X-Git-Url: https://git.draconx.ca/gitweb/cdecl99.git/blobdiff_plain/a9ae295484868287ff1b8f2d613fa4c010c06891..e745d25d7f109057962c6904be98f5f35e874ecc:/src/scan.l

diff --git a/src/scan.l b/src/scan.l
index 5610dc3..7549e6f 100644
--- a/src/scan.l
+++ b/src/scan.l
@@ -19,10 +19,13 @@
 
 #include <config.h>
 #include "parse.h"
+
+#define YY_NO_INPUT 1
+#define YY_NO_UNPUT 1
 }
 
 %option nodefault noyywrap bison-locations reentrant never-interactive
-%option extra-type="_Bool"
+%option extra-type="int"
 %option prefix="cdecl__yy"
 
 %{
@@ -45,68 +48,96 @@
 #define STRTOUMAX strtoul
 #endif
 
-#define dup_token() do { \
-	yylval->strval = malloc(yyleng+1); \
-	if (!yylval->strval) { \
-		cdecl__errmsg(CDECL__ENOMEM); \
-		return T_LEX_ERROR; \
-	} \
-	strcpy(yylval->strval, yytext); \
-} while(0)
-%}
+static char *to_octal(char *dst, unsigned val)
+{
+	unsigned i;
+
+	for (i = 0; i < 3; i++) {
+		*dst++ = '0' + ((val >> 6) & 7u);
+		val <<= 3;
+	}
+
+	return dst;
+}
+
+/*
+ * Convert a single character to a C-style character constant, including quote
+ * characters.  At most 7 bytes are written to the buffer for the longest
+ * octal encoding, e.g., '\177'
+ */
+static void to_readable_ch(char *dst, char c)
+{
+	unsigned char uc = c;
+	unsigned i;
+	char esc;
+
+	/*
+	 * The 7 standard C control characters are contiguous in ASCII,
+	 * permitting a simple and compact lookup table; separating their
+	 * handling from backslash and quote characters hopefully allows
+	 * the compiler to recognize that.
+	 */
+	switch (c) {
+	case '\a': i = 0; break;
+	case '\b': i = 1; break;
+	case '\t': i = 2; break;
+	case '\n': i = 3; break;
+	case '\v': i = 4; break;
+	case '\f': i = 5; break;
+	case '\r': i = 6; break;
+	default:   i = 7; break;
+	}
+	esc = "abtnvfr"[i];
+
+	/* Otherwise printable characters that should still be escaped. */
+	switch (c) {
+	case '\\': case '\'': esc = c; break;
+	}
+
+	*dst++ = '\'';
+	if (esc) {
+		*dst++ = '\\';
+		*dst++ = esc;
+	} else if (isprint(uc)) {
+		*dst++ = c;
+	} else {
+		*dst++ = '\\';
+		dst = to_octal(dst, uc);
+	}
+	*dst++ = '\'';
+	*dst++ = 0;
+}
 
-%s ENGLISH
+%}
 
-IDENT [_[:alpha:]][_[:alnum:]]*
+IDENT [_[:alpha:]][-_[:alnum:]]*
 INTEGER 0x[[:xdigit:]]+|0[0-7]+|[[:digit:]]+
 
 %%
 
 %{
-	if (yyextra) {
-		yyextra = 0;
-		BEGIN(ENGLISH);
-		return T_ENGLISH;
-	}
+	char *c;
 %}
 
-"..." return T_ELLIPSIS;
-";"   return T_SEMICOLON;
-"*"   return T_ASTERISK;
-"("   return T_LPAREN;
-")"   return T_RPAREN;
-"["   return T_LBRACKET;
-"]"   return T_RBRACKET;
-","   return T_COMMA;
-
-"typedef"    return T_TYPEDEF;
-"extern"     return T_EXTERN;
-"static"     return T_STATIC;
-"auto"       return T_AUTO;
-"register"   return T_REGISTER;
-
-"restrict"   return T_RESTRICT;
-"volatile"   return T_VOLATILE;
-"const"      return T_CONST;
-
-"inline"     return T_INLINE;
-
-"void"       return T_VOID;
-"char"       return T_CHAR;
-"short"      return T_SHORT;
-"int"        return T_INT;
-"long"       return T_LONG;
-"float"      return T_FLOAT;
-"double"     return T_DOUBLE;
-"signed"     return T_SIGNED;
-"unsigned"   return T_UNSIGNED;
-"_Bool"      return T_BOOL;
-"_Complex"   return T_COMPLEX;
-"_Imaginary" return T_IMAGINARY;
-
-"struct"     return T_STRUCT;
-"union"      return T_UNION;
-"enum"       return T_ENUM;
+"..."|[][;*(),] {
+	unsigned char *match;
+	static const unsigned char tab[2][8] = {
+		"*[](),.;",
+		{
+			PACK_TOKEN(T_ASTERISK),
+			PACK_TOKEN(T_LBRACKET),
+			PACK_TOKEN(T_RBRACKET),
+			PACK_TOKEN(T_LPAREN),
+			PACK_TOKEN(T_RPAREN),
+			PACK_TOKEN(T_COMMA),
+			PACK_TOKEN(T_ELLIPSIS),
+			PACK_TOKEN(T_SEMICOLON)
+		}
+	};
+
+	match = memchr(&tab, yytext[0], sizeof tab[0]);
+	return UNPACK_TOKEN(match[sizeof tab[0]]);
+}
 
 {INTEGER} {
 	char *end;
@@ -125,46 +156,42 @@ INTEGER 0x[[:xdigit:]]+|0[0-7]+|[[:digit:]]+
 	return T_UINT;
 }
 
-<ENGLISH>{
-	"variable-length" return T_VLA;
-	"type"            return T_TYPE;
-	"declare"         return T_DECLARE;
-	"pointer"         return T_POINTER;
-	"function"        return T_FUNCTION;
-	"returning"       return T_RETURNING;
-	"array"           return T_ARRAY;
-	"to"              return T_TO;
-	"of"              return T_OF;
-	"as"              return T_AS;
+{IDENT} {
+	unsigned x = cdecl__to_keyword(yytext, yyleng, yyextra);
+	int tok;
+
+	yylval->spectype = UNPACK_SPEC(x & 0xff);
+	if ((tok = (x >> 8)) == PACK_TOKEN(T_IDENT)) {
+		/*
+		 * Our IDENT pattern includes hyphens so we can match
+		 * "variable-length" as a keyword.  In all other cases a
+		 * hyphen is an error.
+		 *
+		 * We could use yyless to re-scan the hyphen and hit the
+		 * error catch-all, but jumping straight to the error code
+		 * seems to produce better results with gcc with no obvious
+		 * downsides.
+		 */
+#if 1
+		if ((c = strchr(yytext, '-')))
+			goto invalid_char;
+#else
+		yyless(strcspn(yytext, "-"));
+#endif
+		if (!(yylval->item = cdecl__alloc_item(yyleng+1)))
+			return T_LEX_ERROR; \
+		memcpy(yylval->item->s, yytext, yyleng+1);
+	}
+	return UNPACK_TOKEN(tok);
 }
 
-{IDENT} { dup_token(); return T_IDENT; }
-
 [[:space:]]+
 . {
-	char buf[5] = { yytext[0] };
-	unsigned char c = buf[0];
-
-	if (!isprint(c) || c == '\\' || c == '\'') {
-		/* Encode nonprinting characters with C-style escapes */
-		buf[0] = '\\';
-		switch (c) {
-		case '\a': buf[1] = 'a'; break;
-		case '\b': buf[1] = 'b'; break;
-		case '\f': buf[1] = 'f'; break;
-		case '\n': buf[1] = 'n'; break;
-		case '\r': buf[1] = 'r'; break;
-		case '\t': buf[1] = 't'; break;
-		case '\v': buf[1] = 'v'; break;
-		case '\\': buf[1] = '\\'; break;
-		case '\'': buf[1] = '\''; break;
-		default:
-			buf[1] = '0' + ((c >> 6) & 3);
-			buf[2] = '0' + ((c >> 3) & 7);
-			buf[3] = '0' + ((c >> 0) & 7);
-		}
-	}
+	char buf[8];
 
-	cdecl__err(CDECL_ENOPARSE, _("syntax error, unexpected '%s'"), buf);
+	c = yytext;
+invalid_char:
+	to_readable_ch(buf, *c);
+	cdecl__err(CDECL_ENOPARSE, _("syntax error, unexpected %s"), buf);
 	return T_LEX_ERROR;
 }