New implementation of mb_convert_kana

mb_convert_kana now uses the new text encoding conversion
filters. Microbenchmarking shows speed gains of 50%-150%
across various text encodings and input string lengths.

The behavior is the same as the old mb_convert_kana
except for one fix: if the 'zero codepoint' U+0000 appeared
in the input, the old implementation would sometimes drop
it, not passing it through to the output. This is now
fixed.

Author: alexdowad

ext/mbstring/config.m4

@@ -118,7 +118,6 @@ AC_DEFUN([PHP_MBSTRING_SETUP_LIBMBFL], [
     libmbfl/filters/mbfilter_sjis_mobile.c
     libmbfl/filters/mbfilter_sjis_mac.c
     libmbfl/filters/mbfilter_sjis_2004.c
-    libmbfl/filters/mbfilter_tl_jisx0201_jisx0208.c
     libmbfl/filters/mbfilter_ucs2.c
     libmbfl/filters/mbfilter_ucs4.c
     libmbfl/filters/mbfilter_uhc.c

ext/mbstring/config.w32

@@ -28,8 +28,7 @@ if (PHP_MBSTRING != "no") {
 			mbfilter_utf8_mobile.c mbfilter_uuencode.c \
 			mbfilter_cp5022x.c mbfilter_sjis_mobile.c \
 			mbfilter_sjis_mac.c \
-			mbfilter_iso2022jp_mobile.c mbfilter_singlebyte.c \
-			mbfilter_tl_jisx0201_jisx0208.c", "mbstring");
+			mbfilter_iso2022jp_mobile.c mbfilter_singlebyte.c", "mbstring");
 
 		ADD_SOURCES("ext/mbstring/libmbfl/mbfl", "mbfilter.c mbfilter_8bit.c \
 			mbfilter_pass.c mbfilter_wchar.c mbfl_convert.c mbfl_encoding.c \

ext/mbstring/libmbfl/filters/mbfilter_cp5022x.c

@@ -25,11 +25,11 @@
 #include "mbfilter.h"
 #include "mbfilter_cp5022x.h"
 #include "mbfilter_jis.h"
-#include "mbfilter_tl_jisx0201_jisx0208.h"
 
 #include "unicode_table_cp932_ext.h"
 #include "unicode_table_jis.h"
 #include "cp932_table.h"
+#include "translit_kana_jisx0201_jisx0208.h"
 
 static int mbfl_filt_conv_cp5022x_wchar_flush(mbfl_convert_filter *filter);
 static int mbfl_filt_conv_wchar_cp50220_flush(mbfl_convert_filter *filter);
@@ -40,6 +40,9 @@ static void mb_wchar_to_cp50220(uint32_t *in, size_t len, mb_convert_buf *buf, b
 static void mb_wchar_to_cp50221(uint32_t *in, size_t len, mb_convert_buf *buf, bool end);
 static void mb_wchar_to_cp50222(uint32_t *in, size_t len, mb_convert_buf *buf, bool end);
 
+/* See mbstring.c */
+uint32_t mb_convert_kana_codepoint(uint32_t c, uint32_t next, bool *consumed, uint32_t *second, int mode);
+
 /* Previously, a dubious 'encoding' called 'cp50220raw' was supported
  * This was just CP50220, but the implementation was less strict regarding
  * invalid characters; it would silently pass some through
@@ -336,7 +339,7 @@ static int mbfl_filt_conv_wchar_cp50220(int c, mbfl_convert_filter *filter)
 	bool consumed = false;
 
 	if (filter->cache) {
-		int s = mbfl_convert_kana(filter->cache, c, &consumed, NULL, mode);
+		int s = mb_convert_kana_codepoint(filter->cache, c, &consumed, NULL, mode);
 		filter->cache = consumed ? 0 : c;
 		/* Terrible hack to get CP50220 to emit error markers in the proper
 		 * position, not reordering them with subsequent characters */
@@ -359,7 +362,7 @@ static int mbfl_filt_conv_wchar_cp50220_flush(mbfl_convert_filter *filter)
 	int mode = MBFL_HAN2ZEN_KATAKANA | MBFL_HAN2ZEN_GLUE;
 
 	if (filter->cache) {
-		int s = mbfl_convert_kana(filter->cache, 0, NULL, NULL, mode);
+		int s = mb_convert_kana_codepoint(filter->cache, 0, NULL, NULL, mode);
 		mbfl_filt_conv_wchar_cp50221(s, filter);
 		filter->cache = 0;
 	}
@@ -866,7 +869,7 @@ static void mb_wchar_to_cp50220(uint32_t *in, size_t len, mb_convert_buf *buf, b
 			buf->state |= w << 8;
 			break;
 		} else {
-			w = mbfl_convert_kana(w, len ? *in : 0, &consumed, NULL, MBFL_HAN2ZEN_KATAKANA | MBFL_HAN2ZEN_GLUE);
+			w = mb_convert_kana_codepoint(w, len ? *in : 0, &consumed, NULL, MBFL_HAN2ZEN_KATAKANA | MBFL_HAN2ZEN_GLUE);
 		}
 
 		if (consumed) {