Hello.
To emulate EBCDIC platform, I ran a codelet like below
with #define EBCDIC.
for (uv = 0; uv <= PERL_UNICODE_MAX; uv++) {
memzero(buff,UTF8_MAXBYTES);
d = buff;
d = uvuni_to_utf8(d, uv);
if (!is_utf8_string(buff, d - buff))
++fail;
}
Even though IS_UTF8_CHAR is removed, it still fails in 64 cases
within uv = 0x0000..0x10ffff [uv failed are 00A0..00BF (start = 0x80)
and 0260..027F (start = 0xa0) ]
Results
(1) perl-current:
0x3ff60 failures in uv = 0x0000..0x10ffff.
0x443ff60 failures in uv = 0x0000..0x7fffffff.
(2) change utf8.h but not change utf8.c:
0x40 failures in uv = 0x0000..0x10ffff.
0x4400040 failures in uv = 0x0000..0x7fffffff.
(3) change both utf8.h and is_utf8_char_slow (see a patch below):
no failure in uv = 0x0000..0x10ffff.
no failure in uv = 0x0000..0x7fffffff.
I found that is_utf8_char_slow() in utf.c tries to get a UTF-EBCDIC
value without the conversion of the start octet from UTF-EBCDIC
to I8-sequence. The above failures are due to fail to apply
NATIVE_TO_UTF() to the start octet.
Regards,
sadahiro tomoyuki
! utf8.c utf8.h
diff -ur perl~/utf8.c perl/utf8.c
--- perl~/utf8.c Tue Jul 19 00:53:16 2005
+++ perl/utf8.c Sun Oct 02 16:18:56 2005
@@ -209,6 +209,9 @@
slen = len - 1;
s++;
+#ifdef EBCDIC
+ u = NATIVE_TO_UTF(u);
+#endif
u &= UTF_START_MASK(len);
uv = u;
ouv = uv;
diff -ur perl~/utf8.h perl/utf8.h
--- perl~/utf8.h Wed Jun 08 00:04:28 2005
+++ perl/utf8.h Sun Oct 02 15:47:26 2005
@@ -258,6 +258,9 @@
#endif
#define SHARP_S_SKIP 2
+#ifdef EBCDIC
+/* IS_UTF8_CHAR() is not ported to EBCDIC */
+#else
#define IS_UTF8_CHAR_1(p) \
((p)[0] <= 0x7F)
#define IS_UTF8_CHAR_2(p) \
@@ -329,3 +332,4 @@
#define IS_UTF8_CHAR_FAST(n) ((n) <= 4)
+#endif /* IS_UTF8_CHAR() for UTF-8 */
