MDEV-8102 REGEXP function fails to match hex values when expression is stored as a variable

We don't fix the bug itself, we just make regex functions display errors
returned from pcre_exec() as MariaDB warnings.

mysql-test/r/func_regexp_pcre.result

@@ -845,3 +845,32 @@ SET default_regex_flags=DEFAULT;
 SET default_regex_flags=DEFAULT;
 #
 # MDEV-6965 non-captured group \2 in regexp_replace
+#
+SELECT REGEXP_REPLACE('1 foo and bar', '(\\d+) foo and (\\d+ )?bar', '\\1 this and \\2that');
+REGEXP_REPLACE('1 foo and bar', '(\\d+) foo and (\\d+ )?bar', '\\1 this and \\2that')
+1 this and that
+#
+# MDEV-8102 REGEXP function fails to match hex values when expression is stored as a variable
+#
+# Testing a warning
+SET NAMES latin1;
+SET @regCheck= '\\xE0\\x01';
+SELECT 0xE001 REGEXP @regCheck;
+0xE001 REGEXP @regCheck
+0
+Warnings:
+Warning	1139	Got error 'pcre_exec: Invalid utf8 byte sequence in the subject string' from regexp
+# Testing workaround N1: This makes the pattern to be a binary string:
+SET NAMES latin1;
+SET @regCheck= X'E001';
+SELECT 0xE001 REGEXP @regCheck;
+0xE001 REGEXP @regCheck
+1
+# Testing workaround N2: This also makes the pattern to be a binary string, using a different syntax:
+SET NAMES latin1;
+SET @regCheck= _binary '\\xE0\\x01';
+SELECT 0xE001 REGEXP @regCheck;
+0xE001 REGEXP @regCheck
+1
+# Testing workarond N3: This makes derivation of the subject string stronger (IMLICIT instead of COERCIBLE)
+SET NAMES latin1;

mysql-test/r/func_regexp_pcre_debug.result

@@ -0,0 +1,10 @@
+SET debug_dbug='+d,pcre_exec_error_123';
+SELECT 'a' RLIKE 'a';
+'a' RLIKE 'a'
+0
+Warnings:
+Warning	1139	Got error 'pcre_exec: Internal error (-123)' from regexp
+SET debug_dbug='';
+SELECT 'a' RLIKE 'a';
+'a' RLIKE 'a'
+1

mysql-test/t/func_regexp_pcre.test

@@ -402,3 +402,27 @@ SET default_regex_flags=DEFAULT;
 --echo # MDEV-6965 non-captured group \2 in regexp_replace
 --echo #
 SELECT REGEXP_REPLACE('1 foo and bar', '(\\d+) foo and (\\d+ )?bar', '\\1 this and \\2that');
+
+--echo #
+--echo # MDEV-8102 REGEXP function fails to match hex values when expression is stored as a variable
+--echo #
+
+--echo # Testing a warning
+SET NAMES latin1;
+SET @regCheck= '\\xE0\\x01';
+SELECT 0xE001 REGEXP @regCheck;
+
+--echo # Testing workaround N1: This makes the pattern to be a binary string:
+SET NAMES latin1;
+SET @regCheck= X'E001';
+SELECT 0xE001 REGEXP @regCheck;
+
+--echo # Testing workaround N2: This also makes the pattern to be a binary string, using a different syntax:
+SET NAMES latin1;
+SET @regCheck= _binary '\\xE0\\x01';
+SELECT 0xE001 REGEXP @regCheck;
+
+--echo # Testing workarond N3: This makes derivation of the subject string stronger (IMLICIT instead of COERCIBLE)
+SET NAMES latin1;
+SET @regCheck= '\\xE0\\x01';
+SELECT CAST(0xE001 AS BINARY) REGEXP @regCheck;

mysql-test/t/func_regexp_pcre_debug.test

@@ -0,0 +1,6 @@
+--source include/have_debug.inc
+
+SET debug_dbug='+d,pcre_exec_error_123';
+SELECT 'a' RLIKE 'a';
+SET debug_dbug='';
+SELECT 'a' RLIKE 'a';

sql/item_cmpfunc.cc

@@ -5160,10 +5160,65 @@ bool Regexp_processor_pcre::compile(Item *item, bool send_error)
 }
 
 
+/**
+  Send a warning explaining an error code returned by pcre_exec().
+*/
+void Regexp_processor_pcre::pcre_exec_warn(int rc) const
+{
+  char buf[64];
+  const char *errmsg= NULL;
+  /*
+    Make a descriptive message only for those pcre_exec() error codes
+    that can actually happen in MariaDB.
+  */
+  switch (rc)
+  {
+  case PCRE_ERROR_NOMEMORY:
+    errmsg= "pcre_exec: Out of memory";
+    break;
+  case PCRE_ERROR_BADUTF8:
+    errmsg= "pcre_exec: Invalid utf8 byte sequence in the subject string";
+    break;
+  case PCRE_ERROR_RECURSELOOP:
+    errmsg= "pcre_exec: Recursion loop detected";
+    break;
+  default:
+    /*
+      As other error codes should normally not happen,
+      we just report the error code without textual description
+      of the code.
+    */
+    my_snprintf(buf, sizeof(buf), "pcre_exec: Internal error (%d)", rc);
+    errmsg= buf;
+  }
+  push_warning_printf(current_thd, Sql_condition::WARN_LEVEL_WARN,
+                      ER_REGEXP_ERROR, ER(ER_REGEXP_ERROR), errmsg);
+}
+
+
+/**
+  Call pcre_exec() and send a warning if pcre_exec() returned with an error.
+*/
+int Regexp_processor_pcre::pcre_exec_with_warn(const pcre *code,
+                                               const pcre_extra *extra,
+                                               const char *subject,
+                                               int length, int startoffset,
+                                               int options, int *ovector,
+                                               int ovecsize)
+{
+  int rc= pcre_exec(code, extra, subject, length,
+                    startoffset, options, ovector, ovecsize);
+  DBUG_EXECUTE_IF("pcre_exec_error_123", rc= -123;);
+  if (rc < PCRE_ERROR_NOMATCH)
+    pcre_exec_warn(rc);
+  return rc;
+}
+
+
 bool Regexp_processor_pcre::exec(const char *str, int length, int offset)
 {
-  m_pcre_exec_rc= pcre_exec(m_pcre, NULL, str, length,
+  m_pcre_exec_rc= pcre_exec_with_warn(m_pcre, NULL, str, length, offset, 0,
-                            offset, 0, m_SubStrVec, m_subpatterns_needed * 3);
+                                      m_SubStrVec, m_subpatterns_needed * 3);
   return false;
 }
 
@@ -5173,8 +5228,10 @@ bool Regexp_processor_pcre::exec(String *str, int offset,
 {
   if (!(str= convert_if_needed(str, &subject_converter)))
     return true;
-  m_pcre_exec_rc= pcre_exec(m_pcre, NULL, str->c_ptr_safe(), str->length(),
+  m_pcre_exec_rc= pcre_exec_with_warn(m_pcre, NULL,
-                            offset, 0, m_SubStrVec, m_subpatterns_needed * 3);
+                                      str->c_ptr_safe(), str->length(),
+                                      offset, 0,
+                                      m_SubStrVec, m_subpatterns_needed * 3);
   if (m_pcre_exec_rc > 0)
   {
     uint i;

sql/item_cmpfunc.h

@@ -1549,6 +1549,10 @@ class Regexp_processor_pcre
   int m_pcre_exec_rc;
   int m_SubStrVec[30];
   uint m_subpatterns_needed;
+  void pcre_exec_warn(int rc) const;
+  int pcre_exec_with_warn(const pcre *code, const pcre_extra *extra,
+                          const char *subject, int length, int startoffset,
+                          int options, int *ovector, int ovecsize);
 public:
   String *convert_if_needed(String *src, String *converter);
   String subject_converter;