regcomp: Share two bits in ANYOF flags

It turns out that the INVERT and EOS flags can actually share the same bit, as
explained in the comments, freeing up a bit for other uses.  No code changes
need be made.

diff --git a/regcomp.c b/regcomp.c
index b39cf5c..fa8f44b 100644 (file)
--- a/regcomp.c
+++ b/regcomp.c
@@ -4985,6 +4985,9 @@ reStudy:
        if (ri->regstclass
            && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
            ri->regstclass = NULL;
+
+       /* If the synthetic start class were to ever be used when EOS is set,
+        * that bit would have to be cleared, as it is shared with another */
        if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
            && stclass_flag
            && !(data.start_class->flags & ANYOF_EOS)
@@ -5057,6 +5060,9 @@ reStudy:
 
        r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
                = r->float_substr = r->float_utf8 = NULL;
+
+       /* If the synthetic start class were to ever be used when EOS is set,
+        * that bit would have to be cleared, as it is shared with another */
        if (!(data.start_class->flags & ANYOF_EOS)
            && !cl_is_anything(data.start_class))
        {

diff --git a/regcomp.h b/regcomp.h
index 6adb47a..0dc4374 100644 (file)
--- a/regcomp.h
+++ b/regcomp.h
@@ -313,11 +313,15 @@ struct regnode_charclass_class {
 #define USE_UNI                0x01
 
 /* Flags for node->flags of ANYOF.  These are in short supply, so some games
- * are done to share them, as described below.  For flags that are applicable
- * to the synthetic start class (stc) only, with some work, they could be put
- * in the next-node field, or in an unused bit of the classflags field.  Once
- * the planned change to compile all the above-latin1 code points is done, then
- * the UNICODE_ALL bit can be freed up */
+ * are done to share them, as described below.  If necessary, the ANYOF_LOCALE
+ * and ANYOF_CLASS bits could be shared with a space penalty for locale nodes
+ * (and the code at the time this comment was written, is written so that all
+ * that is necessary to make the change would be to redefine the ANYOF_CLASS
+ * define).  Once the planned change to compile all the above-latin1 code points
+ * is done, then the UNICODE_ALL bit can be freed up.  If flags need to be
+ * added that are applicable to the synthetic start class only, with some work,
+ * they could be put in the next-node field, or in an unused bit of the
+ * classflags field. */
 
 #define ANYOF_LOCALE            0x01
 
@@ -332,6 +336,14 @@ struct regnode_charclass_class {
 
 #define ANYOF_INVERT            0x04
 
+/* EOS, meaning that it can match an empty string too, is used for the
+ * synthetic start class (ssc) only.  It can share the INVERT bit, as the ssc
+ * is never inverted.  The bit just needs to be turned off before regexec.c
+ * gets a hold of it so that regexec.c doesn't think it's inverted, but this
+ * happens automatically, as if the ssc can match an EOS, the ssc is discarded,
+ * and never passed to regexec.c */
+#define ANYOF_EOS              ANYOF_INVERT
+
 /* CLASS is never set unless LOCALE is too: has runtime \d, \w, [:posix:], ...
  * The non-locale ones are resolved at compile-time */
 #define ANYOF_CLASS     0x08
@@ -350,9 +362,6 @@ struct regnode_charclass_class {
 /* Matches every code point 0x100 and above*/
 #define ANYOF_UNICODE_ALL      0x40
 
-/* EOS used for regstclass only */
-#define ANYOF_EOS              0x80    /* Can match an empty string too */
-
 #define ANYOF_FLAGS_ALL                0xff
 
 /* Character classes for node->classflags of ANYOF */