Merge branch '10.1' into 10.2

31 files changed, 825 insertions, 521 deletions

diff --git a/pcre/ChangeLog b/pcre/ChangeLog
index 01511b1327c..590a7542885 100644
--- a/pcre/ChangeLog
+++ b/pcre/ChangeLog
@@ -4,6 +4,53 @@ ChangeLog for PCRE
 Note that the PCRE 8.xx series (PCRE1) is now in a bugfix-only state. All
 development is happening in the PCRE2 10.xx series.
 
+Version 8.41 05-July-2017
+-------------------------
+
+1.  Fixed typo in CMakeLists.txt (wrong number of arguments for
+PCRE_STATIC_RUNTIME (affects MSVC only).
+
+2.  Issue 1 for 8.40 below was not correctly fixed. If pcregrep in multiline
+mode with --only-matching matched several lines, it restarted scanning at the
+next line instead of moving on to the end of the matched string, which can be
+several lines after the start.
+
+3.  Fix a missing else in the JIT compiler reported by 'idaifish'.
+
+4.  A (?# style comment is now ignored between a basic quantifier and a
+following '+' or '?' (example: /X+(?#comment)?Y/.
+
+5.  Avoid use of a potentially overflowing buffer in pcregrep (patch by Petr
+Pisar).
+
+6.  Fuzzers have reported issues in pcretest. These are NOT serious (it is,
+after all, just a test program). However, to stop the reports, some easy ones
+are fixed:
+
+    (a) Check for values < 256 when calling isprint() in pcretest.
+    (b) Give an error for too big a number after \O.
+
+7.  In the 32-bit library in non-UTF mode, an attempt to find a Unicode
+property for a character with a code point greater than 0x10ffff (the Unicode
+maximum) caused a crash.
+
+8. The alternative matching function, pcre_dfa_exec() misbehaved if it
+encountered a character class with a possessive repeat, for example [a-f]{3}+.
+
+9. When pcretest called pcre_copy_substring() in 32-bit mode, it set the buffer
+length incorrectly, which could result in buffer overflow.
+
+10. Remove redundant line of code (accidentally left in ages ago).
+
+11. Applied C++ patch from Irfan Adilovic to guard 'using std::' directives
+with namespace pcrecpp (Bugzilla #2084).
+
+12. Remove a duplication typo in pcre_tables.c.
+
+13. Fix returned offsets from regexec() when REG_STARTEND is used with a
+starting offset greater than zero.
+
+
 Version 8.40 11-January-2017
 ----------------------------
 
diff --git a/pcre/NEWS b/pcre/NEWS
index b92c4f9605e..36be07cb880 100644
--- a/pcre/NEWS
+++ b/pcre/NEWS
@@ -1,6 +1,12 @@
 News about PCRE releases
 ------------------------
 
+Release 8.41 13-June-2017
+-------------------------
+
+This is a bug-fix release.
+
+
 Release 8.40 11-January-2017
 ----------------------------
 
diff --git a/pcre/configure.ac b/pcre/configure.ac
index 24ef7271e05..718a18508c9 100644
--- a/pcre/configure.ac
+++ b/pcre/configure.ac
@@ -9,18 +9,18 @@ dnl The PCRE_PRERELEASE feature is for identifying release candidates. It might
 dnl be defined as -RC2, for example. For real releases, it should be empty.
 
 m4_define(pcre_major, [8])
-m4_define(pcre_minor, [40])
+m4_define(pcre_minor, [41])
 m4_define(pcre_prerelease, [])
-m4_define(pcre_date, [2017-01-11])
+m4_define(pcre_date, [2017-07-05])
 
 # NOTE: The CMakeLists.txt file searches for the above variables in the first
 # 50 lines of this file. Please update that if the variables above are moved.
 
 # Libtool shared library interface versions (current:revision:age)
-m4_define(libpcre_version, [3:8:2])
-m4_define(libpcre16_version, [2:8:2])
-m4_define(libpcre32_version, [0:8:0])
-m4_define(libpcreposix_version, [0:4:0])
+m4_define(libpcre_version, [3:9:2])
+m4_define(libpcre16_version, [2:9:2])
+m4_define(libpcre32_version, [0:9:0])
+m4_define(libpcreposix_version, [0:5:0])
 m4_define(libpcrecpp_version, [0:1:0])
 
 AC_PREREQ(2.57)
diff --git a/pcre/doc/html/pcrejit.html b/pcre/doc/html/pcrejit.html
index 210f1da0262..abb342522f9 100644
--- a/pcre/doc/html/pcrejit.html
+++ b/pcre/doc/html/pcrejit.html
@@ -79,9 +79,12 @@ API that is JIT-specific.
 </P>
 <P>
 If your program may sometimes be linked with versions of PCRE that are older
-than 8.20, but you want to use JIT when it is available, you can test
-the values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT macro such
-as PCRE_CONFIG_JIT, for compile-time control of your code.
+than 8.20, but you want to use JIT when it is available, you can test the
+values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT macro such as
+PCRE_CONFIG_JIT, for compile-time control of your code. Also beware that the
+<b>pcre_jit_exec()</b> function was not available at all before 8.32,
+and may not be available at all if PCRE isn't compiled with
+--enable-jit. See the "JIT FAST PATH API" section below for details.
 </P>
 <br><a name="SEC4" href="#TOC1">SIMPLE USE OF JIT</a><br>
 <P>
@@ -119,6 +122,20 @@ when you call <b>pcre_study()</b>:
   PCRE_STUDY_JIT_PARTIAL_HARD_COMPILE
   PCRE_STUDY_JIT_PARTIAL_SOFT_COMPILE
 </pre>
+If using <b>pcre_jit_exec()</b> and supporting a pre-8.32 version of
+PCRE, you can insert:
+<pre>
+   #if PCRE_MAJOR &#62;= 8 && PCRE_MINOR &#62;= 32
+   pcre_jit_exec(...);
+   #else
+   pcre_exec(...)
+   #endif
+</pre>
+but as described in the "JIT FAST PATH API" section below this assumes
+version 8.32 and later are compiled with --enable-jit, which may
+break.
+<br>
+<br>
 The JIT compiler generates different optimized code for each of the three
 modes (normal, soft partial, hard partial). When <b>pcre_exec()</b> is called,
 the appropriate code is run if it is available. Otherwise, the pattern is
@@ -428,6 +445,36 @@ fast path, and if invalid data is passed, the result is undefined.
 Bypassing the sanity checks and the <b>pcre_exec()</b> wrapping can give
 speedups of more than 10%.
 </P>
+<P>
+Note that the <b>pcre_jit_exec()</b> function is not available in versions of
+PCRE before 8.32 (released in November 2012). If you need to support versions
+that old you must either use the slower <b>pcre_exec()</b>, or switch between
+the two codepaths by checking the values of PCRE_MAJOR and PCRE_MINOR.
+</P>
+<P>
+Due to an unfortunate implementation oversight, even in versions 8.32
+and later there will be no <b>pcre_jit_exec()</b> stub function defined
+when PCRE is compiled with --disable-jit, which is the default, and
+there's no way to detect whether PCRE was compiled with --enable-jit
+via a macro.
+</P>
+<P>
+If you need to support versions older than 8.32, or versions that may
+not build with --enable-jit, you must either use the slower
+<b>pcre_exec()</b>, or switch between the two codepaths by checking the
+values of PCRE_MAJOR and PCRE_MINOR.
+</P>
+<P>
+Switching between the two by checking the version assumes that all the
+versions being targeted are built with --enable-jit. To also support
+builds that may use --disable-jit either <b>pcre_exec()</b> must be
+used, or a compile-time check for JIT via <b>pcre_config()</b> (which
+assumes the runtime environment will be the same), or as the Git
+project decided to do, simply assume that <b>pcre_jit_exec()</b> is
+present in 8.32 or later unless a compile-time flag is provided, see
+the "grep: un-break building with PCRE &#62;= 8.32 without --enable-jit"
+commit in git.git for an example of that.
+</P>
 <br><a name="SEC12" href="#TOC1">SEE ALSO</a><br>
 <P>
 <b>pcreapi</b>(3)
@@ -443,9 +490,9 @@ Cambridge CB2 3QH, England.
 </P>
 <br><a name="SEC14" href="#TOC1">REVISION</a><br>
 <P>
-Last updated: 17 March 2013
+Last updated: 05 July 2017
 <br>
-Copyright &copy; 1997-2013 University of Cambridge.
+Copyright &copy; 1997-2017 University of Cambridge.
 <br>
 <p>
 Return to the <a href="index.html">PCRE index page</a>.
diff --git a/pcre/doc/html/pcretest.html b/pcre/doc/html/pcretest.html
index 839fabf189b..ba540d3c385 100644
--- a/pcre/doc/html/pcretest.html
+++ b/pcre/doc/html/pcretest.html
@@ -74,6 +74,11 @@ newline as data characters. However, in some Windows environments character 26
 maximum portability, therefore, it is safest to use only ASCII characters in
 <b>pcretest</b> input files.
 </P>
+<P>
+The input is processed using using C's string functions, so must not
+contain binary zeroes, even though in Unix-like environments, <b>fgets()</b>
+treats any bytes other than newline as data characters.
+</P>
 <br><a name="SEC3" href="#TOC1">PCRE's 8-BIT, 16-BIT AND 32-BIT LIBRARIES</a><br>
 <P>
 From release 8.30, two separate PCRE libraries can be built. The original one
@@ -1149,9 +1154,9 @@ Cambridge CB2 3QH, England.
 </P>
 <br><a name="SEC17" href="#TOC1">REVISION</a><br>
 <P>
-Last updated: 09 February 2014
+Last updated: 23 February 2017
 <br>
-Copyright &copy; 1997-2014 University of Cambridge.
+Copyright &copy; 1997-2017 University of Cambridge.
 <br>
 <p>
 Return to the <a href="index.html">PCRE index page</a>.
diff --git a/pcre/doc/pcre.txt b/pcre/doc/pcre.txt
index d68d5033ceb..c027538f500 100644
--- a/pcre/doc/pcre.txt
+++ b/pcre/doc/pcre.txt
@@ -8365,7 +8365,11 @@ AVAILABILITY OF JIT SUPPORT
        If  your program may sometimes be linked with versions of PCRE that are
        older than 8.20, but you want to use JIT when it is available, you  can
        test the values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT
-       macro such as PCRE_CONFIG_JIT, for compile-time control of your code.
+       macro such as PCRE_CONFIG_JIT, for compile-time control of  your  code.
+       Also  beware that the pcre_jit_exec() function was not available at all
+       before 8.32, and may not be available at all  if  PCRE  isn't  compiled
+       with  --enable-jit.  See  the  "JIT  FAST  PATH  API" section below for
+       details.
 
 
 SIMPLE USE OF JIT
@@ -8407,6 +8411,18 @@ SIMPLE USE OF JIT
          PCRE_STUDY_JIT_PARTIAL_HARD_COMPILE
          PCRE_STUDY_JIT_PARTIAL_SOFT_COMPILE
 
+       If using pcre_jit_exec() and supporting a pre-8.32 version of PCRE, you
+       can insert:
+
+          #if PCRE_MAJOR >= 8 && PCRE_MINOR >= 32
+          pcre_jit_exec(...);
+          #else
+          pcre_exec(...)
+          #endif
+
+       but  as described in the "JIT FAST PATH API" section below this assumes
+       version 8.32 and later are compiled with --enable-jit, which may break.
+
        The JIT compiler generates different optimized code  for  each  of  the
        three  modes  (normal, soft partial, hard partial). When pcre_exec() is
        called, the appropriate code is run if it is available. Otherwise,  the
@@ -8696,6 +8712,33 @@ JIT FAST PATH API
        Bypassing  the  sanity  checks  and  the  pcre_exec() wrapping can give
        speedups of more than 10%.
 
+       Note that the pcre_jit_exec() function is not available in versions  of
+       PCRE  before  8.32  (released in November 2012). If you need to support
+       versions that old you must either use the slower pcre_exec(), or switch
+       between  the  two  codepaths  by  checking the values of PCRE_MAJOR and
+       PCRE_MINOR.
+
+       Due to an unfortunate implementation oversight, even in  versions  8.32
+       and  later  there will be no pcre_jit_exec() stub function defined when
+       PCRE is compiled with --disable-jit, which is the default, and  there's
+       no  way  to  detect  whether  PCRE was compiled with --enable-jit via a
+       macro.
+
+       If you need to support versions older than 8.32, or versions  that  may
+       not   build   with   --enable-jit,  you  must  either  use  the  slower
+       pcre_exec(), or switch between the two codepaths by checking the values
+       of PCRE_MAJOR and PCRE_MINOR.
+
+       Switching  between the two by checking the version assumes that all the
+       versions being targeted are built with --enable-jit.  To  also  support
+       builds that may use --disable-jit either pcre_exec() must be used, or a
+       compile-time check for JIT via pcre_config() (which assumes the runtime
+       environment  will  be  the  same), or as the Git project decided to do,
+       simply assume that pcre_jit_exec() is present in 8.32 or later unless a
+       compile-time  flag  is  provided, see the "grep: un-break building with
+       PCRE >= 8.32 without --enable-jit" commit in git.git for an example  of
+       that.
+
 
 SEE ALSO
 
@@ -8711,8 +8754,8 @@ AUTHOR
 
 REVISION
 
-       Last updated: 17 March 2013
-       Copyright (c) 1997-2013 University of Cambridge.
+       Last updated: 05 July 2017
+       Copyright (c) 1997-2017 University of Cambridge.
 ------------------------------------------------------------------------------
 
 
diff --git a/pcre/doc/pcrejit.3 b/pcre/doc/pcrejit.3
index 341403f7c84..3b785f0f631 100644
--- a/pcre/doc/pcrejit.3
+++ b/pcre/doc/pcrejit.3
@@ -1,4 +1,4 @@
-.TH PCREJIT 3 "17 March 2013" "PCRE 8.33"
+.TH PCREJIT 3 "05 July 2017" "PCRE 8.41"
 .SH NAME
 PCRE - Perl-compatible regular expressions
 .SH "PCRE JUST-IN-TIME COMPILER SUPPORT"
@@ -54,9 +54,12 @@ programs that need the best possible performance, there is also a "fast path"
 API that is JIT-specific.
 .P
 If your program may sometimes be linked with versions of PCRE that are older
-than 8.20, but you want to use JIT when it is available, you can test
-the values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT macro such
-as PCRE_CONFIG_JIT, for compile-time control of your code.
+than 8.20, but you want to use JIT when it is available, you can test the
+values of PCRE_MAJOR and PCRE_MINOR, or the existence of a JIT macro such as
+PCRE_CONFIG_JIT, for compile-time control of your code. Also beware that the
+\fBpcre_jit_exec()\fP function was not available at all before 8.32,
+and may not be available at all if PCRE isn't compiled with
+--enable-jit. See the "JIT FAST PATH API" section below for details.
 .
 .
 .SH "SIMPLE USE OF JIT"
@@ -96,6 +99,19 @@ when you call \fBpcre_study()\fP:
   PCRE_STUDY_JIT_PARTIAL_HARD_COMPILE
   PCRE_STUDY_JIT_PARTIAL_SOFT_COMPILE
 .sp
+If using \fBpcre_jit_exec()\fP and supporting a pre-8.32 version of
+PCRE, you can insert:
+.sp
+   #if PCRE_MAJOR >= 8 && PCRE_MINOR >= 32
+   pcre_jit_exec(...);
+   #else
+   pcre_exec(...)
+   #endif
+.sp
+but as described in the "JIT FAST PATH API" section below this assumes
+version 8.32 and later are compiled with --enable-jit, which may
+break.
+.sp
 The JIT compiler generates different optimized code for each of the three
 modes (normal, soft partial, hard partial). When \fBpcre_exec()\fP is called,
 the appropriate code is run if it is available. Otherwise, the pattern is
@@ -404,6 +420,32 @@ fast path, and if invalid data is passed, the result is undefined.
 .P
 Bypassing the sanity checks and the \fBpcre_exec()\fP wrapping can give
 speedups of more than 10%.
+.P
+Note that the \fBpcre_jit_exec()\fP function is not available in versions of
+PCRE before 8.32 (released in November 2012). If you need to support versions
+that old you must either use the slower \fBpcre_exec()\fP, or switch between
+the two codepaths by checking the values of PCRE_MAJOR and PCRE_MINOR.
+.P
+Due to an unfortunate implementation oversight, even in versions 8.32
+and later there will be no \fBpcre_jit_exec()\fP stub function defined
+when PCRE is compiled with --disable-jit, which is the default, and
+there's no way to detect whether PCRE was compiled with --enable-jit
+via a macro.
+.P
+If you need to support versions older than 8.32, or versions that may
+not build with --enable-jit, you must either use the slower
+\fBpcre_exec()\fP, or switch between the two codepaths by checking the
+values of PCRE_MAJOR and PCRE_MINOR.
+.P
+Switching between the two by checking the version assumes that all the
+versions being targeted are built with --enable-jit. To also support
+builds that may use --disable-jit either \fBpcre_exec()\fP must be
+used, or a compile-time check for JIT via \fBpcre_config()\fP (which
+assumes the runtime environment will be the same), or as the Git
+project decided to do, simply assume that \fBpcre_jit_exec()\fP is
+present in 8.32 or later unless a compile-time flag is provided, see
+the "grep: un-break building with PCRE >= 8.32 without --enable-jit"
+commit in git.git for an example of that.
 .
 .
 .SH "SEE ALSO"
@@ -426,6 +468,6 @@ Cambridge CB2 3QH, England.
 .rs
 .sp
 .nf
-Last updated: 17 March 2013
-Copyright (c) 1997-2013 University of Cambridge.
+Last updated: 05 July 2017
+Copyright (c) 1997-2017 University of Cambridge.
 .fi
diff --git a/pcre/doc/pcretest.1 b/pcre/doc/pcretest.1
index 92640da8e1b..ea7457c03d0 100644
--- a/pcre/doc/pcretest.1
+++ b/pcre/doc/pcretest.1
@@ -1,4 +1,4 @@
-.TH PCRETEST 1 "09 February 2014" "PCRE 8.35"
+.TH PCRETEST 1 "23 February 2017" "PCRE 8.41"
 .SH NAME
 pcretest - a program for testing Perl-compatible regular expressions.
 .SH SYNOPSIS
@@ -50,6 +50,10 @@ newline as data characters. However, in some Windows environments character 26
 (hex 1A) causes an immediate end of file, and no further data is read. For
 maximum portability, therefore, it is safest to use only ASCII characters in
 \fBpcretest\fP input files.
+.P
+The input is processed using using C's string functions, so must not
+contain binary zeroes, even though in Unix-like environments, \fBfgets()\fP
+treats any bytes other than newline as data characters.
 .
 .
 .SH "PCRE's 8-BIT, 16-BIT AND 32-BIT LIBRARIES"
@@ -1151,6 +1155,6 @@ Cambridge CB2 3QH, England.
 .rs
 .sp
 .nf
-Last updated: 09 February 2014
-Copyright (c) 1997-2014 University of Cambridge.
+Last updated: 23 February 2017
+Copyright (c) 1997-2017 University of Cambridge.
 .fi
diff --git a/pcre/doc/pcretest.txt b/pcre/doc/pcretest.txt
index 55de5024431..6d7305cfe82 100644
--- a/pcre/doc/pcretest.txt
+++ b/pcre/doc/pcretest.txt
@@ -39,6 +39,10 @@ INPUT DATA FORMAT
        For  maximum  portability,  therefore,  it  is safest to use only ASCII
        characters in pcretest input files.
 
+       The input is processed using using C's string functions,  so  must  not
+       contain  binary  zeroes, even though in Unix-like environments, fgets()
+       treats any bytes other than newline as data characters.
+
 
 PCRE's 8-BIT, 16-BIT AND 32-BIT LIBRARIES
 
@@ -1083,5 +1087,5 @@ AUTHOR
 
 REVISION
 
-       Last updated: 09 February 2014
-       Copyright (c) 1997-2014 University of Cambridge.
+       Last updated: 23 February 2017
+       Copyright (c) 1997-2017 University of Cambridge.
diff --git a/pcre/pcre_compile.c b/pcre/pcre_compile.c
index de92313e2f8..42f204cdfff 100644
--- a/pcre/pcre_compile.c
+++ b/pcre/pcre_compile.c
@@ -5739,6 +5739,21 @@ for (;; ptr++)
       ptr = p - 1;    /* Character before the next significant one. */
       }
 
+    /* We also need to skip over (?# comments, which are not dependent on
+    extended mode. */
+
+    if (ptr[1] == CHAR_LEFT_PARENTHESIS && ptr[2] == CHAR_QUESTION_MARK &&
+        ptr[3] == CHAR_NUMBER_SIGN)
+      {
+      ptr += 4;
+      while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
+      if (*ptr == CHAR_NULL)
+        {
+        *errorcodeptr = ERR18;
+        goto FAILED;
+        }
+      }
+
     /* If the next character is '+', we have a possessive quantifier. This
     implies greediness, whatever the setting of the PCRE_UNGREEDY option.
     If the next character is '?' this is a minimizing repeat, by default,
@@ -8210,7 +8225,6 @@ for (;; ptr++)
 
       if (mclength == 1 || req_caseopt == 0)
         {
-        firstchar = mcbuffer[0] | req_caseopt;
         firstchar = mcbuffer[0];
         firstcharflags = req_caseopt;
 
diff --git a/pcre/pcre_dfa_exec.c b/pcre/pcre_dfa_exec.c
index 170ce6a0016..bc09ced3a7c 100644
--- a/pcre/pcre_dfa_exec.c
+++ b/pcre/pcre_dfa_exec.c
@@ -7,7 +7,7 @@ and semantics are as close as possible to those of the Perl 5 language (but see
 below for why this module is different).
 
                        Written by Philip Hazel
-           Copyright (c) 1997-2014 University of Cambridge
+           Copyright (c) 1997-2017 University of Cambridge
 
 -----------------------------------------------------------------------------
 Redistribution and use in source and binary forms, with or without
@@ -2625,7 +2625,7 @@ for (;;)
           if (isinclass)
             {
             int max = (int)GET2(ecode, 1 + IMM2_SIZE);
-            if (*ecode == OP_CRPOSRANGE)
+            if (*ecode == OP_CRPOSRANGE && count >= (int)GET2(ecode, 1))
               {
               active_count--;           /* Remove non-match possibility */
               next_active_state--;
diff --git a/pcre/pcre_exec.c b/pcre/pcre_exec.c
index 24b23ca2864..1a9bdd546ee 100644
--- a/pcre/pcre_exec.c
+++ b/pcre/pcre_exec.c
@@ -669,7 +669,7 @@ if (ecode == NULL)
     return match((PCRE_PUCHAR)&rdepth, NULL, NULL, 0, NULL, NULL, 1);
   else
     {
-    int len = (char *)&rdepth - (char *)eptr;
+    int len = (int)((char *)&rdepth - (char *)eptr);
     return (len > 0)? -len : len;
     }
   }
diff --git a/pcre/pcre_internal.h b/pcre/pcre_internal.h
index 2923b29f82d..97ff55d03b3 100644
--- a/pcre/pcre_internal.h
+++ b/pcre/pcre_internal.h
@@ -2772,6 +2772,9 @@ extern const pcre_uint8  PRIV(ucd_stage1)[];
 extern const pcre_uint16 PRIV(ucd_stage2)[];
 extern const pcre_uint32 PRIV(ucp_gentype)[];
 extern const pcre_uint32 PRIV(ucp_gbtable)[];
+#ifdef COMPILE_PCRE32
+extern const ucd_record  PRIV(dummy_ucd_record)[];
+#endif
 #ifdef SUPPORT_JIT
 extern const int         PRIV(ucp_typerange)[];
 #endif
@@ -2780,10 +2783,16 @@ extern const int         PRIV(ucp_typerange)[];
 /* UCD access macros */
 
 #define UCD_BLOCK_SIZE 128
-#define GET_UCD(ch) (PRIV(ucd_records) + \
+#define REAL_GET_UCD(ch) (PRIV(ucd_records) + \
         PRIV(ucd_stage2)[PRIV(ucd_stage1)[(int)(ch) / UCD_BLOCK_SIZE] * \
         UCD_BLOCK_SIZE + (int)(ch) % UCD_BLOCK_SIZE])
 
+#ifdef COMPILE_PCRE32
+#define GET_UCD(ch) ((ch > 0x10ffff)? PRIV(dummy_ucd_record) : REAL_GET_UCD(ch))
+#else
+#define GET_UCD(ch) REAL_GET_UCD(ch)
+#endif
+
 #define UCD_CHARTYPE(ch)    GET_UCD(ch)->chartype
 #define UCD_SCRIPT(ch)      GET_UCD(ch)->script
 #define UCD_CATEGORY(ch)    PRIV(ucp_gentype)[UCD_CHARTYPE(ch)]
diff --git a/pcre/pcre_jit_compile.c b/pcre/pcre_jit_compile.c
index 46ce6c65d54..249edbe8e7f 100644
--- a/pcre/pcre_jit_compile.c
+++ b/pcre/pcre_jit_compile.c
@@ -487,7 +487,7 @@ typedef struct compare_context {
 #undef CMP
 
 /* Used for accessing the elements of the stack. */
-#define STACK(i)      ((-(i) - 1) * (int)sizeof(sljit_sw))
+#define STACK(i)      ((i) * (int)sizeof(sljit_sw))
 
 #define TMP1          SLJIT_R0
 #define TMP2          SLJIT_R2
@@ -552,13 +552,15 @@ the start pointers when the end of the capturing group has not yet reached. */
   sljit_emit_cmp(compiler, (type), (src1), (src1w), (src2), (src2w))
 #define CMPTO(type, src1, src1w, src2, src2w, label) \
   sljit_set_label(sljit_emit_cmp(compiler, (type), (src1), (src1w), (src2), (src2w)), (label))
-#define OP_FLAGS(op, dst, dstw, src, srcw, type) \
-  sljit_emit_op_flags(compiler, (op), (dst), (dstw), (src), (srcw), (type))
+#define OP_FLAGS(op, dst, dstw, type) \
+  sljit_emit_op_flags(compiler, (op), (dst), (dstw), (type))
 #define GET_LOCAL_BASE(dst, dstw, offset) \
   sljit_get_local_base(compiler, (dst), (dstw), (offset))
 
 #define READ_CHAR_MAX 0x7fffffff
 
+#define INVALID_UTF_CHAR 888
+
 static pcre_uchar *bracketend(pcre_uchar *cc)
 {
 SLJIT_ASSERT((*cc >= OP_ASSERT && *cc <= OP_ASSERTBACK_NOT) || (*cc >= OP_ONCE && *cc <= OP_SCOND));
@@ -784,7 +786,7 @@ switch(*cc)
 
   default:
   /* All opcodes are supported now! */
-  SLJIT_ASSERT_STOP();
+  SLJIT_UNREACHABLE();
   return NULL;
   }
 }
@@ -1660,9 +1662,9 @@ while (cc < ccend)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(0));
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, -OVECTOR(0));
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       setsom_found = TRUE;
       }
     cc += 1;
@@ -1676,9 +1678,9 @@ while (cc < ccend)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), common->mark_ptr);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, -common->mark_ptr);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       setmark_found = TRUE;
       }
     cc += 1 + 2 + cc[1];
@@ -1689,27 +1691,27 @@ while (cc < ccend)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(0));
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, -OVECTOR(0));
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       setsom_found = TRUE;
       }
     if (common->mark_ptr != 0 && !setmark_found)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), common->mark_ptr);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, -common->mark_ptr);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       setmark_found = TRUE;
       }
     if (common->capture_last_ptr != 0 && !capture_last_found)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), common->capture_last_ptr);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, -common->capture_last_ptr);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       capture_last_found = TRUE;
       }
     cc += 1 + LINK_SIZE;
@@ -1723,20 +1725,20 @@ while (cc < ccend)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), common->capture_last_ptr);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, -common->capture_last_ptr);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-      stackpos += (int)sizeof(sljit_sw);
+      stackpos -= (int)sizeof(sljit_sw);
       capture_last_found = TRUE;
       }
     offset = (GET2(cc, 1 + LINK_SIZE)) << 1;
     OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, SLJIT_IMM, OVECTOR(offset));
-    stackpos += (int)sizeof(sljit_sw);
+    stackpos -= (int)sizeof(sljit_sw);
     OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(offset));
     OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(offset + 1));
     OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP1, 0);
-    stackpos += (int)sizeof(sljit_sw);
+    stackpos -= (int)sizeof(sljit_sw);
     OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), stackpos, TMP2, 0);
-    stackpos += (int)sizeof(sljit_sw);
+    stackpos -= (int)sizeof(sljit_sw);
 
     cc += 1 + LINK_SIZE + IMM2_SIZE;
     break;
@@ -1887,18 +1889,17 @@ BOOL tmp1empty = TRUE;
 BOOL tmp2empty = TRUE;
 pcre_uchar *alternative;
 enum {
-  start,
   loop,
   end
 } status;
 
-status = save ? start : loop;
-stackptr = STACK(stackptr - 2);
+status = loop;
+stackptr = STACK(stackptr);
 stacktop = STACK(stacktop - 1);
 
 if (!save)
   {
-  stackptr += (needs_control_head ? 2 : 1) * sizeof(sljit_sw);
+  stacktop -= (needs_control_head ? 2 : 1) * sizeof(sljit_sw);
   if (stackptr < stacktop)
     {
     OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), stackptr);
@@ -1914,196 +1915,186 @@ if (!save)
   /* The tmp1next must be TRUE in either way. */
   }
 
+SLJIT_ASSERT(common->recursive_head_ptr != 0);
+
 do
   {
   count = 0;
-  switch(status)
+  if (cc >= ccend)
     {
-    case start:
-    SLJIT_ASSERT(save && common->recursive_head_ptr != 0);
+    if (!save)
+      break;
+
     count = 1;
     srcw[0] = common->recursive_head_ptr;
     if (needs_control_head)
       {
       SLJIT_ASSERT(common->control_head_ptr != 0);
       count = 2;
-      srcw[1] = common->control_head_ptr;
+      srcw[0] = common->control_head_ptr;
+      srcw[1] = common->recursive_head_ptr;
+      }
+    status = end;
+    }
+  else switch(*cc)
+    {
+    case OP_KET:
+    if (PRIVATE_DATA(cc) != 0)
+      {
+      count = 1;
+      srcw[0] = PRIVATE_DATA(cc);
+      SLJIT_ASSERT(PRIVATE_DATA(cc + 1) != 0);
+      cc += PRIVATE_DATA(cc + 1);
       }
-    status = loop;
+    cc += 1 + LINK_SIZE;
+    break;
+
+    case OP_ASSERT:
+    case OP_ASSERT_NOT:
+    case OP_ASSERTBACK:
+    case OP_ASSERTBACK_NOT:
+    case OP_ONCE:
+    case OP_ONCE_NC:
+    case OP_BRAPOS:
+    case OP_SBRA:
+    case OP_SBRAPOS:
+    case OP_SCOND:
+    count = 1;
+    srcw[0] = PRIVATE_DATA(cc);
+    SLJIT_ASSERT(srcw[0] != 0);
+    cc += 1 + LINK_SIZE;
     break;
 
-    case loop:
-    if (cc >= ccend)
+    case OP_CBRA:
+    case OP_SCBRA:
+    if (common->optimized_cbracket[GET2(cc, 1 + LINK_SIZE)] == 0)
       {
-      status = end;
-      break;
+      count = 1;
+      srcw[0] = OVECTOR_PRIV(GET2(cc, 1 + LINK_SIZE));
       }
+    cc += 1 + LINK_SIZE + IMM2_SIZE;
+    break;
 
-    switch(*cc)
-      {
-      case OP_KET:
-      if (PRIVATE_DATA(cc) != 0)
-        {
-        count = 1;
-        srcw[0] = PRIVATE_DATA(cc);
-        SLJIT_ASSERT(PRIVATE_DATA(cc + 1) != 0);
-        cc += PRIVATE_DATA(cc + 1);
-        }
-      cc += 1 + LINK_SIZE;
-      break;
+    case OP_CBRAPOS:
+    case OP_SCBRAPOS:
+    count = 2;
+    srcw[0] = PRIVATE_DATA(cc);
+    srcw[1] = OVECTOR_PRIV(GET2(cc, 1 + LINK_SIZE));
+    SLJIT_ASSERT(srcw[0] != 0 && srcw[1] != 0);
+    cc += 1 + LINK_SIZE + IMM2_SIZE;
+    break;
 
-      case OP_ASSERT:
-      case OP_ASSERT_NOT:
-      case OP_ASSERTBACK:
-      case OP_ASSERTBACK_NOT:
-      case OP_ONCE:
-      case OP_ONCE_NC:
-      case OP_BRAPOS:
-      case OP_SBRA:
-      case OP_SBRAPOS:
-      case OP_SCOND:
+    case OP_COND:
+    /* Might be a hidden SCOND. */
+    alternative = cc + GET(cc, 1);
+    if (*alternative == OP_KETRMAX || *alternative == OP_KETRMIN)
+      {
       count = 1;
       srcw[0] = PRIVATE_DATA(cc);
       SLJIT_ASSERT(srcw[0] != 0);
-      cc += 1 + LINK_SIZE;
-      break;
-
-      case OP_CBRA:
-      case OP_SCBRA:
-      if (common->optimized_cbracket[GET2(cc, 1 + LINK_SIZE)] == 0)
-        {
-        count = 1;
-        srcw[0] = OVECTOR_PRIV(GET2(cc, 1 + LINK_SIZE));
-        }
-      cc += 1 + LINK_SIZE + IMM2_SIZE;
-      break;
+      }
+    cc += 1 + LINK_SIZE;
+    break;
 
-      case OP_CBRAPOS:
-      case OP_SCBRAPOS:
-      count = 2;
+    CASE_ITERATOR_PRIVATE_DATA_1
+    if (PRIVATE_DATA(cc))
+      {
+      count = 1;
       srcw[0] = PRIVATE_DATA(cc);
-      srcw[1] = OVECTOR_PRIV(GET2(cc, 1 + LINK_SIZE));
-      SLJIT_ASSERT(srcw[0] != 0 && srcw[1] != 0);
-      cc += 1 + LINK_SIZE + IMM2_SIZE;
-      break;
-
-      case OP_COND:
-      /* Might be a hidden SCOND. */
-      alternative = cc + GET(cc, 1);
-      if (*alternative == OP_KETRMAX || *alternative == OP_KETRMIN)
-        {
-        count = 1;
-        srcw[0] = PRIVATE_DATA(cc);
-        SLJIT_ASSERT(srcw[0] != 0);
-        }
-      cc += 1 + LINK_SIZE;
-      break;
-
-      CASE_ITERATOR_PRIVATE_DATA_1
-      if (PRIVATE_DATA(cc))
-        {
-        count = 1;
-        srcw[0] = PRIVATE_DATA(cc);
-        }
-      cc += 2;
+      }
+    cc += 2;
 #ifdef SUPPORT_UTF
-      if (common->utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
+    if (common->utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
 #endif
-      break;
+    break;
 
-      CASE_ITERATOR_PRIVATE_DATA_2A
-      if (PRIVATE_DATA(cc))
-        {
-        count = 2;
-        srcw[0] = PRIVATE_DATA(cc);
-        srcw[1] = PRIVATE_DATA(cc) + sizeof(sljit_sw);
-        }
-      cc += 2;
+    CASE_ITERATOR_PRIVATE_DATA_2A
+    if (PRIVATE_DATA(cc))
+      {
+      count = 2;
+      srcw[0] = PRIVATE_DATA(cc);
+      srcw[1] = PRIVATE_DATA(cc) + sizeof(sljit_sw);
+      }
+    cc += 2;
 #ifdef SUPPORT_UTF
-      if (common->utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
+    if (common->utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
 #endif
-      break;
+    break;
 
-      CASE_ITERATOR_PRIVATE_DATA_2B
-      if (PRIVATE_DATA(cc))
-        {
-        count = 2;
-        srcw[0] = PRIVATE_DATA(cc);
-        srcw[1] = PRIVATE_DATA(cc) + sizeof(sljit_sw);
-        }
-      cc += 2 + IMM2_SIZE;
+    CASE_ITERATOR_PRIVATE_DATA_2B
+    if (PRIVATE_DATA(cc))
+      {
+      count = 2;
+      srcw[0] = PRIVATE_DATA(cc);
+      srcw[1] = PRIVATE_DATA(cc) + sizeof(sljit_sw);
+      }
+    cc += 2 + IMM2_SIZE;
 #ifdef SUPPORT_UTF
-      if (common->utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
+    if (common->utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
 #endif
-      break;
+    break;
 
-      CASE_ITERATOR_TYPE_PRIVATE_DATA_1
-      if (PRIVATE_DATA(cc))
+    CASE_ITERATOR_TYPE_PRIVATE_DATA_1
+    if (PRIVATE_DATA(cc))
+      {
+      count = 1;
+      srcw[0] = PRIVATE_DATA(cc);
+      }
+    cc += 1;
+    break;
+
+    CASE_ITERATOR_TYPE_PRIVATE_DATA_2A
+    if (PRIVATE_DATA(cc))
+      {
+      count = 2;
+      srcw[0] = PRIVATE_DATA(cc);
+      srcw[1] = srcw[0] + sizeof(sljit_sw);
+      }
+    cc += 1;
+    break;
+
+    CASE_ITERATOR_TYPE_PRIVATE_DATA_2B
+    if (PRIVATE_DATA(cc))
+      {
+      count = 2;
+      srcw[0] = PRIVATE_DATA(cc);
+      srcw[1] = srcw[0] + sizeof(sljit_sw);
+      }
+    cc += 1 + IMM2_SIZE;
+    break;
+
+    case OP_CLASS:
+    case OP_NCLASS:
+#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+    case OP_XCLASS:
+    size = (*cc == OP_XCLASS) ? GET(cc, 1) : 1 + 32 / (int)sizeof(pcre_uchar);
+#else
+    size = 1 + 32 / (int)sizeof(pcre_uchar);
+#endif
+    if (PRIVATE_DATA(cc))
+      switch(get_class_iterator_size(cc + size))
         {
+        case 1:
         count = 1;
         srcw[0] = PRIVATE_DATA(cc);
-        }
-      cc += 1;
-      break;
+        break;
 
-      CASE_ITERATOR_TYPE_PRIVATE_DATA_2A
-      if (PRIVATE_DATA(cc))
-        {
+        case 2:
         count = 2;
         srcw[0] = PRIVATE_DATA(cc);
         srcw[1] = srcw[0] + sizeof(sljit_sw);
-        }
-      cc += 1;
-      break;
+        break;
 
-      CASE_ITERATOR_TYPE_PRIVATE_DATA_2B
-      if (PRIVATE_DATA(cc))
-        {
-        count = 2;
-        srcw[0] = PRIVATE_DATA(cc);
-        srcw[1] = srcw[0] + sizeof(sljit_sw);
+        default:
+        SLJIT_UNREACHABLE();
+        break;
         }
-      cc += 1 + IMM2_SIZE;
-      break;
-
-      case OP_CLASS:
-      case OP_NCLASS:
-#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
-      case OP_XCLASS:
-      size = (*cc == OP_XCLASS) ? GET(cc, 1) : 1 + 32 / (int)sizeof(pcre_uchar);
-#else
-      size = 1 + 32 / (int)sizeof(pcre_uchar);
-#endif
-      if (PRIVATE_DATA(cc))
-        switch(get_class_iterator_size(cc + size))
-          {
-          case 1:
-          count = 1;
-          srcw[0] = PRIVATE_DATA(cc);
-          break;
-
-          case 2:
-          count = 2;
-          srcw[0] = PRIVATE_DATA(cc);
-          srcw[1] = srcw[0] + sizeof(sljit_sw);
-          break;
-
-          default:
-          SLJIT_ASSERT_STOP();
-          break;
-          }
-      cc += size;
-      break;
-
-      default:
-      cc = next_opcode(common, cc);
-      SLJIT_ASSERT(cc != NULL);
-      break;
-      }
+    cc += size;
     break;
 
-    case end:
-    SLJIT_ASSERT_STOP();
+    default:
+    cc = next_opcode(common, cc);
+    SLJIT_ASSERT(cc != NULL);
     break;
     }
 
@@ -2312,7 +2303,7 @@ static SLJIT_INLINE void count_match(compiler_common *common)
 {
 DEFINE_COMPILER;
 
-OP2(SLJIT_SUB | SLJIT_SET_E, COUNT_MATCH, 0, COUNT_MATCH, 0, SLJIT_IMM, 1);
+OP2(SLJIT_SUB | SLJIT_SET_Z, COUNT_MATCH, 0, COUNT_MATCH, 0, SLJIT_IMM, 1);
 add_jump(compiler, &common->calllimit, JUMP(SLJIT_ZERO));
 }
 
@@ -2322,7 +2313,7 @@ static SLJIT_INLINE void allocate_stack(compiler_common *common, int size)
 DEFINE_COMPILER;
 
 SLJIT_ASSERT(size > 0);
-OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, size * sizeof(sljit_sw));
+OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, size * sizeof(sljit_sw));
 #ifdef DESTROY_REGISTERS
 OP1(SLJIT_MOV, TMP1, 0, SLJIT_IMM, 12345);
 OP1(SLJIT_MOV, TMP3, 0, TMP1, 0);
@@ -2330,7 +2321,7 @@ OP1(SLJIT_MOV, RETURN_ADDR, 0, TMP1, 0);
 OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), LOCALS0, TMP1, 0);
 OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), LOCALS1, TMP1, 0);
 #endif
-add_stub(common, CMP(SLJIT_GREATER, STACK_TOP, 0, STACK_LIMIT, 0));
+add_stub(common, CMP(SLJIT_LESS, STACK_TOP, 0, STACK_LIMIT, 0));
 }
 
 static SLJIT_INLINE void free_stack(compiler_common *common, int size)
@@ -2338,7 +2329,7 @@ static SLJIT_INLINE void free_stack(compiler_common *common, int size)
 DEFINE_COMPILER;
 
 SLJIT_ASSERT(size > 0);
-OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, size * sizeof(sljit_sw));
+OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, size * sizeof(sljit_sw));
 }
 
 static sljit_uw * allocate_read_only_data(compiler_common *common, sljit_uw size)
@@ -2396,7 +2387,7 @@ else
   OP1(SLJIT_MOV, SLJIT_R2, 0, SLJIT_IMM, length - 1);
   loop = LABEL();
   OP1(SLJIT_MOVU, SLJIT_MEM1(SLJIT_R1), sizeof(sljit_sw), SLJIT_R0, 0);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_R2, 0, SLJIT_R2, 0, SLJIT_IMM, 1);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_R2, 0, SLJIT_R2, 0, SLJIT_IMM, 1);
   JUMPTO(SLJIT_NOT_ZERO, loop);
   }
 }
@@ -2434,7 +2425,7 @@ else
   OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_IMM, length - 2);
   loop = LABEL();
   OP1(SLJIT_MOVU, SLJIT_MEM1(TMP2), sizeof(sljit_sw), TMP1, 0);
-  OP2(SLJIT_SUB | SLJIT_SET_E, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, 1);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, 1);
   JUMPTO(SLJIT_NOT_ZERO, loop);
   }
 
@@ -2452,22 +2443,22 @@ static sljit_sw SLJIT_CALL do_search_mark(sljit_sw *current, const pcre_uchar *s
 {
 while (current != NULL)
   {
-  switch (current[-2])
+  switch (current[1])
     {
     case type_then_trap:
     break;
 
     case type_mark:
-    if (STRCMP_UC_UC(skip_arg, (pcre_uchar *)current[-3]) == 0)
-      return current[-4];
+    if (STRCMP_UC_UC(skip_arg, (pcre_uchar *)current[2]) == 0)
+      return current[3];
     break;
 
     default:
-    SLJIT_ASSERT_STOP();
+    SLJIT_UNREACHABLE();
     break;
     }
-  SLJIT_ASSERT(current > (sljit_sw*)current[-1]);
-  current = (sljit_sw*)current[-1];
+  SLJIT_ASSERT(current[0] == 0 || current < (sljit_sw*)current[0]);
+  current = (sljit_sw*)current[0];
   }
 return -1;
 }
@@ -2501,7 +2492,7 @@ OP2(SLJIT_ADD, SLJIT_S0, 0, SLJIT_S0, 0, SLJIT_IMM, sizeof(sljit_sw));
 OP2(SLJIT_ASHR, SLJIT_S1, 0, SLJIT_S1, 0, SLJIT_IMM, UCHAR_SHIFT);
 #endif
 OP1(SLJIT_MOVU_S32, SLJIT_MEM1(SLJIT_R2), sizeof(int), SLJIT_S1, 0);
-OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_R1, 0, SLJIT_R1, 0, SLJIT_IMM, 1);
+OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_R1, 0, SLJIT_R1, 0, SLJIT_IMM, 1);
 JUMPTO(SLJIT_NOT_ZERO, loop);
 JUMPHERE(early_quit);
 
@@ -3106,8 +3097,8 @@ if (common->utf)
   OP2(SLJIT_SUB, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(1));
   /* Skip low surrogate if necessary. */
   OP2(SLJIT_AND, TMP1, 0, TMP1, 0, SLJIT_IMM, 0xfc00);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xdc00);
-  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xdc00);
+  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_EQUAL);
   OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
   OP2(SLJIT_SUB, STR_PTR, 0, STR_PTR, 0, TMP1, 0);
   return;
@@ -3126,6 +3117,7 @@ struct sljit_jump *jump;
 if (nltype == NLTYPE_ANY)
   {
   add_jump(compiler, &common->anynewline, JUMP(SLJIT_FAST_CALL));
+  sljit_set_current_flags(compiler, SLJIT_SET_Z);
   add_jump(compiler, backtracks, JUMP(jumpifmatch ? SLJIT_NOT_ZERO : SLJIT_ZERO));
   }
 else if (nltype == NLTYPE_ANYCRLF)
@@ -3167,7 +3159,7 @@ OP2(SLJIT_AND, TMP2, 0, TMP2, 0, SLJIT_IMM, 0x3f);
 OP2(SLJIT_OR, TMP1, 0, TMP1, 0, TMP2, 0);
 
 /* Searching for the first zero. */
-OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x800);
+OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x800);
 jump = JUMP(SLJIT_NOT_ZERO);
 /* Two byte sequence. */
 OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(1));
@@ -3181,7 +3173,7 @@ OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, 6);
 OP2(SLJIT_AND, TMP2, 0, TMP2, 0, SLJIT_IMM, 0x3f);
 OP2(SLJIT_OR, TMP1, 0, TMP1, 0, TMP2, 0);
 
-OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x10000);
+OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x10000);
 jump = JUMP(SLJIT_NOT_ZERO);
 /* Three byte sequence. */
 OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(2));
@@ -3215,15 +3207,15 @@ OP2(SLJIT_AND, TMP2, 0, TMP2, 0, SLJIT_IMM, 0x3f);
 OP2(SLJIT_OR, TMP1, 0, TMP1, 0, TMP2, 0);
 
 /* Searching for the first zero. */
-OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x800);
+OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x800);
 jump = JUMP(SLJIT_NOT_ZERO);
 /* Two byte sequence. */
 OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(1));
 sljit_emit_fast_return(compiler, RETURN_ADDR, 0);
 
 JUMPHERE(jump);
-OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x400);
-OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_NOT_ZERO);
+OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x400);
+OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_NOT_ZERO);
 /* This code runs only in 8 bit mode. No need to shift the value. */
 OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, TMP2, 0);
 OP1(MOV_UCHAR, TMP2, 0, SLJIT_MEM1(STR_PTR), IN_UCHARS(1));
@@ -3246,7 +3238,7 @@ struct sljit_jump *compare;
 
 sljit_emit_fast_enter(compiler, RETURN_ADDR, 0);
 
-OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, 0x20);
+OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, 0x20);
 jump = JUMP(SLJIT_NOT_ZERO);
 /* Two byte sequence. */
 OP1(MOV_UCHAR, TMP1, 0, SLJIT_MEM1(STR_PTR), IN_UCHARS(0));
@@ -3287,10 +3279,30 @@ static void do_getucd(compiler_common *common)
 /* Search the UCD record for the character comes in TMP1.
 Returns chartype in TMP1 and UCD offset in TMP2. */
 DEFINE_COMPILER;
+#ifdef COMPILE_PCRE32
+struct sljit_jump *jump;
+#endif
+
+#if defined SLJIT_DEBUG && SLJIT_DEBUG
+/* dummy_ucd_record */
+const ucd_record *record = GET_UCD(INVALID_UTF_CHAR);
+SLJIT_ASSERT(record->script == ucp_Common && record->chartype == ucp_Cn && record->gbprop == ucp_gbOther);
+SLJIT_ASSERT(record->caseset == 0 && record->other_case == 0);
+#endif
 
 SLJIT_ASSERT(UCD_BLOCK_SIZE == 128 && sizeof(ucd_record) == 8);
 
 sljit_emit_fast_enter(compiler, RETURN_ADDR, 0);
+
+#ifdef COMPILE_PCRE32
+if (!common->utf)
+  {
+  jump = CMP(SLJIT_LESS, TMP1, 0, SLJIT_IMM, 0x10ffff + 1);
+  OP1(SLJIT_MOV, TMP1, 0, SLJIT_IMM, INVALID_UTF_CHAR);
+  JUMPHERE(jump);
+  }
+#endif
+
 OP2(SLJIT_LSHR, TMP2, 0, TMP1, 0, SLJIT_IMM, UCD_BLOCK_SHIFT);
 OP1(SLJIT_MOV_U8, TMP2, 0, SLJIT_MEM1(TMP2), (sljit_sw)PRIV(ucd_stage1));
 OP2(SLJIT_AND, TMP1, 0, TMP1, 0, SLJIT_IMM, UCD_BLOCK_MASK);
@@ -3365,8 +3377,8 @@ if (newlinecheck)
   OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(1));
   end = CMP(SLJIT_GREATER_EQUAL, STR_PTR, 0, STR_END, 0);
   OP1(MOV_UCHAR, TMP1, 0, SLJIT_MEM1(STR_PTR), 0);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, common->newline & 0xff);
-  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, common->newline & 0xff);
+  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_EQUAL);
 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
   OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, UCHAR_SHIFT);
 #endif
@@ -3403,8 +3415,8 @@ if (common->utf)
   {
   singlechar = CMP(SLJIT_LESS, TMP1, 0, SLJIT_IMM, 0xd800);
   OP2(SLJIT_AND, TMP1, 0, TMP1, 0, SLJIT_IMM, 0xfc00);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd800);
-  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd800);
+  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_EQUAL);
   OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
   OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, TMP1, 0);
   JUMPHERE(singlechar);
@@ -3853,7 +3865,7 @@ while (TRUE)
   }
 }
 
-#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)
+#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86) && !(defined SUPPORT_VALGRIND)
 
 static sljit_s32 character_to_int32(pcre_uchar chr)
 {
@@ -4019,6 +4031,7 @@ instruction[0] = 0x0f;
 instruction[1] = 0xbc;
 instruction[2] = 0xc0 | (tmp1_ind << 3) | tmp1_ind;
 sljit_emit_op_custom(compiler, instruction, 3);
+sljit_set_current_flags(compiler, SLJIT_SET_Z);
 
 nomatch = JUMP(SLJIT_ZERO);
 
@@ -4119,6 +4132,7 @@ instruction[0] = 0x0f;
 instruction[1] = 0xbc;
 instruction[2] = 0xc0 | (tmp1_ind << 3) | tmp1_ind;
 sljit_emit_op_custom(compiler, instruction, 3);
+sljit_set_current_flags(compiler, SLJIT_SET_Z);
 
 JUMPTO(SLJIT_ZERO, start);
 
@@ -4155,18 +4169,8 @@ if (has_match_end)
   OP1(SLJIT_MOV, TMP3, 0, STR_END, 0);
 
   OP2(SLJIT_ADD, STR_END, 0, SLJIT_MEM1(SLJIT_SP), common->match_end_ptr, SLJIT_IMM, IN_UCHARS(offset + 1));
-#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)
-  if (sljit_x86_is_cmov_available())
-    {
-    OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, STR_END, 0, TMP3, 0);
-    sljit_x86_emit_cmov(compiler, SLJIT_GREATER, STR_END, TMP3, 0);
-    }
-#endif
-    {
-    quit = CMP(SLJIT_LESS_EQUAL, STR_END, 0, TMP3, 0);
-    OP1(SLJIT_MOV, STR_END, 0, TMP3, 0);
-    JUMPHERE(quit);
-    }
+  OP2(SLJIT_SUB | SLJIT_SET_GREATER, SLJIT_UNUSED, 0, STR_END, 0, TMP3, 0);
+  sljit_emit_cmov(compiler, SLJIT_GREATER, STR_END, TMP3, 0);
   }
 
 #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
@@ -4174,11 +4178,11 @@ if (common->utf && offset > 0)
   utf_start = LABEL();
 #endif
 
-#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)
+#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86) && !(defined SUPPORT_VALGRIND)
 
 /* SSE2 accelerated first character search. */
 
-if (sljit_x86_is_sse2_available())
+if (sljit_has_cpu_feature(SLJIT_HAS_SSE2))
   {
   fast_forward_first_char2_sse2(common, char1, char2);
 
@@ -4213,16 +4217,16 @@ if (sljit_x86_is_sse2_available())
     if (offset > 0)
       OP2(SLJIT_SUB, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(offset));
     }
-  else if (sljit_x86_is_cmov_available())
-    {
-    OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, STR_PTR, 0, STR_END, 0);
-    sljit_x86_emit_cmov(compiler, SLJIT_GREATER_EQUAL, STR_PTR, has_match_end ? SLJIT_MEM1(SLJIT_SP) : STR_END, has_match_end ? common->match_end_ptr : 0);
-    }
   else
     {
-    quit = CMP(SLJIT_LESS, STR_PTR, 0, STR_END, 0);
-    OP1(SLJIT_MOV, STR_PTR, 0, has_match_end ? SLJIT_MEM1(SLJIT_SP) : STR_END, has_match_end ? common->match_end_ptr : 0);
-    JUMPHERE(quit);
+    OP2(SLJIT_SUB | SLJIT_SET_GREATER_EQUAL, SLJIT_UNUSED, 0, STR_PTR, 0, STR_END, 0);
+    if (has_match_end)
+      {
+      OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), common->match_end_ptr);
+      sljit_emit_cmov(compiler, SLJIT_GREATER_EQUAL, STR_PTR, TMP1, 0);
+      }
+    else
+      sljit_emit_cmov(compiler, SLJIT_GREATER_EQUAL, STR_PTR, STR_END, 0);
     }
 
   if (has_match_end)
@@ -4249,10 +4253,10 @@ else
     }
   else
     {
-    OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, char1);
-    OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
-    OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, char2);
-    OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, char1);
+    OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, char2);
+    OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_EQUAL);
     found = JUMP(SLJIT_NOT_ZERO);
     }
   }
@@ -4571,8 +4575,8 @@ if (common->nltype == NLTYPE_FIXED && common->newline > 255)
   firstchar = CMP(SLJIT_LESS_EQUAL, STR_PTR, 0, TMP2, 0);
 
   OP2(SLJIT_ADD, TMP1, 0, TMP1, 0, SLJIT_IMM, IN_UCHARS(2));
-  OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, STR_PTR, 0, TMP1, 0);
-  OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_GREATER_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_GREATER_EQUAL, SLJIT_UNUSED, 0, STR_PTR, 0, TMP1, 0);
+  OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_GREATER_EQUAL);
 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
   OP2(SLJIT_SHL, TMP2, 0, TMP2, 0, SLJIT_IMM, UCHAR_SHIFT);
 #endif
@@ -4616,8 +4620,8 @@ if (common->nltype == NLTYPE_ANY || common->nltype == NLTYPE_ANYCRLF)
   JUMPHERE(foundcr);
   notfoundnl = CMP(SLJIT_GREATER_EQUAL, STR_PTR, 0, STR_END, 0);
   OP1(MOV_UCHAR, TMP1, 0, SLJIT_MEM1(STR_PTR), 0);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, CHAR_NL);
-  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, CHAR_NL);
+  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_EQUAL);
 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
   OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, UCHAR_SHIFT);
 #endif
@@ -4670,7 +4674,7 @@ if (!check_class_ranges(common, start_bits, (start_bits[31] & 0x80) != 0, TRUE,
   OP2(SLJIT_LSHR, TMP1, 0, TMP1, 0, SLJIT_IMM, 3);
   OP1(SLJIT_MOV_U8, TMP1, 0, SLJIT_MEM1(TMP1), (sljit_sw)start_bits);
   OP2(SLJIT_SHL, TMP2, 0, SLJIT_IMM, 1, TMP2, 0);
-  OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
+  OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
   found = JUMP(SLJIT_NOT_ZERO);
   }
 
@@ -4692,8 +4696,8 @@ if (common->utf)
   {
   CMPTO(SLJIT_LESS, TMP1, 0, SLJIT_IMM, 0xd800, start);
   OP2(SLJIT_AND, TMP1, 0, TMP1, 0, SLJIT_IMM, 0xfc00);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd800);
-  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd800);
+  OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_EQUAL);
   OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
   OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, TMP1, 0);
   }
@@ -4780,31 +4784,31 @@ struct sljit_jump *jump;
 struct sljit_label *mainloop;
 
 sljit_emit_fast_enter(compiler, RETURN_ADDR, 0);
-OP1(SLJIT_MOV, TMP1, 0, STACK_TOP, 0);
-GET_LOCAL_BASE(TMP3, 0, 0);
+OP1(SLJIT_MOV, TMP3, 0, STACK_TOP, 0);
+GET_LOCAL_BASE(TMP1, 0, 0);
 
 /* Drop frames until we reach STACK_TOP. */
 mainloop = LABEL();
-OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(TMP1), 0);
-OP2(SLJIT_SUB | SLJIT_SET_S, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, 0);
-jump = JUMP(SLJIT_SIG_LESS_EQUAL);
-
-OP2(SLJIT_ADD, TMP2, 0, TMP2, 0, TMP3, 0);
-OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), 0, SLJIT_MEM1(TMP1), sizeof(sljit_sw));
-OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), sizeof(sljit_sw), SLJIT_MEM1(TMP1), 2 * sizeof(sljit_sw));
-OP2(SLJIT_ADD, TMP1, 0, TMP1, 0, SLJIT_IMM, 3 * sizeof(sljit_sw));
+OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), -sizeof(sljit_sw));
+jump = CMP(SLJIT_SIG_LESS_EQUAL, TMP2, 0, SLJIT_IMM, 0);
+
+OP2(SLJIT_ADD, TMP2, 0, TMP2, 0, TMP1, 0);
+OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), 0, SLJIT_MEM1(STACK_TOP), -2 * sizeof(sljit_sw));
+OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), sizeof(sljit_sw), SLJIT_MEM1(STACK_TOP), -3 * sizeof(sljit_sw));
+OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, 3 * sizeof(sljit_sw));
 JUMPTO(SLJIT_JUMP, mainloop);
 
 JUMPHERE(jump);
-jump = JUMP(SLJIT_SIG_LESS);
-/* End of dropping frames. */
+jump = CMP(SLJIT_NOT_ZERO /* SIG_LESS */, TMP2, 0, SLJIT_IMM, 0);
+/* End of reverting values. */
+OP1(SLJIT_MOV, STACK_TOP, 0, TMP3, 0);
 sljit_emit_fast_return(compiler, RETURN_ADDR, 0);
 
 JUMPHERE(jump);
 OP1(SLJIT_NEG, TMP2, 0, TMP2, 0);
-OP2(SLJIT_ADD, TMP2, 0, TMP2, 0, TMP3, 0);
-OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), 0, SLJIT_MEM1(TMP1), sizeof(sljit_sw));
-OP2(SLJIT_ADD, TMP1, 0, TMP1, 0, SLJIT_IMM, 2 * sizeof(sljit_sw));
+OP2(SLJIT_ADD, TMP2, 0, TMP2, 0, TMP1, 0);
+OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), 0, SLJIT_MEM1(STACK_TOP), -2 * sizeof(sljit_sw));
+OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, 2 * sizeof(sljit_sw));
 JUMPTO(SLJIT_JUMP, mainloop);
 }
 
@@ -4837,11 +4841,11 @@ if (common->use_ucp)
   jump = CMP(SLJIT_EQUAL, TMP1, 0, SLJIT_IMM, CHAR_UNDERSCORE);
   add_jump(compiler, &common->getucd, JUMP(SLJIT_FAST_CALL));
   OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, ucp_Ll);
-  OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_Lu - ucp_Ll);
-  OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_Lu - ucp_Ll);
+  OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
   OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, ucp_Nd - ucp_Ll);
-  OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_No - ucp_Nd);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_No - ucp_Nd);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_LESS_EQUAL);
   JUMPHERE(jump);
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), LOCALS1, TMP2, 0);
   }
@@ -4881,11 +4885,11 @@ if (common->use_ucp)
   jump = CMP(SLJIT_EQUAL, TMP1, 0, SLJIT_IMM, CHAR_UNDERSCORE);
   add_jump(compiler, &common->getucd, JUMP(SLJIT_FAST_CALL));
   OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, ucp_Ll);
-  OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_Lu - ucp_Ll);
-  OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_Lu - ucp_Ll);
+  OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
   OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, ucp_Nd - ucp_Ll);
-  OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_No - ucp_Nd);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ucp_No - ucp_Nd);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_LESS_EQUAL);
   JUMPHERE(jump);
   }
 else
@@ -4913,7 +4917,7 @@ else
   }
 set_jumps(skipread_list, LABEL());
 
-OP2(SLJIT_XOR | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_MEM1(SLJIT_SP), LOCALS1);
+OP2(SLJIT_XOR | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_MEM1(SLJIT_SP), LOCALS1);
 sljit_emit_fast_return(compiler, SLJIT_MEM1(SLJIT_SP), LOCALS0);
 }
 
@@ -5064,7 +5068,7 @@ switch(length)
   return TRUE;
 
   default:
-  SLJIT_ASSERT_STOP();
+  SLJIT_UNREACHABLE();
   return FALSE;
   }
 }
@@ -5077,22 +5081,22 @@ DEFINE_COMPILER;
 sljit_emit_fast_enter(compiler, RETURN_ADDR, 0);
 
 OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, 0x0a);
-OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x0d - 0x0a);
-OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
-OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x85 - 0x0a);
+OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x0d - 0x0a);
+OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
+OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x85 - 0x0a);
 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
 #ifdef COMPILE_PCRE8
 if (common->utf)
   {
 #endif
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
   OP2(SLJIT_OR, TMP1, 0, TMP1, 0, SLJIT_IMM, 0x1);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2029 - 0x0a);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2029 - 0x0a);
 #ifdef COMPILE_PCRE8
   }
 #endif
 #endif /* SUPPORT_UTF || COMPILE_PCRE16 || COMPILE_PCRE32 */
-OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_EQUAL);
 sljit_emit_fast_return(compiler, RETURN_ADDR, 0);
 }
 
@@ -5103,34 +5107,34 @@ DEFINE_COMPILER;
 
 sljit_emit_fast_enter(compiler, RETURN_ADDR, 0);
 
-OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x09);
-OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
-OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x20);
-OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xa0);
+OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x09);
+OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
+OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x20);
+OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xa0);
 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
 #ifdef COMPILE_PCRE8
 if (common->utf)
   {
 #endif
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x1680);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x180e);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x1680);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x180e);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
   OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, 0x2000);
-  OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x200A - 0x2000);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x202f - 0x2000);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x205f - 0x2000);
-  OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x3000 - 0x2000);
+  OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x200A - 0x2000);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_LESS_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x202f - 0x2000);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x205f - 0x2000);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x3000 - 0x2000);
 #ifdef COMPILE_PCRE8
   }
 #endif
 #endif /* SUPPORT_UTF || COMPILE_PCRE16 || COMPILE_PCRE32 */
-OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_EQUAL);
 
 sljit_emit_fast_return(compiler, RETURN_ADDR, 0);
 }
@@ -5143,22 +5147,22 @@ DEFINE_COMPILER;
 sljit_emit_fast_enter(compiler, RETURN_ADDR, 0);
 
 OP2(SLJIT_SUB, TMP1, 0, TMP1, 0, SLJIT_IMM, 0x0a);
-OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x0d - 0x0a);
-OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
-OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x85 - 0x0a);
+OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x0d - 0x0a);
+OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
+OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x85 - 0x0a);
 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
 #ifdef COMPILE_PCRE8
 if (common->utf)
   {
 #endif
-  OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+  OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
   OP2(SLJIT_OR, TMP1, 0, TMP1, 0, SLJIT_IMM, 0x1);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2029 - 0x0a);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2029 - 0x0a);
 #ifdef COMPILE_PCRE8
   }
 #endif
 #endif /* SUPPORT_UTF || COMPILE_PCRE16 || COMPILE_PCRE32 */
-OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_EQUAL);
 
 sljit_emit_fast_return(compiler, RETURN_ADDR, 0);
 }
@@ -5183,7 +5187,7 @@ label = LABEL();
 OP1(MOVU_UCHAR, CHAR1, 0, SLJIT_MEM1(TMP1), IN_UCHARS(1));
 OP1(MOVU_UCHAR, CHAR2, 0, SLJIT_MEM1(STR_PTR), IN_UCHARS(1));
 jump = CMP(SLJIT_NOT_EQUAL, CHAR1, 0, CHAR2, 0);
-OP2(SLJIT_SUB | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_IMM, IN_UCHARS(1));
+OP2(SLJIT_SUB | SLJIT_SET_Z, TMP2, 0, TMP2, 0, SLJIT_IMM, IN_UCHARS(1));
 JUMPTO(SLJIT_NOT_ZERO, label);
 
 JUMPHERE(jump);
@@ -5227,7 +5231,7 @@ OP1(SLJIT_MOV_U8, CHAR2, 0, SLJIT_MEM2(LCC_TABLE, CHAR2), 0);
 JUMPHERE(jump);
 #endif
 jump = CMP(SLJIT_NOT_EQUAL, CHAR1, 0, CHAR2, 0);
-OP2(SLJIT_SUB | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_IMM, IN_UCHARS(1));
+OP2(SLJIT_SUB | SLJIT_SET_Z, TMP2, 0, TMP2, 0, SLJIT_IMM, IN_UCHARS(1));
 JUMPTO(SLJIT_NOT_ZERO, label);
 
 JUMPHERE(jump);
@@ -5394,7 +5398,7 @@ do
 #endif
 
       default:
-      SLJIT_ASSERT_STOP();
+      SLJIT_UNREACHABLE();
       break;
       }
     context->ucharptr = 0;
@@ -5568,7 +5572,7 @@ while (*cc != XCL_END)
       break;
 
       default:
-      SLJIT_ASSERT_STOP();
+      SLJIT_UNREACHABLE();
       break;
       }
     cc += 2;
@@ -5592,7 +5596,7 @@ if ((cc[-1] & XCL_HASPROP) == 0)
       OP2(SLJIT_LSHR, TMP1, 0, TMP1, 0, SLJIT_IMM, 3);
       OP1(SLJIT_MOV_U8, TMP1, 0, SLJIT_MEM1(TMP1), (sljit_sw)cc);
       OP2(SLJIT_SHL, TMP2, 0, SLJIT_IMM, 1, TMP2, 0);
-      OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
+      OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
       add_jump(compiler, &found, JUMP(SLJIT_NOT_ZERO));
       }
 
@@ -5625,7 +5629,7 @@ else if ((cc[-1] & XCL_MAP) != 0)
     OP2(SLJIT_LSHR, TMP1, 0, TMP1, 0, SLJIT_IMM, 3);
     OP1(SLJIT_MOV_U8, TMP1, 0, SLJIT_MEM1(TMP1), (sljit_sw)cc);
     OP2(SLJIT_SHL, TMP2, 0, SLJIT_IMM, 1, TMP2, 0);
-    OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
+    OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
     add_jump(compiler, list, JUMP(SLJIT_NOT_ZERO));
 
 #ifdef COMPILE_PCRE8
@@ -5644,6 +5648,15 @@ if (needstype || needsscript)
   if (needschar && !charsaved)
     OP1(SLJIT_MOV, RETURN_ADDR, 0, TMP1, 0);
 
+#ifdef COMPILE_PCRE32
+  if (!common->utf)
+    {
+    jump = CMP(SLJIT_LESS, TMP1, 0, SLJIT_IMM, 0x10ffff + 1);
+    OP1(SLJIT_MOV, TMP1, 0, SLJIT_IMM, INVALID_UTF_CHAR);
+    JUMPHERE(jump);
+    }
+#endif
+
   OP2(SLJIT_LSHR, TMP2, 0, TMP1, 0, SLJIT_IMM, UCD_BLOCK_SHIFT);
   OP1(SLJIT_MOV_U8, TMP2, 0, SLJIT_MEM1(TMP2), (sljit_sw)PRIV(ucd_stage1));
   OP2(SLJIT_AND, TMP1, 0, TMP1, 0, SLJIT_IMM, UCD_BLOCK_MASK);
@@ -5735,14 +5748,14 @@ while (*cc != XCL_END)
 
     if (numberofcmps < 3 && (*cc == XCL_SINGLE || *cc == XCL_RANGE))
       {
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
-      OP_FLAGS(numberofcmps == 0 ? SLJIT_MOV : SLJIT_OR, TMP2, 0, numberofcmps == 0 ? SLJIT_UNUSED : TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
+      OP_FLAGS(numberofcmps == 0 ? SLJIT_MOV : SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
       numberofcmps++;
       }
     else if (numberofcmps > 0)
       {
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       numberofcmps = 0;
       }
@@ -5761,14 +5774,14 @@ while (*cc != XCL_END)
 
     if (numberofcmps < 3 && (*cc == XCL_SINGLE || *cc == XCL_RANGE))
       {
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
-      OP_FLAGS(numberofcmps == 0 ? SLJIT_MOV : SLJIT_OR, TMP2, 0, numberofcmps == 0 ? SLJIT_UNUSED : TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
+      OP_FLAGS(numberofcmps == 0 ? SLJIT_MOV : SLJIT_OR, TMP2, 0, SLJIT_LESS_EQUAL);
       numberofcmps++;
       }
     else if (numberofcmps > 0)
       {
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(c - charoffset));
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_LESS_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       numberofcmps = 0;
       }
@@ -5793,12 +5806,12 @@ while (*cc != XCL_END)
       break;
 
       case PT_LAMP:
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Lu - typeoffset);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Ll - typeoffset);
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Lt - typeoffset);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Lu - typeoffset);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Ll - typeoffset);
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Lt - typeoffset);
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       break;
 
@@ -5820,33 +5833,33 @@ while (*cc != XCL_END)
       case PT_SPACE:
       case PT_PXSPACE:
       SET_CHAR_OFFSET(9);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd - 0x9);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd - 0x9);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
 
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x85 - 0x9);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x85 - 0x9);
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
 
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x180e - 0x9);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x180e - 0x9);
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
 
       SET_TYPE_OFFSET(ucp_Zl);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Zs - ucp_Zl);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Zs - ucp_Zl);
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_LESS_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       break;
 
       case PT_WORD:
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_UNDERSCORE - charoffset));
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_UNDERSCORE - charoffset));
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
       /* Fall through. */
 
       case PT_ALNUM:
       SET_TYPE_OFFSET(ucp_Ll);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Lu - ucp_Ll);
-      OP_FLAGS((*cc == PT_ALNUM) ? SLJIT_MOV : SLJIT_OR, TMP2, 0, (*cc == PT_ALNUM) ? SLJIT_UNUSED : TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Lu - ucp_Ll);
+      OP_FLAGS((*cc == PT_ALNUM) ? SLJIT_MOV : SLJIT_OR, TMP2, 0, SLJIT_LESS_EQUAL);
       SET_TYPE_OFFSET(ucp_Nd);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_No - ucp_Nd);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_No - ucp_Nd);
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_LESS_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       break;
 
@@ -5868,8 +5881,8 @@ while (*cc != XCL_END)
           OP2(SLJIT_ADD, TMP2, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)charoffset);
           OP2(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_IMM, other_cases[1] ^ other_cases[0]);
           }
-        OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, other_cases[1]);
-        OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, other_cases[1]);
+        OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
         other_cases += 2;
         }
       else if (is_powerof2(other_cases[2] ^ other_cases[1]))
@@ -5881,63 +5894,63 @@ while (*cc != XCL_END)
           OP2(SLJIT_ADD, TMP2, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)charoffset);
           OP2(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_IMM, other_cases[1] ^ other_cases[0]);
           }
-        OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, other_cases[2]);
-        OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP2, 0, SLJIT_IMM, other_cases[2]);
+        OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
 
-        OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(other_cases[0] - charoffset));
-        OP_FLAGS(SLJIT_OR | ((other_cases[3] == NOTACHAR) ? SLJIT_SET_E : 0), TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(other_cases[0] - charoffset));
+        OP_FLAGS(SLJIT_OR | ((other_cases[3] == NOTACHAR) ? SLJIT_SET_Z : 0), TMP2, 0, SLJIT_EQUAL);
 
         other_cases += 3;
         }
       else
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(*other_cases++ - charoffset));
-        OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(*other_cases++ - charoffset));
+        OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
         }
 
       while (*other_cases != NOTACHAR)
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(*other_cases++ - charoffset));
-        OP_FLAGS(SLJIT_OR | ((*other_cases == NOTACHAR) ? SLJIT_SET_E : 0), TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(*other_cases++ - charoffset));
+        OP_FLAGS(SLJIT_OR | ((*other_cases == NOTACHAR) ? SLJIT_SET_Z : 0), TMP2, 0, SLJIT_EQUAL);
         }
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       break;
 
       case PT_UCNC:
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_DOLLAR_SIGN - charoffset));
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_COMMERCIAL_AT - charoffset));
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_GRAVE_ACCENT - charoffset));
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_DOLLAR_SIGN - charoffset));
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_COMMERCIAL_AT - charoffset));
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(CHAR_GRAVE_ACCENT - charoffset));
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
 
       SET_CHAR_OFFSET(0xa0);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(0xd7ff - charoffset));
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (sljit_sw)(0xd7ff - charoffset));
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_LESS_EQUAL);
       SET_CHAR_OFFSET(0);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xe000 - 0);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_GREATER_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_GREATER_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xe000 - 0);
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_GREATER_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       break;
 
       case PT_PXGRAPH:
       /* C and Z groups are the farthest two groups. */
       SET_TYPE_OFFSET(ucp_Ll);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_So - ucp_Ll);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_GREATER);
+      OP2(SLJIT_SUB | SLJIT_SET_GREATER, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_So - ucp_Ll);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_GREATER);
 
       jump = CMP(SLJIT_NOT_EQUAL, typereg, 0, SLJIT_IMM, ucp_Cf - ucp_Ll);
 
       /* In case of ucp_Cf, we overwrite the result. */
       SET_CHAR_OFFSET(0x2066);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2069 - 0x2066);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2069 - 0x2066);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
 
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x061c - 0x2066);
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x061c - 0x2066);
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
 
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x180e - 0x2066);
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x180e - 0x2066);
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
 
       JUMPHERE(jump);
       jump = CMP(SLJIT_ZERO ^ invertcmp, TMP2, 0, SLJIT_IMM, 0);
@@ -5946,21 +5959,21 @@ while (*cc != XCL_END)
       case PT_PXPRINT:
       /* C and Z groups are the farthest two groups. */
       SET_TYPE_OFFSET(ucp_Ll);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_So - ucp_Ll);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_GREATER);
+      OP2(SLJIT_SUB | SLJIT_SET_GREATER, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_So - ucp_Ll);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_GREATER);
 
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Zs - ucp_Ll);
-      OP_FLAGS(SLJIT_AND, TMP2, 0, TMP2, 0, SLJIT_NOT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Zs - ucp_Ll);
+      OP_FLAGS(SLJIT_AND, TMP2, 0, SLJIT_NOT_EQUAL);
 
       jump = CMP(SLJIT_NOT_EQUAL, typereg, 0, SLJIT_IMM, ucp_Cf - ucp_Ll);
 
       /* In case of ucp_Cf, we overwrite the result. */
       SET_CHAR_OFFSET(0x2066);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2069 - 0x2066);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x2069 - 0x2066);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
 
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x061c - 0x2066);
-      OP_FLAGS(SLJIT_OR, TMP2, 0, TMP2, 0, SLJIT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x061c - 0x2066);
+      OP_FLAGS(SLJIT_OR, TMP2, 0, SLJIT_EQUAL);
 
       JUMPHERE(jump);
       jump = CMP(SLJIT_ZERO ^ invertcmp, TMP2, 0, SLJIT_IMM, 0);
@@ -5968,21 +5981,21 @@ while (*cc != XCL_END)
 
       case PT_PXPUNCT:
       SET_TYPE_OFFSET(ucp_Sc);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_So - ucp_Sc);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_So - ucp_Sc);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS_EQUAL);
 
       SET_CHAR_OFFSET(0);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x7f);
-      OP_FLAGS(SLJIT_AND, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0x7f);
+      OP_FLAGS(SLJIT_AND, TMP2, 0, SLJIT_LESS_EQUAL);
 
       SET_TYPE_OFFSET(ucp_Pc);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Ps - ucp_Pc);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_LESS_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS_EQUAL, SLJIT_UNUSED, 0, typereg, 0, SLJIT_IMM, ucp_Ps - ucp_Pc);
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_LESS_EQUAL);
       jump = JUMP(SLJIT_NOT_ZERO ^ invertcmp);
       break;
 
       default:
-      SLJIT_ASSERT_STOP();
+      SLJIT_UNREACHABLE();
       break;
       }
     cc += 2;
@@ -6028,6 +6041,7 @@ switch(type)
   case OP_NOT_WORD_BOUNDARY:
   case OP_WORD_BOUNDARY:
   add_jump(compiler, &common->wordboundary, JUMP(SLJIT_FAST_CALL));
+  sljit_set_current_flags(compiler, SLJIT_SET_Z);
   add_jump(compiler, backtracks, JUMP(type == OP_NOT_WORD_BOUNDARY ? SLJIT_NOT_ZERO : SLJIT_ZERO));
   return cc;
 
@@ -6043,10 +6057,10 @@ switch(type)
     else
       {
       jump[1] = CMP(SLJIT_EQUAL, TMP2, 0, STR_END, 0);
-      OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP2, 0, STR_END, 0);
-      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_UNUSED, 0, SLJIT_LESS);
-      OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (common->newline >> 8) & 0xff);
-      OP_FLAGS(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_NOT_EQUAL);
+      OP2(SLJIT_SUB | SLJIT_SET_LESS, SLJIT_UNUSED, 0, TMP2, 0, STR_END, 0);
+      OP_FLAGS(SLJIT_MOV, TMP2, 0, SLJIT_LESS);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, (common->newline >> 8) & 0xff);
+      OP_FLAGS(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, SLJIT_NOT_EQUAL);
       add_jump(compiler, backtracks, JUMP(SLJIT_NOT_EQUAL));
       check_partial(common, TRUE);
       add_jump(compiler, backtracks, JUMP(SLJIT_JUMP));
@@ -6068,9 +6082,9 @@ switch(type)
     OP1(MOV_UCHAR, TMP1, 0, SLJIT_MEM1(STR_PTR), IN_UCHARS(0));
     jump[1] = CMP(SLJIT_NOT_EQUAL, TMP1, 0, SLJIT_IMM, CHAR_CR);
     OP2(SLJIT_ADD, TMP2, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(2));
-    OP2(SLJIT_SUB | SLJIT_SET_U, SLJIT_UNUSED, 0, TMP2, 0, STR_END, 0);
+    OP2(SLJIT_SUB | SLJIT_SET_Z | SLJIT_SET_GREATER, SLJIT_UNUSED, 0, TMP2, 0, STR_END, 0);
     jump[2] = JUMP(SLJIT_GREATER);
-    add_jump(compiler, backtracks, JUMP(SLJIT_LESS));
+    add_jump(compiler, backtracks, JUMP(SLJIT_NOT_EQUAL) /* LESS */);
     /* Equal. */
     OP1(MOV_UCHAR, TMP1, 0, SLJIT_MEM1(STR_PTR), IN_UCHARS(1));
     jump[3] = CMP(SLJIT_EQUAL, TMP1, 0, SLJIT_IMM, CHAR_NL);
@@ -6089,6 +6103,7 @@ switch(type)
       read_char_range(common, common->nlmin, common->nlmax, TRUE);
       add_jump(compiler, backtracks, CMP(SLJIT_NOT_EQUAL, STR_PTR, 0, STR_END, 0));
       add_jump(compiler, &common->anynewline, JUMP(SLJIT_FAST_CALL));
+      sljit_set_current_flags(compiler, SLJIT_SET_Z);
       add_jump(compiler, backtracks, JUMP(SLJIT_ZERO));
       OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(SLJIT_SP), LOCALS1);
       }
@@ -6204,7 +6219,7 @@ switch(type)
     label = LABEL();
     add_jump(compiler, backtracks, CMP(SLJIT_LESS_EQUAL, STR_PTR, 0, TMP3, 0));
     skip_char_back(common);
-    OP2(SLJIT_SUB | SLJIT_SET_E, TMP2, 0, TMP2, 0, SLJIT_IMM, 1);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, TMP2, 0, TMP2, 0, SLJIT_IMM, 1);
     JUMPTO(SLJIT_NOT_ZERO, label);
     }
   else
@@ -6217,7 +6232,7 @@ switch(type)
   check_start_used_ptr(common);
   return cc + LINK_SIZE;
   }
-SLJIT_ASSERT_STOP();
+SLJIT_UNREACHABLE();
 return cc;
 }
 
@@ -6250,7 +6265,7 @@ switch(type)
 #endif
     read_char8_type(common, type == OP_NOT_DIGIT);
     /* Flip the starting bit in the negative case. */
-  OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ctype_digit);
+  OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ctype_digit);
   add_jump(compiler, backtracks, JUMP(type == OP_DIGIT ? SLJIT_ZERO : SLJIT_NOT_ZERO));
   return cc;
 
@@ -6264,7 +6279,7 @@ switch(type)
   else
 #endif
     read_char8_type(common, type == OP_NOT_WHITESPACE);
-  OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ctype_space);
+  OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ctype_space);
   add_jump(compiler, backtracks, JUMP(type == OP_WHITESPACE ? SLJIT_ZERO : SLJIT_NOT_ZERO));
   return cc;
 
@@ -6278,7 +6293,7 @@ switch(type)
   else
 #endif
     read_char8_type(common, type == OP_NOT_WORDCHAR);
-  OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ctype_word);
+  OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, ctype_word);
   add_jump(compiler, backtracks, JUMP(type == OP_WORDCHAR ? SLJIT_ZERO : SLJIT_NOT_ZERO));
   return cc;
 
@@ -6320,8 +6335,8 @@ switch(type)
 #elif defined COMPILE_PCRE16
     jump[0] = CMP(SLJIT_LESS, TMP1, 0, SLJIT_IMM, 0xd800);
     OP2(SLJIT_AND, TMP1, 0, TMP1, 0, SLJIT_IMM, 0xfc00);
-    OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd800);
-    OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_UNUSED, 0, SLJIT_EQUAL);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, SLJIT_IMM, 0xd800);
+    OP_FLAGS(SLJIT_MOV, TMP1, 0, SLJIT_EQUAL);
     OP2(SLJIT_SHL, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
     OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, TMP1, 0);
 #endif
@@ -6383,6 +6398,7 @@ switch(type)
     detect_partial_match(common, backtracks);
   read_char_range(common, 0x9, 0x3000, type == OP_NOT_HSPACE);
   add_jump(compiler, &common->hspace, JUMP(SLJIT_FAST_CALL));
+  sljit_set_current_flags(compiler, SLJIT_SET_Z);
   add_jump(compiler, backtracks, JUMP(type == OP_NOT_HSPACE ? SLJIT_NOT_ZERO : SLJIT_ZERO));
   return cc;
 
@@ -6392,6 +6408,7 @@ switch(type)
     detect_partial_match(common, backtracks);
   read_char_range(common, 0xa, 0x2029, type == OP_NOT_VSPACE);
   add_jump(compiler, &common->vspace, JUMP(SLJIT_FAST_CALL));
+  sljit_set_current_flags(compiler, SLJIT_SET_Z);
   add_jump(compiler, backtracks, JUMP(type == OP_NOT_VSPACE ? SLJIT_NOT_ZERO : SLJIT_ZERO));
   return cc;
 
@@ -6418,7 +6435,7 @@ switch(type)
   OP1(SLJIT_MOV_U32, TMP1, 0, SLJIT_MEM1(STACK_TOP), (sljit_sw)PRIV(ucp_gbtable));
   OP1(SLJIT_MOV, STACK_TOP, 0, TMP2, 0);
   OP2(SLJIT_SHL, TMP2, 0, SLJIT_IMM, 1, TMP2, 0);
-  OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
+  OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
   JUMPTO(SLJIT_NOT_ZERO, label);
 
   OP1(SLJIT_MOV, STR_PTR, 0, TMP3, 0);
@@ -6587,7 +6604,7 @@ switch(type)
   OP2(SLJIT_LSHR, TMP1, 0, TMP1, 0, SLJIT_IMM, 3);
   OP1(SLJIT_MOV_U8, TMP1, 0, SLJIT_MEM1(TMP1), (sljit_sw)cc);
   OP2(SLJIT_SHL, TMP2, 0, SLJIT_IMM, 1, TMP2, 0);
-  OP2(SLJIT_AND | SLJIT_SET_E, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
+  OP2(SLJIT_AND | SLJIT_SET_Z, SLJIT_UNUSED, 0, TMP1, 0, TMP2, 0);
   add_jump(compiler, backtracks, JUMP(SLJIT_ZERO));
 
 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
@@ -6604,7 +6621,7 @@ switch(type)
   return cc + GET(cc, 0) - 1;
 #endif
   }
-SLJIT_ASSERT_STOP();
+SLJIT_UNREACHABLE();
 return cc;
 }
 
@@ -6790,9 +6807,9 @@ else
 #endif /* SUPPORT_UTF && SUPPORT_UCP */
   {
   if (ref)
-    OP2(SLJIT_SUB | SLJIT_SET_E, TMP2, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(offset + 1), TMP1, 0);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, TMP2, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(offset + 1), TMP1, 0);
   else
-    OP2(SLJIT_SUB | SLJIT_SET_E, TMP2, 0, SLJIT_MEM1(TMP2), sizeof(sljit_sw), TMP1, 0);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, TMP2, 0, SLJIT_MEM1(TMP2), sizeof(sljit_sw), TMP1, 0);
 
   if (withchecks)
     jump = JUMP(SLJIT_ZERO);
@@ -6883,7 +6900,7 @@ switch(type)
   cc += 1 + IMM2_SIZE + 1 + 2 * IMM2_SIZE;
   break;
   default:
-  SLJIT_ASSERT_STOP();
+  SLJIT_UNREACHABLE();
   break;
   }
 
@@ -6897,7 +6914,7 @@ if (!minimize)
     OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(0), STR_PTR, 0);
     OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(1), SLJIT_IMM, 0);
     /* Temporary release of STR_PTR. */
-    OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
+    OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
     /* Handles both invalid and empty cases. Since the minimum repeat,
     is zero the invalid case is basically the same as an empty case. */
     if (ref)
@@ -6910,7 +6927,7 @@ if (!minimize)
       zerolength = CMP(SLJIT_EQUAL, TMP1, 0, SLJIT_MEM1(TMP2), sizeof(sljit_sw));
       }
     /* Restore if not zero length. */
-    OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
+    OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
     }
   else
     {
@@ -7157,7 +7174,7 @@ return (*PUBL(callout))(callout_block);
     (((int)sizeof(PUBL(callout_block)) + 7) & ~7)
 
 #define CALLOUT_ARG_OFFSET(arg) \
-    (-CALLOUT_ARG_SIZE + SLJIT_OFFSETOF(PUBL(callout_block), arg))
+    SLJIT_OFFSETOF(PUBL(callout_block), arg)
 
 static SLJIT_INLINE pcre_uchar *compile_callout_matchingpath(compiler_common *common, pcre_uchar *cc, backtrack_common *parent)
 {
@@ -7187,7 +7204,8 @@ OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), CALLOUT_ARG_OFFSET(mark), (common->mark_pt
 
 /* Needed to save important temporary registers. */
 OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), LOCALS0, STACK_TOP, 0);
-OP2(SLJIT_SUB, SLJIT_R1, 0, STACK_TOP, 0, SLJIT_IMM, CALLOUT_ARG_SIZE);
+/* SLJIT_R0 = arguments */
+OP1(SLJIT_MOV, SLJIT_R1, 0, STACK_TOP, 0);
 GET_LOCAL_BASE(SLJIT_R2, 0, OVECTOR_START);
 sljit_emit_ijump(compiler, SLJIT_CALL3, SLJIT_IMM, SLJIT_FUNC_OFFSET(do_callout));
 OP1(SLJIT_MOV_S32, SLJIT_RETURN_REG, 0, SLJIT_RETURN_REG, 0);
@@ -7195,12 +7213,12 @@ OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), LOCALS0);
 free_stack(common, CALLOUT_ARG_SIZE / sizeof(sljit_sw));
 
 /* Check return value. */
-OP2(SLJIT_SUB | SLJIT_SET_S, SLJIT_UNUSED, 0, SLJIT_RETURN_REG, 0, SLJIT_IMM, 0);
+OP2(SLJIT_SUB | SLJIT_SET_Z | SLJIT_SET_SIG_GREATER, SLJIT_UNUSED, 0, SLJIT_RETURN_REG, 0, SLJIT_IMM, 0);
 add_jump(compiler, &backtrack->topbacktracks, JUMP(SLJIT_SIG_GREATER));
 if (common->forced_quit_label == NULL)
-  add_jump(compiler, &common->forced_quit, JUMP(SLJIT_SIG_LESS));
+  add_jump(compiler, &common->forced_quit, JUMP(SLJIT_NOT_EQUAL) /* SIG_LESS */);
 else
-  JUMPTO(SLJIT_SIG_LESS, common->forced_quit_label);
+  JUMPTO(SLJIT_NOT_EQUAL /* SIG_LESS */, common->forced_quit_label);
 return cc + 2 + 2 * LINK_SIZE;
 }
 
@@ -7321,7 +7339,7 @@ else
   allocate_stack(common, framesize + extrasize);
 
   OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
-  OP2(SLJIT_SUB, TMP2, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + extrasize) * sizeof(sljit_sw));
+  OP2(SLJIT_ADD, TMP2, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + extrasize) * sizeof(sljit_sw));
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, TMP2, 0);
   if (needs_control_head)
     OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr);
@@ -7392,22 +7410,22 @@ while (1)
       free_stack(common, extrasize);
 
     if (needs_control_head)
-      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_MEM1(STACK_TOP), 0);
+      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_MEM1(STACK_TOP), STACK(-1));
     }
   else
     {
     if ((opcode != OP_ASSERT_NOT && opcode != OP_ASSERTBACK_NOT) || conditional)
       {
       /* We don't need to keep the STR_PTR, only the previous private_data_ptr. */
-      OP2(SLJIT_ADD, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + 1) * sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + 1) * sizeof(sljit_sw));
       if (needs_control_head)
-        OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_MEM1(STACK_TOP), 0);
+        OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_MEM1(STACK_TOP), STACK(-1));
       }
     else
       {
       OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
       if (needs_control_head)
-        OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_MEM1(STACK_TOP), (framesize + 1) * sizeof(sljit_sw));
+        OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_MEM1(STACK_TOP), STACK(-framesize - 2));
       add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
       }
     }
@@ -7418,25 +7436,25 @@ while (1)
     if (conditional)
       {
       if (extrasize > 0)
-        OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), needs_control_head ? sizeof(sljit_sw) : 0);
+        OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), needs_control_head ? STACK(-2) : STACK(-1));
       }
     else if (bra == OP_BRAZERO)
       {
       if (framesize < 0)
-        OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), (extrasize - 1) * sizeof(sljit_sw));
+        OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), STACK(-extrasize));
       else
         {
-        OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), framesize * sizeof(sljit_sw));
-        OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), (framesize + extrasize - 1) * sizeof(sljit_sw));
+        OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), STACK(-framesize - 1));
+        OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), STACK(-framesize - extrasize));
         OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, TMP1, 0);
         }
-      OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(0), SLJIT_IMM, 0);
       }
     else if (framesize >= 0)
       {
       /* For OP_BRA and OP_BRAMINZERO. */
-      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), framesize * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-framesize - 1));
       }
     }
   add_jump(compiler, found, JUMP(SLJIT_JUMP));
@@ -7480,12 +7498,12 @@ if (common->positive_assert_quit != NULL)
   set_jumps(common->positive_assert_quit, LABEL());
   SLJIT_ASSERT(framesize != no_stack);
   if (framesize < 0)
-    OP2(SLJIT_ADD, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, extrasize * sizeof(sljit_sw));
+    OP2(SLJIT_SUB, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, extrasize * sizeof(sljit_sw));
   else
     {
     OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
     add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
-    OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + extrasize) * sizeof(sljit_sw));
+    OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + extrasize) * sizeof(sljit_sw));
     }
   JUMPHERE(jump);
   }
@@ -7534,18 +7552,18 @@ if (opcode == OP_ASSERT || opcode == OP_ASSERTBACK)
     {
     /* We know that STR_PTR was stored on the top of the stack. */
     if (extrasize > 0)
-      OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), (extrasize - 1) * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), STACK(-extrasize));
 
     /* Keep the STR_PTR on the top of the stack. */
     if (bra == OP_BRAZERO)
       {
-      OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
       if (extrasize == 2)
         OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(0), STR_PTR, 0);
       }
     else if (bra == OP_BRAMINZERO)
       {
-      OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(0), SLJIT_IMM, 0);
       }
     }
@@ -7554,13 +7572,13 @@ if (opcode == OP_ASSERT || opcode == OP_ASSERTBACK)
     if (bra == OP_BRA)
       {
       /* We don't need to keep the STR_PTR, only the previous private_data_ptr. */
-      OP2(SLJIT_ADD, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + 1) * sizeof(sljit_sw));
-      OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), (extrasize - 2) * sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + 1) * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), STACK(-extrasize + 1));
       }
     else
       {
       /* We don't need to keep the STR_PTR, only the previous private_data_ptr. */
-      OP2(SLJIT_ADD, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + 2) * sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + 2) * sizeof(sljit_sw));
       if (extrasize == 2)
         {
         OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(STACK_TOP), STACK(0));
@@ -7588,7 +7606,7 @@ if (opcode == OP_ASSERT || opcode == OP_ASSERTBACK)
       {
       OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
       add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
-      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), framesize * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-framesize - 1));
       }
     set_jumps(backtrack->common.topbacktracks, LABEL());
     }
@@ -7675,23 +7693,23 @@ if (framesize < 0)
     }
 
   if (needs_control_head)
-    OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), (ket != OP_KET || has_alternatives) ? sizeof(sljit_sw) : 0);
+    OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), (ket != OP_KET || has_alternatives) ? STACK(-2) : STACK(-1));
 
   /* TMP2 which is set here used by OP_KETRMAX below. */
   if (ket == OP_KETRMAX)
-    OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), 0);
+    OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), STACK(-1));
   else if (ket == OP_KETRMIN)
     {
     /* Move the STR_PTR to the private_data_ptr. */
-    OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), 0);
+    OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-1));
     }
   }
 else
   {
   stacksize = (ket != OP_KET || has_alternatives) ? 2 : 1;
-  OP2(SLJIT_ADD, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + stacksize) * sizeof(sljit_sw));
+  OP2(SLJIT_SUB, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, (framesize + stacksize) * sizeof(sljit_sw));
   if (needs_control_head)
-    OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), 0);
+    OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), STACK(-1));
 
   if (ket == OP_KETRMAX)
     {
@@ -7927,7 +7945,7 @@ if (bra == OP_BRAMINZERO)
         {
         /* Except when the whole stack frame must be saved. */
         OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
-        braminzero = CMP(SLJIT_EQUAL, STR_PTR, 0, SLJIT_MEM1(TMP1), (BACKTRACK_AS(bracket_backtrack)->u.framesize + 1) * sizeof(sljit_sw));
+        braminzero = CMP(SLJIT_EQUAL, STR_PTR, 0, SLJIT_MEM1(TMP1), STACK(-BACKTRACK_AS(bracket_backtrack)->u.framesize - 2));
         }
       JUMPHERE(skip);
       }
@@ -8000,7 +8018,7 @@ if (opcode == OP_ONCE)
         OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(stacksize), STR_PTR, 0);
       if (BACKTRACK_AS(bracket_backtrack)->u.framesize == no_frame)
-        OP2(SLJIT_SUB, SLJIT_MEM1(SLJIT_SP), private_data_ptr, STACK_TOP, 0, SLJIT_IMM, needs_control_head ? (2 * sizeof(sljit_sw)) : sizeof(sljit_sw));
+        OP2(SLJIT_ADD, SLJIT_MEM1(SLJIT_SP), private_data_ptr, STACK_TOP, 0, SLJIT_IMM, needs_control_head ? (2 * sizeof(sljit_sw)) : sizeof(sljit_sw));
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(stacksize + 1), TMP2, 0);
       }
     else if (ket == OP_KETRMAX || has_alternatives)
@@ -8018,7 +8036,7 @@ if (opcode == OP_ONCE)
       OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(0), TMP2, 0);
 
     OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
-    OP2(SLJIT_SUB, TMP2, 0, STACK_TOP, 0, SLJIT_IMM, stacksize * sizeof(sljit_sw));
+    OP2(SLJIT_ADD, TMP2, 0, STACK_TOP, 0, SLJIT_IMM, stacksize * sizeof(sljit_sw));
 
     stacksize = needs_control_head ? 1 : 0;
     if (ket != OP_KET || has_alternatives)
@@ -8090,13 +8108,13 @@ if (opcode == OP_COND || opcode == OP_SCOND)
     slot = common->name_table + GET2(matchingpath, 1) * common->name_entry_size;
     OP1(SLJIT_MOV, TMP3, 0, STR_PTR, 0);
     OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(1));
-    OP2(SLJIT_SUB | SLJIT_SET_E, TMP2, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(GET2(slot, 0) << 1), TMP1, 0);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, TMP2, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(GET2(slot, 0) << 1), TMP1, 0);
     slot += common->name_entry_size;
     i--;
     while (i-- > 0)
       {
       OP2(SLJIT_SUB, STR_PTR, 0, SLJIT_MEM1(SLJIT_SP), OVECTOR(GET2(slot, 0) << 1), TMP1, 0);
-      OP2(SLJIT_OR | SLJIT_SET_E, TMP2, 0, TMP2, 0, STR_PTR, 0);
+      OP2(SLJIT_OR | SLJIT_SET_Z, TMP2, 0, TMP2, 0, STR_PTR, 0);
       slot += common->name_entry_size;
       }
     OP1(SLJIT_MOV, STR_PTR, 0, TMP3, 0);
@@ -8111,7 +8129,7 @@ if (opcode == OP_COND || opcode == OP_SCOND)
 
     if (*matchingpath == OP_FAIL)
       stacksize = 0;
-    if (*matchingpath == OP_RREF)
+    else if (*matchingpath == OP_RREF)
       {
       stacksize = GET2(matchingpath, 1);
       if (common->currententry == NULL)
@@ -8244,7 +8262,7 @@ if (ket == OP_KETRMAX)
     {
     if (has_alternatives)
       BACKTRACK_AS(bracket_backtrack)->alternative_matchingpath = LABEL();
-    OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_IMM, 1);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_IMM, 1);
     JUMPTO(SLJIT_NOT_ZERO, rmax_label);
     /* Drop STR_PTR for greedy plus quantifier. */
     if (opcode != OP_ONCE)
@@ -8274,7 +8292,7 @@ if (ket == OP_KETRMAX)
 if (repeat_type == OP_EXACT)
   {
   count_match(common);
-  OP2(SLJIT_SUB | SLJIT_SET_E, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_IMM, 1);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_MEM1(SLJIT_SP), repeat_ptr, SLJIT_IMM, 1);
   JUMPTO(SLJIT_NOT_ZERO, rmax_label);
   }
 else if (repeat_type == OP_UPTO)
@@ -8374,7 +8392,7 @@ switch(opcode)
   break;
 
   default:
-  SLJIT_ASSERT_STOP();
+  SLJIT_UNREACHABLE();
   break;
   }
 
@@ -8452,7 +8470,7 @@ else
   OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
   if (needs_control_head)
     OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr);
-  OP2(SLJIT_SUB, SLJIT_MEM1(SLJIT_SP), private_data_ptr, STACK_TOP, 0, SLJIT_IMM, -STACK(stacksize - 1));
+  OP2(SLJIT_ADD, SLJIT_MEM1(SLJIT_SP), private_data_ptr, STACK_TOP, 0, SLJIT_IMM, stacksize * sizeof(sljit_sw));
 
   stack = 0;
   if (!zero)
@@ -8524,7 +8542,7 @@ while (*cc != OP_KETRPOS)
     {
     if (offset != 0)
       {
-      OP2(SLJIT_ADD, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, stacksize * sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_IMM, stacksize * sizeof(sljit_sw));
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), cbraprivptr);
       OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), OVECTOR(offset + 1), STR_PTR, 0);
       OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), cbraprivptr, STR_PTR, 0);
@@ -8535,10 +8553,10 @@ while (*cc != OP_KETRPOS)
     else
       {
       OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
-      OP2(SLJIT_ADD, STACK_TOP, 0, TMP2, 0, SLJIT_IMM, stacksize * sizeof(sljit_sw));
+      OP2(SLJIT_SUB, STACK_TOP, 0, TMP2, 0, SLJIT_IMM, stacksize * sizeof(sljit_sw));
       if (opcode == OP_SBRAPOS)
-        OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(TMP2), (framesize + 1) * sizeof(sljit_sw));
-      OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), (framesize + 1) * sizeof(sljit_sw), STR_PTR, 0);
+        OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(TMP2), STACK(-framesize - 2));
+      OP1(SLJIT_MOV, SLJIT_MEM1(TMP2), STACK(-framesize - 2), STR_PTR, 0);
       }
 
     /* Even if the match is empty, we need to reset the control head. */
@@ -8584,7 +8602,7 @@ while (*cc != OP_KETRPOS)
     else
       {
       OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
-      OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(TMP2), (framesize + 1) * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(TMP2), STACK(-framesize - 2));
       }
     }
 
@@ -8601,7 +8619,7 @@ if (!zero)
   if (framesize < 0)
     add_jump(compiler, &backtrack->topbacktracks, CMP(SLJIT_NOT_EQUAL, SLJIT_MEM1(STACK_TOP), STACK(stacksize - 1), SLJIT_IMM, 0));
   else /* TMP2 is set to [private_data_ptr] above. */
-    add_jump(compiler, &backtrack->topbacktracks, CMP(SLJIT_NOT_EQUAL, SLJIT_MEM1(TMP2), (stacksize - 1) * sizeof(sljit_sw), SLJIT_IMM, 0));
+    add_jump(compiler, &backtrack->topbacktracks, CMP(SLJIT_NOT_EQUAL, SLJIT_MEM1(TMP2), STACK(-stacksize), SLJIT_IMM, 0));
   }
 
 /* None of them matched. */
@@ -8824,7 +8842,7 @@ if (exact > 1)
     OP1(SLJIT_MOV, tmp_base, tmp_offset, SLJIT_IMM, exact);
     label = LABEL();
     compile_char1_matchingpath(common, type, cc, &backtrack->topbacktracks, FALSE);
-    OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
     JUMPTO(SLJIT_NOT_ZERO, label);
     }
   else
@@ -8832,7 +8850,7 @@ if (exact > 1)
     OP1(SLJIT_MOV, tmp_base, tmp_offset, SLJIT_IMM, exact);
     label = LABEL();
     compile_char1_matchingpath(common, type, cc, &backtrack->topbacktracks, TRUE);
-    OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
     JUMPTO(SLJIT_NOT_ZERO, label);
     }
   }
@@ -8862,7 +8880,7 @@ switch(opcode)
     if (opcode == OP_UPTO)
       {
       OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), POSSESSIVE0);
-      OP2(SLJIT_SUB | SLJIT_SET_E, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
+      OP2(SLJIT_SUB | SLJIT_SET_Z, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
       jump = JUMP(SLJIT_ZERO);
       OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), POSSESSIVE0, TMP1, 0);
       }
@@ -8924,7 +8942,7 @@ switch(opcode)
       label = LABEL();
       if (opcode == OP_UPTO)
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
         add_jump(compiler, &backtrack->topbacktracks, JUMP(SLJIT_ZERO));
         }
       compile_char1_matchingpath(common, type, cc, &backtrack->topbacktracks, FALSE);
@@ -8944,7 +8962,7 @@ switch(opcode)
       OP1(SLJIT_MOV, base, offset1, STR_PTR, 0);
       if (opcode == OP_UPTO)
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
         add_jump(compiler, &no_match, JUMP(SLJIT_ZERO));
         }
 
@@ -8971,7 +8989,7 @@ switch(opcode)
 
       if (opcode == OP_UPTO)
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
         JUMPTO(SLJIT_NOT_ZERO, label);
         }
       else
@@ -9000,7 +9018,7 @@ switch(opcode)
 
       if (opcode == OP_UPTO)
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
         JUMPTO(SLJIT_NOT_ZERO, label);
         }
       else
@@ -9026,7 +9044,7 @@ switch(opcode)
       compile_char1_matchingpath(common, type, cc, &no_char1_match, FALSE);
       if (opcode == OP_UPTO)
         {
-        OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+        OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
         JUMPTO(SLJIT_NOT_ZERO, label);
         OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(1));
         }
@@ -9113,7 +9131,7 @@ switch(opcode)
     label = LABEL();
     compile_char1_matchingpath(common, type, cc, &no_match, TRUE);
     OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), POSSESSIVE1, STR_PTR, 0);
-    OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+    OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
     JUMPTO(SLJIT_NOT_ZERO, label);
     set_jumps(no_match, LABEL());
     OP1(SLJIT_MOV, STR_PTR, 0, SLJIT_MEM1(SLJIT_SP), POSSESSIVE1);
@@ -9124,7 +9142,7 @@ switch(opcode)
   label = LABEL();
   detect_partial_match(common, &no_match);
   compile_char1_matchingpath(common, type, cc, &no_char1_match, FALSE);
-  OP2(SLJIT_SUB | SLJIT_SET_E, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, tmp_base, tmp_offset, tmp_base, tmp_offset, SLJIT_IMM, 1);
   JUMPTO(SLJIT_NOT_ZERO, label);
   OP2(SLJIT_ADD, STR_PTR, 0, STR_PTR, 0, SLJIT_IMM, IN_UCHARS(1));
   set_jumps(no_char1_match, LABEL());
@@ -9142,7 +9160,7 @@ switch(opcode)
   break;
 
   default:
-  SLJIT_ASSERT_STOP();
+  SLJIT_UNREACHABLE();
   break;
   }
 
@@ -9264,7 +9282,7 @@ size = 3 + (size < 0 ? 0 : size);
 OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr);
 allocate_stack(common, size);
 if (size > 3)
-  OP2(SLJIT_SUB, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, STACK_TOP, 0, SLJIT_IMM, (size - 3) * sizeof(sljit_sw));
+  OP2(SLJIT_ADD, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, STACK_TOP, 0, SLJIT_IMM, (size - 3) * sizeof(sljit_sw));
 else
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, STACK_TOP, 0);
 OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(size - 1), SLJIT_IMM, BACKTRACK_AS(then_trap_backtrack)->start);
@@ -9569,7 +9587,7 @@ while (cc < ccend)
     break;
 
     default:
-    SLJIT_ASSERT_STOP();
+    SLJIT_UNREACHABLE();
     return;
     }
   if (cc == NULL)
@@ -9677,7 +9695,7 @@ switch(opcode)
   case OP_MINUPTO:
   OP1(SLJIT_MOV, TMP1, 0, base, offset1);
   OP1(SLJIT_MOV, STR_PTR, 0, base, offset0);
-  OP2(SLJIT_SUB | SLJIT_SET_E, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
+  OP2(SLJIT_SUB | SLJIT_SET_Z, TMP1, 0, TMP1, 0, SLJIT_IMM, 1);
   add_jump(compiler, &jumplist, JUMP(SLJIT_ZERO));
 
   OP1(SLJIT_MOV, base, offset1, TMP1, 0);
@@ -9723,7 +9741,7 @@ switch(opcode)
   break;
 
   default:
-  SLJIT_ASSERT_STOP();
+  SLJIT_UNREACHABLE();
   break;
   }
 
@@ -9831,7 +9849,7 @@ if (*cc == OP_ASSERT || *cc == OP_ASSERTBACK)
   {
   OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), CURRENT_AS(assert_backtrack)->private_data_ptr);
   add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
-  OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), CURRENT_AS(assert_backtrack)->private_data_ptr, SLJIT_MEM1(STACK_TOP), CURRENT_AS(assert_backtrack)->framesize * sizeof(sljit_sw));
+  OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), CURRENT_AS(assert_backtrack)->private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-CURRENT_AS(assert_backtrack)->framesize - 1));
 
   set_jumps(current->topbacktracks, LABEL());
   }
@@ -9841,7 +9859,7 @@ else
 if (bra == OP_BRAZERO)
   {
   /* We know there is enough place on the stack. */
-  OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
+  OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, sizeof(sljit_sw));
   OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(0), SLJIT_IMM, 0);
   JUMPTO(SLJIT_JUMP, CURRENT_AS(assert_backtrack)->matchingpath);
   JUMPHERE(brajump);
@@ -9954,7 +9972,7 @@ else if (ket == OP_KETRMIN)
       else
         {
         OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(SLJIT_SP), private_data_ptr);
-        CMPTO(SLJIT_NOT_EQUAL, STR_PTR, 0, SLJIT_MEM1(TMP1), (CURRENT_AS(bracket_backtrack)->u.framesize + 1) * sizeof(sljit_sw), CURRENT_AS(bracket_backtrack)->recursive_matchingpath);
+        CMPTO(SLJIT_NOT_EQUAL, STR_PTR, 0, SLJIT_MEM1(TMP1), STACK(-CURRENT_AS(bracket_backtrack)->u.framesize - 2), CURRENT_AS(bracket_backtrack)->recursive_matchingpath);
         }
       /* Drop STR_PTR for non-greedy plus quantifier. */
       if (opcode != OP_ONCE)
@@ -10060,7 +10078,7 @@ if (SLJIT_UNLIKELY(opcode == OP_COND) || SLJIT_UNLIKELY(opcode == OP_SCOND))
       {
       OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), assert->private_data_ptr);
       add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
-      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), assert->private_data_ptr, SLJIT_MEM1(STACK_TOP), assert->framesize * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), assert->private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-assert->framesize - 1));
       }
     cond = JUMP(SLJIT_JUMP);
     set_jumps(CURRENT_AS(bracket_backtrack)->u.assert->condfailed, LABEL());
@@ -10201,7 +10219,7 @@ if (has_alternatives)
       {
       OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), assert->private_data_ptr);
       add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
-      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), assert->private_data_ptr, SLJIT_MEM1(STACK_TOP), assert->framesize * sizeof(sljit_sw));
+      OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), assert->private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-assert->framesize - 1));
       }
     JUMPHERE(cond);
     }
@@ -10256,7 +10274,7 @@ else if (opcode == OP_ONCE)
   JUMPHERE(once);
   /* Restore previous private_data_ptr */
   if (CURRENT_AS(bracket_backtrack)->u.framesize >= 0)
-    OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), CURRENT_AS(bracket_backtrack)->u.framesize * sizeof(sljit_sw));
+    OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-CURRENT_AS(bracket_backtrack)->u.framesize - 1));
   else if (ket == OP_KETRMIN)
     {
     OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), STACK(1));
@@ -10346,7 +10364,7 @@ if (current->topbacktracks)
   free_stack(common, CURRENT_AS(bracketpos_backtrack)->stacksize);
   JUMPHERE(jump);
   }
-OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), CURRENT_AS(bracketpos_backtrack)->private_data_ptr, SLJIT_MEM1(STACK_TOP), CURRENT_AS(bracketpos_backtrack)->framesize * sizeof(sljit_sw));
+OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), CURRENT_AS(bracketpos_backtrack)->private_data_ptr, SLJIT_MEM1(STACK_TOP), STACK(-CURRENT_AS(bracketpos_backtrack)->framesize - 1));
 }
 
 static SLJIT_INLINE void compile_braminzero_backtrackingpath(compiler_common *common, struct backtrack_common *current)
@@ -10392,10 +10410,10 @@ if (opcode == OP_THEN || opcode == OP_THEN_ARG)
     jump = JUMP(SLJIT_JUMP);
 
     loop = LABEL();
-    OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(STACK_TOP), -(int)sizeof(sljit_sw));
+    OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(STACK_TOP), STACK(0));
     JUMPHERE(jump);
-    CMPTO(SLJIT_NOT_EQUAL, SLJIT_MEM1(STACK_TOP), -(int)(2 * sizeof(sljit_sw)), TMP1, 0, loop);
-    CMPTO(SLJIT_NOT_EQUAL, SLJIT_MEM1(STACK_TOP), -(int)(3 * sizeof(sljit_sw)), TMP2, 0, loop);
+    CMPTO(SLJIT_NOT_EQUAL, SLJIT_MEM1(STACK_TOP), STACK(1), TMP1, 0, loop);
+    CMPTO(SLJIT_NOT_EQUAL, SLJIT_MEM1(STACK_TOP), STACK(2), TMP2, 0, loop);
     add_jump(compiler, &common->then_trap->quit, JUMP(SLJIT_JUMP));
     return;
     }
@@ -10645,7 +10663,7 @@ while (current)
     break;
 
     default:
-    SLJIT_ASSERT_STOP();
+    SLJIT_UNREACHABLE();
     break;
     }
   current = current->prev;
@@ -10684,7 +10702,7 @@ sljit_emit_fast_enter(compiler, TMP2, 0);
 count_match(common);
 allocate_stack(common, private_data_size + framesize + alternativesize);
 OP1(SLJIT_MOV, SLJIT_MEM1(STACK_TOP), STACK(private_data_size + framesize + alternativesize - 1), TMP2, 0);
-copy_private_data(common, ccbegin, ccend, TRUE, private_data_size + framesize + alternativesize, framesize + alternativesize, needs_control_head);
+copy_private_data(common, ccbegin, ccend, TRUE, framesize + alternativesize, private_data_size + framesize + alternativesize, needs_control_head);
 if (needs_control_head)
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, SLJIT_IMM, 0);
 OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->recursive_head_ptr, STACK_TOP, 0);
@@ -10737,9 +10755,9 @@ if (common->quit != NULL)
   OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), common->recursive_head_ptr);
   if (needs_frame)
     {
-    OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + alternativesize) * sizeof(sljit_sw));
-    add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
     OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + alternativesize) * sizeof(sljit_sw));
+    add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
+    OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + alternativesize) * sizeof(sljit_sw));
     }
   OP1(SLJIT_MOV, TMP3, 0, SLJIT_IMM, 0);
   common->quit = NULL;
@@ -10750,32 +10768,32 @@ set_jumps(common->accept, LABEL());
 OP1(SLJIT_MOV, STACK_TOP, 0, SLJIT_MEM1(SLJIT_SP), common->recursive_head_ptr);
 if (needs_frame)
   {
-  OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + alternativesize) * sizeof(sljit_sw));
-  add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
   OP2(SLJIT_ADD, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + alternativesize) * sizeof(sljit_sw));
+  add_jump(compiler, &common->revertframes, JUMP(SLJIT_FAST_CALL));
+  OP2(SLJIT_SUB, STACK_TOP, 0, STACK_TOP, 0, SLJIT_IMM, (framesize + alternativesize) * sizeof(sljit_sw));
   }
 OP1(SLJIT_MOV, TMP3, 0, SLJIT_IMM, 1);
 
 JUMPHERE(jump);
 if (common->quit != NULL)
   set_jumps(common->quit, LABEL());
-copy_private_data(common, ccbegin, ccend, FALSE, private_data_size + framesize + alternativesize, framesize + alternativesize, needs_control_head);
+copy_private_data(common, ccbegin, ccend, FALSE, framesize + alternativesize, private_data_size + framesize + alternativesize, needs_control_head);
 free_stack(common, private_data_size + framesize + alternativesize);
 if (needs_control_head)
   {
-  OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), 2 * sizeof(sljit_sw));
-  OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), sizeof(sljit_sw));
+  OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(STACK_TOP), STACK(-3));
+  OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), STACK(-2));
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->recursive_head_ptr, TMP1, 0);
   OP1(SLJIT_MOV, TMP1, 0, TMP3, 0);
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->control_head_ptr, TMP2, 0);
   }
 else
   {
-  OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), sizeof(sljit_sw));
+  OP1(SLJIT_MOV, TMP2, 0, SLJIT_MEM1(STACK_TOP), STACK(-2));
   OP1(SLJIT_MOV, TMP1, 0, TMP3, 0);
   OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), common->recursive_head_ptr, TMP2, 0);
   }
-sljit_emit_fast_return(compiler, SLJIT_MEM1(STACK_TOP), 0);
+sljit_emit_fast_return(compiler, SLJIT_MEM1(STACK_TOP), STACK(-1));
 }
 
 #undef COMPILE_BACKTRACKINGPATH
@@ -11237,7 +11255,7 @@ OP1(SLJIT_MOV, SLJIT_MEM1(SLJIT_SP), LOCALS1, TMP2, 0);
 OP1(SLJIT_MOV, TMP1, 0, ARGUMENTS, 0);
 OP1(SLJIT_MOV, TMP1, 0, SLJIT_MEM1(TMP1), SLJIT_OFFSETOF(jit_arguments, stack));
 OP1(SLJIT_MOV, SLJIT_MEM1(TMP1), SLJIT_OFFSETOF(struct sljit_stack, top), STACK_TOP, 0);
-OP2(SLJIT_ADD, TMP2, 0, SLJIT_MEM1(TMP1), SLJIT_OFFSETOF(struct sljit_stack, limit), SLJIT_IMM, STACK_GROWTH_RATE);
+OP2(SLJIT_SUB, TMP2, 0, SLJIT_MEM1(TMP1), SLJIT_OFFSETOF(struct sljit_stack, limit), SLJIT_IMM, STACK_GROWTH_RATE);
 
 sljit_emit_ijump(compiler, SLJIT_CALL2, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_stack_resize));
 jump = CMP(SLJIT_NOT_EQUAL, SLJIT_RETURN_REG, 0, SLJIT_IMM, 0);
@@ -11391,10 +11409,10 @@ union {
 sljit_u8 local_space[MACHINE_STACK_SIZE];
 struct sljit_stack local_stack;
 
-local_stack.top = (sljit_sw)&local_space;
-local_stack.base = local_stack.top;
-local_stack.limit = local_stack.base + MACHINE_STACK_SIZE;
-local_stack.max_limit = local_stack.limit;
+local_stack.max_limit = local_space;
+local_stack.limit = local_space;
+local_stack.base = local_space + MACHINE_STACK_SIZE;
+local_stack.top = local_space + MACHINE_STACK_SIZE;
 arguments->stack = &local_stack;
 convert_executable_func.executable_func = executable_func;
 return convert_executable_func.call_executable_func(arguments);
diff --git a/pcre/pcre_scanner_unittest.cc b/pcre/pcre_scanner_unittest.cc
index c00312c4f63..623e2afda80 100644
--- a/pcre/pcre_scanner_unittest.cc
+++ b/pcre/pcre_scanner_unittest.cc
@@ -57,6 +57,7 @@
 } while (0)
 
 using std::vector;
+using std::string;
 using pcrecpp::StringPiece;
 using pcrecpp::Scanner;
 
diff --git a/pcre/pcre_stringpiece.h.in b/pcre/pcre_stringpiece.h.in
index eb25826b453..f54f3f3b31b 100644
--- a/pcre/pcre_stringpiece.h.in
+++ b/pcre/pcre_stringpiece.h.in
@@ -52,12 +52,12 @@
 
 #include <pcre.h>
 
+namespace pcrecpp {
+
 using std::memcmp;
 using std::strlen;
 using std::string;
 
-namespace pcrecpp {
-
 class PCRECPP_EXP_DEFN StringPiece {
  private:
   const char*   ptr_;
diff --git a/pcre/pcre_stringpiece_unittest.cc b/pcre/pcre_stringpiece_unittest.cc
index 1c4759da3b0..88e73a1f976 100644
--- a/pcre/pcre_stringpiece_unittest.cc
+++ b/pcre/pcre_stringpiece_unittest.cc
@@ -24,6 +24,7 @@
   }                                                     \
 } while (0)
 
+using std::string;
 using pcrecpp::StringPiece;
 
 static void CheckSTLComparator() {
diff --git a/pcre/pcre_tables.c b/pcre/pcre_tables.c
index 4960af57c4d..5e18e8cf904 100644
--- a/pcre/pcre_tables.c
+++ b/pcre/pcre_tables.c
@@ -6,7 +6,7 @@
 and semantics are as close as possible to those of the Perl 5 language.
 
                        Written by Philip Hazel
-           Copyright (c) 1997-2012 University of Cambridge
+           Copyright (c) 1997-2017 University of Cambridge
 
 -----------------------------------------------------------------------------
 Redistribution and use in source and binary forms, with or without
@@ -161,7 +161,7 @@ const pcre_uint32 PRIV(ucp_gbtable[]) = {
 
    (1<<ucp_gbExtend)|(1<<ucp_gbSpacingMark),                /*  5 SpacingMark */
    (1<<ucp_gbExtend)|(1<<ucp_gbSpacingMark)|(1<<ucp_gbL)|   /*  6 L */
-     (1<<ucp_gbL)|(1<<ucp_gbV)|(1<<ucp_gbLV)|(1<<ucp_gbLVT),
+     (1<<ucp_gbV)|(1<<ucp_gbLV)|(1<<ucp_gbLVT),
 
    (1<<ucp_gbExtend)|(1<<ucp_gbSpacingMark)|(1<<ucp_gbV)|   /*  7 V */
      (1<<ucp_gbT),
diff --git a/pcre/pcre_ucd.c b/pcre/pcre_ucd.c
index 69c4fd42c34..f22f826c4c2 100644
--- a/pcre/pcre_ucd.c
+++ b/pcre/pcre_ucd.c
@@ -38,6 +38,20 @@ const pcre_uint16 PRIV(ucd_stage2)[] = {0};
 const pcre_uint32 PRIV(ucd_caseless_sets)[] = {0};
 #else
 
+/* If the 32-bit library is run in non-32-bit mode, character values
+greater than 0x10ffff may be encountered. For these we set up a
+special record. */
+
+#ifdef COMPILE_PCRE32
+const ucd_record PRIV(dummy_ucd_record)[] = {{
+  ucp_Common,    /* script */
+  ucp_Cn,        /* type unassigned */
+  ucp_gbOther,   /* grapheme break property */
+  0,             /* case set */
+  0,             /* other case */
+  }};
+#endif
+
 /* When recompiling tables with a new Unicode version, please check the
 types in this structure definition from pcre_internal.h (the actual
 field names will be different):
diff --git a/pcre/pcrecpp_unittest.cc b/pcre/pcrecpp_unittest.cc
index 92cae8fbea5..4b15fbef1c3 100644
--- a/pcre/pcrecpp_unittest.cc
+++ b/pcre/pcrecpp_unittest.cc
@@ -43,6 +43,7 @@
 #include <vector>
 #include "pcrecpp.h"
 
+using std::string;
 using pcrecpp::StringPiece;
 using pcrecpp::RE;
 using pcrecpp::RE_Options;
diff --git a/pcre/pcregrep.c b/pcre/pcregrep.c
index fd2a67622ba..317f7454e13 100644
--- a/pcre/pcregrep.c
+++ b/pcre/pcregrep.c
@@ -1804,11 +1804,6 @@ while (ptr < endptr)
         if (line_buffered) fflush(stdout);
         rc = 0;                      /* Had some success */
 
-        /* If the current match ended past the end of the line (only possible
-        in multiline mode), we are done with this line. */
-
-        if ((unsigned int)offsets[1] > linelength) goto END_ONE_MATCH;
-
         startoffset = offsets[1];    /* Restart after the match */
         if (startoffset <= oldstartoffset)
           {
@@ -1818,6 +1813,22 @@ while (ptr < endptr)
           if (utf8)
             while ((matchptr[startoffset] & 0xc0) == 0x80) startoffset++;
           }
+
+        /* If the current match ended past the end of the line (only possible
+        in multiline mode), we must move on to the line in which it did end
+        before searching for more matches. */
+
+        while (startoffset > (int)linelength)
+          {
+          matchptr = ptr += linelength + endlinelength;
+          filepos += (int)(linelength + endlinelength);
+          linenumber++;
+          startoffset -= (int)(linelength + endlinelength);
+          t = end_of_line(ptr, endptr, &endlinelength);
+          linelength = t - ptr - endlinelength;
+          length = (size_t)(endptr - ptr);
+          }
+
         goto ONLY_MATCHING_RESTART;
         }
       }
@@ -3179,9 +3190,11 @@ for (j = 1, cp = patterns; cp != NULL; j++, cp = cp->next)
   cp->hint = pcre_study(cp->compiled, study_options, &error);
   if (error != NULL)
     {
-    char s[16];
-    if (patterns->next == NULL) s[0] = 0; else sprintf(s, " number %d", j);
-    fprintf(stderr, "pcregrep: Error while studying regex%s: %s\n", s, error);
+    if (patterns->next == NULL)
+      fprintf(stderr, "pcregrep: Error while studying regex: %s\n", error);
+    else
+      fprintf(stderr, "pcregrep: Error while studying regex number %d: %s\n",
+        j, error);
     goto EXIT2;
     }
 #ifdef SUPPORT_PCREGREP_JIT
diff --git a/pcre/pcreposix.c b/pcre/pcreposix.c
index cf75588c40d..7b404a71100 100644
--- a/pcre/pcreposix.c
+++ b/pcre/pcreposix.c
@@ -6,7 +6,7 @@
 and semantics are as close as possible to those of the Perl 5 language.
 
                        Written by Philip Hazel
-           Copyright (c) 1997-2016 University of Cambridge
+           Copyright (c) 1997-2017 University of Cambridge
 
 -----------------------------------------------------------------------------
 Redistribution and use in source and binary forms, with or without
@@ -389,8 +389,8 @@ if (rc >= 0)
     {
     for (i = 0; i < (size_t)rc; i++)
       {
-      pmatch[i].rm_so = ovector[i*2];
-      pmatch[i].rm_eo = ovector[i*2+1];
+      pmatch[i].rm_so = ovector[i*2] + so;
+      pmatch[i].rm_eo = ovector[i*2+1] + so;
       }
     if (allocated_ovector) free(ovector);
     for (; i < nmatch; i++) pmatch[i].rm_so = pmatch[i].rm_eo = -1;
diff --git a/pcre/pcretest.c b/pcre/pcretest.c
index 5b73a918075..f1303037281 100644
--- a/pcre/pcretest.c
+++ b/pcre/pcretest.c
@@ -177,7 +177,7 @@ that differ in their output from isprint() even in the "C" locale. */
 #define PRINTABLE(c) ((c) >= 32 && (c) < 127)
 #endif
 
-#define PRINTOK(c) (locale_set? isprint(c) : PRINTABLE(c))
+#define PRINTOK(c) (locale_set? (((c) < 256) && isprint(c)) : PRINTABLE(c))
 
 /* Posix support is disabled in 16 or 32 bit only mode. */
 #if !defined SUPPORT_PCRE8 && !defined NOPOSIX
@@ -426,11 +426,11 @@ argument, the casting might be incorrectly applied. */
 #define PCRE_COPY_NAMED_SUBSTRING32(rc, re, bptr, offsets, count, \
     namesptr, cbuffer, size) \
   rc = pcre32_copy_named_substring((pcre32 *)re, (PCRE_SPTR32)bptr, offsets, \
-    count, (PCRE_SPTR32)namesptr, (PCRE_UCHAR32 *)cbuffer, size/2)
+    count, (PCRE_SPTR32)namesptr, (PCRE_UCHAR32 *)cbuffer, size/4)
 
 #define PCRE_COPY_SUBSTRING32(rc, bptr, offsets, count, i, cbuffer, size) \
   rc = pcre32_copy_substring((PCRE_SPTR32)bptr, offsets, count, i, \
-    (PCRE_UCHAR32 *)cbuffer, size/2)
+    (PCRE_UCHAR32 *)cbuffer, size/4)
 
 #define PCRE_DFA_EXEC32(count, re, extra, bptr, len, start_offset, options, \
     offsets, size_offsets, workspace, size_workspace) \
@@ -4834,7 +4834,16 @@ while (!done)
         continue;
 
         case 'O':
-        while(isdigit(*p)) n = n * 10 + *p++ - '0';
+        while(isdigit(*p))
+          {
+          if (n > (INT_MAX-10)/10)   /* Hack to stop fuzzers */
+            {
+            printf("** \\O argument is too big\n");
+            yield = 1;
+            goto EXIT;
+            }
+          n = n * 10 + *p++ - '0';
+          }
         if (n > size_offsets_max)
           {
           size_offsets_max = n;
diff --git a/pcre/testdata/testinput1 b/pcre/testdata/testinput1
index 93abab3c851..5c23f41fa81 100644
--- a/pcre/testdata/testinput1
+++ b/pcre/testdata/testinput1
@@ -5739,4 +5739,7 @@ AbcdCBefgBhiBqz
 /(?=.*X)X$/ 
     \  X
      
+/X+(?#comment)?/
+    >XXX<
+
 /-- End of testinput1 --/
diff --git a/pcre/testdata/testinput12 b/pcre/testdata/testinput12
index 944be6943f5..89ed4564bcd 100644
--- a/pcre/testdata/testinput12
+++ b/pcre/testdata/testinput12
@@ -104,4 +104,6 @@ and a couple of things that are different with JIT. --/
 /(.|.)*?bx/
     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaabax
 
+/((?(?!))x)(?'name')(?1)/S++
+
 /-- End of testinput12 --/
diff --git a/pcre/testdata/testinput15 b/pcre/testdata/testinput15
index 83e26773471..c065105b1ec 100644
--- a/pcre/testdata/testinput15
+++ b/pcre/testdata/testinput15
@@ -363,4 +363,7 @@ correctly, but that messes up comparisons). --/
 
 /abc/89
 
+//8+L
+    \xf1\xad\xae\xae
+
 /-- End of testinput15 --/
diff --git a/pcre/testdata/testinput8 b/pcre/testdata/testinput8
index 7f8fa8292c5..e931410479f 100644
--- a/pcre/testdata/testinput8
+++ b/pcre/testdata/testinput8
@@ -4845,4 +4845,7 @@
     aaa\D
     a\D
 
+/(02-)?[0-9]{3}-[0-9]{3}/
+    02-123-123
+
 /-- End of testinput8 --/
diff --git a/pcre/testdata/testoutput1 b/pcre/testdata/testoutput1
index a2b3cffe9d4..eff8ecc948c 100644
--- a/pcre/testdata/testoutput1
+++ b/pcre/testdata/testoutput1
@@ -9442,4 +9442,8 @@ No match
     \  X
  0: X
      
+/X+(?#comment)?/
+    >XXX<
+ 0: X
+
 /-- End of testinput1 --/
diff --git a/pcre/testdata/testoutput12 b/pcre/testdata/testoutput12
index 87911086f49..7632c4e5801 100644
--- a/pcre/testdata/testoutput12
+++ b/pcre/testdata/testoutput12
@@ -201,4 +201,6 @@ No match, mark = m (JIT)
     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaabax
 Error -8 (match limit exceeded)
 
+/((?(?!))x)(?'name')(?1)/S++
+
 /-- End of testinput12 --/
diff --git a/pcre/testdata/testoutput15 b/pcre/testdata/testoutput15
index bad2807c2fd..e4e123c3983 100644
--- a/pcre/testdata/testoutput15
+++ b/pcre/testdata/testoutput15
@@ -1136,4 +1136,9 @@ Failed: setting UTF is disabled by the application at offset 0
 /abc/89
 Failed: setting UTF is disabled by the application at offset 0
 
+//8+L
+    \xf1\xad\xae\xae
+ 0: 
+ 0+ \x{6dbae}
+
 /-- End of testinput15 --/
diff --git a/pcre/testdata/testoutput8 b/pcre/testdata/testoutput8
index 17b667a980c..4984376d3c8 100644
--- a/pcre/testdata/testoutput8
+++ b/pcre/testdata/testoutput8
@@ -7801,4 +7801,8 @@ No match
 ** Show all captures ignored after DFA matching
  0: a
 
+/(02-)?[0-9]{3}-[0-9]{3}/
+    02-123-123
+ 0: 02-123-123
+
 /-- End of testinput8 --/