Introduce REPEAT_ALWAYS model for {0,} castle repeats

As Castle guards the repeats, no more state is needed for these repeats

src/nfa/castlecompile.cpp

@@ -316,7 +316,7 @@ void buildSubcastles(const CastleProto &proto, vector<SubCastle> &subs,
         bool is_reset = repeatInfoPair[i].second;
 
         enum RepeatType rtype = chooseRepeatType(pr.bounds.min, pr.bounds.max,
-                                                 min_period, is_reset);
+                                                 min_period, is_reset, true);
         RepeatStateInfo rsi(rtype, pr.bounds.min, pr.bounds.max, min_period);
 
         DEBUG_PRINTF("sub %u: selected %s model for %s repeat\n", i,

src/nfa/lbr.c

@@ -130,6 +130,9 @@ char repeatIsDead(const struct RepeatInfo *info,
         return lstate->ctrl.ring.offset == REPEAT_DEAD;
     case REPEAT_TRAILER:
         return lstate->ctrl.trailer.offset == REPEAT_DEAD;
+    case REPEAT_ALWAYS:
+        assert(!"REPEAT_ALWAYS should only be used by Castle");
+        return 0;
     }
 
     assert(0);

src/nfa/repeat.c

@@ -922,6 +922,11 @@ void repeatPackOffset(char *dest, const struct RepeatInfo *info,
                       const union RepeatControl *ctrl, u64a offset) {
     const struct RepeatOffsetControl *xs = &ctrl->offset;
     DEBUG_PRINTF("packing offset %llu [h %u]\n", xs->offset, info->horizon);
+    if (!info->packedCtrlSize) {
+        assert(info->type == REPEAT_ALWAYS);
+        DEBUG_PRINTF("externally guarded .*\n");
+        return;
+    }
     storePackedRelative(dest, xs->offset, offset, info->horizon,
                         info->packedCtrlSize);
 }
@@ -1040,6 +1045,9 @@ void repeatPack(char *dest, const struct RepeatInfo *info,
     case REPEAT_TRAILER:
         repeatPackTrailer(dest, info, ctrl, offset);
         break;
+    case REPEAT_ALWAYS:
+        /* nothing to do - no state */
+        break;
     }
 }
 
@@ -1072,7 +1080,13 @@ static
 void repeatUnpackOffset(const char *src, const struct RepeatInfo *info,
                         u64a offset, union RepeatControl *ctrl) {
     struct RepeatOffsetControl *xs = &ctrl->offset;
-    xs->offset = loadPackedRelative(src, offset, info->packedCtrlSize);
+    if (!info->packedCtrlSize) {
+        assert(info->type == REPEAT_ALWAYS);
+        DEBUG_PRINTF("externally guarded .*\n");
+        xs->offset = 0;
+    } else {
+        xs->offset = loadPackedRelative(src, offset, info->packedCtrlSize);
+    }
     DEBUG_PRINTF("unpacking offset %llu [h%u]\n", xs->offset,
                  info->horizon);
 }
@@ -1149,6 +1163,9 @@ void repeatUnpack(const char *src, const struct RepeatInfo *info, u64a offset,
     case REPEAT_TRAILER:
         repeatUnpackTrailer(src, info, offset, ctrl);
         break;
+    case REPEAT_ALWAYS:
+        /* nothing to do - no state */
+        break;
     }
 }
 

src/nfa/repeat.h

@@ -135,6 +135,8 @@ u64a repeatLastTop(const struct RepeatInfo *info,
         return repeatLastTopSparseOptimalP(info, ctrl, state);
     case REPEAT_TRAILER:
         return repeatLastTopTrailer(info, ctrl);
+    case REPEAT_ALWAYS:
+        return 0;
     }
 
     DEBUG_PRINTF("bad repeat type %u\n", info->type);
@@ -200,6 +202,8 @@ u64a repeatNextMatch(const struct RepeatInfo *info,
         return repeatNextMatchSparseOptimalP(info, ctrl, state, offset);
     case REPEAT_TRAILER:
         return repeatNextMatchTrailer(info, ctrl, offset);
+    case REPEAT_ALWAYS:
+        return offset + 1;
     }
 
     DEBUG_PRINTF("bad repeat type %u\n", info->type);
@@ -275,6 +279,9 @@ void repeatStore(const struct RepeatInfo *info, union RepeatControl *ctrl,
     case REPEAT_TRAILER:
         repeatStoreTrailer(info, ctrl, offset, is_alive);
         break;
+    case REPEAT_ALWAYS:
+        /* nothing to do - no state */
+        break;
     }
 }
 
@@ -348,6 +355,8 @@ enum RepeatMatch repeatHasMatch(const struct RepeatInfo *info,
         return repeatHasMatchSparseOptimalP(info, ctrl, state, offset);
     case REPEAT_TRAILER:
         return repeatHasMatchTrailer(info, ctrl, offset);
+    case REPEAT_ALWAYS:
+        return REPEAT_MATCH;
     }
 
     assert(0);

src/nfa/repeat_internal.h

@@ -47,26 +47,26 @@ enum RepeatType {
     /** General mechanism for tracking {N,M} repeats. Stores the first top as
      * an absolute offset, then subsequent tops in the {N,M} range as a ring of
      * relative top indices stored in a multibit. */
-    REPEAT_RING = 0,
+    REPEAT_RING,
 
     /** Used to track {N,} repeats. Uses the \ref RepeatOffsetControl structure,
      * since only the first top encountered needs to be stored. */
-    REPEAT_FIRST = 1,
+    REPEAT_FIRST,
 
     /** Used to track {0,N} repeats. Much like ::REPEAT_FIRST, except that we
      * store the most recent top encountered. */
-    REPEAT_LAST = 2,
+    REPEAT_LAST,
 
     /** Like ::REPEAT_RING, this is also used for {N,M} repeats, but for cases
      * where there is a large difference between N and M, and developed to
      * reduce the state requirements of this case (relative to the RING model).
      * Uses a small ordered array of top indices relative to \ref
      * RepeatRangeControl::offset. */
-    REPEAT_RANGE = 3,
+    REPEAT_RANGE,
 
     /** Used for {N,M} repeats where 0 < M <= 64. Uses the \ref
      * RepeatBitmapControl structure at runtime. */
-    REPEAT_BITMAP = 4,
+    REPEAT_BITMAP,
 
     /** Optimal mechanism for tracking {N,M} repeats when there is a bound on
      * how frequently they can be retriggered.
@@ -78,13 +78,17 @@ enum RepeatType {
      * referencing a table that stores values from f(0, min) to f(repeat, min)
      * eg: repeat = 5, min = 2.  10001 => f(4,2) + f(0,2) = 9.
      * We search the optimal patch size between min and repeat in advance and
-     * use the scheme above to do encoding and decoding to reduce stream state size
-     * */
-    REPEAT_SPARSE_OPTIMAL_P = 5,
+     * use the scheme above to do encoding and decoding to reduce stream state
+     * size. */
+    REPEAT_SPARSE_OPTIMAL_P,
 
-    /** Used for {N,M} repeats where 0 < N < 64. Uses the \ref RepeatTrailerControl
-     * structure at runtime. */
-    REPEAT_TRAILER = 6,
+    /** Used for {N,M} repeats where 0 < N < 64. Uses the
+     * \ref RepeatTrailerControl structure at runtime. */
+    REPEAT_TRAILER,
+
+    /** Degenerate repeat that always returns true. Used by castle for pseudo
+     * [^X]* repeats. */
+    REPEAT_ALWAYS,
 };
 
 /**
@@ -204,6 +208,8 @@ const char *repeatTypeName(u8 type) {
         return "SPARSE_OPTIMAL_P";
     case REPEAT_TRAILER:
         return "TRAILER";
+    case REPEAT_ALWAYS:
+        return "ALWAYS";
     }
     assert(0);
     return "UNKNOWN";

src/nfa/repeatcompile.cpp

@@ -206,6 +206,13 @@ RepeatStateInfo::RepeatStateInfo(enum RepeatType type, const depth &repeatMin,
         packedFieldSizes[1] = repeatMin;
         packedCtrlSize = (packedFieldSizes[0] + packedFieldSizes[1] + 7U) / 8U;
         break;
+    case REPEAT_ALWAYS:
+        assert(repeatMin == 0ULL);
+        assert(repeatMax.is_infinite());
+        stateSize = 0; // everything is in the control block.
+        horizon = 0;
+        packedCtrlSize = 0;
+        break;
     }
     DEBUG_PRINTF("stateSize=%u, packedCtrlSize=%u, horizon=%u\n", stateSize,
                  packedCtrlSize, horizon);
@@ -232,9 +239,14 @@ u32 streamStateSize(enum RepeatType type, const depth &repeatMin,
 }
 
 enum RepeatType chooseRepeatType(const depth &repeatMin, const depth &repeatMax,
-                                 u32 minPeriod, bool is_reset) {
+                                 u32 minPeriod, bool is_reset,
+                                 bool has_external_guard) {
     if (repeatMax.is_infinite()) {
-        return REPEAT_FIRST;
+        if (has_external_guard && !repeatMin) {
+            return REPEAT_ALWAYS;
+        } else {
+            return REPEAT_FIRST;
+        }
     }
 
     if (repeatMin == depth(0) || is_reset) {

src/nfa/repeatcompile.h

@@ -68,7 +68,8 @@ struct RepeatStateInfo {
  * type.
  */
 enum RepeatType chooseRepeatType(const depth &repeatMin, const depth &repeatMax,
-                                 u32 minPeriod, bool is_reset);
+                                 u32 minPeriod, bool is_reset,
+                                 bool has_external_guard = false);
 
 u32 calcPackedBytes(u64a val);
 

unit/internal/repeat.cpp

@@ -193,7 +193,9 @@ static const RepeatTestInfo repeatTests[] = {
     { REPEAT_FIRST, 100, depth::infinity() },
     { REPEAT_FIRST, 1000, depth::infinity() },
     { REPEAT_FIRST, 3000, depth::infinity() },
-    { REPEAT_FIRST, 10000, depth::infinity() }
+    { REPEAT_FIRST, 10000, depth::infinity() },
+    // {,} repeats -- always
+    { REPEAT_ALWAYS, 0, depth::infinity() },
 };
 
 INSTANTIATE_TEST_CASE_P(Repeat, RepeatTest, ValuesIn(repeatTests));
@@ -289,6 +291,10 @@ TEST_P(RepeatTest, FillRing) {
 
 TEST_P(RepeatTest, FindTops) {
     SCOPED_TRACE(testing::Message() << "Repeat: " << info);
+    /* REPEAT_ALWAYS has no state and so does not track top locations */
+    if (info.type == REPEAT_ALWAYS) {
+        return;
+    }
 
     repeatStore(&info, ctrl, state, 1000, 0);
     ASSERT_EQ(1000, repeatLastTop(&info, ctrl, state));
@@ -364,7 +370,8 @@ TEST_P(RepeatTest, TwoTops) {
     SCOPED_TRACE(testing::Message() << "Repeat: " << info);
 
     // Only appropriate for tests that store more than one top.
-    if (info.type == REPEAT_FIRST || info.type == REPEAT_LAST) {
+    if (info.type == REPEAT_FIRST || info.type == REPEAT_LAST
+        || info.type == REPEAT_ALWAYS) {
         return;
     }