mergeins(), moveins(), and sortins().

◆ sortouts()

static void sortouts ( struct nfa nfa,
struct state s 
)
static

Definition at line 687 of file regc_nfa.c.

689{
690 struct arc **sortarray;
691 struct arc *a;
692 int n = s->nouts;
693 int i;
694
695 if (n <= 1)
696 return; /* nothing to do */
697 /* make an array of arc pointers ... */
698 sortarray = (struct arc **) MALLOC(n * sizeof(struct arc *));
699 if (sortarray == NULL)
700 {
702 return;
703 }
704 i = 0;
705 for (a = s->outs; a != NULL; a = a->outchain)
706 sortarray[i++] = a;
707 assert(i == n);
708 /* ... sort the array */
709 qsort(sortarray, n, sizeof(struct arc *), sortouts_cmp);
710 /* ... and rebuild arc list in order */
711 /* it seems worth special-casing first and last items to simplify loop */
712 a = sortarray[0];
713 s->outs = a;
714 a->outchain = sortarray[1];
715 a->outchainRev = NULL;
716 for (i = 1; i < n - 1; i++)
717 {
718 a = sortarray[i];
719 a->outchain = sortarray[i + 1];
720 a->outchainRev = sortarray[i - 1];
721 }
722 a = sortarray[i];
723 a->outchain = NULL;
724 a->outchainRev = sortarray[i - 1];
725 FREE(sortarray);
726}

References a, assert, FREE, i, MALLOC, NERR, state::nouts, state::outs, qsort, REG_ESPACE, and sortouts_cmp().

Referenced by copyouts(), and moveouts().

◆ sortouts_cmp()

static int sortouts_cmp ( const void *  a,
const void *  b 
)
static

Definition at line 729 of file regc_nfa.c.

730{
731 const struct arc *aa = *((const struct arc *const *) a);
732 const struct arc *bb = *((const struct arc *const *) b);
733
734 /* we check the fields in the order they are most likely to be different */
735 if (aa->to->no < bb->to->no)
736 return -1;
737 if (aa->to->no > bb->to->no)
738 return 1;
739 if (aa->co < bb->co)
740 return -1;
741 if (aa->co > bb->co)
742 return 1;
743 if (aa->type < bb->type)
744 return -1;
745 if (aa->type > bb->type)
746 return 1;
747 return 0;
748}

References a, b, arc::co, state::no, arc::to, and arc::type.

Referenced by copyouts(), moveouts(), and sortouts().

◆ specialcolors()

static void specialcolors ( struct nfa nfa)
static

Definition at line 1555 of file regc_nfa.c.

1556{
1557 /* false colors for BOS, BOL, EOS, EOL */
1558 if (nfa->parent == NULL)
1559 {
1560 nfa->bos[0] = pseudocolor(nfa->cm);
1561 nfa->bos[1] = pseudocolor(nfa->cm);
1562 nfa->eos[0] = pseudocolor(nfa->cm);
1563 nfa->eos[1] = pseudocolor(nfa->cm);
1564 }
1565 else
1566 {
1567 assert(nfa->parent->bos[0] != COLORLESS);
1568 nfa->bos[0] = nfa->parent->bos[0];
1569 assert(nfa->parent->bos[1] != COLORLESS);
1570 nfa->bos[1] = nfa->parent->bos[1];
1571 assert(nfa->parent->eos[0] != COLORLESS);
1572 nfa->eos[0] = nfa->parent->eos[0];
1573 assert(nfa->parent->eos[1] != COLORLESS);
1574 nfa->eos[1] = nfa->parent->eos[1];
1575 }
1576}
static color pseudocolor(struct colormap *cm)
Definition: regc_color.c:312

References assert, nfa::bos, nfa::cm, COLORLESS, nfa::eos, nfa::parent, and pseudocolor().