xref: /freebsd/contrib/bmake/cond.c (revision 36d6566e5985030fd2f1100bd9c1387bbe0bd290)
1 /*	$NetBSD: cond.c,v 1.173 2020/10/30 20:30:44 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 /* Handling of conditionals in a makefile.
73  *
74  * Interface:
75  *	Cond_EvalLine	Evaluate the conditional.
76  *
77  *	Cond_EvalCondition
78  *			Evaluate the conditional, which is either the argument
79  *			of one of the .if directives or the condition in a
80  *			':?then:else' variable modifier.
81  *
82  *	Cond_save_depth
83  *	Cond_restore_depth
84  *			Save and restore the nesting of the conditions, at
85  *			the start and end of including another makefile, to
86  *			ensure that in each makefile the conditional
87  *			directives are well-balanced.
88  */
89 
90 #include <errno.h>
91 
92 #include "make.h"
93 #include "dir.h"
94 
95 /*	"@(#)cond.c	8.2 (Berkeley) 1/2/94"	*/
96 MAKE_RCSID("$NetBSD: cond.c,v 1.173 2020/10/30 20:30:44 rillig Exp $");
97 
98 /*
99  * The parsing of conditional expressions is based on this grammar:
100  *	E -> F || E
101  *	E -> F
102  *	F -> T && F
103  *	F -> T
104  *	T -> defined(variable)
105  *	T -> make(target)
106  *	T -> exists(file)
107  *	T -> empty(varspec)
108  *	T -> target(name)
109  *	T -> commands(name)
110  *	T -> symbol
111  *	T -> $(varspec) op value
112  *	T -> $(varspec) == "string"
113  *	T -> $(varspec) != "string"
114  *	T -> "string"
115  *	T -> ( E )
116  *	T -> ! T
117  *	op -> == | != | > | < | >= | <=
118  *
119  * 'symbol' is some other symbol to which the default function is applied.
120  *
121  * The tokens are scanned by CondToken, which returns:
122  *	TOK_AND		for '&' or '&&'
123  *	TOK_OR		for '|' or '||'
124  *	TOK_NOT		for '!'
125  *	TOK_LPAREN	for '('
126  *	TOK_RPAREN	for ')'
127  * Other terminal symbols are evaluated using either the default function or
128  * the function given in the terminal, they return either TOK_TRUE or
129  * TOK_FALSE.
130  *
131  * TOK_FALSE is 0 and TOK_TRUE 1 so we can directly assign C comparisons.
132  *
133  * All non-terminal functions (CondParser_Expr, CondParser_Factor and
134  * CondParser_Term) return either TOK_FALSE, TOK_TRUE, or TOK_ERROR on error.
135  */
136 typedef enum Token {
137     TOK_FALSE = 0, TOK_TRUE = 1, TOK_AND, TOK_OR, TOK_NOT,
138     TOK_LPAREN, TOK_RPAREN, TOK_EOF, TOK_NONE, TOK_ERROR
139 } Token;
140 
141 typedef struct CondParser {
142     const struct If *if_info;	/* Info for current statement */
143     const char *p;		/* The remaining condition to parse */
144     Token curr;			/* Single push-back token used in parsing */
145 
146     /* Whether an error message has already been printed for this condition.
147      * The first available error message is usually the most specific one,
148      * therefore it makes sense to suppress the standard "Malformed
149      * conditional" message. */
150     Boolean printedError;
151 } CondParser;
152 
153 static Token CondParser_Expr(CondParser *par, Boolean);
154 
155 static unsigned int cond_depth = 0;	/* current .if nesting level */
156 static unsigned int cond_min_depth = 0;	/* depth at makefile open */
157 
158 /*
159  * Indicate when we should be strict about lhs of comparisons.
160  * In strict mode, the lhs must be a variable expression or a string literal
161  * in quotes. In non-strict mode it may also be an unquoted string literal.
162  *
163  * TRUE when CondEvalExpression is called from Cond_EvalLine (.if etc)
164  * FALSE when CondEvalExpression is called from ApplyModifier_IfElse
165  * since lhs is already expanded and we cannot tell if
166  * it was a variable reference or not.
167  */
168 static Boolean lhsStrict;
169 
170 static int
171 is_token(const char *str, const char *tok, size_t len)
172 {
173     return strncmp(str, tok, len) == 0 && !ch_isalpha(str[len]);
174 }
175 
176 /* Push back the most recent token read. We only need one level of this. */
177 static void
178 CondParser_PushBack(CondParser *par, Token t)
179 {
180     assert(par->curr == TOK_NONE);
181     assert(t != TOK_NONE);
182 
183     par->curr = t;
184 }
185 
186 static void
187 CondParser_SkipWhitespace(CondParser *par)
188 {
189     cpp_skip_whitespace(&par->p);
190 }
191 
192 /* Parse the argument of a built-in function.
193  *
194  * Arguments:
195  *	*pp initially points at the '(',
196  *	upon successful return it points right after the ')'.
197  *
198  *	*out_arg receives the argument as string.
199  *
200  *	func says whether the argument belongs to an actual function, or
201  *	whether the parsed argument is passed to the default function.
202  *
203  * Return the length of the argument. */
204 static size_t
205 ParseFuncArg(const char **pp, Boolean doEval, const char *func,
206 	     char **out_arg) {
207     const char *p = *pp;
208     Buffer argBuf;
209     int paren_depth;
210     size_t argLen;
211 
212     if (func != NULL)
213 	p++;			/* Skip opening '(' - verified by caller */
214 
215     if (*p == '\0') {
216 	/*
217 	 * No arguments whatsoever. Because 'make' and 'defined' aren't really
218 	 * "reserved words", we don't print a message. I think this is better
219 	 * than hitting the user with a warning message every time s/he uses
220 	 * the word 'make' or 'defined' at the beginning of a symbol...
221 	 */
222 	*out_arg = NULL;
223 	return 0;
224     }
225 
226     while (*p == ' ' || *p == '\t') {
227 	p++;
228     }
229 
230     Buf_Init(&argBuf, 16);
231 
232     paren_depth = 0;
233     for (;;) {
234 	char ch = *p;
235 	if (ch == 0 || ch == ' ' || ch == '\t')
236 	    break;
237 	if ((ch == '&' || ch == '|') && paren_depth == 0)
238 	    break;
239 	if (*p == '$') {
240 	    /*
241 	     * Parse the variable spec and install it as part of the argument
242 	     * if it's valid. We tell Var_Parse to complain on an undefined
243 	     * variable, so we don't need to do it. Nor do we return an error,
244 	     * though perhaps we should...
245 	     */
246 	    void *nestedVal_freeIt;
247 	    VarEvalFlags eflags = VARE_UNDEFERR | (doEval ? VARE_WANTRES : 0);
248 	    const char *nestedVal;
249 	    (void)Var_Parse(&p, VAR_CMDLINE, eflags, &nestedVal,
250 			    &nestedVal_freeIt);
251 	    /* TODO: handle errors */
252 	    Buf_AddStr(&argBuf, nestedVal);
253 	    free(nestedVal_freeIt);
254 	    continue;
255 	}
256 	if (ch == '(')
257 	    paren_depth++;
258 	else if (ch == ')' && --paren_depth < 0)
259 	    break;
260 	Buf_AddByte(&argBuf, *p);
261 	p++;
262     }
263 
264     *out_arg = Buf_GetAll(&argBuf, &argLen);
265     Buf_Destroy(&argBuf, FALSE);
266 
267     while (*p == ' ' || *p == '\t') {
268 	p++;
269     }
270 
271     if (func != NULL && *p++ != ')') {
272 	Parse_Error(PARSE_WARNING, "Missing closing parenthesis for %s()",
273 		    func);
274 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
275 	return 0;
276     }
277 
278     *pp = p;
279     return argLen;
280 }
281 
282 /* Test whether the given variable is defined. */
283 static Boolean
284 FuncDefined(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
285 {
286     void *freeIt;
287     Boolean result = Var_Value(arg, VAR_CMDLINE, &freeIt) != NULL;
288     bmake_free(freeIt);
289     return result;
290 }
291 
292 /* See if the given target is being made. */
293 static Boolean
294 FuncMake(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
295 {
296     StringListNode *ln;
297 
298     for (ln = opts.create->first; ln != NULL; ln = ln->next)
299 	if (Str_Match(ln->datum, arg))
300 	    return TRUE;
301     return FALSE;
302 }
303 
304 /* See if the given file exists. */
305 static Boolean
306 FuncExists(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
307 {
308     Boolean result;
309     char *path;
310 
311     path = Dir_FindFile(arg, dirSearchPath);
312     DEBUG2(COND, "exists(%s) result is \"%s\"\n", arg, path ? path : "");
313     if (path != NULL) {
314 	result = TRUE;
315 	free(path);
316     } else {
317 	result = FALSE;
318     }
319     return result;
320 }
321 
322 /* See if the given node exists and is an actual target. */
323 static Boolean
324 FuncTarget(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
325 {
326     GNode *gn = Targ_FindNode(arg);
327     return gn != NULL && GNode_IsTarget(gn);
328 }
329 
330 /* See if the given node exists and is an actual target with commands
331  * associated with it. */
332 static Boolean
333 FuncCommands(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
334 {
335     GNode *gn = Targ_FindNode(arg);
336     return gn != NULL && GNode_IsTarget(gn) && !Lst_IsEmpty(gn->commands);
337 }
338 
339 /*-
340  * Convert the given number into a double.
341  * We try a base 10 or 16 integer conversion first, if that fails
342  * then we try a floating point conversion instead.
343  *
344  * Results:
345  *	Sets 'value' to double value of string.
346  *	Returns TRUE if the conversion succeeded.
347  */
348 static Boolean
349 TryParseNumber(const char *str, double *value)
350 {
351     char *eptr, ech;
352     unsigned long l_val;
353     double d_val;
354 
355     errno = 0;
356     if (!*str) {
357 	*value = 0.0;
358 	return TRUE;
359     }
360     l_val = strtoul(str, &eptr, str[1] == 'x' ? 16 : 10);
361     ech = *eptr;
362     if (ech == '\0' && errno != ERANGE) {
363 	d_val = str[0] == '-' ? -(double)-l_val : (double)l_val;
364     } else {
365 	if (ech != '\0' && ech != '.' && ech != 'e' && ech != 'E')
366 	    return FALSE;
367 	d_val = strtod(str, &eptr);
368 	if (*eptr)
369 	    return FALSE;
370     }
371 
372     *value = d_val;
373     return TRUE;
374 }
375 
376 static Boolean
377 is_separator(char ch)
378 {
379     return ch == '\0' || ch_isspace(ch) || strchr("!=><)", ch) != NULL;
380 }
381 
382 /*-
383  * Parse a string from a variable reference or an optionally quoted
384  * string.  This is called for the lhs and rhs of string comparisons.
385  *
386  * Results:
387  *	Returns the string, absent any quotes, or NULL on error.
388  *	Sets quoted if the string was quoted.
389  *	Sets freeIt if needed.
390  */
391 /* coverity:[+alloc : arg-*4] */
392 static const char *
393 CondParser_String(CondParser *par, Boolean doEval, Boolean strictLHS,
394 		  Boolean *quoted, void **freeIt)
395 {
396     Buffer buf;
397     const char *str;
398     Boolean atStart;
399     const char *nested_p;
400     Boolean qt;
401     const char *start;
402     VarEvalFlags eflags;
403     VarParseResult parseResult;
404 
405     Buf_Init(&buf, 0);
406     str = NULL;
407     *freeIt = NULL;
408     *quoted = qt = par->p[0] == '"' ? 1 : 0;
409     start = par->p;
410     if (qt)
411 	par->p++;
412     while (par->p[0] && str == NULL) {
413 	switch (par->p[0]) {
414 	case '\\':
415 	    par->p++;
416 	    if (par->p[0] != '\0') {
417 		Buf_AddByte(&buf, par->p[0]);
418 		par->p++;
419 	    }
420 	    continue;
421 	case '"':
422 	    if (qt) {
423 		par->p++;	/* we don't want the quotes */
424 		goto got_str;
425 	    }
426 	    Buf_AddByte(&buf, par->p[0]); /* likely? */
427 	    par->p++;
428 	    continue;
429 	case ')':
430 	case '!':
431 	case '=':
432 	case '>':
433 	case '<':
434 	case ' ':
435 	case '\t':
436 	    if (!qt)
437 		goto got_str;
438 	    Buf_AddByte(&buf, par->p[0]);
439 	    par->p++;
440 	    continue;
441 	case '$':
442 	    /* if we are in quotes, an undefined variable is ok */
443 	    eflags = ((!qt && doEval) ? VARE_UNDEFERR : 0) |
444 		     (doEval ? VARE_WANTRES : 0);
445 	    nested_p = par->p;
446 	    atStart = nested_p == start;
447 	    parseResult = Var_Parse(&nested_p, VAR_CMDLINE, eflags, &str,
448 				    freeIt);
449 	    /* TODO: handle errors */
450 	    if (str == var_Error) {
451 		if (parseResult & VPR_ANY_MSG)
452 		    par->printedError = TRUE;
453 		if (*freeIt) {
454 		    free(*freeIt);
455 		    *freeIt = NULL;
456 		}
457 		/*
458 		 * Even if !doEval, we still report syntax errors, which
459 		 * is what getting var_Error back with !doEval means.
460 		 */
461 		str = NULL;
462 		goto cleanup;
463 	    }
464 	    par->p = nested_p;
465 
466 	    /*
467 	     * If the '$' started the string literal (which means no quotes),
468 	     * and the variable expression is followed by a space, looks like
469 	     * a comparison operator or is the end of the expression, we are
470 	     * done.
471 	     */
472 	    if (atStart && is_separator(par->p[0]))
473 		goto cleanup;
474 
475 	    Buf_AddStr(&buf, str);
476 	    if (*freeIt) {
477 		free(*freeIt);
478 		*freeIt = NULL;
479 	    }
480 	    str = NULL;		/* not finished yet */
481 	    continue;
482 	default:
483 	    if (strictLHS && !qt && *start != '$' && !ch_isdigit(*start)) {
484 		/* lhs must be quoted, a variable reference or number */
485 		if (*freeIt) {
486 		    free(*freeIt);
487 		    *freeIt = NULL;
488 		}
489 		str = NULL;
490 		goto cleanup;
491 	    }
492 	    Buf_AddByte(&buf, par->p[0]);
493 	    par->p++;
494 	    continue;
495 	}
496     }
497 got_str:
498     *freeIt = Buf_GetAll(&buf, NULL);
499     str = *freeIt;
500 cleanup:
501     Buf_Destroy(&buf, FALSE);
502     return str;
503 }
504 
505 /* The different forms of .if directives. */
506 static const struct If {
507     const char *form;		/* Form of if */
508     size_t formlen;		/* Length of form */
509     Boolean doNot;		/* TRUE if default function should be negated */
510     Boolean (*defProc)(size_t, const char *); /* Default function to apply */
511 } ifs[] = {
512     { "def",   3, FALSE, FuncDefined },
513     { "ndef",  4, TRUE,  FuncDefined },
514     { "make",  4, FALSE, FuncMake },
515     { "nmake", 5, TRUE,  FuncMake },
516     { "",      0, FALSE, FuncDefined },
517     { NULL,    0, FALSE, NULL }
518 };
519 
520 /* Evaluate a "comparison without operator", such as in ".if ${VAR}" or
521  * ".if 0". */
522 static Token
523 EvalNotEmpty(CondParser *par, const char *lhs, Boolean lhsQuoted)
524 {
525     double left;
526 
527     /* For .ifxxx "..." check for non-empty string. */
528     if (lhsQuoted)
529 	return lhs[0] != '\0';
530 
531     /* For .ifxxx <number> compare against zero */
532     if (TryParseNumber(lhs, &left))
533 	return left != 0.0;
534 
535     /* For .if ${...} check for non-empty string (defProc is ifdef). */
536     if (par->if_info->form[0] == '\0')
537 	return lhs[0] != 0;
538 
539     /* Otherwise action default test ... */
540     return par->if_info->defProc(strlen(lhs), lhs) == !par->if_info->doNot;
541 }
542 
543 /* Evaluate a numerical comparison, such as in ".if ${VAR} >= 9". */
544 static Token
545 EvalCompareNum(double lhs, const char *op, double rhs)
546 {
547     DEBUG3(COND, "lhs = %f, rhs = %f, op = %.2s\n", lhs, rhs, op);
548 
549     switch (op[0]) {
550     case '!':
551 	if (op[1] != '=') {
552 	    Parse_Error(PARSE_WARNING, "Unknown operator");
553 	    /* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
554 	    return TOK_ERROR;
555 	}
556 	return lhs != rhs;
557     case '=':
558 	if (op[1] != '=') {
559 	    Parse_Error(PARSE_WARNING, "Unknown operator");
560 	    /* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
561 	    return TOK_ERROR;
562 	}
563 	return lhs == rhs;
564     case '<':
565 	return op[1] == '=' ? lhs <= rhs : lhs < rhs;
566     case '>':
567 	return op[1] == '=' ? lhs >= rhs : lhs > rhs;
568     }
569     return TOK_ERROR;
570 }
571 
572 static Token
573 EvalCompareStr(const char *lhs, const char *op, const char *rhs)
574 {
575     if (!((op[0] == '!' || op[0] == '=') && op[1] == '=')) {
576 	Parse_Error(PARSE_WARNING,
577 		    "String comparison operator must be either == or !=");
578 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
579 	return TOK_ERROR;
580     }
581 
582     DEBUG3(COND, "lhs = \"%s\", rhs = \"%s\", op = %.2s\n", lhs, rhs, op);
583     return (*op == '=') == (strcmp(lhs, rhs) == 0);
584 }
585 
586 /* Evaluate a comparison, such as "${VAR} == 12345". */
587 static Token
588 EvalCompare(const char *lhs, Boolean lhsQuoted, const char *op,
589 	    const char *rhs, Boolean rhsQuoted)
590 {
591     double left, right;
592 
593     if (!rhsQuoted && !lhsQuoted)
594 	if (TryParseNumber(lhs, &left) && TryParseNumber(rhs, &right))
595 	    return EvalCompareNum(left, op, right);
596 
597     return EvalCompareStr(lhs, op, rhs);
598 }
599 
600 /* Parse a comparison condition such as:
601  *
602  *	0
603  *	${VAR:Mpattern}
604  *	${VAR} == value
605  *	${VAR:U0} < 12345
606  */
607 static Token
608 CondParser_Comparison(CondParser *par, Boolean doEval)
609 {
610     Token t = TOK_ERROR;
611     const char *lhs, *op, *rhs;
612     void *lhsFree, *rhsFree;
613     Boolean lhsQuoted, rhsQuoted;
614 
615     rhs = NULL;
616     lhsFree = rhsFree = NULL;
617     lhsQuoted = rhsQuoted = FALSE;
618 
619     /*
620      * Parse the variable spec and skip over it, saving its
621      * value in lhs.
622      */
623     lhs = CondParser_String(par, doEval, lhsStrict, &lhsQuoted, &lhsFree);
624     if (!lhs)
625 	goto done;
626 
627     CondParser_SkipWhitespace(par);
628 
629     /*
630      * Make sure the operator is a valid one. If it isn't a
631      * known relational operator, pretend we got a
632      * != 0 comparison.
633      */
634     op = par->p;
635     switch (par->p[0]) {
636     case '!':
637     case '=':
638     case '<':
639     case '>':
640 	if (par->p[1] == '=') {
641 	    par->p += 2;
642 	} else {
643 	    par->p++;
644 	}
645 	break;
646     default:
647 	t = doEval ? EvalNotEmpty(par, lhs, lhsQuoted) : TOK_FALSE;
648 	goto done;
649     }
650 
651     CondParser_SkipWhitespace(par);
652 
653     if (par->p[0] == '\0') {
654 	Parse_Error(PARSE_WARNING, "Missing right-hand-side of operator");
655 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
656 	goto done;
657     }
658 
659     rhs = CondParser_String(par, doEval, FALSE, &rhsQuoted, &rhsFree);
660     if (rhs == NULL)
661 	goto done;
662 
663     if (!doEval) {
664 	t = TOK_FALSE;
665 	goto done;
666     }
667 
668     t = EvalCompare(lhs, lhsQuoted, op, rhs, rhsQuoted);
669 
670 done:
671     free(lhsFree);
672     free(rhsFree);
673     return t;
674 }
675 
676 static size_t
677 ParseEmptyArg(const char **linePtr, Boolean doEval,
678 	      const char *func MAKE_ATTR_UNUSED, char **argPtr)
679 {
680     void *val_freeIt;
681     const char *val;
682     size_t magic_res;
683 
684     /* We do all the work here and return the result as the length */
685     *argPtr = NULL;
686 
687     (*linePtr)--;		/* Make (*linePtr)[1] point to the '('. */
688     (void)Var_Parse(linePtr, VAR_CMDLINE, doEval ? VARE_WANTRES : 0,
689 		    &val, &val_freeIt);
690     /* TODO: handle errors */
691     /* If successful, *linePtr points beyond the closing ')' now. */
692 
693     if (val == var_Error) {
694 	free(val_freeIt);
695 	return (size_t)-1;
696     }
697 
698     /* A variable is empty when it just contains spaces... 4/15/92, christos */
699     cpp_skip_whitespace(&val);
700 
701     /*
702      * For consistency with the other functions we can't generate the
703      * true/false here.
704      */
705     magic_res = *val != '\0' ? 2 : 1;
706     free(val_freeIt);
707     return magic_res;
708 }
709 
710 static Boolean
711 FuncEmpty(size_t arglen, const char *arg MAKE_ATTR_UNUSED)
712 {
713     /* Magic values ahead, see ParseEmptyArg. */
714     return arglen == 1;
715 }
716 
717 static Token
718 CondParser_Func(CondParser *par, Boolean doEval)
719 {
720     static const struct fn_def {
721 	const char *fn_name;
722 	size_t fn_name_len;
723 	size_t (*fn_parse)(const char **, Boolean, const char *, char **);
724 	Boolean (*fn_eval)(size_t, const char *);
725     } fn_defs[] = {
726 	{ "defined",  7, ParseFuncArg,  FuncDefined },
727 	{ "make",     4, ParseFuncArg,  FuncMake },
728 	{ "exists",   6, ParseFuncArg,  FuncExists },
729 	{ "empty",    5, ParseEmptyArg, FuncEmpty },
730 	{ "target",   6, ParseFuncArg,  FuncTarget },
731 	{ "commands", 8, ParseFuncArg,  FuncCommands },
732 	{ NULL,       0, NULL, NULL },
733     };
734     const struct fn_def *fn_def;
735     Token t;
736     char *arg = NULL;
737     size_t arglen;
738     const char *cp = par->p;
739     const char *cp1;
740 
741     for (fn_def = fn_defs; fn_def->fn_name != NULL; fn_def++) {
742 	if (!is_token(cp, fn_def->fn_name, fn_def->fn_name_len))
743 	    continue;
744 	cp += fn_def->fn_name_len;
745 	/* There can only be whitespace before the '(' */
746 	cpp_skip_whitespace(&cp);
747 	if (*cp != '(')
748 	    break;
749 
750 	arglen = fn_def->fn_parse(&cp, doEval, fn_def->fn_name, &arg);
751 	if (arglen == 0 || arglen == (size_t)-1) {
752 	    par->p = cp;
753 	    return arglen == 0 ? TOK_FALSE : TOK_ERROR;
754 	}
755 	/* Evaluate the argument using the required function. */
756 	t = !doEval || fn_def->fn_eval(arglen, arg);
757 	free(arg);
758 	par->p = cp;
759 	return t;
760     }
761 
762     /* Push anything numeric through the compare expression */
763     cp = par->p;
764     if (ch_isdigit(cp[0]) || strchr("+-", cp[0]))
765 	return CondParser_Comparison(par, doEval);
766 
767     /*
768      * Most likely we have a naked token to apply the default function to.
769      * However ".if a == b" gets here when the "a" is unquoted and doesn't
770      * start with a '$'. This surprises people.
771      * If what follows the function argument is a '=' or '!' then the syntax
772      * would be invalid if we did "defined(a)" - so instead treat as an
773      * expression.
774      */
775     arglen = ParseFuncArg(&cp, doEval, NULL, &arg);
776     cp1 = cp;
777     cpp_skip_whitespace(&cp1);
778     if (*cp1 == '=' || *cp1 == '!')
779 	return CondParser_Comparison(par, doEval);
780     par->p = cp;
781 
782     /*
783      * Evaluate the argument using the default function.
784      * This path always treats .if as .ifdef. To get here, the character
785      * after .if must have been taken literally, so the argument cannot
786      * be empty - even if it contained a variable expansion.
787      */
788     t = !doEval || par->if_info->defProc(arglen, arg) == !par->if_info->doNot;
789     free(arg);
790     return t;
791 }
792 
793 /* Return the next token or comparison result from the parser. */
794 static Token
795 CondParser_Token(CondParser *par, Boolean doEval)
796 {
797     Token t;
798 
799     t = par->curr;
800     if (t != TOK_NONE) {
801 	par->curr = TOK_NONE;
802 	return t;
803     }
804 
805     while (par->p[0] == ' ' || par->p[0] == '\t') {
806 	par->p++;
807     }
808 
809     switch (par->p[0]) {
810 
811     case '(':
812 	par->p++;
813 	return TOK_LPAREN;
814 
815     case ')':
816 	par->p++;
817 	return TOK_RPAREN;
818 
819     case '|':
820 	par->p++;
821 	if (par->p[0] == '|') {
822 	    par->p++;
823 	}
824 	return TOK_OR;
825 
826     case '&':
827 	par->p++;
828 	if (par->p[0] == '&') {
829 	    par->p++;
830 	}
831 	return TOK_AND;
832 
833     case '!':
834 	par->p++;
835 	return TOK_NOT;
836 
837     case '#':
838     case '\n':
839     case '\0':
840 	return TOK_EOF;
841 
842     case '"':
843     case '$':
844 	return CondParser_Comparison(par, doEval);
845 
846     default:
847 	return CondParser_Func(par, doEval);
848     }
849 }
850 
851 /* Parse a single term in the expression. This consists of a terminal symbol
852  * or TOK_NOT and a term (not including the binary operators):
853  *
854  *	T -> defined(variable) | make(target) | exists(file) | symbol
855  *	T -> ! T | ( E )
856  *
857  * Results:
858  *	TOK_TRUE, TOK_FALSE or TOK_ERROR.
859  */
860 static Token
861 CondParser_Term(CondParser *par, Boolean doEval)
862 {
863     Token t;
864 
865     t = CondParser_Token(par, doEval);
866 
867     if (t == TOK_EOF) {
868 	/*
869 	 * If we reached the end of the expression, the expression
870 	 * is malformed...
871 	 */
872 	t = TOK_ERROR;
873     } else if (t == TOK_LPAREN) {
874 	/*
875 	 * T -> ( E )
876 	 */
877 	t = CondParser_Expr(par, doEval);
878 	if (t != TOK_ERROR) {
879 	    if (CondParser_Token(par, doEval) != TOK_RPAREN) {
880 		t = TOK_ERROR;
881 	    }
882 	}
883     } else if (t == TOK_NOT) {
884 	t = CondParser_Term(par, doEval);
885 	if (t == TOK_TRUE) {
886 	    t = TOK_FALSE;
887 	} else if (t == TOK_FALSE) {
888 	    t = TOK_TRUE;
889 	}
890     }
891     return t;
892 }
893 
894 /* Parse a conjunctive factor (nice name, wot?)
895  *
896  *	F -> T && F | T
897  *
898  * Results:
899  *	TOK_TRUE, TOK_FALSE or TOK_ERROR
900  */
901 static Token
902 CondParser_Factor(CondParser *par, Boolean doEval)
903 {
904     Token l, o;
905 
906     l = CondParser_Term(par, doEval);
907     if (l != TOK_ERROR) {
908 	o = CondParser_Token(par, doEval);
909 
910 	if (o == TOK_AND) {
911 	    /*
912 	     * F -> T && F
913 	     *
914 	     * If T is TOK_FALSE, the whole thing will be TOK_FALSE, but we
915 	     * have to parse the r.h.s. anyway (to throw it away).
916 	     * If T is TOK_TRUE, the result is the r.h.s., be it a TOK_ERROR
917 	     * or not.
918 	     */
919 	    if (l == TOK_TRUE) {
920 		l = CondParser_Factor(par, doEval);
921 	    } else {
922 		(void)CondParser_Factor(par, FALSE);
923 	    }
924 	} else {
925 	    /*
926 	     * F -> T
927 	     */
928 	    CondParser_PushBack(par, o);
929 	}
930     }
931     return l;
932 }
933 
934 /* Main expression production.
935  *
936  *	E -> F || E | F
937  *
938  * Results:
939  *	TOK_TRUE, TOK_FALSE or TOK_ERROR.
940  */
941 static Token
942 CondParser_Expr(CondParser *par, Boolean doEval)
943 {
944     Token l, o;
945 
946     l = CondParser_Factor(par, doEval);
947     if (l != TOK_ERROR) {
948 	o = CondParser_Token(par, doEval);
949 
950 	if (o == TOK_OR) {
951 	    /*
952 	     * E -> F || E
953 	     *
954 	     * A similar thing occurs for ||, except that here we make sure
955 	     * the l.h.s. is TOK_FALSE before we bother to evaluate the r.h.s.
956 	     * Once again, if l is TOK_FALSE, the result is the r.h.s. and once
957 	     * again if l is TOK_TRUE, we parse the r.h.s. to throw it away.
958 	     */
959 	    if (l == TOK_FALSE) {
960 		l = CondParser_Expr(par, doEval);
961 	    } else {
962 		(void)CondParser_Expr(par, FALSE);
963 	    }
964 	} else {
965 	    /*
966 	     * E -> F
967 	     */
968 	    CondParser_PushBack(par, o);
969 	}
970     }
971     return l;
972 }
973 
974 static CondEvalResult
975 CondParser_Eval(CondParser *par, Boolean *value)
976 {
977     Token res;
978 
979     DEBUG1(COND, "CondParser_Eval: %s\n", par->p);
980 
981     res = CondParser_Expr(par, TRUE);
982     if (res != TOK_FALSE && res != TOK_TRUE)
983 	return COND_INVALID;
984 
985     if (CondParser_Token(par, TRUE /* XXX: Why TRUE? */) != TOK_EOF)
986 	return COND_INVALID;
987 
988     *value = res == TOK_TRUE;
989     return COND_PARSE;
990 }
991 
992 /* Evaluate the condition, including any side effects from the variable
993  * expressions in the condition. The condition consists of &&, ||, !,
994  * function(arg), comparisons and parenthetical groupings thereof.
995  *
996  * Results:
997  *	COND_PARSE	if the condition was valid grammatically
998  *	COND_INVALID	if not a valid conditional.
999  *
1000  *	(*value) is set to the boolean value of the condition
1001  */
1002 static CondEvalResult
1003 CondEvalExpression(const struct If *info, const char *cond, Boolean *value,
1004 		    Boolean eprint, Boolean strictLHS)
1005 {
1006     static const struct If *dflt_info;
1007     CondParser par;
1008     int rval;
1009 
1010     lhsStrict = strictLHS;
1011 
1012     while (*cond == ' ' || *cond == '\t')
1013 	cond++;
1014 
1015     if (info == NULL && (info = dflt_info) == NULL) {
1016 	/* Scan for the entry for .if - it can't be first */
1017 	for (info = ifs;; info++)
1018 	    if (info->form[0] == 0)
1019 		break;
1020 	dflt_info = info;
1021     }
1022     assert(info != NULL);
1023 
1024     par.if_info = info;
1025     par.p = cond;
1026     par.curr = TOK_NONE;
1027     par.printedError = FALSE;
1028 
1029     rval = CondParser_Eval(&par, value);
1030 
1031     if (rval == COND_INVALID && eprint && !par.printedError)
1032 	Parse_Error(PARSE_FATAL, "Malformed conditional (%s)", cond);
1033 
1034     return rval;
1035 }
1036 
1037 CondEvalResult
1038 Cond_EvalCondition(const char *cond, Boolean *out_value)
1039 {
1040 	return CondEvalExpression(NULL, cond, out_value, FALSE, FALSE);
1041 }
1042 
1043 /* Evaluate the conditional in the passed line. The line looks like this:
1044  *	.<cond-type> <expr>
1045  * In this line, <cond-type> is any of if, ifmake, ifnmake, ifdef, ifndef,
1046  * elif, elifmake, elifnmake, elifdef, elifndef.
1047  * In this line, <expr> consists of &&, ||, !, function(arg), comparisons
1048  * and parenthetical groupings thereof.
1049  *
1050  * Note that the states IF_ACTIVE and ELSE_ACTIVE are only different in order
1051  * to detect spurious .else lines (as are SKIP_TO_ELSE and SKIP_TO_ENDIF),
1052  * otherwise .else could be treated as '.elif 1'.
1053  *
1054  * Results:
1055  *	COND_PARSE	to continue parsing the lines after the conditional
1056  *			(when .if or .else returns TRUE)
1057  *	COND_SKIP	to skip the lines after the conditional
1058  *			(when .if or .elif returns FALSE, or when a previous
1059  *			branch has already been taken)
1060  *	COND_INVALID	if the conditional was not valid, either because of
1061  *			a syntax error or because some variable was undefined
1062  *			or because the condition could not be evaluated
1063  */
1064 CondEvalResult
1065 Cond_EvalLine(const char *line)
1066 {
1067     enum { MAXIF = 128 };	/* maximum depth of .if'ing */
1068     enum { MAXIF_BUMP = 32 };	/* how much to grow by */
1069     enum if_states {
1070 	IF_ACTIVE,		/* .if or .elif part active */
1071 	ELSE_ACTIVE,		/* .else part active */
1072 	SEARCH_FOR_ELIF,	/* searching for .elif/else to execute */
1073 	SKIP_TO_ELSE,		/* has been true, but not seen '.else' */
1074 	SKIP_TO_ENDIF		/* nothing else to execute */
1075     };
1076     static enum if_states *cond_state = NULL;
1077     static unsigned int max_if_depth = MAXIF;
1078 
1079     const struct If *ifp;
1080     Boolean isElif;
1081     Boolean value;
1082     enum if_states state;
1083 
1084     if (!cond_state) {
1085 	cond_state = bmake_malloc(max_if_depth * sizeof(*cond_state));
1086 	cond_state[0] = IF_ACTIVE;
1087     }
1088     /* skip leading character (the '.') and any whitespace */
1089     for (line++; *line == ' ' || *line == '\t'; line++)
1090 	continue;
1091 
1092     /* Find what type of if we're dealing with.  */
1093     if (line[0] == 'e') {
1094 	if (line[1] != 'l') {
1095 	    if (!is_token(line + 1, "ndif", 4))
1096 		return COND_INVALID;
1097 	    /* End of conditional section */
1098 	    if (cond_depth == cond_min_depth) {
1099 		Parse_Error(PARSE_FATAL, "if-less endif");
1100 		return COND_PARSE;
1101 	    }
1102 	    /* Return state for previous conditional */
1103 	    cond_depth--;
1104 	    return cond_state[cond_depth] <= ELSE_ACTIVE
1105 		   ? COND_PARSE : COND_SKIP;
1106 	}
1107 
1108 	/* Quite likely this is 'else' or 'elif' */
1109 	line += 2;
1110 	if (is_token(line, "se", 2)) {
1111 	    /* It is else... */
1112 	    if (cond_depth == cond_min_depth) {
1113 		Parse_Error(PARSE_FATAL, "if-less else");
1114 		return COND_PARSE;
1115 	    }
1116 
1117 	    state = cond_state[cond_depth];
1118 	    switch (state) {
1119 	    case SEARCH_FOR_ELIF:
1120 		state = ELSE_ACTIVE;
1121 		break;
1122 	    case ELSE_ACTIVE:
1123 	    case SKIP_TO_ENDIF:
1124 		Parse_Error(PARSE_WARNING, "extra else");
1125 		/* FALLTHROUGH */
1126 	    default:
1127 	    case IF_ACTIVE:
1128 	    case SKIP_TO_ELSE:
1129 		state = SKIP_TO_ENDIF;
1130 		break;
1131 	    }
1132 	    cond_state[cond_depth] = state;
1133 	    return state <= ELSE_ACTIVE ? COND_PARSE : COND_SKIP;
1134 	}
1135 	/* Assume for now it is an elif */
1136 	isElif = TRUE;
1137     } else
1138 	isElif = FALSE;
1139 
1140     if (line[0] != 'i' || line[1] != 'f')
1141 	/* Not an ifxxx or elifxxx line */
1142 	return COND_INVALID;
1143 
1144     /*
1145      * Figure out what sort of conditional it is -- what its default
1146      * function is, etc. -- by looking in the table of valid "ifs"
1147      */
1148     line += 2;
1149     for (ifp = ifs;; ifp++) {
1150 	if (ifp->form == NULL)
1151 	    return COND_INVALID;
1152 	if (is_token(ifp->form, line, ifp->formlen)) {
1153 	    line += ifp->formlen;
1154 	    break;
1155 	}
1156     }
1157 
1158     /* Now we know what sort of 'if' it is... */
1159 
1160     if (isElif) {
1161 	if (cond_depth == cond_min_depth) {
1162 	    Parse_Error(PARSE_FATAL, "if-less elif");
1163 	    return COND_PARSE;
1164 	}
1165 	state = cond_state[cond_depth];
1166 	if (state == SKIP_TO_ENDIF || state == ELSE_ACTIVE) {
1167 	    Parse_Error(PARSE_WARNING, "extra elif");
1168 	    cond_state[cond_depth] = SKIP_TO_ENDIF;
1169 	    return COND_SKIP;
1170 	}
1171 	if (state != SEARCH_FOR_ELIF) {
1172 	    /* Either just finished the 'true' block, or already SKIP_TO_ELSE */
1173 	    cond_state[cond_depth] = SKIP_TO_ELSE;
1174 	    return COND_SKIP;
1175 	}
1176     } else {
1177 	/* Normal .if */
1178 	if (cond_depth + 1 >= max_if_depth) {
1179 	    /*
1180 	     * This is rare, but not impossible.
1181 	     * In meta mode, dirdeps.mk (only runs at level 0)
1182 	     * can need more than the default.
1183 	     */
1184 	    max_if_depth += MAXIF_BUMP;
1185 	    cond_state = bmake_realloc(cond_state,
1186 				       max_if_depth * sizeof(*cond_state));
1187 	}
1188 	state = cond_state[cond_depth];
1189 	cond_depth++;
1190 	if (state > ELSE_ACTIVE) {
1191 	    /* If we aren't parsing the data, treat as always false */
1192 	    cond_state[cond_depth] = SKIP_TO_ELSE;
1193 	    return COND_SKIP;
1194 	}
1195     }
1196 
1197     /* And evaluate the conditional expression */
1198     if (CondEvalExpression(ifp, line, &value, TRUE, TRUE) == COND_INVALID) {
1199 	/* Syntax error in conditional, error message already output. */
1200 	/* Skip everything to matching .endif */
1201 	cond_state[cond_depth] = SKIP_TO_ELSE;
1202 	return COND_SKIP;
1203     }
1204 
1205     if (!value) {
1206 	cond_state[cond_depth] = SEARCH_FOR_ELIF;
1207 	return COND_SKIP;
1208     }
1209     cond_state[cond_depth] = IF_ACTIVE;
1210     return COND_PARSE;
1211 }
1212 
1213 void
1214 Cond_restore_depth(unsigned int saved_depth)
1215 {
1216     unsigned int open_conds = cond_depth - cond_min_depth;
1217 
1218     if (open_conds != 0 || saved_depth > cond_depth) {
1219 	Parse_Error(PARSE_FATAL, "%u open conditional%s", open_conds,
1220 		    open_conds == 1 ? "" : "s");
1221 	cond_depth = cond_min_depth;
1222     }
1223 
1224     cond_min_depth = saved_depth;
1225 }
1226 
1227 unsigned int
1228 Cond_save_depth(void)
1229 {
1230     unsigned int depth = cond_min_depth;
1231 
1232     cond_min_depth = cond_depth;
1233     return depth;
1234 }
1235