Home | History | Annotate | Line # | Download | only in make
cond.c revision 1.168
      1 /*	$NetBSD: cond.c,v 1.168 2020/10/24 04:51:19 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 1.168 2020/10/24 04:51:19 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_CMD, eflags, &nestedVal, &nestedVal_freeIt);
    250 	    /* TODO: handle errors */
    251 	    Buf_AddStr(&argBuf, nestedVal);
    252 	    free(nestedVal_freeIt);
    253 	    continue;
    254 	}
    255 	if (ch == '(')
    256 	    paren_depth++;
    257 	else if (ch == ')' && --paren_depth < 0)
    258 	    break;
    259 	Buf_AddByte(&argBuf, *p);
    260 	p++;
    261     }
    262 
    263     *out_arg = Buf_GetAll(&argBuf, &argLen);
    264     Buf_Destroy(&argBuf, FALSE);
    265 
    266     while (*p == ' ' || *p == '\t') {
    267 	p++;
    268     }
    269 
    270     if (func != NULL && *p++ != ')') {
    271 	Parse_Error(PARSE_WARNING, "Missing closing parenthesis for %s()",
    272 		    func);
    273 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    274 	return 0;
    275     }
    276 
    277     *pp = p;
    278     return argLen;
    279 }
    280 
    281 /* Test whether the given variable is defined. */
    282 static Boolean
    283 FuncDefined(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
    284 {
    285     char *freeIt;
    286     Boolean result = Var_Value(arg, VAR_CMD, &freeIt) != NULL;
    287     bmake_free(freeIt);
    288     return result;
    289 }
    290 
    291 /* See if the given target is being made. */
    292 static Boolean
    293 FuncMake(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
    294 {
    295     StringListNode *ln;
    296 
    297     for (ln = create->first; ln != NULL; ln = ln->next)
    298 	if (Str_Match(ln->datum, arg))
    299 	    return TRUE;
    300     return FALSE;
    301 }
    302 
    303 /* See if the given file exists. */
    304 static Boolean
    305 FuncExists(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
    306 {
    307     Boolean result;
    308     char *path;
    309 
    310     path = Dir_FindFile(arg, dirSearchPath);
    311     DEBUG2(COND, "exists(%s) result is \"%s\"\n", arg, path ? path : "");
    312     if (path != NULL) {
    313 	result = TRUE;
    314 	free(path);
    315     } else {
    316 	result = FALSE;
    317     }
    318     return result;
    319 }
    320 
    321 /* See if the given node exists and is an actual target. */
    322 static Boolean
    323 FuncTarget(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
    324 {
    325     GNode *gn = Targ_FindNode(arg);
    326     return gn != NULL && GNode_IsTarget(gn);
    327 }
    328 
    329 /* See if the given node exists and is an actual target with commands
    330  * associated with it. */
    331 static Boolean
    332 FuncCommands(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
    333 {
    334     GNode *gn = Targ_FindNode(arg);
    335     return gn != NULL && GNode_IsTarget(gn) && !Lst_IsEmpty(gn->commands);
    336 }
    337 
    338 /*-
    339  * Convert the given number into a double.
    340  * We try a base 10 or 16 integer conversion first, if that fails
    341  * then we try a floating point conversion instead.
    342  *
    343  * Results:
    344  *	Sets 'value' to double value of string.
    345  *	Returns TRUE if the conversion succeeded.
    346  */
    347 static Boolean
    348 TryParseNumber(const char *str, double *value)
    349 {
    350     char *eptr, ech;
    351     unsigned long l_val;
    352     double d_val;
    353 
    354     errno = 0;
    355     if (!*str) {
    356 	*value = 0.0;
    357 	return TRUE;
    358     }
    359     l_val = strtoul(str, &eptr, str[1] == 'x' ? 16 : 10);
    360     ech = *eptr;
    361     if (ech == '\0' && errno != ERANGE) {
    362 	d_val = str[0] == '-' ? -(double)-l_val : (double)l_val;
    363     } else {
    364 	if (ech != '\0' && ech != '.' && ech != 'e' && ech != 'E')
    365 	    return FALSE;
    366 	d_val = strtod(str, &eptr);
    367 	if (*eptr)
    368 	    return FALSE;
    369     }
    370 
    371     *value = d_val;
    372     return TRUE;
    373 }
    374 
    375 static Boolean
    376 is_separator(char ch)
    377 {
    378     return ch == '\0' || ch_isspace(ch) || strchr("!=><)", ch);
    379 }
    380 
    381 /*-
    382  * Parse a string from a variable reference or an optionally quoted
    383  * string.  This is called for the lhs and rhs of string comparisons.
    384  *
    385  * Results:
    386  *	Returns the string, absent any quotes, or NULL on error.
    387  *	Sets quoted if the string was quoted.
    388  *	Sets freeIt if needed.
    389  */
    390 /* coverity:[+alloc : arg-*4] */
    391 static const char *
    392 CondParser_String(CondParser *par, Boolean doEval, Boolean strictLHS,
    393 		  Boolean *quoted, void **freeIt)
    394 {
    395     Buffer buf;
    396     const char *str;
    397     Boolean atStart;
    398     const char *nested_p;
    399     Boolean qt;
    400     const char *start;
    401     VarEvalFlags eflags;
    402     VarParseResult parseResult;
    403 
    404     Buf_Init(&buf, 0);
    405     str = NULL;
    406     *freeIt = NULL;
    407     *quoted = qt = par->p[0] == '"' ? 1 : 0;
    408     if (qt)
    409 	par->p++;
    410     start = par->p;
    411     while (par->p[0] && str == NULL) {
    412 	switch (par->p[0]) {
    413 	case '\\':
    414 	    par->p++;
    415 	    if (par->p[0] != '\0') {
    416 		Buf_AddByte(&buf, par->p[0]);
    417 		par->p++;
    418 	    }
    419 	    continue;
    420 	case '"':
    421 	    if (qt) {
    422 		par->p++;	/* we don't want the quotes */
    423 		goto got_str;
    424 	    }
    425 	    Buf_AddByte(&buf, par->p[0]); /* likely? */
    426 	    par->p++;
    427 	    continue;
    428 	case ')':
    429 	case '!':
    430 	case '=':
    431 	case '>':
    432 	case '<':
    433 	case ' ':
    434 	case '\t':
    435 	    if (!qt)
    436 		goto got_str;
    437 	    Buf_AddByte(&buf, par->p[0]);
    438 	    par->p++;
    439 	    continue;
    440 	case '$':
    441 	    /* if we are in quotes, an undefined variable is ok */
    442 	    eflags = ((!qt && doEval) ? VARE_UNDEFERR : 0) |
    443 		     (doEval ? VARE_WANTRES : 0);
    444 	    nested_p = par->p;
    445 	    atStart = nested_p == start;
    446 	    parseResult = Var_Parse(&nested_p, VAR_CMD, eflags, &str, freeIt);
    447 	    /* TODO: handle errors */
    448 	    if (str == var_Error) {
    449 		if (parseResult & VPR_ANY_MSG)
    450 		    par->printedError = TRUE;
    451 		if (*freeIt) {
    452 		    free(*freeIt);
    453 		    *freeIt = NULL;
    454 		}
    455 		/*
    456 		 * Even if !doEval, we still report syntax errors, which
    457 		 * is what getting var_Error back with !doEval means.
    458 		 */
    459 		str = NULL;
    460 		goto cleanup;
    461 	    }
    462 	    par->p = nested_p;
    463 
    464 	    /*
    465 	     * If the '$' started the string literal (which means no quotes),
    466 	     * and the variable expression is followed by a space, looks like
    467 	     * a comparison operator or is the end of the expression, we are
    468 	     * done.
    469 	     */
    470 	    if (atStart && is_separator(par->p[0]))
    471 		goto cleanup;
    472 
    473 	    Buf_AddStr(&buf, str);
    474 	    if (*freeIt) {
    475 		free(*freeIt);
    476 		*freeIt = NULL;
    477 	    }
    478 	    str = NULL;		/* not finished yet */
    479 	    continue;
    480 	default:
    481 	    if (strictLHS && !qt && *start != '$' && !ch_isdigit(*start)) {
    482 		/* lhs must be quoted, a variable reference or number */
    483 		if (*freeIt) {
    484 		    free(*freeIt);
    485 		    *freeIt = NULL;
    486 		}
    487 		str = NULL;
    488 		goto cleanup;
    489 	    }
    490 	    Buf_AddByte(&buf, par->p[0]);
    491 	    par->p++;
    492 	    continue;
    493 	}
    494     }
    495 got_str:
    496     *freeIt = Buf_GetAll(&buf, NULL);
    497     str = *freeIt;
    498 cleanup:
    499     Buf_Destroy(&buf, FALSE);
    500     return str;
    501 }
    502 
    503 /* The different forms of .if directives. */
    504 static const struct If {
    505     const char *form;		/* Form of if */
    506     size_t formlen;		/* Length of form */
    507     Boolean doNot;		/* TRUE if default function should be negated */
    508     Boolean (*defProc)(size_t, const char *); /* Default function to apply */
    509 } ifs[] = {
    510     { "def",   3, FALSE, FuncDefined },
    511     { "ndef",  4, TRUE,  FuncDefined },
    512     { "make",  4, FALSE, FuncMake },
    513     { "nmake", 5, TRUE,  FuncMake },
    514     { "",      0, FALSE, FuncDefined },
    515     { NULL,    0, FALSE, NULL }
    516 };
    517 
    518 /* Evaluate a "comparison without operator", such as in ".if ${VAR}" or
    519  * ".if 0". */
    520 static Token
    521 EvalNotEmpty(CondParser *par, const char *lhs, Boolean lhsQuoted)
    522 {
    523     double left;
    524 
    525     /* For .ifxxx "..." check for non-empty string. */
    526     if (lhsQuoted)
    527 	return lhs[0] != '\0';
    528 
    529     /* For .ifxxx <number> compare against zero */
    530     if (TryParseNumber(lhs, &left))
    531 	return left != 0.0;
    532 
    533     /* For .if ${...} check for non-empty string (defProc is ifdef). */
    534     if (par->if_info->form[0] == '\0')
    535 	return lhs[0] != 0;
    536 
    537     /* Otherwise action default test ... */
    538     return par->if_info->defProc(strlen(lhs), lhs) == !par->if_info->doNot;
    539 }
    540 
    541 /* Evaluate a numerical comparison, such as in ".if ${VAR} >= 9". */
    542 static Token
    543 EvalCompareNum(double lhs, const char *op, double rhs)
    544 {
    545     DEBUG3(COND, "lhs = %f, rhs = %f, op = %.2s\n", lhs, rhs, op);
    546 
    547     switch (op[0]) {
    548     case '!':
    549 	if (op[1] != '=') {
    550 	    Parse_Error(PARSE_WARNING, "Unknown operator");
    551 	    /* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    552 	    return TOK_ERROR;
    553 	}
    554 	return lhs != rhs;
    555     case '=':
    556 	if (op[1] != '=') {
    557 	    Parse_Error(PARSE_WARNING, "Unknown operator");
    558 	    /* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    559 	    return TOK_ERROR;
    560 	}
    561 	return lhs == rhs;
    562     case '<':
    563 	return op[1] == '=' ? lhs <= rhs : lhs < rhs;
    564     case '>':
    565 	return op[1] == '=' ? lhs >= rhs : lhs > rhs;
    566     }
    567     return TOK_ERROR;
    568 }
    569 
    570 static Token
    571 EvalCompareStr(const char *lhs, const char *op, const char *rhs)
    572 {
    573     if (!((op[0] == '!' || op[0] == '=') && op[1] == '=')) {
    574 	Parse_Error(PARSE_WARNING,
    575 		    "String comparison operator must be either == or !=");
    576 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    577 	return TOK_ERROR;
    578     }
    579 
    580     DEBUG3(COND, "lhs = \"%s\", rhs = \"%s\", op = %.2s\n", lhs, rhs, op);
    581     return (*op == '=') == (strcmp(lhs, rhs) == 0);
    582 }
    583 
    584 /* Evaluate a comparison, such as "${VAR} == 12345". */
    585 static Token
    586 EvalCompare(const char *lhs, Boolean lhsQuoted, const char *op,
    587 	    const char *rhs, Boolean rhsQuoted)
    588 {
    589     double left, right;
    590 
    591     if (!rhsQuoted && !lhsQuoted)
    592 	if (TryParseNumber(lhs, &left) && TryParseNumber(rhs, &right))
    593 	    return EvalCompareNum(left, op, right);
    594 
    595     return EvalCompareStr(lhs, op, rhs);
    596 }
    597 
    598 /* Parse a comparison condition such as:
    599  *
    600  *	0
    601  *	${VAR:Mpattern}
    602  *	${VAR} == value
    603  *	${VAR:U0} < 12345
    604  */
    605 static Token
    606 CondParser_Comparison(CondParser *par, Boolean doEval)
    607 {
    608     Token t = TOK_ERROR;
    609     const char *lhs, *op, *rhs;
    610     void *lhsFree, *rhsFree;
    611     Boolean lhsQuoted, rhsQuoted;
    612 
    613     rhs = NULL;
    614     lhsFree = rhsFree = NULL;
    615     lhsQuoted = rhsQuoted = FALSE;
    616 
    617     /*
    618      * Parse the variable spec and skip over it, saving its
    619      * value in lhs.
    620      */
    621     lhs = CondParser_String(par, doEval, lhsStrict, &lhsQuoted, &lhsFree);
    622     if (!lhs)
    623 	goto done;
    624 
    625     CondParser_SkipWhitespace(par);
    626 
    627     /*
    628      * Make sure the operator is a valid one. If it isn't a
    629      * known relational operator, pretend we got a
    630      * != 0 comparison.
    631      */
    632     op = par->p;
    633     switch (par->p[0]) {
    634     case '!':
    635     case '=':
    636     case '<':
    637     case '>':
    638 	if (par->p[1] == '=') {
    639 	    par->p += 2;
    640 	} else {
    641 	    par->p++;
    642 	}
    643 	break;
    644     default:
    645 	t = doEval ? EvalNotEmpty(par, lhs, lhsQuoted) : TOK_FALSE;
    646 	goto done;
    647     }
    648 
    649     CondParser_SkipWhitespace(par);
    650 
    651     if (par->p[0] == '\0') {
    652 	Parse_Error(PARSE_WARNING, "Missing right-hand-side of operator");
    653 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    654 	goto done;
    655     }
    656 
    657     rhs = CondParser_String(par, doEval, FALSE, &rhsQuoted, &rhsFree);
    658     if (rhs == NULL)
    659 	goto done;
    660 
    661     if (!doEval) {
    662 	t = TOK_FALSE;
    663 	goto done;
    664     }
    665 
    666     t = EvalCompare(lhs, lhsQuoted, op, rhs, rhsQuoted);
    667 
    668 done:
    669     free(lhsFree);
    670     free(rhsFree);
    671     return t;
    672 }
    673 
    674 static size_t
    675 ParseEmptyArg(const char **linePtr, Boolean doEval,
    676 	      const char *func MAKE_ATTR_UNUSED, char **argPtr)
    677 {
    678     void *val_freeIt;
    679     const char *val;
    680     size_t magic_res;
    681 
    682     /* We do all the work here and return the result as the length */
    683     *argPtr = NULL;
    684 
    685     (*linePtr)--;		/* Make (*linePtr)[1] point to the '('. */
    686     (void)Var_Parse(linePtr, VAR_CMD, doEval ? VARE_WANTRES : 0,
    687 		    &val, &val_freeIt);
    688     /* TODO: handle errors */
    689     /* If successful, *linePtr points beyond the closing ')' now. */
    690 
    691     if (val == var_Error) {
    692 	free(val_freeIt);
    693 	return (size_t)-1;
    694     }
    695 
    696     /* A variable is empty when it just contains spaces... 4/15/92, christos */
    697     cpp_skip_whitespace(&val);
    698 
    699     /*
    700      * For consistency with the other functions we can't generate the
    701      * true/false here.
    702      */
    703     magic_res = *val != '\0' ? 2 : 1;
    704     free(val_freeIt);
    705     return magic_res;
    706 }
    707 
    708 static Boolean
    709 FuncEmpty(size_t arglen, const char *arg MAKE_ATTR_UNUSED)
    710 {
    711     /* Magic values ahead, see ParseEmptyArg. */
    712     return arglen == 1;
    713 }
    714 
    715 static Token
    716 CondParser_Func(CondParser *par, Boolean doEval)
    717 {
    718     static const struct fn_def {
    719 	const char *fn_name;
    720 	size_t fn_name_len;
    721 	size_t (*fn_parse)(const char **, Boolean, const char *, char **);
    722 	Boolean (*fn_eval)(size_t, const char *);
    723     } fn_defs[] = {
    724 	{ "defined",  7, ParseFuncArg,  FuncDefined },
    725 	{ "make",     4, ParseFuncArg,  FuncMake },
    726 	{ "exists",   6, ParseFuncArg,  FuncExists },
    727 	{ "empty",    5, ParseEmptyArg, FuncEmpty },
    728 	{ "target",   6, ParseFuncArg,  FuncTarget },
    729 	{ "commands", 8, ParseFuncArg,  FuncCommands },
    730 	{ NULL,       0, NULL, NULL },
    731     };
    732     const struct fn_def *fn_def;
    733     Token t;
    734     char *arg = NULL;
    735     size_t arglen;
    736     const char *cp = par->p;
    737     const char *cp1;
    738 
    739     for (fn_def = fn_defs; fn_def->fn_name != NULL; fn_def++) {
    740 	if (!is_token(cp, fn_def->fn_name, fn_def->fn_name_len))
    741 	    continue;
    742 	cp += fn_def->fn_name_len;
    743 	/* There can only be whitespace before the '(' */
    744 	cpp_skip_whitespace(&cp);
    745 	if (*cp != '(')
    746 	    break;
    747 
    748 	arglen = fn_def->fn_parse(&cp, doEval, fn_def->fn_name, &arg);
    749 	if (arglen == 0 || arglen == (size_t)-1) {
    750 	    par->p = cp;
    751 	    return arglen == 0 ? TOK_FALSE : TOK_ERROR;
    752 	}
    753 	/* Evaluate the argument using the required function. */
    754 	t = !doEval || fn_def->fn_eval(arglen, arg);
    755 	free(arg);
    756 	par->p = cp;
    757 	return t;
    758     }
    759 
    760     /* Push anything numeric through the compare expression */
    761     cp = par->p;
    762     if (ch_isdigit(cp[0]) || strchr("+-", cp[0]))
    763 	return CondParser_Comparison(par, doEval);
    764 
    765     /*
    766      * Most likely we have a naked token to apply the default function to.
    767      * However ".if a == b" gets here when the "a" is unquoted and doesn't
    768      * start with a '$'. This surprises people.
    769      * If what follows the function argument is a '=' or '!' then the syntax
    770      * would be invalid if we did "defined(a)" - so instead treat as an
    771      * expression.
    772      */
    773     arglen = ParseFuncArg(&cp, doEval, NULL, &arg);
    774     cp1 = cp;
    775     cpp_skip_whitespace(&cp1);
    776     if (*cp1 == '=' || *cp1 == '!')
    777 	return CondParser_Comparison(par, doEval);
    778     par->p = cp;
    779 
    780     /*
    781      * Evaluate the argument using the default function.
    782      * This path always treats .if as .ifdef. To get here, the character
    783      * after .if must have been taken literally, so the argument cannot
    784      * be empty - even if it contained a variable expansion.
    785      */
    786     t = !doEval || par->if_info->defProc(arglen, arg) == !par->if_info->doNot;
    787     free(arg);
    788     return t;
    789 }
    790 
    791 /* Return the next token or comparison result from the parser. */
    792 static Token
    793 CondParser_Token(CondParser *par, Boolean doEval)
    794 {
    795     Token t;
    796 
    797     t = par->curr;
    798     if (t != TOK_NONE) {
    799 	par->curr = TOK_NONE;
    800 	return t;
    801     }
    802 
    803     while (par->p[0] == ' ' || par->p[0] == '\t') {
    804 	par->p++;
    805     }
    806 
    807     switch (par->p[0]) {
    808 
    809     case '(':
    810 	par->p++;
    811 	return TOK_LPAREN;
    812 
    813     case ')':
    814 	par->p++;
    815 	return TOK_RPAREN;
    816 
    817     case '|':
    818 	par->p++;
    819 	if (par->p[0] == '|') {
    820 	    par->p++;
    821 	}
    822 	return TOK_OR;
    823 
    824     case '&':
    825 	par->p++;
    826 	if (par->p[0] == '&') {
    827 	    par->p++;
    828 	}
    829 	return TOK_AND;
    830 
    831     case '!':
    832 	par->p++;
    833 	return TOK_NOT;
    834 
    835     case '#':
    836     case '\n':
    837     case '\0':
    838 	return TOK_EOF;
    839 
    840     case '"':
    841     case '$':
    842 	return CondParser_Comparison(par, doEval);
    843 
    844     default:
    845 	return CondParser_Func(par, doEval);
    846     }
    847 }
    848 
    849 /* Parse a single term in the expression. This consists of a terminal symbol
    850  * or TOK_NOT and a term (not including the binary operators):
    851  *
    852  *	T -> defined(variable) | make(target) | exists(file) | symbol
    853  *	T -> ! T | ( E )
    854  *
    855  * Results:
    856  *	TOK_TRUE, TOK_FALSE or TOK_ERROR.
    857  */
    858 static Token
    859 CondParser_Term(CondParser *par, Boolean doEval)
    860 {
    861     Token t;
    862 
    863     t = CondParser_Token(par, doEval);
    864 
    865     if (t == TOK_EOF) {
    866 	/*
    867 	 * If we reached the end of the expression, the expression
    868 	 * is malformed...
    869 	 */
    870 	t = TOK_ERROR;
    871     } else if (t == TOK_LPAREN) {
    872 	/*
    873 	 * T -> ( E )
    874 	 */
    875 	t = CondParser_Expr(par, doEval);
    876 	if (t != TOK_ERROR) {
    877 	    if (CondParser_Token(par, doEval) != TOK_RPAREN) {
    878 		t = TOK_ERROR;
    879 	    }
    880 	}
    881     } else if (t == TOK_NOT) {
    882 	t = CondParser_Term(par, doEval);
    883 	if (t == TOK_TRUE) {
    884 	    t = TOK_FALSE;
    885 	} else if (t == TOK_FALSE) {
    886 	    t = TOK_TRUE;
    887 	}
    888     }
    889     return t;
    890 }
    891 
    892 /* Parse a conjunctive factor (nice name, wot?)
    893  *
    894  *	F -> T && F | T
    895  *
    896  * Results:
    897  *	TOK_TRUE, TOK_FALSE or TOK_ERROR
    898  */
    899 static Token
    900 CondParser_Factor(CondParser *par, Boolean doEval)
    901 {
    902     Token l, o;
    903 
    904     l = CondParser_Term(par, doEval);
    905     if (l != TOK_ERROR) {
    906 	o = CondParser_Token(par, doEval);
    907 
    908 	if (o == TOK_AND) {
    909 	    /*
    910 	     * F -> T && F
    911 	     *
    912 	     * If T is TOK_FALSE, the whole thing will be TOK_FALSE, but we
    913 	     * have to parse the r.h.s. anyway (to throw it away).
    914 	     * If T is TOK_TRUE, the result is the r.h.s., be it a TOK_ERROR
    915 	     * or not.
    916 	     */
    917 	    if (l == TOK_TRUE) {
    918 		l = CondParser_Factor(par, doEval);
    919 	    } else {
    920 		(void)CondParser_Factor(par, FALSE);
    921 	    }
    922 	} else {
    923 	    /*
    924 	     * F -> T
    925 	     */
    926 	    CondParser_PushBack(par, o);
    927 	}
    928     }
    929     return l;
    930 }
    931 
    932 /* Main expression production.
    933  *
    934  *	E -> F || E | F
    935  *
    936  * Results:
    937  *	TOK_TRUE, TOK_FALSE or TOK_ERROR.
    938  */
    939 static Token
    940 CondParser_Expr(CondParser *par, Boolean doEval)
    941 {
    942     Token l, o;
    943 
    944     l = CondParser_Factor(par, doEval);
    945     if (l != TOK_ERROR) {
    946 	o = CondParser_Token(par, doEval);
    947 
    948 	if (o == TOK_OR) {
    949 	    /*
    950 	     * E -> F || E
    951 	     *
    952 	     * A similar thing occurs for ||, except that here we make sure
    953 	     * the l.h.s. is TOK_FALSE before we bother to evaluate the r.h.s.
    954 	     * Once again, if l is TOK_FALSE, the result is the r.h.s. and once
    955 	     * again if l is TOK_TRUE, we parse the r.h.s. to throw it away.
    956 	     */
    957 	    if (l == TOK_FALSE) {
    958 		l = CondParser_Expr(par, doEval);
    959 	    } else {
    960 		(void)CondParser_Expr(par, FALSE);
    961 	    }
    962 	} else {
    963 	    /*
    964 	     * E -> F
    965 	     */
    966 	    CondParser_PushBack(par, o);
    967 	}
    968     }
    969     return l;
    970 }
    971 
    972 static CondEvalResult
    973 CondParser_Eval(CondParser *par, Boolean *value)
    974 {
    975     Token res;
    976 
    977     DEBUG1(COND, "CondParser_Eval: %s\n", par->p);
    978 
    979     res = CondParser_Expr(par, TRUE);
    980     if (res != TOK_FALSE && res != TOK_TRUE)
    981 	return COND_INVALID;
    982 
    983     if (CondParser_Token(par, TRUE /* XXX: Why TRUE? */) != TOK_EOF)
    984 	return COND_INVALID;
    985 
    986     *value = res == TOK_TRUE;
    987     return COND_PARSE;
    988 }
    989 
    990 /* Evaluate the condition, including any side effects from the variable
    991  * expressions in the condition. The condition consists of &&, ||, !,
    992  * function(arg), comparisons and parenthetical groupings thereof.
    993  *
    994  * Results:
    995  *	COND_PARSE	if the condition was valid grammatically
    996  *	COND_INVALID	if not a valid conditional.
    997  *
    998  *	(*value) is set to the boolean value of the condition
    999  */
   1000 static CondEvalResult
   1001 CondEvalExpression(const struct If *info, const char *cond, Boolean *value,
   1002 		    Boolean eprint, Boolean strictLHS)
   1003 {
   1004     static const struct If *dflt_info;
   1005     CondParser par;
   1006     int rval;
   1007 
   1008     lhsStrict = strictLHS;
   1009 
   1010     while (*cond == ' ' || *cond == '\t')
   1011 	cond++;
   1012 
   1013     if (info == NULL && (info = dflt_info) == NULL) {
   1014 	/* Scan for the entry for .if - it can't be first */
   1015 	for (info = ifs;; info++)
   1016 	    if (info->form[0] == 0)
   1017 		break;
   1018 	dflt_info = info;
   1019     }
   1020     assert(info != NULL);
   1021 
   1022     par.if_info = info;
   1023     par.p = cond;
   1024     par.curr = TOK_NONE;
   1025     par.printedError = FALSE;
   1026 
   1027     rval = CondParser_Eval(&par, value);
   1028 
   1029     if (rval == COND_INVALID && eprint && !par.printedError)
   1030 	Parse_Error(PARSE_FATAL, "Malformed conditional (%s)", cond);
   1031 
   1032     return rval;
   1033 }
   1034 
   1035 CondEvalResult
   1036 Cond_EvalCondition(const char *cond, Boolean *out_value)
   1037 {
   1038 	return CondEvalExpression(NULL, cond, out_value, FALSE, FALSE);
   1039 }
   1040 
   1041 /* Evaluate the conditional in the passed line. The line looks like this:
   1042  *	.<cond-type> <expr>
   1043  * In this line, <cond-type> is any of if, ifmake, ifnmake, ifdef, ifndef,
   1044  * elif, elifmake, elifnmake, elifdef, elifndef.
   1045  * In this line, <expr> consists of &&, ||, !, function(arg), comparisons
   1046  * and parenthetical groupings thereof.
   1047  *
   1048  * Note that the states IF_ACTIVE and ELSE_ACTIVE are only different in order
   1049  * to detect spurious .else lines (as are SKIP_TO_ELSE and SKIP_TO_ENDIF),
   1050  * otherwise .else could be treated as '.elif 1'.
   1051  *
   1052  * Results:
   1053  *	COND_PARSE	to continue parsing the lines after the conditional
   1054  *			(when .if or .else returns TRUE)
   1055  *	COND_SKIP	to skip the lines after the conditional
   1056  *			(when .if or .elif returns FALSE, or when a previous
   1057  *			branch has already been taken)
   1058  *	COND_INVALID	if the conditional was not valid, either because of
   1059  *			a syntax error or because some variable was undefined
   1060  *			or because the condition could not be evaluated
   1061  */
   1062 CondEvalResult
   1063 Cond_EvalLine(const char *line)
   1064 {
   1065     enum { MAXIF = 128 };	/* maximum depth of .if'ing */
   1066     enum { MAXIF_BUMP = 32 };	/* how much to grow by */
   1067     enum if_states {
   1068 	IF_ACTIVE,		/* .if or .elif part active */
   1069 	ELSE_ACTIVE,		/* .else part active */
   1070 	SEARCH_FOR_ELIF,	/* searching for .elif/else to execute */
   1071 	SKIP_TO_ELSE,		/* has been true, but not seen '.else' */
   1072 	SKIP_TO_ENDIF		/* nothing else to execute */
   1073     };
   1074     static enum if_states *cond_state = NULL;
   1075     static unsigned int max_if_depth = MAXIF;
   1076 
   1077     const struct If *ifp;
   1078     Boolean isElif;
   1079     Boolean value;
   1080     enum if_states state;
   1081 
   1082     if (!cond_state) {
   1083 	cond_state = bmake_malloc(max_if_depth * sizeof(*cond_state));
   1084 	cond_state[0] = IF_ACTIVE;
   1085     }
   1086     /* skip leading character (the '.') and any whitespace */
   1087     for (line++; *line == ' ' || *line == '\t'; line++)
   1088 	continue;
   1089 
   1090     /* Find what type of if we're dealing with.  */
   1091     if (line[0] == 'e') {
   1092 	if (line[1] != 'l') {
   1093 	    if (!is_token(line + 1, "ndif", 4))
   1094 		return COND_INVALID;
   1095 	    /* End of conditional section */
   1096 	    if (cond_depth == cond_min_depth) {
   1097 		Parse_Error(PARSE_FATAL, "if-less endif");
   1098 		return COND_PARSE;
   1099 	    }
   1100 	    /* Return state for previous conditional */
   1101 	    cond_depth--;
   1102 	    return cond_state[cond_depth] <= ELSE_ACTIVE
   1103 		   ? COND_PARSE : COND_SKIP;
   1104 	}
   1105 
   1106 	/* Quite likely this is 'else' or 'elif' */
   1107 	line += 2;
   1108 	if (is_token(line, "se", 2)) {
   1109 	    /* It is else... */
   1110 	    if (cond_depth == cond_min_depth) {
   1111 		Parse_Error(PARSE_FATAL, "if-less else");
   1112 		return COND_PARSE;
   1113 	    }
   1114 
   1115 	    state = cond_state[cond_depth];
   1116 	    switch (state) {
   1117 	    case SEARCH_FOR_ELIF:
   1118 		state = ELSE_ACTIVE;
   1119 		break;
   1120 	    case ELSE_ACTIVE:
   1121 	    case SKIP_TO_ENDIF:
   1122 		Parse_Error(PARSE_WARNING, "extra else");
   1123 		/* FALLTHROUGH */
   1124 	    default:
   1125 	    case IF_ACTIVE:
   1126 	    case SKIP_TO_ELSE:
   1127 		state = SKIP_TO_ENDIF;
   1128 		break;
   1129 	    }
   1130 	    cond_state[cond_depth] = state;
   1131 	    return state <= ELSE_ACTIVE ? COND_PARSE : COND_SKIP;
   1132 	}
   1133 	/* Assume for now it is an elif */
   1134 	isElif = TRUE;
   1135     } else
   1136 	isElif = FALSE;
   1137 
   1138     if (line[0] != 'i' || line[1] != 'f')
   1139 	/* Not an ifxxx or elifxxx line */
   1140 	return COND_INVALID;
   1141 
   1142     /*
   1143      * Figure out what sort of conditional it is -- what its default
   1144      * function is, etc. -- by looking in the table of valid "ifs"
   1145      */
   1146     line += 2;
   1147     for (ifp = ifs;; ifp++) {
   1148 	if (ifp->form == NULL)
   1149 	    return COND_INVALID;
   1150 	if (is_token(ifp->form, line, ifp->formlen)) {
   1151 	    line += ifp->formlen;
   1152 	    break;
   1153 	}
   1154     }
   1155 
   1156     /* Now we know what sort of 'if' it is... */
   1157 
   1158     if (isElif) {
   1159 	if (cond_depth == cond_min_depth) {
   1160 	    Parse_Error(PARSE_FATAL, "if-less elif");
   1161 	    return COND_PARSE;
   1162 	}
   1163 	state = cond_state[cond_depth];
   1164 	if (state == SKIP_TO_ENDIF || state == ELSE_ACTIVE) {
   1165 	    Parse_Error(PARSE_WARNING, "extra elif");
   1166 	    cond_state[cond_depth] = SKIP_TO_ENDIF;
   1167 	    return COND_SKIP;
   1168 	}
   1169 	if (state != SEARCH_FOR_ELIF) {
   1170 	    /* Either just finished the 'true' block, or already SKIP_TO_ELSE */
   1171 	    cond_state[cond_depth] = SKIP_TO_ELSE;
   1172 	    return COND_SKIP;
   1173 	}
   1174     } else {
   1175 	/* Normal .if */
   1176 	if (cond_depth + 1 >= max_if_depth) {
   1177 	    /*
   1178 	     * This is rare, but not impossible.
   1179 	     * In meta mode, dirdeps.mk (only runs at level 0)
   1180 	     * can need more than the default.
   1181 	     */
   1182 	    max_if_depth += MAXIF_BUMP;
   1183 	    cond_state = bmake_realloc(cond_state,
   1184 				       max_if_depth * sizeof(*cond_state));
   1185 	}
   1186 	state = cond_state[cond_depth];
   1187 	cond_depth++;
   1188 	if (state > ELSE_ACTIVE) {
   1189 	    /* If we aren't parsing the data, treat as always false */
   1190 	    cond_state[cond_depth] = SKIP_TO_ELSE;
   1191 	    return COND_SKIP;
   1192 	}
   1193     }
   1194 
   1195     /* And evaluate the conditional expression */
   1196     if (CondEvalExpression(ifp, line, &value, TRUE, TRUE) == COND_INVALID) {
   1197 	/* Syntax error in conditional, error message already output. */
   1198 	/* Skip everything to matching .endif */
   1199 	cond_state[cond_depth] = SKIP_TO_ELSE;
   1200 	return COND_SKIP;
   1201     }
   1202 
   1203     if (!value) {
   1204 	cond_state[cond_depth] = SEARCH_FOR_ELIF;
   1205 	return COND_SKIP;
   1206     }
   1207     cond_state[cond_depth] = IF_ACTIVE;
   1208     return COND_PARSE;
   1209 }
   1210 
   1211 void
   1212 Cond_restore_depth(unsigned int saved_depth)
   1213 {
   1214     unsigned int open_conds = cond_depth - cond_min_depth;
   1215 
   1216     if (open_conds != 0 || saved_depth > cond_depth) {
   1217 	Parse_Error(PARSE_FATAL, "%u open conditional%s", open_conds,
   1218 		    open_conds == 1 ? "" : "s");
   1219 	cond_depth = cond_min_depth;
   1220     }
   1221 
   1222     cond_min_depth = saved_depth;
   1223 }
   1224 
   1225 unsigned int
   1226 Cond_save_depth(void)
   1227 {
   1228     unsigned int depth = cond_min_depth;
   1229 
   1230     cond_min_depth = cond_depth;
   1231     return depth;
   1232 }
   1233