ack/lang/m2/comp/class.h
1986-09-25 19:39:06 +00:00

39 lines
1.5 KiB
C

/* U S E O F C H A R A C T E R C L A S S E S */
/* $Header$ */
/* As a starter, chars are divided into classes, according to which
token they can be the start of.
At present such a class number is supposed to fit in 4 bits.
*/
#define class(ch) (tkclass[ch])
/* Being the start of a token is, fortunately, a mutual exclusive
property, so, as there are less than 16 classes they can be
packed in 4 bits.
*/
#define STSKIP 0 /* spaces and so on: skipped characters */
#define STNL 1 /* newline character(s): update linenumber etc. */
#define STGARB 2 /* garbage ascii character: not allowed */
#define STSIMP 3 /* this character can occur as token */
#define STCOMP 4 /* this one can start a compound token */
#define STIDF 5 /* being the initial character of an identifier */
#define STCHAR 6 /* the starter of a character constant */
#define STSTR 7 /* the starter of a string */
#define STNUM 8 /* the starter of a numeric constant */
#define STEOI 9 /* End-Of-Information mark */
/* But occurring inside a token is not, so we need 1 bit for each
class. This is implemented as a collection of tables to speed up
the decision whether a character has a special meaning.
*/
#define in_idf(ch) ((unsigned)ch < 0177 && inidf[ch])
#define is_oct(ch) ((unsigned)ch < 0177 && isoct[ch])
#define is_dig(ch) ((unsigned)ch < 0177 && isdig[ch])
#define is_hex(ch) ((unsigned)ch < 0177 && ishex[ch])
extern char tkclass[];
extern char inidf[], isoct[], isdig[], ishex[];