dfa: process all MBCSET constructs via glibc's matcher

The DFA matcher does not support collating symbols or equivalence
classes, so ensure that any MBCSET reference is handled by the glibc
matcher.  dfa.c already handled this in one case, but not the other,
so that a command like "printf '\0' |src/grep -aE '^\s?$'" would
mistakenly end up using dfa.c's match_mb_charset function rather
than glibc's matcher.

* src/dfa.c (dfaexec_main): Move that code into the
State_transition macro.  This renders the match_mb_charset
unused by grep.
* tests/multibyte-white-space: Add a test to exercise the
just-rendered-inaccessible code path.

2 files changed, 20 insertions, 10 deletions

diff --git a/src/dfa.c b/src/dfa.c
index 58a4b832..de836891 100644
--- a/src/dfa.c
+++ b/src/dfa.c
@@ -3338,20 +3338,20 @@ dfaexec_main (struct dfa *d, char const *begin, char *end,
                   continue;
                 }
 
-              /* Falling back to the glibc matcher in this case gives
-                 better performance (up to 25% better on [a-z], for
-                 example) and enables support for collating symbols and
-                 equivalence classes.  */
-              if (d->states[s].has_mbcset && backref)
-                {
-                  *backref = 1;
-                  goto done;
-                }
-
               /* The following code is used twice.
                  Use a macro to avoid the risk that they diverge.  */
 #define State_transition()                                              \
   do {                                                                  \
+              /* Falling back to the glibc matcher in this case gives   \
+                 better performance (up to 25% better on [a-z], for     \
+                 example) and enables support for collating symbols and \
+                 equivalence classes.  */                               \
+              if (d->states[s].has_mbcset && backref)                   \
+                {                                                       \
+                  *backref = 1;                                         \
+                  goto done;                                            \
+                }                                                       \
+                                                                        \
               /* Can match with a multibyte character (and multi-character \
                  collating element).  Transition table might be updated.  */ \
               s = transit_state (d, s, &p, (unsigned char *) end);      \
diff --git a/tests/multibyte-white-space b/tests/multibyte-white-space
index c9b3d1fa..58166431 100755
--- a/tests/multibyte-white-space
+++ b/tests/multibyte-white-space
@@ -73,4 +73,14 @@ for i in $utf8_space_characters; do
       || { warn_ "$i vs. \\S FAILED"; fail=1; }
 done
 
+
+# This is a separate test, only nominally related to \s.
+# It is solely to get coverage of a code path (exercising dfa.c's
+# match_mb_charset function) that would have otherwise been untouched.
+# However, as of the change-set adding this new test, match_mb_charset
+# is unreachable via grep.
+printf '\0' | grep -aE '^\s?$' > out 2>&1
+test $? = 1 || fail=1
+compare /dev/null out
+
 Exit $fail