Almost drop direct use of LEVELS

getCharCE appeared in R-2.7, making it possible to check for strings
_marked_ as UTF-8 or Latin-1.

There is no marking as ASCII, so fixing IS_ASCII will have to wait for R
>= 4.5.

Author: aitap

src/bmerge.c

@@ -17,9 +17,6 @@ Differences over standard binary search (e.g. bsearch in stdlib.h) :
   o non equi joins (no != yet) since 1.9.8
 */
 
-#define ENC_KNOWN(x) (LEVELS(x) & 12)
-// 12 = LATIN1_MASK (1<<2) | UTF8_MASK (1<<3)  // Would use these definitions from Defn.h, but that appears to be private to R. Hence 12.
-
 #define EQ 1
 #define LE 2
 #define LT 3

src/data.table.h

@@ -33,9 +33,10 @@
 // #include <signal.h> // the debugging machinery + breakpoint aidee
 // raise(SIGINT);
 
-#define IS_UTF8(x)  (LEVELS(x) & 8)
-#define IS_ASCII(x) (LEVELS(x) & 64)
-#define IS_LATIN(x) (LEVELS(x) & 4)
+/* we mean the encoding bits, not CE_NATIVE in a UTF-8 locale */
+#define IS_UTF8(x)  (getCharCE(x) == CE_UTF8)
+#define IS_LATIN(x) (getCharCE(x) == CE_LATIN1)
+#define IS_ASCII(x) (LEVELS(x) & 64) // API expected in R >= 4.5
 #define IS_TRUE(x)  (TYPEOF(x)==LGLSXP && LENGTH(x)==1 && LOGICAL(x)[0]==TRUE)
 #define IS_FALSE(x) (TYPEOF(x)==LGLSXP && LENGTH(x)==1 && LOGICAL(x)[0]==FALSE)
 #define IS_TRUE_OR_FALSE(x) (TYPEOF(x)==LGLSXP && LENGTH(x)==1 && LOGICAL(x)[0]!=NA_LOGICAL)