* src/regex-emacs.c (analyze_first): Fix incorrect optimization

The optimization was incorrect in a particular corner case.
In this fix I just disable it conservatively for more cases because
it's not obvious how to fix it while preserving the "good" cases.

We may find a better fix by using an approach like the one
in `mutually_exhaustive_aux`, but for now this is good enough,
especially since \{..\} repetitions are not used very frequently.

* test/src/regex-resources/PTESTS: New test.

2 files changed, 5 insertions, 2 deletions

diff --git a/src/regex-emacs.c b/src/regex-emacs.c
index 338323cf79e..ae82dd63917 100644
--- a/src/regex-emacs.c
+++ b/src/regex-emacs.c
@@ -3086,8 +3086,10 @@ analyze_first (re_char *p, re_char *pend, char *fastmap, bool multibyte)
           /* We only care about one iteration of the loop, so we don't
              need to consider the case where this behaves like an
              on_failure_jump.  */
-          continue;
-
+          /* FIXME: Sadly, the above is not true when the loop's body
+             can match the empty string :-(  */
+          /* continue; */
+          return -1;
 
         case set_number_at:
           p += 4;
diff --git a/test/src/regex-resources/PTESTS b/test/src/regex-resources/PTESTS
index 68acc314d37..59dd4b3bc21 100644
--- a/test/src/regex-resources/PTESTS
+++ b/test/src/regex-resources/PTESTS
@@ -269,6 +269,7 @@
 #W the expected result for \([a-c]*\)\{2,\} is failure which isn't correct
 1¦3¦\([a-c]*\)\{2,\}¦abcdefg¦
 1¦3¦\([a-c]*\)\{1,\}¦abcdefg¦
+0¦0¦\([a-c]*\)\{2,\}¦gabcdefg¦
 -1¦-1¦a\{64,\}¦aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa¦
 # GA142
 1¦3¦a\{2,3\}¦aaaa¦