ARROW-7404: [C++][Gandiva] Fix utf8 char length error on Arm64

cyb70289 · Pindikura Ravindra · commit 0d00decd049c · 2019-12-17T10:03:13.000+05:30
Current code checks if a UTF-8 eight-bit code unit is within 0x00~0x7F by "if (c >= 0)", where c is defined as "char". This checking assumes char is always signed, which is not true[1]. On Arm64, char is unsigned by default and causes some Gandiva unit tests fail. Fix it by casting to "signed char" explicitly. [1] Cited from https://en.cppreference.com/w/cpp/language/types The signedness of char depends on the compiler and the target platform: the defaults for ARM and PowerPC are typically unsigned, the defaults for x86 and x64 are typically signed. Signed-off-by: Yibo Cai <yibo.cai@arm.com> Closes apache#6043 from cyb70289/utf8_char_len and squashes the following commits: a18f43d <Yibo Cai> ARROW-7404: Fix utf8 char length error on Arm64 Authored-by: Yibo Cai <yibo.cai@arm.com> Signed-off-by: Pindikura Ravindra <ravindra@dremio.com>
diff --git a/cpp/src/gandiva/precompiled/string_ops.cc b/cpp/src/gandiva/precompiled/string_ops.cc
@@ -92,7 +92,7 @@ bool ends_with_utf8_utf8(const char* data, int32 data_len, const char* suffix,
 
 FORCE_INLINE
 int32 utf8_char_length(char c) {
-  if (c >= 0) {  // 1-byte char
+  if ((signed char)c >= 0) {  // 1-byte char (0x00 ~ 0x7F)
     return 1;
   } else if ((c & 0xE0) == 0xC0) {  // 2-byte char
     return 2;
