xref: /freebsd/contrib/bmake/cond.c (revision 6ef644f5889afbd0f681b08ed1a2f369524af83e)
1 /*	$NetBSD: cond.c,v 1.353 2023/06/23 05:21:10 rillig Exp $	*/
2 
3 /*
4  * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Adam de Boor.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * Copyright (c) 1988, 1989 by Adam de Boor
37  * Copyright (c) 1989 by Berkeley Softworks
38  * All rights reserved.
39  *
40  * This code is derived from software contributed to Berkeley by
41  * Adam de Boor.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by the University of
54  *	California, Berkeley and its contributors.
55  * 4. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  */
71 
72 /*
73  * Handling of conditionals in a makefile.
74  *
75  * Interface:
76  *	Cond_EvalLine   Evaluate the conditional directive, such as
77  *			'.if <cond>', '.elifnmake <cond>', '.else', '.endif'.
78  *
79  *	Cond_EvalCondition
80  *			Evaluate the conditional, which is either the argument
81  *			of one of the .if directives or the condition in a
82  *			':?then:else' variable modifier.
83  *
84  *	Cond_EndFile
85  *			At the end of reading a makefile, ensure that the
86  *			conditional directives are well-balanced.
87  */
88 
89 #include <errno.h>
90 
91 #include "make.h"
92 #include "dir.h"
93 
94 /*	"@(#)cond.c	8.2 (Berkeley) 1/2/94"	*/
95 MAKE_RCSID("$NetBSD: cond.c,v 1.353 2023/06/23 05:21:10 rillig Exp $");
96 
97 /*
98  * Conditional expressions conform to this grammar:
99  *	Or -> And ('||' And)*
100  *	And -> Term ('&&' Term)*
101  *	Term -> Function '(' Argument ')'
102  *	Term -> Leaf Operator Leaf
103  *	Term -> Leaf
104  *	Term -> '(' Or ')'
105  *	Term -> '!' Term
106  *	Leaf -> "string"
107  *	Leaf -> Number
108  *	Leaf -> VariableExpression
109  *	Leaf -> BareWord
110  *	Operator -> '==' | '!=' | '>' | '<' | '>=' | '<='
111  *
112  * BareWord is an unquoted string literal, its evaluation depends on the kind
113  * of '.if' directive.
114  *
115  * The tokens are scanned by CondParser_Token, which returns:
116  *	TOK_AND		for '&&'
117  *	TOK_OR		for '||'
118  *	TOK_NOT		for '!'
119  *	TOK_LPAREN	for '('
120  *	TOK_RPAREN	for ')'
121  *
122  * Other terminal symbols are evaluated using either the default function or
123  * the function given in the terminal, they return either TOK_TRUE, TOK_FALSE
124  * or TOK_ERROR.
125  */
126 typedef enum Token {
127 	TOK_FALSE, TOK_TRUE, TOK_AND, TOK_OR, TOK_NOT,
128 	TOK_LPAREN, TOK_RPAREN, TOK_EOF, TOK_NONE, TOK_ERROR
129 } Token;
130 
131 typedef enum ComparisonOp {
132 	LT, LE, GT, GE, EQ, NE
133 } ComparisonOp;
134 
135 typedef struct CondParser {
136 
137 	/*
138 	 * The plain '.if ${VAR}' evaluates to true if the value of the
139 	 * expression has length > 0 and is not numerically zero.  The other
140 	 * '.if' variants delegate to evalBare instead, for example '.ifdef
141 	 * ${VAR}' is equivalent to '.if defined(${VAR})', checking whether
142 	 * the variable named by the expression '${VAR}' is defined.
143 	 */
144 	bool plain;
145 
146 	/* The function to apply on unquoted bare words. */
147 	bool (*evalBare)(const char *);
148 	bool negateEvalBare;
149 
150 	/*
151 	 * Whether the left-hand side of a comparison may be an unquoted
152 	 * string.  This is allowed for expressions of the form
153 	 * ${condition:?:}, see ApplyModifier_IfElse.  Such a condition is
154 	 * expanded before it is evaluated, due to ease of implementation.
155 	 * This means that at the point where the condition is evaluated,
156 	 * make cannot know anymore whether the left-hand side had originally
157 	 * been a variable expression or a plain word.
158 	 *
159 	 * In conditional directives like '.if', the left-hand side must
160 	 * either be a variable expression, a quoted string or a number.
161 	 */
162 	bool leftUnquotedOK;
163 
164 	const char *p;		/* The remaining condition to parse */
165 	Token curr;		/* Single push-back token used in parsing */
166 
167 	/*
168 	 * Whether an error message has already been printed for this
169 	 * condition. The first available error message is usually the most
170 	 * specific one, therefore it makes sense to suppress the standard
171 	 * "Malformed conditional" message.
172 	 */
173 	bool printedError;
174 } CondParser;
175 
176 static CondResult CondParser_Or(CondParser *, bool);
177 
178 unsigned int cond_depth = 0;	/* current .if nesting level */
179 
180 /* Names for ComparisonOp. */
181 static const char opname[][3] = { "<", "<=", ">", ">=", "==", "!=" };
182 
183 MAKE_INLINE bool
184 skip_string(const char **pp, const char *str)
185 {
186 	size_t len = strlen(str);
187 	bool ok = strncmp(*pp, str, len) == 0;
188 	if (ok)
189 		*pp += len;
190 	return ok;
191 }
192 
193 static Token
194 ToToken(bool cond)
195 {
196 	return cond ? TOK_TRUE : TOK_FALSE;
197 }
198 
199 static void
200 CondParser_SkipWhitespace(CondParser *par)
201 {
202 	cpp_skip_whitespace(&par->p);
203 }
204 
205 /*
206  * Parse a single word, taking into account balanced parentheses as well as
207  * embedded expressions.  Used for the argument of a built-in function as
208  * well as for bare words, which are then passed to the default function.
209  */
210 static char *
211 ParseWord(const char **pp, bool doEval)
212 {
213 	const char *p = *pp;
214 	Buffer word;
215 	int paren_depth;
216 
217 	Buf_InitSize(&word, 16);
218 
219 	paren_depth = 0;
220 	for (;;) {
221 		char ch = *p;
222 		if (ch == '\0' || ch == ' ' || ch == '\t')
223 			break;
224 		if ((ch == '&' || ch == '|') && paren_depth == 0)
225 			break;
226 		if (ch == '$') {
227 			/*
228 			 * Parse the variable expression and install it as
229 			 * part of the argument if it's valid. We tell
230 			 * Var_Parse to complain on an undefined variable,
231 			 * (XXX: but Var_Parse ignores that request)
232 			 * so we don't need to do it. Nor do we return an
233 			 * error, though perhaps we should.
234 			 */
235 			VarEvalMode emode = doEval
236 			    ? VARE_UNDEFERR
237 			    : VARE_PARSE_ONLY;
238 			FStr nestedVal = Var_Parse(&p, SCOPE_CMDLINE, emode);
239 			/* TODO: handle errors */
240 			Buf_AddStr(&word, nestedVal.str);
241 			FStr_Done(&nestedVal);
242 			continue;
243 		}
244 		if (ch == '(')
245 			paren_depth++;
246 		else if (ch == ')' && --paren_depth < 0)
247 			break;
248 		Buf_AddByte(&word, ch);
249 		p++;
250 	}
251 
252 	cpp_skip_hspace(&p);
253 	*pp = p;
254 
255 	return Buf_DoneData(&word);
256 }
257 
258 /* Parse the function argument, including the surrounding parentheses. */
259 static char *
260 ParseFuncArg(CondParser *par, const char **pp, bool doEval, const char *func)
261 {
262 	const char *p = *pp;
263 	char *res;
264 
265 	p++;			/* Skip opening '(' - verified by caller */
266 	cpp_skip_hspace(&p);
267 	res = ParseWord(&p, doEval);
268 	cpp_skip_hspace(&p);
269 
270 	if (*p++ != ')') {
271 		int len = 0;
272 		while (ch_isalpha(func[len]))
273 			len++;
274 
275 		Parse_Error(PARSE_FATAL,
276 		    "Missing closing parenthesis for %.*s()", len, func);
277 		par->printedError = true;
278 		free(res);
279 		return NULL;
280 	}
281 
282 	*pp = p;
283 	return res;
284 }
285 
286 /* See if the given variable is defined. */
287 static bool
288 FuncDefined(const char *var)
289 {
290 	return Var_Exists(SCOPE_CMDLINE, var);
291 }
292 
293 /* See if a target matching targetPattern is requested to be made. */
294 static bool
295 FuncMake(const char *targetPattern)
296 {
297 	StringListNode *ln;
298 	bool warned = false;
299 
300 	for (ln = opts.create.first; ln != NULL; ln = ln->next) {
301 		StrMatchResult res = Str_Match(ln->datum, targetPattern);
302 		if (res.error != NULL && !warned) {
303 			warned = true;
304 			Parse_Error(PARSE_WARNING,
305 			    "%s in pattern argument '%s' to function 'make'",
306 			    res.error, targetPattern);
307 		}
308 		if (res.matched)
309 			return true;
310 	}
311 	return false;
312 }
313 
314 /* See if the given file exists. */
315 static bool
316 FuncExists(const char *file)
317 {
318 	bool result;
319 	char *path;
320 
321 	path = Dir_FindFile(file, &dirSearchPath);
322 	DEBUG2(COND, "exists(%s) result is \"%s\"\n",
323 	    file, path != NULL ? path : "");
324 	result = path != NULL;
325 	free(path);
326 	return result;
327 }
328 
329 /* See if the given node exists and is an actual target. */
330 static bool
331 FuncTarget(const char *node)
332 {
333 	GNode *gn = Targ_FindNode(node);
334 	return gn != NULL && GNode_IsTarget(gn);
335 }
336 
337 /*
338  * See if the given node exists and is an actual target with commands
339  * associated with it.
340  */
341 static bool
342 FuncCommands(const char *node)
343 {
344 	GNode *gn = Targ_FindNode(node);
345 	return gn != NULL && GNode_IsTarget(gn) &&
346 	       !Lst_IsEmpty(&gn->commands);
347 }
348 
349 /*
350  * Convert the string to a floating point number.  Accepted formats are
351  * base-10 integer, base-16 integer and finite floating point numbers.
352  */
353 static bool
354 TryParseNumber(const char *str, double *out_value)
355 {
356 	char *end;
357 	unsigned long ul_val;
358 	double dbl_val;
359 
360 	if (str[0] == '\0') {	/* XXX: why is an empty string a number? */
361 		*out_value = 0.0;
362 		return true;
363 	}
364 
365 	errno = 0;
366 	ul_val = strtoul(str, &end, str[1] == 'x' ? 16 : 10);
367 	if (*end == '\0' && errno != ERANGE) {
368 		*out_value = str[0] == '-' ? -(double)-ul_val : (double)ul_val;
369 		return true;
370 	}
371 
372 	if (*end != '\0' && *end != '.' && *end != 'e' && *end != 'E')
373 		return false;	/* skip the expensive strtod call */
374 	dbl_val = strtod(str, &end);
375 	if (*end != '\0')
376 		return false;
377 
378 	*out_value = dbl_val;
379 	return true;
380 }
381 
382 static bool
383 is_separator(char ch)
384 {
385 	return ch == '\0' || ch_isspace(ch) || ch == '!' || ch == '=' ||
386 	       ch == '>' || ch == '<' || ch == ')' /* but not '(' */;
387 }
388 
389 /*
390  * In a quoted or unquoted string literal or a number, parse a variable
391  * expression and add its value to the buffer.
392  *
393  * Return whether to continue parsing the leaf.
394  *
395  * Example: .if x${CENTER}y == "${PREFIX}${SUFFIX}" || 0x${HEX}
396  */
397 static bool
398 CondParser_StringExpr(CondParser *par, const char *start,
399 		      bool doEval, bool quoted,
400 		      Buffer *buf, FStr *inout_str)
401 {
402 	VarEvalMode emode;
403 	const char *p;
404 	bool atStart;
405 
406 	emode = doEval && quoted ? VARE_WANTRES
407 	    : doEval ? VARE_UNDEFERR
408 	    : VARE_PARSE_ONLY;
409 
410 	p = par->p;
411 	atStart = p == start;
412 	*inout_str = Var_Parse(&p, SCOPE_CMDLINE, emode);
413 	/* TODO: handle errors */
414 	if (inout_str->str == var_Error) {
415 		FStr_Done(inout_str);
416 		*inout_str = FStr_InitRefer(NULL);
417 		return false;
418 	}
419 	par->p = p;
420 
421 	/*
422 	 * If the '$' started the string literal (which means no quotes), and
423 	 * the expression is followed by a space, a comparison operator or
424 	 * the end of the expression, we are done.
425 	 */
426 	if (atStart && is_separator(par->p[0]))
427 		return false;
428 
429 	Buf_AddStr(buf, inout_str->str);
430 	FStr_Done(inout_str);
431 	*inout_str = FStr_InitRefer(NULL);	/* not finished yet */
432 	return true;
433 }
434 
435 /*
436  * Parse a string from a variable expression or an optionally quoted string,
437  * on the left-hand and right-hand sides of comparisons.
438  *
439  * Results:
440  *	Returns the string without any enclosing quotes, or NULL on error.
441  *	Sets out_quoted if the leaf was a quoted string literal.
442  */
443 static void
444 CondParser_Leaf(CondParser *par, bool doEval, bool unquotedOK,
445 		  FStr *out_str, bool *out_quoted)
446 {
447 	Buffer buf;
448 	FStr str;
449 	bool quoted;
450 	const char *start;
451 
452 	Buf_Init(&buf);
453 	str = FStr_InitRefer(NULL);
454 	*out_quoted = quoted = par->p[0] == '"';
455 	start = par->p;
456 	if (quoted)
457 		par->p++;
458 
459 	while (par->p[0] != '\0' && str.str == NULL) {
460 		switch (par->p[0]) {
461 		case '\\':
462 			par->p++;
463 			if (par->p[0] != '\0') {
464 				Buf_AddByte(&buf, par->p[0]);
465 				par->p++;
466 			}
467 			continue;
468 		case '"':
469 			par->p++;
470 			if (quoted)
471 				goto return_buf;	/* skip the closing quote */
472 			Buf_AddByte(&buf, '"');
473 			continue;
474 		case ')':	/* see is_separator */
475 		case '!':
476 		case '=':
477 		case '>':
478 		case '<':
479 		case ' ':
480 		case '\t':
481 			if (!quoted)
482 				goto return_buf;
483 			Buf_AddByte(&buf, par->p[0]);
484 			par->p++;
485 			continue;
486 		case '$':
487 			if (!CondParser_StringExpr(par,
488 			    start, doEval, quoted, &buf, &str))
489 				goto return_str;
490 			continue;
491 		default:
492 			if (!unquotedOK && !quoted && *start != '$' &&
493 			    !ch_isdigit(*start)) {
494 				/*
495 				 * The left-hand side must be quoted,
496 				 * a variable expression or a number.
497 				 */
498 				str = FStr_InitRefer(NULL);
499 				goto return_str;
500 			}
501 			Buf_AddByte(&buf, par->p[0]);
502 			par->p++;
503 			continue;
504 		}
505 	}
506 return_buf:
507 	str = FStr_InitOwn(buf.data);
508 	buf.data = NULL;
509 return_str:
510 	Buf_Done(&buf);
511 	*out_str = str;
512 }
513 
514 /*
515  * Evaluate a "comparison without operator", such as in ".if ${VAR}" or
516  * ".if 0".
517  */
518 static bool
519 EvalTruthy(CondParser *par, const char *value, bool quoted)
520 {
521 	double num;
522 
523 	/* For .ifxxx "...", check for non-empty string. */
524 	if (quoted)
525 		return value[0] != '\0';
526 
527 	/* For .ifxxx <number>, compare against zero */
528 	if (TryParseNumber(value, &num))
529 		return num != 0.0;
530 
531 	/*
532 	 * For .if ${...}, check for non-empty string.  This is different
533 	 * from the evaluation function from that .if variant, which would
534 	 * test whether a variable of the given name were defined.
535 	 */
536 	/*
537 	 * XXX: Whitespace should count as empty, just as in
538 	 * CondParser_FuncCallEmpty.
539 	 */
540 	if (par->plain)
541 		return value[0] != '\0';
542 
543 	return par->evalBare(value) != par->negateEvalBare;
544 }
545 
546 /* Evaluate a numerical comparison, such as in ".if ${VAR} >= 9". */
547 static bool
548 EvalCompareNum(double lhs, ComparisonOp op, double rhs)
549 {
550 	DEBUG3(COND, "Comparing %f %s %f\n", lhs, opname[op], rhs);
551 
552 	switch (op) {
553 	case LT:
554 		return lhs < rhs;
555 	case LE:
556 		return lhs <= rhs;
557 	case GT:
558 		return lhs > rhs;
559 	case GE:
560 		return lhs >= rhs;
561 	case EQ:
562 		return lhs == rhs;
563 	default:
564 		return lhs != rhs;
565 	}
566 }
567 
568 static Token
569 EvalCompareStr(CondParser *par, const char *lhs,
570 	       ComparisonOp op, const char *rhs)
571 {
572 	if (op != EQ && op != NE) {
573 		Parse_Error(PARSE_FATAL,
574 		    "Comparison with '%s' requires both operands "
575 		    "'%s' and '%s' to be numeric",
576 		    opname[op], lhs, rhs);
577 		par->printedError = true;
578 		return TOK_ERROR;
579 	}
580 
581 	DEBUG3(COND, "Comparing \"%s\" %s \"%s\"\n", lhs, opname[op], rhs);
582 	return ToToken((op == EQ) == (strcmp(lhs, rhs) == 0));
583 }
584 
585 /* Evaluate a comparison, such as "${VAR} == 12345". */
586 static Token
587 EvalCompare(CondParser *par, const char *lhs, bool lhsQuoted,
588 	    ComparisonOp op, const char *rhs, bool rhsQuoted)
589 {
590 	double left, right;
591 
592 	if (!rhsQuoted && !lhsQuoted)
593 		if (TryParseNumber(lhs, &left) && TryParseNumber(rhs, &right))
594 			return ToToken(EvalCompareNum(left, op, right));
595 
596 	return EvalCompareStr(par, lhs, op, rhs);
597 }
598 
599 static bool
600 CondParser_ComparisonOp(CondParser *par, ComparisonOp *out_op)
601 {
602 	const char *p = par->p;
603 
604 	if (p[0] == '<' && p[1] == '=')
605 		return par->p += 2, *out_op = LE, true;
606 	if (p[0] == '<')
607 		return par->p += 1, *out_op = LT, true;
608 	if (p[0] == '>' && p[1] == '=')
609 		return par->p += 2, *out_op = GE, true;
610 	if (p[0] == '>')
611 		return par->p += 1, *out_op = GT, true;
612 	if (p[0] == '=' && p[1] == '=')
613 		return par->p += 2, *out_op = EQ, true;
614 	if (p[0] == '!' && p[1] == '=')
615 		return par->p += 2, *out_op = NE, true;
616 	return false;
617 }
618 
619 /*
620  * Parse a comparison condition such as:
621  *
622  *	0
623  *	${VAR:Mpattern}
624  *	${VAR} == value
625  *	${VAR:U0} < 12345
626  */
627 static Token
628 CondParser_Comparison(CondParser *par, bool doEval)
629 {
630 	Token t = TOK_ERROR;
631 	FStr lhs, rhs;
632 	ComparisonOp op;
633 	bool lhsQuoted, rhsQuoted;
634 
635 	CondParser_Leaf(par, doEval, par->leftUnquotedOK, &lhs, &lhsQuoted);
636 	if (lhs.str == NULL)
637 		goto done_lhs;
638 
639 	CondParser_SkipWhitespace(par);
640 
641 	if (!CondParser_ComparisonOp(par, &op)) {
642 		/* Unknown operator, compare against an empty string or 0. */
643 		t = ToToken(doEval && EvalTruthy(par, lhs.str, lhsQuoted));
644 		goto done_lhs;
645 	}
646 
647 	CondParser_SkipWhitespace(par);
648 
649 	if (par->p[0] == '\0') {
650 		Parse_Error(PARSE_FATAL,
651 		    "Missing right-hand side of operator '%s'", opname[op]);
652 		par->printedError = true;
653 		goto done_lhs;
654 	}
655 
656 	CondParser_Leaf(par, doEval, true, &rhs, &rhsQuoted);
657 	t = rhs.str == NULL ? TOK_ERROR
658 	    : !doEval ? TOK_FALSE
659 	    : EvalCompare(par, lhs.str, lhsQuoted, op, rhs.str, rhsQuoted);
660 	FStr_Done(&rhs);
661 
662 done_lhs:
663 	FStr_Done(&lhs);
664 	return t;
665 }
666 
667 /*
668  * The argument to empty() is a variable name, optionally followed by
669  * variable modifiers.
670  */
671 static bool
672 CondParser_FuncCallEmpty(CondParser *par, bool doEval, Token *out_token)
673 {
674 	const char *cp = par->p;
675 	Token tok;
676 	FStr val;
677 
678 	if (!skip_string(&cp, "empty"))
679 		return false;
680 
681 	cpp_skip_whitespace(&cp);
682 	if (*cp != '(')
683 		return false;
684 
685 	cp--;			/* Make cp[1] point to the '('. */
686 	val = Var_Parse(&cp, SCOPE_CMDLINE,
687 	    doEval ? VARE_WANTRES : VARE_PARSE_ONLY);
688 	/* TODO: handle errors */
689 
690 	if (val.str == var_Error)
691 		tok = TOK_ERROR;
692 	else {
693 		cpp_skip_whitespace(&val.str);
694 		tok = ToToken(doEval && val.str[0] == '\0');
695 	}
696 
697 	FStr_Done(&val);
698 	*out_token = tok;
699 	par->p = cp;
700 	return true;
701 }
702 
703 /* Parse a function call expression, such as 'exists(${file})'. */
704 static bool
705 CondParser_FuncCall(CondParser *par, bool doEval, Token *out_token)
706 {
707 	char *arg;
708 	const char *p = par->p;
709 	bool (*fn)(const char *);
710 	const char *fn_name = p;
711 
712 	if (skip_string(&p, "defined"))
713 		fn = FuncDefined;
714 	else if (skip_string(&p, "make"))
715 		fn = FuncMake;
716 	else if (skip_string(&p, "exists"))
717 		fn = FuncExists;
718 	else if (skip_string(&p, "target"))
719 		fn = FuncTarget;
720 	else if (skip_string(&p, "commands"))
721 		fn = FuncCommands;
722 	else
723 		return false;
724 
725 	cpp_skip_whitespace(&p);
726 	if (*p != '(')
727 		return false;
728 
729 	arg = ParseFuncArg(par, &p, doEval, fn_name);
730 	*out_token = ToToken(doEval &&
731 	    arg != NULL && arg[0] != '\0' && fn(arg));
732 	free(arg);
733 
734 	par->p = p;
735 	return true;
736 }
737 
738 /*
739  * Parse a comparison that neither starts with '"' nor '$', such as the
740  * unusual 'bare == right' or '3 == ${VAR}', or a simple leaf without
741  * operator, which is a number, a variable expression or a string literal.
742  *
743  * TODO: Can this be merged into CondParser_Comparison?
744  */
745 static Token
746 CondParser_ComparisonOrLeaf(CondParser *par, bool doEval)
747 {
748 	Token t;
749 	char *arg;
750 	const char *cp;
751 
752 	/* Push anything numeric through the compare expression */
753 	cp = par->p;
754 	if (ch_isdigit(cp[0]) || cp[0] == '-' || cp[0] == '+')
755 		return CondParser_Comparison(par, doEval);
756 
757 	/*
758 	 * Most likely we have a naked token to apply the default function to.
759 	 * However ".if a == b" gets here when the "a" is unquoted and doesn't
760 	 * start with a '$'. This surprises people.
761 	 * If what follows the function argument is a '=' or '!' then the
762 	 * syntax would be invalid if we did "defined(a)" - so instead treat
763 	 * as an expression.
764 	 */
765 	/*
766 	 * XXX: In edge cases, a variable expression may be evaluated twice,
767 	 *  see cond-token-plain.mk, keyword 'twice'.
768 	 */
769 	arg = ParseWord(&cp, doEval);
770 	assert(arg[0] != '\0');
771 
772 	if (*cp == '=' || *cp == '!' || *cp == '<' || *cp == '>')
773 		return CondParser_Comparison(par, doEval);
774 	par->p = cp;
775 
776 	/*
777 	 * Evaluate the argument using the default function.
778 	 * This path always treats .if as .ifdef. To get here, the character
779 	 * after .if must have been taken literally, so the argument cannot
780 	 * be empty - even if it contained a variable expansion.
781 	 */
782 	t = ToToken(doEval && par->evalBare(arg) != par->negateEvalBare);
783 	free(arg);
784 	return t;
785 }
786 
787 /* Return the next token or comparison result from the parser. */
788 static Token
789 CondParser_Token(CondParser *par, bool doEval)
790 {
791 	Token t;
792 
793 	t = par->curr;
794 	if (t != TOK_NONE) {
795 		par->curr = TOK_NONE;
796 		return t;
797 	}
798 
799 	cpp_skip_hspace(&par->p);
800 
801 	switch (par->p[0]) {
802 
803 	case '(':
804 		par->p++;
805 		return TOK_LPAREN;
806 
807 	case ')':
808 		par->p++;
809 		return TOK_RPAREN;
810 
811 	case '|':
812 		par->p++;
813 		if (par->p[0] == '|')
814 			par->p++;
815 		else if (opts.strict) {
816 			Parse_Error(PARSE_FATAL, "Unknown operator '|'");
817 			par->printedError = true;
818 			return TOK_ERROR;
819 		}
820 		return TOK_OR;
821 
822 	case '&':
823 		par->p++;
824 		if (par->p[0] == '&')
825 			par->p++;
826 		else if (opts.strict) {
827 			Parse_Error(PARSE_FATAL, "Unknown operator '&'");
828 			par->printedError = true;
829 			return TOK_ERROR;
830 		}
831 		return TOK_AND;
832 
833 	case '!':
834 		par->p++;
835 		return TOK_NOT;
836 
837 	case '#':		/* XXX: see unit-tests/cond-token-plain.mk */
838 	case '\n':		/* XXX: why should this end the condition? */
839 		/* Probably obsolete now, from 1993-03-21. */
840 	case '\0':
841 		return TOK_EOF;
842 
843 	case '"':
844 	case '$':
845 		return CondParser_Comparison(par, doEval);
846 
847 	default:
848 		if (CondParser_FuncCallEmpty(par, doEval, &t))
849 			return t;
850 		if (CondParser_FuncCall(par, doEval, &t))
851 			return t;
852 		return CondParser_ComparisonOrLeaf(par, doEval);
853 	}
854 }
855 
856 /* Skip the next token if it equals t. */
857 static bool
858 CondParser_Skip(CondParser *par, Token t)
859 {
860 	Token actual;
861 
862 	actual = CondParser_Token(par, false);
863 	if (actual == t)
864 		return true;
865 
866 	assert(par->curr == TOK_NONE);
867 	assert(actual != TOK_NONE);
868 	par->curr = actual;
869 	return false;
870 }
871 
872 /*
873  * Term -> '(' Or ')'
874  * Term -> '!' Term
875  * Term -> Leaf Operator Leaf
876  * Term -> Leaf
877  */
878 static CondResult
879 CondParser_Term(CondParser *par, bool doEval)
880 {
881 	CondResult res;
882 	Token t;
883 
884 	t = CondParser_Token(par, doEval);
885 	if (t == TOK_TRUE)
886 		return CR_TRUE;
887 	if (t == TOK_FALSE)
888 		return CR_FALSE;
889 
890 	if (t == TOK_LPAREN) {
891 		res = CondParser_Or(par, doEval);
892 		if (res == CR_ERROR)
893 			return CR_ERROR;
894 		if (CondParser_Token(par, doEval) != TOK_RPAREN)
895 			return CR_ERROR;
896 		return res;
897 	}
898 
899 	if (t == TOK_NOT) {
900 		res = CondParser_Term(par, doEval);
901 		if (res == CR_TRUE)
902 			res = CR_FALSE;
903 		else if (res == CR_FALSE)
904 			res = CR_TRUE;
905 		return res;
906 	}
907 
908 	return CR_ERROR;
909 }
910 
911 /*
912  * And -> Term ('&&' Term)*
913  */
914 static CondResult
915 CondParser_And(CondParser *par, bool doEval)
916 {
917 	CondResult res, rhs;
918 
919 	res = CR_TRUE;
920 	do {
921 		if ((rhs = CondParser_Term(par, doEval)) == CR_ERROR)
922 			return CR_ERROR;
923 		if (rhs == CR_FALSE) {
924 			res = CR_FALSE;
925 			doEval = false;
926 		}
927 	} while (CondParser_Skip(par, TOK_AND));
928 
929 	return res;
930 }
931 
932 /*
933  * Or -> And ('||' And)*
934  */
935 static CondResult
936 CondParser_Or(CondParser *par, bool doEval)
937 {
938 	CondResult res, rhs;
939 
940 	res = CR_FALSE;
941 	do {
942 		if ((rhs = CondParser_And(par, doEval)) == CR_ERROR)
943 			return CR_ERROR;
944 		if (rhs == CR_TRUE) {
945 			res = CR_TRUE;
946 			doEval = false;
947 		}
948 	} while (CondParser_Skip(par, TOK_OR));
949 
950 	return res;
951 }
952 
953 static CondResult
954 CondParser_Eval(CondParser *par)
955 {
956 	CondResult res;
957 
958 	DEBUG1(COND, "CondParser_Eval: %s\n", par->p);
959 
960 	res = CondParser_Or(par, true);
961 	if (res != CR_ERROR && CondParser_Token(par, false) != TOK_EOF)
962 		return CR_ERROR;
963 
964 	return res;
965 }
966 
967 /*
968  * Evaluate the condition, including any side effects from the variable
969  * expressions in the condition. The condition consists of &&, ||, !,
970  * function(arg), comparisons and parenthetical groupings thereof.
971  */
972 static CondResult
973 CondEvalExpression(const char *cond, bool plain,
974 		   bool (*evalBare)(const char *), bool negate,
975 		   bool eprint, bool leftUnquotedOK)
976 {
977 	CondParser par;
978 	CondResult rval;
979 
980 	cpp_skip_hspace(&cond);
981 
982 	par.plain = plain;
983 	par.evalBare = evalBare;
984 	par.negateEvalBare = negate;
985 	par.leftUnquotedOK = leftUnquotedOK;
986 	par.p = cond;
987 	par.curr = TOK_NONE;
988 	par.printedError = false;
989 
990 	rval = CondParser_Eval(&par);
991 
992 	if (rval == CR_ERROR && eprint && !par.printedError)
993 		Parse_Error(PARSE_FATAL, "Malformed conditional (%s)", cond);
994 
995 	return rval;
996 }
997 
998 /*
999  * Evaluate a condition in a :? modifier, such as
1000  * ${"${VAR}" == value:?yes:no}.
1001  */
1002 CondResult
1003 Cond_EvalCondition(const char *cond)
1004 {
1005 	return CondEvalExpression(cond, true,
1006 	    FuncDefined, false, false, true);
1007 }
1008 
1009 static bool
1010 IsEndif(const char *p)
1011 {
1012 	return p[0] == 'e' && p[1] == 'n' && p[2] == 'd' &&
1013 	       p[3] == 'i' && p[4] == 'f' && !ch_isalpha(p[5]);
1014 }
1015 
1016 static bool
1017 DetermineKindOfConditional(const char **pp, bool *out_plain,
1018 			   bool (**out_evalBare)(const char *),
1019 			   bool *out_negate)
1020 {
1021 	const char *p = *pp + 2;
1022 
1023 	*out_plain = false;
1024 	*out_evalBare = FuncDefined;
1025 	*out_negate = skip_string(&p, "n");
1026 
1027 	if (skip_string(&p, "def")) {		/* .ifdef and .ifndef */
1028 	} else if (skip_string(&p, "make"))	/* .ifmake and .ifnmake */
1029 		*out_evalBare = FuncMake;
1030 	else if (!*out_negate)			/* plain .if */
1031 		*out_plain = true;
1032 	else
1033 		goto unknown_directive;
1034 	if (ch_isalpha(*p))
1035 		goto unknown_directive;
1036 
1037 	*pp = p;
1038 	return true;
1039 
1040 unknown_directive:
1041 	/*
1042 	 * TODO: Add error message about unknown directive, since there is no
1043 	 * other known directive that starts with 'el' or 'if'.
1044 	 *
1045 	 * Example: .elifx 123
1046 	 */
1047 	return false;
1048 }
1049 
1050 /*
1051  * Evaluate the conditional directive in the line, which is one of:
1052  *
1053  *	.if <cond>
1054  *	.ifmake <cond>
1055  *	.ifnmake <cond>
1056  *	.ifdef <cond>
1057  *	.ifndef <cond>
1058  *	.elif <cond>
1059  *	.elifmake <cond>
1060  *	.elifnmake <cond>
1061  *	.elifdef <cond>
1062  *	.elifndef <cond>
1063  *	.else
1064  *	.endif
1065  *
1066  * In these directives, <cond> consists of &&, ||, !, function(arg),
1067  * comparisons, expressions, bare words, numbers and strings, and
1068  * parenthetical groupings thereof.
1069  *
1070  * Results:
1071  *	CR_TRUE		to continue parsing the lines that follow the
1072  *			conditional (when <cond> evaluates to true)
1073  *	CR_FALSE	to skip the lines after the conditional
1074  *			(when <cond> evaluates to false, or when a previous
1075  *			branch has already been taken)
1076  *	CR_ERROR	if the conditional was not valid, either because of
1077  *			a syntax error or because some variable was undefined
1078  *			or because the condition could not be evaluated
1079  */
1080 CondResult
1081 Cond_EvalLine(const char *line)
1082 {
1083 	typedef enum IfState {
1084 
1085 		/* None of the previous <cond> evaluated to true. */
1086 		IFS_INITIAL	= 0,
1087 
1088 		/*
1089 		 * The previous <cond> evaluated to true. The lines following
1090 		 * this condition are interpreted.
1091 		 */
1092 		IFS_ACTIVE	= 1 << 0,
1093 
1094 		/* The previous directive was an '.else'. */
1095 		IFS_SEEN_ELSE	= 1 << 1,
1096 
1097 		/* One of the previous <cond> evaluated to true. */
1098 		IFS_WAS_ACTIVE	= 1 << 2
1099 
1100 	} IfState;
1101 
1102 	static enum IfState *cond_states = NULL;
1103 	static unsigned int cond_states_cap = 128;
1104 
1105 	bool plain;
1106 	bool (*evalBare)(const char *);
1107 	bool negate;
1108 	bool isElif;
1109 	CondResult res;
1110 	IfState state;
1111 	const char *p = line;
1112 
1113 	if (cond_states == NULL) {
1114 		cond_states = bmake_malloc(
1115 		    cond_states_cap * sizeof *cond_states);
1116 		cond_states[0] = IFS_ACTIVE;
1117 	}
1118 
1119 	p++;			/* skip the leading '.' */
1120 	cpp_skip_hspace(&p);
1121 
1122 	if (IsEndif(p)) {	/* It is an '.endif'. */
1123 		if (p[5] != '\0') {
1124 			Parse_Error(PARSE_FATAL,
1125 			    "The .endif directive does not take arguments");
1126 		}
1127 
1128 		if (cond_depth == CurFile_CondMinDepth()) {
1129 			Parse_Error(PARSE_FATAL, "if-less endif");
1130 			return CR_TRUE;
1131 		}
1132 
1133 		/* Return state for previous conditional */
1134 		cond_depth--;
1135 		Parse_GuardEndif();
1136 		return cond_states[cond_depth] & IFS_ACTIVE
1137 		    ? CR_TRUE : CR_FALSE;
1138 	}
1139 
1140 	/* Parse the name of the directive, such as 'if', 'elif', 'endif'. */
1141 	if (p[0] == 'e') {
1142 		if (p[1] != 'l') {
1143 			/*
1144 			 * Unknown directive.  It might still be a
1145 			 * transformation rule like '.err.txt',
1146 			 * therefore no error message here.
1147 			 */
1148 			return CR_ERROR;
1149 		}
1150 
1151 		/* Quite likely this is 'else' or 'elif' */
1152 		p += 2;
1153 		if (strncmp(p, "se", 2) == 0 && !ch_isalpha(p[2])) {
1154 			if (p[2] != '\0')
1155 				Parse_Error(PARSE_FATAL,
1156 				    "The .else directive "
1157 				    "does not take arguments");
1158 
1159 			if (cond_depth == CurFile_CondMinDepth()) {
1160 				Parse_Error(PARSE_FATAL, "if-less else");
1161 				return CR_TRUE;
1162 			}
1163 			Parse_GuardElse();
1164 
1165 			state = cond_states[cond_depth];
1166 			if (state == IFS_INITIAL) {
1167 				state = IFS_ACTIVE | IFS_SEEN_ELSE;
1168 			} else {
1169 				if (state & IFS_SEEN_ELSE)
1170 					Parse_Error(PARSE_WARNING,
1171 					    "extra else");
1172 				state = IFS_WAS_ACTIVE | IFS_SEEN_ELSE;
1173 			}
1174 			cond_states[cond_depth] = state;
1175 
1176 			return state & IFS_ACTIVE ? CR_TRUE : CR_FALSE;
1177 		}
1178 		/* Assume for now it is an elif */
1179 		isElif = true;
1180 	} else
1181 		isElif = false;
1182 
1183 	if (p[0] != 'i' || p[1] != 'f') {
1184 		/*
1185 		 * Unknown directive.  It might still be a transformation rule
1186 		 * like '.elisp.scm', therefore no error message here.
1187 		 */
1188 		return CR_ERROR;	/* Not an ifxxx or elifxxx line */
1189 	}
1190 
1191 	if (!DetermineKindOfConditional(&p, &plain, &evalBare, &negate))
1192 		return CR_ERROR;
1193 
1194 	if (isElif) {
1195 		if (cond_depth == CurFile_CondMinDepth()) {
1196 			Parse_Error(PARSE_FATAL, "if-less elif");
1197 			return CR_TRUE;
1198 		}
1199 		Parse_GuardElse();
1200 		state = cond_states[cond_depth];
1201 		if (state & IFS_SEEN_ELSE) {
1202 			Parse_Error(PARSE_WARNING, "extra elif");
1203 			cond_states[cond_depth] =
1204 			    IFS_WAS_ACTIVE | IFS_SEEN_ELSE;
1205 			return CR_FALSE;
1206 		}
1207 		if (state != IFS_INITIAL) {
1208 			cond_states[cond_depth] = IFS_WAS_ACTIVE;
1209 			return CR_FALSE;
1210 		}
1211 	} else {
1212 		/* Normal .if */
1213 		if (cond_depth + 1 >= cond_states_cap) {
1214 			/*
1215 			 * This is rare, but not impossible.
1216 			 * In meta mode, dirdeps.mk (only runs at level 0)
1217 			 * can need more than the default.
1218 			 */
1219 			cond_states_cap += 32;
1220 			cond_states = bmake_realloc(cond_states,
1221 			    cond_states_cap * sizeof *cond_states);
1222 		}
1223 		state = cond_states[cond_depth];
1224 		cond_depth++;
1225 		if (!(state & IFS_ACTIVE)) {
1226 			/*
1227 			 * If we aren't parsing the data,
1228 			 * treat as always false.
1229 			 */
1230 			cond_states[cond_depth] = IFS_WAS_ACTIVE;
1231 			return CR_FALSE;
1232 		}
1233 	}
1234 
1235 	/* And evaluate the conditional expression */
1236 	res = CondEvalExpression(p, plain, evalBare, negate, true, false);
1237 	if (res == CR_ERROR) {
1238 		/* Syntax error, error message already output. */
1239 		/* Skip everything to the matching '.endif'. */
1240 		/* An extra '.else' is not detected in this case. */
1241 		cond_states[cond_depth] = IFS_WAS_ACTIVE;
1242 		return CR_FALSE;
1243 	}
1244 
1245 	cond_states[cond_depth] = res == CR_TRUE ? IFS_ACTIVE : IFS_INITIAL;
1246 	return res;
1247 }
1248 
1249 static bool
1250 ParseVarnameGuard(const char **pp, const char **varname)
1251 {
1252 	const char *p = *pp;
1253 
1254 	if (ch_isalpha(*p) || *p == '_') {
1255 		while (ch_isalnum(*p) || *p == '_')
1256 			p++;
1257 		*varname = *pp;
1258 		*pp = p;
1259 		return true;
1260 	}
1261 	return false;
1262 }
1263 
1264 /* Extracts the multiple-inclusion guard from a conditional, if any. */
1265 Guard *
1266 Cond_ExtractGuard(const char *line)
1267 {
1268 	const char *p, *varname;
1269 	Substring dir;
1270 	enum GuardKind kind;
1271 	Guard *guard;
1272 
1273 	p = line + 1;		/* skip the '.' */
1274 	cpp_skip_hspace(&p);
1275 
1276 	dir.start = p;
1277 	while (ch_isalpha(*p))
1278 		p++;
1279 	dir.end = p;
1280 	cpp_skip_hspace(&p);
1281 
1282 	if (Substring_Equals(dir, "if")) {
1283 		if (skip_string(&p, "!defined(")) {
1284 			if (ParseVarnameGuard(&p, &varname)
1285 			    && strcmp(p, ")") == 0)
1286 				goto found_variable;
1287 		} else if (skip_string(&p, "!target(")) {
1288 			const char *arg_p = p;
1289 			free(ParseWord(&p, false));
1290 			if (strcmp(p, ")") == 0) {
1291 				char *target = ParseWord(&arg_p, true);
1292 				guard = bmake_malloc(sizeof(*guard));
1293 				guard->kind = GK_TARGET;
1294 				guard->name = target;
1295 				return guard;
1296 			}
1297 		}
1298 	} else if (Substring_Equals(dir, "ifndef")) {
1299 		if (ParseVarnameGuard(&p, &varname) && *p == '\0')
1300 			goto found_variable;
1301 	}
1302 	return NULL;
1303 
1304 found_variable:
1305 	kind = GK_VARIABLE;
1306 	guard = bmake_malloc(sizeof(*guard));
1307 	guard->kind = kind;
1308 	guard->name = bmake_strsedup(varname, p);
1309 	return guard;
1310 }
1311 
1312 void
1313 Cond_EndFile(void)
1314 {
1315 	unsigned int open_conds = cond_depth - CurFile_CondMinDepth();
1316 
1317 	if (open_conds != 0) {
1318 		Parse_Error(PARSE_FATAL, "%u open conditional%s",
1319 		    open_conds, open_conds == 1 ? "" : "s");
1320 		cond_depth = CurFile_CondMinDepth();
1321 	}
1322 }
1323