regex.c (mutually_exclusive_aux) <wordbound>: Remove optimization

Another case that was too optimistic.  Better use \> or \< rather
than \b if you want your regexp to be handled efficiently.

* src/regex-emacs.c (mutually_exclusive_aux) <wordbound>: Cancel optimization.
* test/src/regex-emacs-tests.el (regexp-tests-backtrack-optimization):
New test.

diff --git a/src/regex-emacs.c b/src/regex-emacs.c
index ae82dd63917e1056d0351ce349861e6771c7a8a2..ad140908609e728478043fd70afc3e53bbfad441 100644 (file)
--- a/src/regex-emacs.c
+++ b/src/regex-emacs.c
@@ -3874,9 +3874,22 @@ mutually_exclusive_aux (struct re_pattern_buffer *bufp, re_char *p1,
       return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
 
     case wordbound:
+      /* FIXME: This optimization seems correct after the first iteration
+         of the loop, but not for the very first :-(
+         IOW we'd need to pull out the first iteration and do:
+
+            syntaxspec w
+            on_failure_keep_string_jump end
+          loop:
+            syntaxspec w
+            goto loop
+          end:
+            wordbound
+
       return (((re_opcode_t) *p1 == notsyntaxspec
-              || (re_opcode_t) *p1 == syntaxspec)
-             && p1[1] == Sword);
+               || (re_opcode_t) *p1 == syntaxspec)
+              && p1[1] == Sword);  */
+      return false;
 
     case categoryspec:
       return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);

diff --git a/test/src/regex-emacs-tests.el b/test/src/regex-emacs-tests.el
index d78f70ba409cf034795e5ce1c1f94d77c6f01579..621e4dbe2c029683ac0d2345d00ffd4e3082a836 100644 (file)
--- a/test/src/regex-emacs-tests.el
+++ b/test/src/regex-emacs-tests.el
@@ -907,6 +907,7 @@ This evaluates the TESTS test cases from glibc."
     ;; Regression check for overly optimistic optimization.
     (should (eq 0 (string-match "\\(ca*\\|ab\\)+d" "cabd")))
     (should (string-match "\\(aa*\\|b\\)*c" "ababc"))
+    (should (string-match " \\sw*\\bfoo" " foo"))
     ))
 
 (ert-deftest regexp-tests-zero-width-assertion-repetition ()