Allow anchored patterns to use "first code unit" optimization.

ChangeLog

@@ -205,6 +205,11 @@ JIT.
 subjects from 1000 to 2000 for 8-bit searches, since they use memchr() and are 
 much faster.
 
+46. Arrange for anchored patterns to record and use "first code unit" data,
+because this can give a fast "no match" without searching for a "required code 
+unit". Previously only non-anchored patterns did this.
+
+
 
 Version 10.23 14-February-2017
 ------------------------------

src/pcre2_compile.c

@@ -9632,14 +9632,19 @@ if ((re->overall_options & PCRE2_ANCHORED) == 0 &&
      is_anchored(codestart, 0, &cb, 0, FALSE))
   re->overall_options |= PCRE2_ANCHORED;
 
-/* If the pattern is still not anchored and we do not have a first code unit,
+/* Set up the first code unit or startline flag, the required code unit, and
-see if there is one that is asserted (these are not saved during the compile
+then study the pattern. This code need not be obeyed if PCRE2_NO_START_OPTIMIZE
-because they can cause conflicts with actual literals that follow). This code
+is set, as the data it would create will not be used. Note that a first code
-need not be obeyed if PCRE2_NO_START_OPTIMIZE is set, as the data it would
+unit (but not the startline flag) is useful for anchored patterns because it
-create will not be used. */
+can still give a quick "no match" and also avoid searching for a last code
+unit. */
 
-if ((re->overall_options & (PCRE2_ANCHORED|PCRE2_NO_START_OPTIMIZE)) == 0)
+if ((re->overall_options & PCRE2_NO_START_OPTIMIZE) == 0)
   {
+  /* If we do not have a first code unit, see if there is one that is asserted
+  (these are not saved during the compile because they can cause conflicts with
+  actual literals that follow). */
+
   if (firstcuflags < 0)
     firstcu = find_firstassertedcu(codestart, &firstcuflags, FALSE);
 
@@ -9672,22 +9677,21 @@ if ((re->overall_options & (PCRE2_ANCHORED|PCRE2_NO_START_OPTIMIZE)) == 0)
       }
     }
 
-  /* When there is no first code unit, see if we can set the PCRE2_STARTLINE
+  /* When there is no first code unit, for non-anchored patterns, see if we can
-  flag. This is helpful for multiline matches when all branches start with ^
+  set the PCRE2_STARTLINE flag. This is helpful for multiline matches when all
-  and also when all branches start with non-atomic .* for non-DOTALL matches
+  branches start with ^ and also when all branches start with non-atomic .* for
-  when *PRUNE and SKIP are not present. (There is an option that disables this
+  non-DOTALL matches when *PRUNE and SKIP are not present. (There is an option
-  case.) */
+  that disables this case.) */
 
-  else if (is_startline(codestart, 0, &cb, 0, FALSE))
+  else if ((re->overall_options & PCRE2_ANCHORED) == 0 &&
+           is_startline(codestart, 0, &cb, 0, FALSE))
     re->flags |= PCRE2_STARTLINE;
-  }
 
-/* Handle the "required code unit", if one is set. In the case of an anchored
+  /* Handle the "required code unit", if one is set. In the case of an anchored
-pattern, do this only if it follows a variable length item in the pattern.
+  pattern, do this only if it follows a variable length item in the pattern. */
-Again, skip this if PCRE2_NO_START_OPTIMIZE is set. */
 
-if (reqcuflags >= 0 &&
+  if (reqcuflags >= 0 &&
-     ((re->overall_options & (PCRE2_ANCHORED|PCRE2_NO_START_OPTIMIZE)) == 0 ||
+       ((re->overall_options & PCRE2_ANCHORED) == 0 ||
         (reqcuflags & REQ_VARY) != 0))
     {
     re->last_codeunit = reqcu;
@@ -9708,16 +9712,15 @@ if (reqcuflags >= 0 &&
       }
     }
 
-/* Finally, unless PCRE2_NO_START_OPTIMIZE is set, study the compiled pattern
+  /* Finally, study the compiled pattern to set up information such as a bitmap
-to set up information such as a bitmap of starting code units and a minimum
+  of starting code units and a minimum matching length. */
-matching length. */
 
-if ((re->overall_options & PCRE2_NO_START_OPTIMIZE) == 0 &&
+  if (PRIV(study)(re) != 0)
-    PRIV(study)(re) != 0)
     {
     errorcode = ERR31;
     goto HAD_CB_ERROR;
     }
+  }   /* End of start-of-match optimizations. */
 
 /* Control ends up here in all cases. When running under valgrind, make a
 pattern's terminating zero defined again. If memory was obtained for the parsed

src/pcre2_dfa_match.c

@@ -3341,15 +3341,10 @@ if (utf && (options & PCRE2_NO_UTF_CHECK) == 0)
   }
 #endif  /* SUPPORT_UNICODE */
 
-/* Set up the first code unit to match, if available. The first_codeunit value
+/* Set up the first code unit to match, if available. If there's no first code
-is never set for an anchored regular expression, but the anchoring may be
+unit there may be a bitmap of possible first characters. */
-forced at run time, so we have to test for anchoring. The first code unit may
-be unset for an unanchored pattern, of course. If there's no first code unit
-there may be a bitmap of possible first characters. */
 
-if (!anchored)
+if ((re->flags & PCRE2_FIRSTSET) != 0)
-  {
-  if ((re->flags & PCRE2_FIRSTSET) != 0)
   {
   has_first_cu = TRUE;
   first_cu = first_cu2 = (PCRE2_UCHAR)(re->first_codeunit);
@@ -3362,13 +3357,11 @@ if (!anchored)
 #endif
     }
   }
-  else
+else
   if (!startline && (re->flags & PCRE2_FIRSTMAPSET) != 0)
     start_bits = re->start_bitmap;
-  }
 
-/* For anchored or unanchored matches, there may be a "last known required
+/* There may be a "last known required code unit" set. */
-character" set. */
 
 if ((re->flags & PCRE2_LASTSET) != 0)
   {
@@ -3414,8 +3407,8 @@ for (;;)
     /* If firstline is TRUE, the start of the match is constrained to the first
     line of a multiline string. That is, the match must be before or at the
     first newline. Implement this by temporarily adjusting end_subject so that
-    we stop the optimization scans at a newline. If the match fails at the
+    we stop the optimization scans for a first code unit at a newline. If the
-    newline, later code breaks this loop. */
+    match fails at the newline, later code breaks this loop. */
 
     if (firstline)
       {
@@ -3435,21 +3428,87 @@ for (;;)
       end_subject = t;
       }
       
-    /* Advance to a unique first code unit if there is one. */
+    /* Anchored: check the first code unit if one is recorded. This may seem
+    pointless but it can help in detecting a no match case without scanning for
+    the required code unit. */
 
-    if (has_first_cu)
+    if (anchored)
       {
-      PCRE2_UCHAR smc;
+      if (has_first_cu || start_bits != NULL)
-      if (first_cu != first_cu2)
+        {
-        while (start_match < end_subject &&
+        BOOL ok = start_match < end_subject;
-          (smc = UCHAR21TEST(start_match)) != first_cu && smc != first_cu2)
+        if (ok)
-          start_match++;
+          {
-      else
+          PCRE2_UCHAR c = UCHAR21TEST(start_match);
-        while (start_match < end_subject && UCHAR21TEST(start_match) != first_cu)
+          ok = has_first_cu && (c == first_cu || c == first_cu2);
-          start_match++;
+          if (!ok && start_bits != NULL)
+            {
+#if PCRE2_CODE_UNIT_WIDTH != 8
+            if (c > 255) c = 255;
+#endif
+            ok = (start_bits[c/8] & (1 << (c&7))) != 0;
+            }
+          }
+        if (!ok) break;
+        }
       }
 
-    /* Or to just after a linebreak for a multiline match */
+    /* Not anchored. Advance to a unique first code unit if there is one. In
+    8-bit mode, the use of memchr() gives a big speed up, even though we have
+    to call it twice in caseless mode, in order to find the earliest occurrence
+    of the character in either of its cases. */
+
+    else
+      {
+      if (has_first_cu)
+        {
+        if (first_cu != first_cu2)  /* Caseless */
+          {
+#if PCRE2_CODE_UNIT_WIDTH != 8
+          PCRE2_UCHAR smc;
+          while (start_match < end_subject &&
+                (smc = UCHAR21TEST(start_match)) != first_cu &&
+                  smc != first_cu2)
+            start_match++;
+#else  /* 8-bit code units */
+          PCRE2_SPTR pp1 =
+            memchr(start_match, first_cu, end_subject-start_match);
+          PCRE2_SPTR pp2 =
+            memchr(start_match, first_cu2, end_subject-start_match);
+          if (pp1 == NULL)
+            start_match = (pp2 == NULL)? end_subject : pp2;
+          else
+            start_match = (pp2 == NULL || pp1 < pp2)? pp1 : pp2;
+#endif
+          }
+
+        /* The caseful case */
+
+        else
+          {
+#if PCRE2_CODE_UNIT_WIDTH != 8
+          while (start_match < end_subject && UCHAR21TEST(start_match) !=
+                 first_cu)
+            start_match++;
+#else
+          start_match = memchr(start_match, first_cu, end_subject - start_match);
+          if (start_match == NULL) start_match = end_subject;
+#endif
+          }
+
+        /* If we can't find the required code unit, break the bumpalong loop,
+        to force a match failure, except when doing partial matching, when we
+        let the next cycle run at the end of the subject. To see why, consider
+        the pattern /(?<=abc)def/, which partially matches "abc", even though
+        the string does not contain the starting character "d". */
+
+        if ((mb->moptions & (PCRE2_PARTIAL_HARD|PCRE2_PARTIAL_SOFT)) == 0 &&
+            start_match >= end_subject)
+          break;
+        }
+
+      /* If there's no first code unit, advance to just after a linebreak for a
+      multiline match if required. */
 
       else if (startline)
         {
@@ -3471,8 +3530,8 @@ for (;;)
             start_match++;
 
           /* If we have just passed a CR and the newline option is ANY or
-        ANYCRLF, and we are now at a LF, advance the match position by one more
+          ANYCRLF, and we are now at a LF, advance the match position by one
-        code unit. */
+          more code unit. */
 
           if (start_match[-1] == CHAR_CR &&
                (mb->nltype == NLTYPE_ANY || mb->nltype == NLTYPE_ANYCRLF) &&
@@ -3482,9 +3541,10 @@ for (;;)
           }
         }
 
-    /* Or to a non-unique first code unit if any have been identified. The
+      /* If there's no first code unit or a requirement for a multiline line
-    bitmap contains only 256 bits. When code units are 16 or 32 bits wide, all
+      start, advance to a non-unique first code unit if any have been
-    code units greater than 254 set the 255 bit. */
+      identified. The bitmap contains only 256 bits. When code units are 16 or
+      32 bits wide, all code units greater than 254 set the 255 bit. */
 
       else if (start_bits != NULL)
         {
@@ -3498,6 +3558,7 @@ for (;;)
           start_match++;
           }
         }
+      }  /* End of first code unit handling */
 
     /* Restore fudged end_subject */
 

src/pcre2_match.c

@@ -6333,15 +6333,10 @@ mb->lcc = re->tables + lcc_offset;
 mb->fcc = re->tables + fcc_offset;
 mb->ctypes = re->tables + ctypes_offset;
 
-/* Set up the first code unit to match, if available. The first_codeunit value
+/* Set up the first code unit to match, if available. If there's no first code
-is never set for an anchored regular expression, but the anchoring may be
+unit there may be a bitmap of possible first characters. */
-forced at run time, so we have to test for anchoring. The first code unit may
-be unset for an unanchored pattern, of course. If there's no first code unit
-there may be a bitmap of possible first characters. */
 
-if (!anchored)
+if ((re->flags & PCRE2_FIRSTSET) != 0)
-  {
-  if ((re->flags & PCRE2_FIRSTSET) != 0)
   {
   has_first_cu = TRUE;
   first_cu = first_cu2 = (PCRE2_UCHAR)(re->first_codeunit);
@@ -6353,13 +6348,11 @@ if (!anchored)
 #endif
     }
   }
-  else
+else
   if (!startline && (re->flags & PCRE2_FIRSTMAPSET) != 0)
     start_bits = re->start_bitmap;
-  }
 
-/* For anchored or unanchored matches, there may be a "last known required
+/* There may also be a "last known required character" set. */
-character" set. */
 
 if ((re->flags & PCRE2_LASTSET) != 0)
   {
@@ -6398,8 +6391,8 @@ for(;;)
     /* If firstline is TRUE, the start of the match is constrained to the first
     line of a multiline string. That is, the match must be before or at the
     first newline. Implement this by temporarily adjusting end_subject so that
-    we stop the optimization scans at a newline. If the match fails at the
+    we stop the optimization scans for a first code unit at a newline. If the
-    newline, later code breaks this loop. */
+    match fails at the newline, later code breaks this loop. */
 
     if (firstline)
       {
@@ -6419,11 +6412,42 @@ for(;;)
       end_subject = t;
       }
 
-    /* Advance to a unique first code unit if there is one. In 8-bit mode, the
+    /* Anchored: check the first code unit if one is recorded. This may seem
-    use of memchr() gives a big speed up, even though we have to call it twice
+    pointless but it can help in detecting a no match case without scanning for
-    in caseless mode, in order to find the first occurrence of the character in
+    the required code unit. */
-    either of its cases. */
 
+    if (anchored)
+      {
+      if (has_first_cu || start_bits != NULL)
+        {
+        BOOL ok = start_match < end_subject;
+        if (ok)
+          {
+          PCRE2_UCHAR c = UCHAR21TEST(start_match);
+          ok = has_first_cu && (c == first_cu || c == first_cu2);
+          if (!ok && start_bits != NULL)
+            {
+#if PCRE2_CODE_UNIT_WIDTH != 8
+            if (c > 255) c = 255;
+#endif
+            ok = (start_bits[c/8] & (1 << (c&7))) != 0;
+            }
+          }
+        if (!ok)
+          {
+          rc = MATCH_NOMATCH;
+          break;
+          }
+        }
+      }
+
+    /* Not anchored. Advance to a unique first code unit if there is one. In
+    8-bit mode, the use of memchr() gives a big speed up, even though we have
+    to call it twice in caseless mode, in order to find the earliest occurrence
+    of the character in either of its cases. */
+
+    else
+      {
       if (has_first_cu)
         {
         if (first_cu != first_cu2)  /* Caseless */
@@ -6431,11 +6455,14 @@ for(;;)
 #if PCRE2_CODE_UNIT_WIDTH != 8
           PCRE2_UCHAR smc;
           while (start_match < end_subject &&
-              (smc = UCHAR21TEST(start_match)) != first_cu && smc != first_cu2)
+                (smc = UCHAR21TEST(start_match)) != first_cu &&
+                  smc != first_cu2)
             start_match++;
 #else  /* 8-bit code units */
-        PCRE2_SPTR pp1 = memchr(start_match, first_cu, end_subject-start_match);
+          PCRE2_SPTR pp1 =
-        PCRE2_SPTR pp2 = memchr(start_match, first_cu2, end_subject-start_match);
+            memchr(start_match, first_cu, end_subject-start_match);
+          PCRE2_SPTR pp2 =
+            memchr(start_match, first_cu2, end_subject-start_match);
           if (pp1 == NULL)
             start_match = (pp2 == NULL)? end_subject : pp2;
           else
@@ -6448,7 +6475,8 @@ for(;;)
         else
           {
 #if PCRE2_CODE_UNIT_WIDTH != 8
-        while (start_match < end_subject && UCHAR21TEST(start_match) != first_cu)
+          while (start_match < end_subject && UCHAR21TEST(start_match) !=
+                 first_cu)
             start_match++;
 #else
           start_match = memchr(start_match, first_cu, end_subject - start_match);
@@ -6456,11 +6484,11 @@ for(;;)
 #endif
           }
 
-      /* If we can't find the required code unit, break the bumpalong loop, to
+        /* If we can't find the required code unit, break the bumpalong loop,
-      force a match failure, except when doing partial matching, when we let
+        to force a match failure, except when doing partial matching, when we
-      the next cycle run at the end of the subject. To see why, consider the
+        let the next cycle run at the end of the subject. To see why, consider
-      pattern /(?<=abc)def/, which partially matches "abc", even though the
+        the pattern /(?<=abc)def/, which partially matches "abc", even though
-      string does not contain the starting character "d". */
+        the string does not contain the starting character "d". */
 
         if (!mb->partial && start_match >= end_subject)
           {
@@ -6492,8 +6520,8 @@ for(;;)
             start_match++;
 
           /* If we have just passed a CR and the newline option is ANY or
-        ANYCRLF, and we are now at a LF, advance the match position by one more
+          ANYCRLF, and we are now at a LF, advance the match position by one
-        code unit. */
+          more code unit. */
 
           if (start_match[-1] == CHAR_CR &&
                (mb->nltype == NLTYPE_ANY || mb->nltype == NLTYPE_ANYCRLF) &&
@@ -6504,9 +6532,9 @@ for(;;)
         }
 
       /* If there's no first code unit or a requirement for a multiline line
-    start, advance to a non-unique first code unit if any have been identified.
+      start, advance to a non-unique first code unit if any have been
-    The bitmap contains only 256 bits. When code units are 16 or 32 bits wide,
+      identified. The bitmap contains only 256 bits. When code units are 16 or
-    all code units greater than 254 set the 255 bit. */
+      32 bits wide, all code units greater than 254 set the 255 bit. */
 
       else if (start_bits != NULL)
         {
@@ -6520,6 +6548,7 @@ for(;;)
           start_match++;
           }
         }
+      }   /* End first code unit handling */
 
     /* Restore fudged end_subject */
 

src/pcre2_study.c

@@ -953,7 +953,6 @@ do
       case OP_ALLANY:
       case OP_ANY:
       case OP_ANYBYTE:
-      case OP_CIRC:
       case OP_CIRCM:
       case OP_CLOSE:
       case OP_COMMIT:
@@ -1021,6 +1020,13 @@ do
       case OP_THEN_ARG:
       return SSB_FAIL;
 
+      /* OP_CIRC happens only at the start of an anchored branch (multiline ^
+      uses OP_CIRCM). Skip over it. */
+
+      case OP_CIRC:
+      tcode += PRIV(OP_lengths)[OP_CIRC];
+      break;
+
       /* A "real" property test implies no starting bits, but the fake property
       PT_CLIST identifies a list of characters. These lists are short, as they
       are used for characters with more than one "other case", so there is no
@@ -1579,12 +1585,11 @@ BOOL utf = (re->overall_options & PCRE2_UTF) != 0;
 code = (PCRE2_UCHAR *)((uint8_t *)re + sizeof(pcre2_real_code)) +
   re->name_entry_size * re->name_count;
 
-/* For an anchored pattern, or an unanchored pattern that has a first code
+/* For a pattern that has a first code unit, or a multiline pattern that
-unit, or a multiline pattern that matches only at "line start", there is no
+matches only at "line start", there is no point in seeking a list of starting
-point in seeking a list of starting code units. */
+code units. */
 
-if ((re->overall_options & PCRE2_ANCHORED) == 0 &&
+if ((re->flags & (PCRE2_FIRSTSET|PCRE2_STARTLINE)) == 0)
-    (re->flags & (PCRE2_FIRSTSET|PCRE2_STARTLINE)) == 0)
   {
   int rc = set_start_bits(re, code, utf);
   if (rc == SSB_UNKNOWN) return 1;

testdata/testinput10

@@ -467,4 +467,13 @@
 /\udfff\o{157401}/utf,alt_bsux,allow_surrogate_escapes
     \x{dfff}\x{df01}\=no_utf_check
     
+# This has different starting code units in 8-bit mode. 
+
+/^[^ab]/IB,utf
+    c
+    \x{ff}
+    \x{100}
+\= Expect no match
+    aaa
+
 # End of testinput10

testdata/testinput12

@@ -373,4 +373,13 @@
 /\udfff\o{157401}/utf,alt_bsux,allow_surrogate_escapes
     \x{dfff}\x{df01}\=no_utf_check
 
+# This has different starting code units in 8-bit mode. 
+
+/^[^ab]/IB,utf
+    c
+    \x{ff}
+    \x{100}
+\= Expect no match
+    aaa
+
 # End of testinput12

testdata/testinput2

@@ -5256,6 +5256,9 @@ a)"xI
     XAB     
 
 /^(?!A(?C1)B)C/
+    ABC\=callout_error=1,no_jit
+
+/^(?!A(?C1)B)C/no_start_optimize
     ABC\=callout_error=1
 
 /^(?(?!A(?C1)B)C)/

testdata/testinput5

@@ -120,13 +120,6 @@
     \x{ff}
     \x{100}
 
-/^[^ab]/IB,utf
-    c
-    \x{ff}
-    \x{100}
-\= Expect no match
-    aaa
-
 /\x{100}*(\d+|"(?1)")/utf
     1234
     "1234"
@@ -190,7 +183,10 @@
 /\w/utf
     \x{100}X
 
-/^\ሴ/IB,utf
+# Use no_start_optimize because the first code unit is different in 8-bit from
+# the wider modes.
+
+/^\ሴ/IB,utf,no_start_optimize
 
 /()()()()()()()()()()
  ()()()()()()()()()()

testdata/testoutput10

@@ -1586,4 +1586,38 @@ Failed: error 176 at offset 259: name is too long in (*MARK), (*PRUNE), (*SKIP),
     \x{dfff}\x{df01}\=no_utf_check
  0: \x{dfff}\x{df01}
     
+# This has different starting code units in 8-bit mode. 
+
+/^[^ab]/IB,utf
+------------------------------------------------------------------
+        Bra
+        ^
+        [\x00-`c-\xff] (neg)
+        Ket
+        End
+------------------------------------------------------------------
+Capturing subpattern count = 0
+Compile options: utf
+Overall options: anchored utf
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 
+  5 6 7 8 9 : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y 
+  Z [ \ ] ^ _ ` c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x7f 
+  \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb \xcc \xcd \xce \xcf \xd0 
+  \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda \xdb \xdc \xdd \xde \xdf 
+  \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 \xea \xeb \xec \xed \xee 
+  \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 \xf9 \xfa \xfb \xfc \xfd 
+  \xfe \xff 
+Subject length lower bound = 1
+    c
+ 0: c
+    \x{ff}
+ 0: \x{ff}
+    \x{100}
+ 0: \x{100}
+\= Expect no match
+    aaa
+No match
+
 # End of testinput10

testdata/testoutput12-16

@@ -1433,4 +1433,42 @@ Failed: error 191 at offset 0: PCRE2_EXTRA_ALLOW_SURROGATE_ESCAPES is not allowe
 Failed: error 191 at offset 0: PCRE2_EXTRA_ALLOW_SURROGATE_ESCAPES is not allowed in UTF-16 mode
     \x{dfff}\x{df01}\=no_utf_check
 
+# This has different starting code units in 8-bit mode. 
+
+/^[^ab]/IB,utf
+------------------------------------------------------------------
+        Bra
+        ^
+        [\x00-`c-\xff] (neg)
+        Ket
+        End
+------------------------------------------------------------------
+Capturing subpattern count = 0
+Compile options: utf
+Overall options: anchored utf
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 
+  5 6 7 8 9 : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y 
+  Z [ \ ] ^ _ ` c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x7f 
+  \x80 \x81 \x82 \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e 
+  \x8f \x90 \x91 \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d 
+  \x9e \x9f \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac 
+  \xad \xae \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb 
+  \xbc \xbd \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca 
+  \xcb \xcc \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 
+  \xda \xdb \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 
+  \xe9 \xea \xeb \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 
+  \xf8 \xf9 \xfa \xfb \xfc \xfd \xfe \xff 
+Subject length lower bound = 1
+    c
+ 0: c
+    \x{ff}
+ 0: \x{ff}
+    \x{100}
+ 0: \x{100}
+\= Expect no match
+    aaa
+No match
+
 # End of testinput12

testdata/testoutput12-32

@@ -1425,4 +1425,42 @@ No match
     \x{dfff}\x{df01}\=no_utf_check
  0: \x{dfff}\x{df01}
 
+# This has different starting code units in 8-bit mode. 
+
+/^[^ab]/IB,utf
+------------------------------------------------------------------
+        Bra
+        ^
+        [\x00-`c-\xff] (neg)
+        Ket
+        End
+------------------------------------------------------------------
+Capturing subpattern count = 0
+Compile options: utf
+Overall options: anchored utf
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 
+  5 6 7 8 9 : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y 
+  Z [ \ ] ^ _ ` c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x7f 
+  \x80 \x81 \x82 \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e 
+  \x8f \x90 \x91 \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d 
+  \x9e \x9f \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac 
+  \xad \xae \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb 
+  \xbc \xbd \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca 
+  \xcb \xcc \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 
+  \xda \xdb \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 
+  \xe9 \xea \xeb \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 
+  \xf8 \xf9 \xfa \xfb \xfc \xfd \xfe \xff 
+Subject length lower bound = 1
+    c
+ 0: c
+    \x{ff}
+ 0: \x{ff}
+    \x{100}
+ 0: \x{100}
+\= Expect no match
+    aaa
+No match
+
 # End of testinput12

testdata/testoutput17

@@ -368,6 +368,7 @@ No match
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 6
 JIT compilation was successful
 #pop jitverify
@@ -379,6 +380,7 @@ JIT compilation was successful
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 6
 JIT compilation was successful
 #save testsaved1

testdata/testoutput2

@@ -72,6 +72,7 @@ No match
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
     abc
  0: abc
@@ -110,6 +111,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
     abc
  0: abc
@@ -339,6 +341,7 @@ Subject length lower bound = 19
 /the quick brown fox/I,anchored
 Capturing subpattern count = 0
 Options: anchored
+First code unit = 't'
 Subject length lower bound = 19
     the quick brown fox
  0: the quick brown fox
@@ -351,6 +354,7 @@ Failed: error 111 at offset 4: unrecognized character after (? or (?-
 
 /^abc|def/I
 Capturing subpattern count = 0
+Starting code units: a d 
 Subject length lower bound = 3
     abcdef
  0: abc
@@ -495,12 +499,14 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = '1'
 Subject length lower bound = 4
 
 /(^b|(?i)^d)/I
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+Starting code units: D b d 
 Subject length lower bound = 1
 
 /(?s).*/I
@@ -624,6 +630,7 @@ Capturing subpattern count = 0
 Max lookbehind = 1
 Compile options: multiline
 Overall options: anchored multiline
+First code unit = 'a'
 Subject length lower bound = 3
 
 /^abc/Im
@@ -637,6 +644,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 5
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
   aaaaabbbbbcccccdef
  0: aaaaabbbbbcccccdef
@@ -808,6 +816,7 @@ Capturing subpattern count = 1
 Max back reference = 1
 Compile options: <none>
 Overall options: anchored
+Starting code units: a 
 Subject length lower bound = 4
 \= Expect no match
     aaaa
@@ -1004,6 +1013,7 @@ Subject length lower bound = 16
 Capturing subpattern count = 3
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b 
 Subject length lower bound = 4
     adef\=get=1,get=2,get=3,get=4,getall
  0: adef
@@ -1042,6 +1052,7 @@ Get substring 4 failed (-49): unknown substring
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 7
     abc\00def\=copy=0,getall
  0: abc\x00def
@@ -1227,6 +1238,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'i'
 Subject length lower bound = 3
     ississippi
  0: iss
@@ -1286,6 +1298,7 @@ Capturing subpattern count = 0
 Contains explicit CR or LF match
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
     ab\nab\ncd
  0: ab\x0a
@@ -1776,6 +1789,8 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: 0 1 2 3 4 5 6 7 8 9 A B C D E F G H I J K L M N O P 
+  Q R S T U V W X Y Z a b c d e f g h i j k l m n o p q r s t u v w x y z 
 Subject length lower bound = 1
 
 /^[[:^alnum:]]/IB
@@ -1789,6 +1804,18 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / : ; < = > 
+  ? @ [ \ ] ^ _ ` { | } ~ \x7f \x80 \x81 \x82 \x83 \x84 \x85 \x86 \x87 \x88 
+  \x89 \x8a \x8b \x8c \x8d \x8e \x8f \x90 \x91 \x92 \x93 \x94 \x95 \x96 \x97 
+  \x98 \x99 \x9a \x9b \x9c \x9d \x9e \x9f \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 
+  \xa7 \xa8 \xa9 \xaa \xab \xac \xad \xae \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 
+  \xb6 \xb7 \xb8 \xb9 \xba \xbb \xbc \xbd \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 
+  \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb \xcc \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 
+  \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda \xdb \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 
+  \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 \xea \xeb \xec \xed \xee \xef \xf0 \xf1 
+  \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 \xf9 \xfa \xfb \xfc \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /^[[:alpha:]]/IB
@@ -1802,6 +1829,8 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z 
+  a b c d e f g h i j k l m n o p q r s t u v w x y z 
 Subject length lower bound = 1
 
 /^[[:^alpha:]]/IB
@@ -1815,6 +1844,19 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 
+  5 6 7 8 9 : ; < = > ? @ [ \ ] ^ _ ` { | } ~ \x7f \x80 \x81 \x82 \x83 \x84 
+  \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e \x8f \x90 \x91 \x92 \x93 
+  \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d \x9e \x9f \xa0 \xa1 \xa2 
+  \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac \xad \xae \xaf \xb0 \xb1 
+  \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb \xbc \xbd \xbe \xbf \xc0 
+  \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb \xcc \xcd \xce \xcf 
+  \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda \xdb \xdc \xdd \xde 
+  \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 \xea \xeb \xec \xed 
+  \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 \xf9 \xfa \xfb \xfc 
+  \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /[_[:alpha:]]/I
@@ -1834,6 +1876,12 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 
+  5 6 7 8 9 : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y 
+  Z [ \ ] ^ _ ` a b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ 
+  \x7f 
 Subject length lower bound = 1
 
 /^[[:^ascii:]]/IB
@@ -1847,6 +1895,15 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x80 \x81 \x82 \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a 
+  \x8b \x8c \x8d \x8e \x8f \x90 \x91 \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 
+  \x9a \x9b \x9c \x9d \x9e \x9f \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 
+  \xa9 \xaa \xab \xac \xad \xae \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 
+  \xb8 \xb9 \xba \xbb \xbc \xbd \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 
+  \xc7 \xc8 \xc9 \xca \xcb \xcc \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 
+  \xd6 \xd7 \xd8 \xd9 \xda \xdb \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 
+  \xe5 \xe6 \xe7 \xe8 \xe9 \xea \xeb \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 
+  \xf4 \xf5 \xf6 \xf7 \xf8 \xf9 \xfa \xfb \xfc \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /^[[:blank:]]/IB
@@ -1860,6 +1917,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x09 \x20 
 Subject length lower bound = 1
 
 /^[[:^blank:]]/IB
@@ -1873,6 +1931,20 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x0a \x0b 
+  \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 \x1a 
+  \x1b \x1c \x1d \x1e \x1f ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 9 
+  : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z [ \ ] ^ 
+  _ ` a b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x7f \x80 
+  \x81 \x82 \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e \x8f 
+  \x90 \x91 \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d \x9e 
+  \x9f \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac \xad 
+  \xae \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb \xbc 
+  \xbd \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb 
+  \xcc \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda 
+  \xdb \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 
+  \xea \xeb \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 
+  \xf9 \xfa \xfb \xfc \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /[\n\x0b\x0c\x0d[:blank:]]/I
@@ -1892,6 +1964,9 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x7f 
 Subject length lower bound = 1
 
 /^[[:digit:]]/IB
@@ -1905,6 +1980,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: 0 1 2 3 4 5 6 7 8 9 
 Subject length lower bound = 1
 
 /^[[:graph:]]/IB
@@ -1918,6 +1994,9 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 9 : 
+  ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z [ \ ] ^ _ 
+  ` a b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ 
 Subject length lower bound = 1
 
 /^[[:lower:]]/IB
@@ -1931,6 +2010,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b c d e f g h i j k l m n o p q r s t u v w x y z 
 Subject length lower bound = 1
 
 /^[[:print:]]/IB
@@ -1944,6 +2024,9 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 
+  9 : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z [ \ ] 
+  ^ _ ` a b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ 
 Subject length lower bound = 1
 
 /^[[:punct:]]/IB
@@ -1957,6 +2040,8 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: ! " # $ % & ' ( ) * + , - . / : ; < = > ? @ [ \ ] ^ 
+  _ ` { | } ~ 
 Subject length lower bound = 1
 
 /^[[:space:]]/IB
@@ -1970,6 +2055,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x09 \x0a \x0b \x0c \x0d \x20 
 Subject length lower bound = 1
 
 /^[[:upper:]]/IB
@@ -1983,6 +2069,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z 
 Subject length lower bound = 1
 
 /^[[:xdigit:]]/IB
@@ -1996,6 +2083,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: 0 1 2 3 4 5 6 7 8 9 A B C D E F a b c d e f 
 Subject length lower bound = 1
 
 /^[[:word:]]/IB
@@ -2009,6 +2097,8 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: 0 1 2 3 4 5 6 7 8 9 A B C D E F G H I J K L M N O P 
+  Q R S T U V W X Y Z _ a b c d e f g h i j k l m n o p q r s t u v w x y z 
 Subject length lower bound = 1
 
 /^[[:^cntrl:]]/IB
@@ -2022,6 +2112,18 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x20 ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 
+  9 : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z [ \ ] 
+  ^ _ ` a b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x80 \x81 
+  \x82 \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e \x8f \x90 
+  \x91 \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d \x9e \x9f 
+  \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac \xad \xae 
+  \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb \xbc \xbd 
+  \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb \xcc 
+  \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda \xdb 
+  \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 \xea 
+  \xeb \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 \xf9 
+  \xfa \xfb \xfc \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /^[12[:^digit:]]/IB
@@ -2035,6 +2137,20 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x09 \x0a 
+  \x0b \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 
+  \x1a \x1b \x1c \x1d \x1e \x1f \x20 ! " # $ % & ' ( ) * + , - . / 1 2 : ; < 
+  = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z [ \ ] ^ _ ` a 
+  b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x7f \x80 \x81 \x82 
+  \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e \x8f \x90 \x91 
+  \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d \x9e \x9f \xa0 
+  \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac \xad \xae \xaf 
+  \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb \xbc \xbd \xbe 
+  \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb \xcc \xcd 
+  \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda \xdb \xdc 
+  \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 \xea \xeb 
+  \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 \xf9 \xfa 
+  \xfb \xfc \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /^[[:^blank:]]/IB
@@ -2048,6 +2164,20 @@ Subject length lower bound = 1
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: \x00 \x01 \x02 \x03 \x04 \x05 \x06 \x07 \x08 \x0a \x0b 
+  \x0c \x0d \x0e \x0f \x10 \x11 \x12 \x13 \x14 \x15 \x16 \x17 \x18 \x19 \x1a 
+  \x1b \x1c \x1d \x1e \x1f ! " # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 9 
+  : ; < = > ? @ A B C D E F G H I J K L M N O P Q R S T U V W X Y Z [ \ ] ^ 
+  _ ` a b c d e f g h i j k l m n o p q r s t u v w x y z { | } ~ \x7f \x80 
+  \x81 \x82 \x83 \x84 \x85 \x86 \x87 \x88 \x89 \x8a \x8b \x8c \x8d \x8e \x8f 
+  \x90 \x91 \x92 \x93 \x94 \x95 \x96 \x97 \x98 \x99 \x9a \x9b \x9c \x9d \x9e 
+  \x9f \xa0 \xa1 \xa2 \xa3 \xa4 \xa5 \xa6 \xa7 \xa8 \xa9 \xaa \xab \xac \xad 
+  \xae \xaf \xb0 \xb1 \xb2 \xb3 \xb4 \xb5 \xb6 \xb7 \xb8 \xb9 \xba \xbb \xbc 
+  \xbd \xbe \xbf \xc0 \xc1 \xc2 \xc3 \xc4 \xc5 \xc6 \xc7 \xc8 \xc9 \xca \xcb 
+  \xcc \xcd \xce \xcf \xd0 \xd1 \xd2 \xd3 \xd4 \xd5 \xd6 \xd7 \xd8 \xd9 \xda 
+  \xdb \xdc \xdd \xde \xdf \xe0 \xe1 \xe2 \xe3 \xe4 \xe5 \xe6 \xe7 \xe8 \xe9 
+  \xea \xeb \xec \xed \xee \xef \xf0 \xf1 \xf2 \xf3 \xf4 \xf5 \xf6 \xf7 \xf8 
+  \xf9 \xfa \xfb \xfc \xfd \xfe \xff 
 Subject length lower bound = 1
 
 /[01[:alpha:]%]/IB
@@ -2418,6 +2548,7 @@ Subject length lower bound = 4
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 1
     aba
  0: aba
@@ -2428,6 +2559,7 @@ Subject length lower bound = 1
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2438,6 +2570,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2448,6 +2581,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2458,6 +2592,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2467,6 +2602,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 3
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2478,6 +2614,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2488,6 +2625,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2497,6 +2635,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbbaa
  0: aabbbaa
@@ -2506,6 +2645,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbbaa
  0: aabbbaa
@@ -2515,6 +2655,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbaa
  0: aabbaa
@@ -2524,6 +2665,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbbaa
  0: aabbbaa
@@ -2533,6 +2675,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 3
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbbaa
  0: aabbbaa
@@ -2544,6 +2687,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 3
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
     aabbbbaa
  0: aabbbbaa
@@ -3052,6 +3196,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 5
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
 
 /^x(?U)a+b/IB
@@ -3067,6 +3212,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'x'
 Last code unit = 'b'
 Subject length lower bound = 3
 
@@ -3085,6 +3231,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'x'
 Last code unit = 'b'
 Subject length lower bound = 3
 
@@ -3725,6 +3872,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
 
 /(?C)a|b/I
@@ -3785,6 +3933,7 @@ No match
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = '>'
 Last code unit = '<'
 Subject length lower bound = 10
    >abc>123<xyz<
@@ -3835,6 +3984,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+Starting code units: ( - 0 1 2 3 4 5 6 7 8 9 
 Subject length lower bound = 1
     12
  0: 12
@@ -3854,6 +4004,7 @@ No match
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+First code unit = 'x'
 Subject length lower bound = 3
     xyz
  0: xyz
@@ -3913,6 +4064,7 @@ Failed: error 114 at offset 10: missing closing parenthesis
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 9
     abcdefabc
  0: abcdefabc
@@ -3922,6 +4074,7 @@ Subject length lower bound = 9
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b c 
 Subject length lower bound = 2
     a=a
  0: a=a
@@ -3937,6 +4090,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 2
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b c 
 Subject length lower bound = 2
     a=a
  0: a=a
@@ -5173,6 +5327,7 @@ No match
 Capturing subpattern count = 3
 Compile options: <none>
 Overall options: anchored
+Starting code units: 0 1 2 3 4 5 6 7 8 9 
 Last code unit = '/'
 Subject length lower bound = 6
     13/05/04\=ps
@@ -5270,6 +5425,7 @@ Partial match: c12
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: 0 1 2 3 4 5 6 7 8 9 
 Last code unit = 'X'
 Subject length lower bound = 4
     1\=ps
@@ -5643,6 +5799,7 @@ Named capturing subpatterns:
   A   3
 Compile options: dupnames
 Overall options: anchored dupnames
+First code unit = 'a'
 Subject length lower bound = 2
     a1b\=copy=A
  0: a1
@@ -5680,6 +5837,7 @@ Named capturing subpatterns:
   A   2
 Compile options: dupnames
 Overall options: anchored dupnames
+First code unit = 'a'
 Subject length lower bound = 2
     ab\=copy=A
  0: ab
@@ -5693,6 +5851,7 @@ Named capturing subpatterns:
   A   1
   A   2
 Options: dupnames
+Starting code units: a c 
 Subject length lower bound = 2
     ab\=copy=A
  0: ab
@@ -5711,6 +5870,7 @@ Named capturing subpatterns:
   A   3
   A   4
 Options: dupnames
+Starting code units: a c 
 Subject length lower bound = 2
     cdefgh\=copy=A
  0: cdefgh
@@ -5727,6 +5887,7 @@ Named capturing subpatterns:
   A   3
 Compile options: dupnames
 Overall options: anchored dupnames
+First code unit = 'a'
 Subject length lower bound = 2
     a1b\=get=A
  0: a1
@@ -5754,6 +5915,7 @@ Named capturing subpatterns:
   A   2
 Compile options: dupnames
 Overall options: anchored dupnames
+First code unit = 'a'
 Subject length lower bound = 2
     ab\=get=A
  0: ab
@@ -5767,6 +5929,7 @@ Named capturing subpatterns:
   A   1
   A   2
 Options: dupnames
+Starting code units: a c 
 Subject length lower bound = 2
     ab\=get=A
  0: ab
@@ -5785,6 +5948,7 @@ Named capturing subpatterns:
   A   3
   A   4
 Options: dupnames
+Starting code units: a c 
 Subject length lower bound = 2
     cdefgh\=get=A
  0: cdefgh
@@ -5802,6 +5966,7 @@ Named capturing subpatterns:
 Compile options: <none>
 Overall options: anchored
 Duplicate name status changes
+First code unit = 'a'
 Subject length lower bound = 2
     a1b\=copy=A
  0: a1
@@ -5832,6 +5997,7 @@ Named capturing subpatterns:
 Compile options: <none>
 Overall options: anchored
 Duplicate name status changes
+First code unit = 'a'
 Subject length lower bound = 6
     a bc d\=copy=A,copy=B,copy=C
  0: a bc d
@@ -6233,6 +6399,7 @@ Subject length lower bound = 4
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b 
 Last code unit = 'b'
 Subject length lower bound = 2
 
@@ -6249,6 +6416,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b 
 Last code unit = 'b'
 Subject length lower bound = 2
 
@@ -6265,6 +6433,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+Starting code units: a b 
 Last code unit = 'b'
 Subject length lower bound = 2
 
@@ -6281,6 +6450,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Last code unit = 'A'
 Subject length lower bound = 3
     aaaA5
@@ -6302,6 +6472,7 @@ No match
 Capturing subpattern count = 0
 Compile options: caseless
 Overall options: anchored caseless
+Starting code units: A a 
 Last code unit = 'A' (caseless)
 Subject length lower bound = 2
     aaaA5
@@ -9540,6 +9711,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'F'
 Last code unit = ':'
 Subject length lower bound = 22
 
@@ -9691,6 +9863,7 @@ Named capturing subpatterns:
   D   1
 Compile options: dupnames extended
 Overall options: anchored dupnames extended
+Starting code units: a e 
 Subject length lower bound = 2
     abcdX
  0: abcdX
@@ -10445,12 +10618,14 @@ Failed: error 125 at offset 0: lookbehind assertion is not fixed length
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
 
 /(^ab)++/I
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
 
 /(^ab|^)+/I
@@ -10471,12 +10646,14 @@ Subject length lower bound = 0
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
 
 /(?:^ab)++/I
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 2
 
 /(?:^ab|^)+/I
@@ -11586,6 +11763,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: dotall
 Overall options: anchored dotall
+First code unit = 'a'
 Subject length lower bound = 2
 
 /.*?a(*SKIP)b/I
@@ -11608,6 +11786,7 @@ Subject length lower bound = 2
 Capturing subpattern count = 0
 Compile options: dotall
 Overall options: anchored dotall
+First code unit = 'a'
 Subject length lower bound = 2
 
 /(?>.*?)(?<=(abcd)|(wxyz))/I
@@ -13375,7 +13554,6 @@ Subject length lower bound = 1
 /(|ab)*?d/I,no_start_optimize
 Capturing subpattern count = 1
 Options: no_start_optimize
-Last code unit = 'd'
 Subject length lower bound = 0
    abd
  0: abd
@@ -13641,12 +13819,14 @@ get substring list failed (-2): partial match
 Capturing subpattern count = 0
 Compile options: <none>
 Overall options: anchored
+First code unit = 'a'
 Subject length lower bound = 3
 
 /^abc/info,no_dotstar_anchor
 Capturing subpattern count = 0
 Compile options: no_dotstar_anchor
 Overall options: anchored no_dotstar_anchor
+First code unit = 'a'
 Subject length lower bound = 3
 
 /.*\d/info,auto_callout
@@ -14684,6 +14864,7 @@ Capturing subpattern count = 2
 Max back reference = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'o'
 Last code unit = '}'
 Subject length lower bound = 65535
 
@@ -15607,6 +15788,7 @@ No match
 Capturing subpattern count = 1
 Compile options: <none>
 Overall options: anchored
+First code unit = 'b'
 Subject length lower bound = 2
 
 /(a){0}.*bc/sI
@@ -15885,6 +16067,10 @@ No match
 No match
 
 /^(?!A(?C1)B)C/
+    ABC\=callout_error=1,no_jit
+No match
+
+/^(?!A(?C1)B)C/no_start_optimize
     ABC\=callout_error=1
 --->ABC
   1 ^^      B

testdata/testoutput5

@@ -194,6 +194,7 @@ Subject length lower bound = 3
 Capturing subpattern count = 0
 Compile options: utf
 Overall options: anchored utf
+Starting code units: a b 
 Subject length lower bound = 1
     bar
  0: b
@@ -205,28 +206,6 @@ No match
     \x{100}
 No match
 
-/^[^ab]/IB,utf
-------------------------------------------------------------------
-        Bra
-        ^
-        [\x00-`c-\xff] (neg)
-        Ket
-        End
-------------------------------------------------------------------
-Capturing subpattern count = 0
-Compile options: utf
-Overall options: anchored utf
-Subject length lower bound = 1
-    c
- 0: c
-    \x{ff}
- 0: \x{ff}
-    \x{100}
- 0: \x{100}
-\= Expect no match
-    aaa
-No match
-
 /\x{100}*(\d+|"(?1)")/utf
     1234
  0: 1234
@@ -479,7 +458,10 @@ Subject length lower bound = 0
     \x{100}X
  0: X
 
-/^\ሴ/IB,utf
+# Use no_start_optimize because the first code unit is different in 8-bit from
+# the wider modes.
+
+/^\ሴ/IB,utf,no_start_optimize
 ------------------------------------------------------------------
         Bra
         ^
@@ -488,9 +470,9 @@ Subject length lower bound = 0
         End
 ------------------------------------------------------------------
 Capturing subpattern count = 0
-Compile options: utf
+Compile options: no_start_optimize utf
-Overall options: anchored utf
+Overall options: anchored no_start_optimize utf
-Subject length lower bound = 1
+Subject length lower bound = 0
 
 /()()()()()()()()()()
  ()()()()()()()()()()