/* Convert branch sequences to more efficient trie ops? */
#define PERL_ENABLE_TRIE_OPTIMISATION 1
-/* Be really agressive about optimising patterns with trie sequences? */
+/* Be really aggressive about optimising patterns with trie sequences? */
#define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1
/* Use old style unicode mappings for perl and posix character classes
* both match or neither match.
* NOTE: Disabling this will cause various backwards compatibility issues to rear
* their head, and tests to fail. However it will make the charclass behaviour
- * consistant regardless of internal string type, and make character class inversions
- * consistant. The tests that fail in the regex engine are basically broken tests.
+ * consistent regardless of internal string type, and make character class inversions
+ * consistent. The tests that fail in the regex engine are basically broken tests.
*
* Personally I think 5.12 should disable this for sure. Its a bit more debatable for
* 5.10, so for now im leaving it enabled.
/* Flags for node->flags of ANYOF */
-#define ANYOF_LOCALE 0x01
-#define ANYOF_FOLD 0x02
-#define ANYOF_INVERT 0x04
+#define ANYOF_LOCALE 0x01
+
+/* The fold is calculated and stored in the bitmap where possible at compile
+ * time. However there are two cases where it isn't possible. These share
+ * this bit: 1) under locale, where the actual folding varies depending on
+ * what the locale is at the time of execution; and 2) where the folding is
+ * specified in a swash, not the bitmap, such as characters which aren't
+ * specified in the bitmap, or properties that aren't looked at at compile time
+ */
+#define ANYOF_LOC_NONBITMAP_FOLD 0x02
+
+#define ANYOF_INVERT 0x04
/* CLASS is never set unless LOCALE is too: has runtime \d, \w, [:posix:], ... */
-/* For now, set it always when LOCALE is set, to save a bit for other uses. */
-#define ANYOF_CLASS ANYOF_LOCALE
+#define ANYOF_CLASS 0x08
#define ANYOF_LARGE ANYOF_CLASS /* Same; name retained for back compat */
-/* EOS used for regstclass only */
-#define ANYOF_EOS 0x10 /* Can match an empty string too */
+/* Can match something outside the bitmap that is expressible only in utf8 */
+#define ANYOF_UTF8 0x10
+
+/* Can match something outside the bitmap that isn't in utf8 */
+#define ANYOF_NONBITMAP_NON_UTF8 0x20
/* Set if the bitmap doesn't fully represent what this node can match */
-#define ANYOF_NONBITMAP 0x20
+#define ANYOF_NONBITMAP (ANYOF_UTF8|ANYOF_NONBITMAP_NON_UTF8)
#define ANYOF_UNICODE ANYOF_NONBITMAP /* old name, for back compat */
-#define ANYOF_UNICODE_ALL 0x40 /* Matches 0x100 - infinity */
+/* 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
#define ANYOF_CLASS_CLEAR(p, c) (ANYOF_CLASS_BYTE(p, c) &= ~ANYOF_BIT(c))
#define ANYOF_CLASS_TEST(p, c) (ANYOF_CLASS_BYTE(p, c) & ANYOF_BIT(c))
-/* Quicker way to see if there are actually any tests. This is because
- * currently the set of tests can be empty even when the class bitmap is
- * allocated */
-#define ANYOF_CLASS_TEST_ANY_SET(p) /* assumes sizeof(p) = 4 */ \
- memNE (((struct regnode_charclass_class*)(p))->classflags, "0000", ANYOF_CLASS_SIZE)
-
#define ANYOF_CLASS_ZERO(ret) Zero(((struct regnode_charclass_class*)(ret))->classflags, ANYOF_CLASSBITMAP_SIZE, char)
#define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
#define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode))
#define ANYOF_CLASS_SKIP ((ANYOF_CLASS_SIZE - 1)/sizeof(regnode))
-#define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP)
+
+/* The class bit can be set to the locale one if necessary to save bits at the
+ * expense of having locale ANYOF nodes always have a class bit map, and hence
+ * take up extra space. This allows convenient changing it as development
+ * proceeds on this */
+#if ANYOF_CLASS == ANYOF_LOCALE
+# undef ANYOF_CLASS_ADD_SKIP
+# define ANYOF_ADD_LOC_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP)
+
+ /* Quicker way to see if there are actually any tests. This is because
+ * currently the set of tests can be empty even when the class bitmap is
+ * allocated */
+# if ANYOF_CLASSBITMAP_SIZE != 4
+# error ANYOF_CLASSBITMAP_SIZE is expected to be 4
+# endif
+# define ANYOF_CLASS_TEST_ANY_SET(p) /* assumes sizeof(p) = 4 */ \
+ memNE (((struct regnode_charclass_class*)(p))->classflags, \
+ "\0\0\0\0", ANYOF_CLASSBITMAP_SIZE)
+#else
+# define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP)
+# undef ANYOF_ADD_LOC_SKIP
+# define ANYOF_CLASS_TEST_ANY_SET(p) (ANYOF_FLAGS(p) & ANYOF_CLASS)
+#endif
/*
};
typedef struct _reg_ac_data reg_ac_data;
-/* ANY_BIT doesnt use the structure, so we can borrow it here.
+/* ANY_BIT doesn't use the structure, so we can borrow it here.
This is simpler than refactoring all of it as wed end up with
three different sets... */
const char * const rpv = \
pv_pretty((dsv), (pv), (l), (m), \
PL_colors[(c1)],PL_colors[(c2)], \
- PERL_PV_ESCAPE_RE |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
+ PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
const int rlen = SvCUR(dsv)
#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
const char * const rpv = \
pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
PL_colors[(c1)],PL_colors[(c2)], \
- PERL_PV_ESCAPE_RE |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
+ PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
const char * const rpv = \
pv_pretty((dsv), (pv), (l), (m), \
PL_colors[0], PL_colors[1], \
- ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_PRETTY_ELLIPSES | \
+ ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
)