dt_lex.l

来自「Sun Solaris 10 中的 DTrace 组件的源代码。请参看: htt」· L 代码 · 共 794 行 · 第 1/2 页

#else
			bcopy(s, ((char *)&yylval.l_int) +
			    sizeof (yylval.l_int) - nbytes, nbytes);
#endif
			return (DT_TOK_INT);
		}

<S0>"/*"	|
<S2>"/*"	{
			yypcb->pcb_cstate = (YYSTATE);
			BEGIN(S1);
		}

<S0>{RGX_INTERP} |
<S2>{RGX_INTERP}	; /* discard any #! lines */

<S0>{RGX_CTL}	|
<S2>{RGX_CTL}	{
			assert(yypragma == NULL);
			yypcb->pcb_cstate = (YYSTATE);
			BEGIN(S3);
		}

<S0>"/"		{
			int c, tok;

			/*
			 * The use of "/" as the predicate delimiter and as the
			 * integer division symbol requires special lookahead
			 * to avoid a shift/reduce conflict in the D grammar.
			 * We look ahead to the next non-whitespace character.
			 * If we encounter EOF, ";", "{", or "/", then this "/"
			 * closes the predicate and we return DT_TOK_EPRED.
			 * If we encounter anything else, it's DT_TOK_DIV.
			 */
			while ((c = input()) != 0) {
				if (strchr("\f\n\r\t\v ", c) == NULL)
					break;
			}

			if (c == 0 || c == ';' || c == '{' || c == '/') {
				if (yypcb->pcb_parens != 0) {
					yyerror("closing ) expected in "
					    "predicate before /\n");
				}
				if (yypcb->pcb_brackets != 0) {
					yyerror("closing ] expected in "
					    "predicate before /\n");
				}
				tok = DT_TOK_EPRED;
			} else
				tok = DT_TOK_DIV;

			unput(c);
			return (tok);
		}

<S0>"("		{
			yypcb->pcb_parens++;
			return (DT_TOK_LPAR);
		}

<S0>")"		{
			if (--yypcb->pcb_parens < 0)
				yyerror("extra ) in input stream\n");
			return (DT_TOK_RPAR);
		}

<S0>"["		{
			yypcb->pcb_brackets++;
			return (DT_TOK_LBRAC);
		}

<S0>"]"		{
			if (--yypcb->pcb_brackets < 0)
				yyerror("extra ] in input stream\n");
			return (DT_TOK_RBRAC);
		}

<S0>"{"		|
<S2>"{"		{
			yypcb->pcb_braces++;
			return ('{');
		}

<S0>"}"		{
			if (--yypcb->pcb_braces < 0)
				yyerror("extra } in input stream\n");
			return ('}');
		}

<S0>"|"		return (DT_TOK_BOR);
<S0>"^"		return (DT_TOK_XOR);
<S0>"&"		return (DT_TOK_BAND);
<S0>"&&"	return (DT_TOK_LAND);
<S0>"^^"	return (DT_TOK_LXOR);
<S0>"||"	return (DT_TOK_LOR);
<S0>"=="	return (DT_TOK_EQU);
<S0>"!="	return (DT_TOK_NEQ);
<S0>"<"		return (DT_TOK_LT);
<S0>"<="	return (DT_TOK_LE);
<S0>">"		return (DT_TOK_GT);
<S0>">="	return (DT_TOK_GE);
<S0>"<<"	return (DT_TOK_LSH);
<S0>">>"	return (DT_TOK_RSH);
<S0>"+"		return (DT_TOK_ADD);
<S0>"-"		return (DT_TOK_SUB);
<S0>"*"		return (DT_TOK_MUL);
<S0>"%"		return (DT_TOK_MOD);
<S0>"~"		return (DT_TOK_BNEG);
<S0>"!"		return (DT_TOK_LNEG);
<S0>"?"		return (DT_TOK_QUESTION);
<S0>":"		return (DT_TOK_COLON);
<S0>"."		return (DT_TOK_DOT);
<S0>"->"	return (DT_TOK_PTR);
<S0>"="		return (DT_TOK_ASGN);
<S0>"+="	return (DT_TOK_ADD_EQ);
<S0>"-="	return (DT_TOK_SUB_EQ);
<S0>"*="	return (DT_TOK_MUL_EQ);
<S0>"/="	return (DT_TOK_DIV_EQ);
<S0>"%="	return (DT_TOK_MOD_EQ);
<S0>"&="	return (DT_TOK_AND_EQ);
<S0>"^="	return (DT_TOK_XOR_EQ);
<S0>"|="	return (DT_TOK_OR_EQ);
<S0>"<<="	return (DT_TOK_LSH_EQ);
<S0>">>="	return (DT_TOK_RSH_EQ);
<S0>"++"	return (DT_TOK_ADDADD);
<S0>"--"	return (DT_TOK_SUBSUB);
<S0>"..."	return (DT_TOK_ELLIPSIS);
<S0>","		return (DT_TOK_COMMA);
<S0>";"		return (';');
<S0>{RGX_WS}	; /* discard */
<S0>"\\"\n	; /* discard */
<S0>.		yyerror("syntax error near \"%c\"\n", yytext[0]);

<S1>"*/"	BEGIN(yypcb->pcb_cstate);
<S1>.|\n	; /* discard */

<S2>{RGX_PSPEC}	{
			/*
			 * S2 has an ambiguity because RGX_PSPEC includes '*'
			 * as a glob character and '*' also can be DT_TOK_STAR.
			 * Since lex always matches the longest token, this
			 * rule can be matched by an input string like "int*",
			 * which could begin a global variable declaration such
			 * as "int*x;" or could begin a RGX_PSPEC with globbing
			 * such as "int* { trace(timestamp); }".  If C_PSPEC is
			 * not set, we must resolve the ambiguity in favor of
			 * the type and perform lexer pushback if the fragment
			 * before '*' or entire fragment matches a type name.
			 * If C_PSPEC is set, we always return a PSPEC token.
			 * If C_PSPEC is off, the user can avoid ambiguity by
			 * including a ':' delimiter in the specifier, which
			 * they should be doing anyway to specify the provider.
			 */
			if (!(yypcb->pcb_cflags & DTRACE_C_PSPEC) &&
			    strchr(yytext, ':') == NULL) {

				char *p = strchr(yytext, '*');
				char *q = yytext + yyleng - 1;

				if (p != NULL && p > yytext)
					*p = '\0'; /* prune yytext */

				if (dt_type_lookup(yytext, NULL) == 0) {
					yylval.l_str = strdup(yytext);

					if (yylval.l_str == NULL) {
						longjmp(yypcb->pcb_jmpbuf,
						    EDT_NOMEM);
					}

					if (p != NULL && p > yytext) {
						for (*p = '*'; q >= p; q--)
							unput(*q);
					}

					yybegin(YYS_EXPR);
					return (DT_TOK_TNAME);
				}

				if (p != NULL && p > yytext)
					*p = '*'; /* restore yytext */
			}

			if ((yylval.l_str = strdup(yytext)) == NULL)
				longjmp(yypcb->pcb_jmpbuf, EDT_NOMEM);

			return (DT_TOK_PSPEC);
		}

<S2>"/"		return (DT_TOK_DIV);
<S2>","		return (DT_TOK_COMMA);

<S2>{RGX_WS}	; /* discard */
<S2>.		yyerror("syntax error near \"%c\"\n", yytext[0]);

<S3>\n		{
			dt_pragma(yypragma);
			yypragma = NULL;
			BEGIN(yypcb->pcb_cstate);
		}

<S3>[\f\t\v ]+	; /* discard */

<S3>[^\f\n\t\v "]+ {
			dt_node_t *dnp;

			if ((yylval.l_str = strdup(yytext)) == NULL)
				longjmp(yypcb->pcb_jmpbuf, EDT_NOMEM);

			/*
			 * We want to call dt_node_ident() here, but we can't
			 * because it will expand inlined identifiers, which we
			 * don't want to do from #pragma context in order to
			 * support pragmas that apply to the ident itself.  We
			 * call dt_node_string() and then reset dn_op instead.
			 */
			dnp = dt_node_string(yylval.l_str);
			dnp->dn_kind = DT_NODE_IDENT;
			dnp->dn_op = DT_TOK_IDENT;
			yypragma = dt_node_link(yypragma, dnp);
		}

<S3>.		yyerror("syntax error near \"%c\"\n", yytext[0]);

%%

/*
 * yybegin provides a wrapper for use from C code around the lex BEGIN() macro.
 * We use two main states for lexing because probe descriptions use a syntax
 * that is incompatible with the normal D tokens (e.g. names can contain "-").
 * yybegin also handles the job of switching between two lists of dt_nodes
 * as we allocate persistent definitions, like inlines, and transient nodes
 * that will be freed once we are done parsing the current program file.
 */
void
yybegin(yystate_t state)
{
#ifdef	YYDEBUG
	yydebug = _dtrace_debug;
#endif
	if (yypcb->pcb_yystate == state)
		return; /* nothing to do if we're in the state already */

	if (yypcb->pcb_yystate == YYS_DEFINE) {
        	yypcb->pcb_list = yypcb->pcb_hold;
        	yypcb->pcb_hold = NULL;
	}

	switch (state) {
	case YYS_CLAUSE:
		BEGIN(S2);
		break;
	case YYS_DEFINE:
		assert(yypcb->pcb_hold == NULL);
		yypcb->pcb_hold = yypcb->pcb_list;
		yypcb->pcb_list = NULL;
		/*FALLTHRU*/
	case YYS_EXPR:
		BEGIN(S0);
		break;
	default:
		xyerror(D_UNKNOWN, "internal error -- bad yystate %d\n", state);
	}

	yypcb->pcb_yystate = state;
}

void
yyinit(dt_pcb_t *pcb)
{
	yypcb = pcb;
	yylineno = 1;
	yypragma = NULL;
	yysptr = yysbuf;
}

/*
 * Given a lexeme 's' (typically yytext), set yylval and return an appropriate
 * token to the parser indicating either an identifier or a typedef name.
 * User-defined global variables always take precedence over types, but we do
 * use some heuristics because D programs can look at an ever-changing set of
 * kernel types and also can implicitly instantiate variables by assignment,
 * unlike in C.  The code here is ordered carefully as lookups are not cheap.
 */
static int
id_or_type(const char *s)
{
	dtrace_hdl_t *dtp = yypcb->pcb_hdl;
	int c0, c1, ttok = DT_TOK_TNAME;
	dt_ident_t *idp;

	if ((s = yylval.l_str = strdup(s)) == NULL)
		longjmp(yypcb->pcb_jmpbuf, EDT_NOMEM);

	/*
	 * If the lexeme is a global variable or likely identifier or *not* a
	 * type_name, then it is an identifier token.
	 */
	if (dt_idhash_lookup(dtp->dt_globals, s) != NULL ||
	    dt_idhash_lookup(yypcb->pcb_idents, s) != NULL ||
	    dt_type_lookup(s, NULL) != 0)
		return (DT_TOK_IDENT);

	/*
	 * If the lexeme is a type name and we are not in a program clause,
	 * then always interpret it as a type and return DT_TOK_TNAME.
	 */
	if ((YYSTATE) != S0)
		return (DT_TOK_TNAME);

	/*
	 * If the lexeme matches a type name but is in a program clause, then
	 * it could be a type or it could be an undefined variable.  Peek at
	 * the next token to decide.  If we see ++, --, [, or =, we know there
	 * might be an assignment that is trying to create a global variable,
	 * so we optimistically return DT_TOK_IDENT.  There is no harm in being
	 * wrong: a type_name followed by ++, --, [, or = is a syntax error.
	 */
	while ((c0 = input()) != 0) {
		if (strchr("\f\n\r\t\v ", c0) == NULL)
			break;
	}

	switch (c0) {
	case '+':
	case '-':
		if ((c1 = input()) == c0)
			ttok = DT_TOK_IDENT;
		unput(c1);
		break;

	case '=':
		if ((c1 = input()) != c0)
			ttok = DT_TOK_IDENT;
		unput(c1);
		break;
	case '[':
		ttok = DT_TOK_IDENT;
		break;
	}

	if (ttok == DT_TOK_IDENT) {
		idp = dt_idhash_insert(yypcb->pcb_idents, s, DT_IDENT_SCALAR, 0,
		    0, _dtrace_defattr, 0, &dt_idops_thaw, NULL, dtp->dt_gen);

		if (idp == NULL)
			longjmp(yypcb->pcb_jmpbuf, EDT_NOMEM);
	}

	unput(c0);
	return (ttok);
}

static int
input(void)
{
	int c;

	if (yysptr > yysbuf)
		c = *--yysptr;
	else if (yypcb->pcb_fileptr != NULL)
		c = fgetc(yypcb->pcb_fileptr);
	else if (yypcb->pcb_strptr < yypcb->pcb_string + yypcb->pcb_strlen)
		c = *yypcb->pcb_strptr++;
	else
		c = EOF;

	if (c == '\n')
		yylineno++;

	if (c != EOF)
		return (c);

	if ((YYSTATE) == S1)
		yyerror("end-of-file encountered before matching */\n");

	if ((YYSTATE) == S3)
		yyerror("end-of-file encountered before end of control line\n");

	if (yypcb->pcb_fileptr != NULL && ferror(yypcb->pcb_fileptr))
		longjmp(yypcb->pcb_jmpbuf, EDT_FIO);

	return (0); /* EOF */
}

static void
unput(int c)
{
	if (c == '\n')
		yylineno--;

	*yysptr++ = c;
	yytchar = c;
}