Home | History | Annotate | Line # | Download | only in make
cond.c revision 1.144
      1 /*	$NetBSD: cond.c,v 1.144 2020/09/13 15:15:51 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 in the passed line.
     76  *
     77  *	Cond_EvalCondition
     78  *			Evaluate the conditional in the passed line, which
     79  *			is either the argument of one of the .if directives
     80  *			or the condition in a :?true:false 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 1.144 2020/09/13 15:15:51 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 {
    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 {
    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 } CondParser;
    146 
    147 static Token CondParser_Expr(CondParser *par, Boolean);
    148 static CondEvalResult CondParser_Eval(CondParser *par, Boolean *value);
    149 
    150 static unsigned int cond_depth = 0;	/* current .if nesting level */
    151 static unsigned int cond_min_depth = 0;	/* depth at makefile open */
    152 
    153 /*
    154  * Indicate when we should be strict about lhs of comparisons.
    155  * In strict mode, the lhs must be a variable expression or a string literal
    156  * in quotes. In non-strict mode it may also be an unquoted string literal.
    157  *
    158  * TRUE when CondEvalExpression is called from Cond_EvalLine (.if etc)
    159  * FALSE when CondEvalExpression is called from ApplyModifier_IfElse
    160  * since lhs is already expanded and we cannot tell if
    161  * it was a variable reference or not.
    162  */
    163 static Boolean lhsStrict;
    164 
    165 static int
    166 is_token(const char *str, const char *tok, size_t len)
    167 {
    168     return strncmp(str, tok, len) == 0 && !ch_isalpha(str[len]);
    169 }
    170 
    171 /* Push back the most recent token read. We only need one level of this. */
    172 static void
    173 CondParser_PushBack(CondParser *par, Token t)
    174 {
    175     assert(par->curr == TOK_NONE);
    176     assert(t != TOK_NONE);
    177 
    178     par->curr = t;
    179 }
    180 
    181 static void
    182 CondParser_SkipWhitespace(CondParser *par)
    183 {
    184     while (ch_isspace(par->p[0]))
    185 	par->p++;
    186 }
    187 
    188 /* Parse the argument of a built-in function.
    189  *
    190  * Arguments:
    191  *	*linePtr initially points at the '(', upon successful return points
    192  *	right after the ')'.
    193  *
    194  *	*out_arg receives the argument as string.
    195  *
    196  *	func says whether the argument belongs to an actual function, or
    197  *	whether the parsed argument is passed to the default function.
    198  *
    199  * Return the length of the argument. */
    200 static int
    201 ParseFuncArg(const char **linePtr, Boolean doEval, const char *func,
    202 	     char **out_arg) {
    203     const char *cp;
    204     Buffer buf;
    205     int paren_depth;
    206     char ch;
    207     size_t argLen;
    208 
    209     cp = *linePtr;
    210     if (func != NULL)
    211 	/* Skip opening '(' - verified by caller */
    212 	cp++;
    213 
    214     if (*cp == '\0') {
    215 	/*
    216 	 * No arguments whatsoever. Because 'make' and 'defined' aren't really
    217 	 * "reserved words", we don't print a message. I think this is better
    218 	 * than hitting the user with a warning message every time s/he uses
    219 	 * the word 'make' or 'defined' at the beginning of a symbol...
    220 	 */
    221 	*out_arg = NULL;
    222 	return 0;
    223     }
    224 
    225     while (*cp == ' ' || *cp == '\t') {
    226 	cp++;
    227     }
    228 
    229     /*
    230      * Create a buffer for the argument and start it out at 16 characters
    231      * long. Why 16? Why not?
    232      */
    233     Buf_Init(&buf, 16);
    234 
    235     paren_depth = 0;
    236     for (;;) {
    237 	ch = *cp;
    238 	if (ch == 0 || ch == ' ' || ch == '\t')
    239 	    break;
    240 	if ((ch == '&' || ch == '|') && paren_depth == 0)
    241 	    break;
    242 	if (*cp == '$') {
    243 	    /*
    244 	     * Parse the variable spec and install it as part of the argument
    245 	     * if it's valid. We tell Var_Parse to complain on an undefined
    246 	     * variable, so we don't need to do it. Nor do we return an error,
    247 	     * though perhaps we should...
    248 	     */
    249 	    void *freeIt;
    250 	    VarEvalFlags eflags = VARE_UNDEFERR | (doEval ? VARE_WANTRES : 0);
    251 	    const char *cp2 = Var_Parse(&cp, VAR_CMD, eflags, &freeIt);
    252 	    Buf_AddStr(&buf, cp2);
    253 	    free(freeIt);
    254 	    continue;
    255 	}
    256 	if (ch == '(')
    257 	    paren_depth++;
    258 	else if (ch == ')' && --paren_depth < 0)
    259 	    break;
    260 	Buf_AddByte(&buf, *cp);
    261 	cp++;
    262     }
    263 
    264     *out_arg = Buf_GetAll(&buf, &argLen);
    265     Buf_Destroy(&buf, FALSE);
    266 
    267     while (*cp == ' ' || *cp == '\t') {
    268 	cp++;
    269     }
    270 
    271     if (func != NULL && *cp++ != ')') {
    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     *linePtr = cp;
    279     return argLen;
    280 }
    281 
    282 /* Test whether the given variable is defined. */
    283 static Boolean
    284 FuncDefined(int argLen MAKE_ATTR_UNUSED, const char *arg)
    285 {
    286     char *freeIt;
    287     Boolean result = Var_Value(arg, VAR_CMD, &freeIt) != NULL;
    288     bmake_free(freeIt);
    289     return result;
    290 }
    291 
    292 /* Wrapper around Str_Match, to be used by Lst_Find. */
    293 static Boolean
    294 CondFindStrMatch(const void *string, const void *pattern)
    295 {
    296     return Str_Match(string, pattern);
    297 }
    298 
    299 /* See if the given target is being made. */
    300 static Boolean
    301 FuncMake(int argLen MAKE_ATTR_UNUSED, const char *arg)
    302 {
    303     return Lst_Find(create, CondFindStrMatch, arg) != NULL;
    304 }
    305 
    306 /* See if the given file exists. */
    307 static Boolean
    308 FuncExists(int argLen MAKE_ATTR_UNUSED, const char *arg)
    309 {
    310     Boolean result;
    311     char *path;
    312 
    313     path = Dir_FindFile(arg, dirSearchPath);
    314     if (DEBUG(COND)) {
    315 	fprintf(debug_file, "exists(%s) result is \"%s\"\n",
    316 		arg, path ? path : "");
    317     }
    318     if (path != NULL) {
    319 	result = TRUE;
    320 	free(path);
    321     } else {
    322 	result = FALSE;
    323     }
    324     return result;
    325 }
    326 
    327 /* See if the given node exists and is an actual target. */
    328 static Boolean
    329 FuncTarget(int argLen MAKE_ATTR_UNUSED, const char *arg)
    330 {
    331     GNode *gn;
    332 
    333     gn = Targ_FindNode(arg, TARG_NOCREATE);
    334     return gn != NULL && !OP_NOP(gn->type);
    335 }
    336 
    337 /* See if the given node exists and is an actual target with commands
    338  * associated with it. */
    339 static Boolean
    340 FuncCommands(int argLen MAKE_ATTR_UNUSED, const char *arg)
    341 {
    342     GNode *gn;
    343 
    344     gn = Targ_FindNode(arg, TARG_NOCREATE);
    345     return gn != NULL && !OP_NOP(gn->type) && !Lst_IsEmpty(gn->commands);
    346 }
    347 
    348 /*-
    349  * Convert the given number into a double.
    350  * We try a base 10 or 16 integer conversion first, if that fails
    351  * then we try a floating point conversion instead.
    352  *
    353  * Results:
    354  *	Sets 'value' to double value of string.
    355  *	Returns TRUE if the conversion succeeded.
    356  */
    357 static Boolean
    358 TryParseNumber(const char *str, double *value)
    359 {
    360     char *eptr, ech;
    361     unsigned long l_val;
    362     double d_val;
    363 
    364     errno = 0;
    365     if (!*str) {
    366 	*value = (double)0;
    367 	return TRUE;
    368     }
    369     l_val = strtoul(str, &eptr, str[1] == 'x' ? 16 : 10);
    370     ech = *eptr;
    371     if (ech == '\0' && errno != ERANGE) {
    372 	d_val = str[0] == '-' ? -(double)-l_val : (double)l_val;
    373     } else {
    374 	if (ech != '\0' && ech != '.' && ech != 'e' && ech != 'E')
    375 	    return FALSE;
    376 	d_val = strtod(str, &eptr);
    377 	if (*eptr)
    378 	    return FALSE;
    379     }
    380 
    381     *value = d_val;
    382     return TRUE;
    383 }
    384 
    385 static Boolean
    386 is_separator(char ch)
    387 {
    388     return ch == '\0' || ch_isspace(ch) || strchr("!=><)", ch);
    389 }
    390 
    391 /*-
    392  * Parse a string from a variable reference or an optionally quoted
    393  * string.  This is called for the lhs and rhs of string comparisons.
    394  *
    395  * Results:
    396  *	Returns the string, absent any quotes, or NULL on error.
    397  *	Sets quoted if the string was quoted.
    398  *	Sets freeIt if needed.
    399  */
    400 /* coverity:[+alloc : arg-*3] */
    401 static const char *
    402 CondParser_String(CondParser *par, Boolean doEval, Boolean strictLHS,
    403 		  Boolean *quoted, void **freeIt)
    404 {
    405     Buffer buf;
    406     const char *str;
    407     Boolean atStart;
    408     const char *nested_p;
    409     Boolean qt;
    410     const char *start;
    411     VarEvalFlags eflags;
    412 
    413     Buf_Init(&buf, 0);
    414     str = NULL;
    415     *freeIt = NULL;
    416     *quoted = qt = par->p[0] == '"' ? 1 : 0;
    417     if (qt)
    418 	par->p++;
    419     start = par->p;
    420     while (par->p[0] && str == NULL) {
    421 	switch (par->p[0]) {
    422 	case '\\':
    423 	    par->p++;
    424 	    if (par->p[0] != '\0') {
    425 		Buf_AddByte(&buf, par->p[0]);
    426 		par->p++;
    427 	    }
    428 	    continue;
    429 	case '"':
    430 	    if (qt) {
    431 		par->p++;	/* we don't want the quotes */
    432 		goto got_str;
    433 	    }
    434 	    Buf_AddByte(&buf, par->p[0]); /* likely? */
    435 	    par->p++;
    436 	    continue;
    437 	case ')':
    438 	case '!':
    439 	case '=':
    440 	case '>':
    441 	case '<':
    442 	case ' ':
    443 	case '\t':
    444 	    if (!qt)
    445 		goto got_str;
    446 	    Buf_AddByte(&buf, par->p[0]);
    447 	    par->p++;
    448 	    continue;
    449 	case '$':
    450 	    /* if we are in quotes, an undefined variable is ok */
    451 	    eflags = ((!qt && doEval) ? VARE_UNDEFERR : 0) |
    452 		     (doEval ? VARE_WANTRES : 0);
    453 	    nested_p = par->p;
    454 	    atStart = nested_p == start;
    455 	    str = Var_Parse(&nested_p, VAR_CMD, eflags, freeIt);
    456 	    if (str == var_Error) {
    457 		if (*freeIt) {
    458 		    free(*freeIt);
    459 		    *freeIt = NULL;
    460 		}
    461 		/*
    462 		 * Even if !doEval, we still report syntax errors, which
    463 		 * is what getting var_Error back with !doEval means.
    464 		 */
    465 		str = NULL;
    466 		goto cleanup;
    467 	    }
    468 	    par->p = nested_p;
    469 
    470 	    /*
    471 	     * If the '$' started the string literal (which means no quotes),
    472 	     * and the variable expression is followed by a space, looks like
    473 	     * a comparison operator or is the end of the expression, we are
    474 	     * done.
    475 	     */
    476 	    if (atStart && is_separator(par->p[0]))
    477 		goto cleanup;
    478 
    479 	    Buf_AddStr(&buf, str);
    480 	    if (*freeIt) {
    481 		free(*freeIt);
    482 		*freeIt = NULL;
    483 	    }
    484 	    str = NULL;		/* not finished yet */
    485 	    continue;
    486 	default:
    487 	    if (strictLHS && !qt && *start != '$' && !ch_isdigit(*start)) {
    488 		/* lhs must be quoted, a variable reference or number */
    489 		if (*freeIt) {
    490 		    free(*freeIt);
    491 		    *freeIt = NULL;
    492 		}
    493 		str = NULL;
    494 		goto cleanup;
    495 	    }
    496 	    Buf_AddByte(&buf, par->p[0]);
    497 	    par->p++;
    498 	    continue;
    499 	}
    500     }
    501 got_str:
    502     *freeIt = Buf_GetAll(&buf, NULL);
    503     str = *freeIt;
    504 cleanup:
    505     Buf_Destroy(&buf, FALSE);
    506     return str;
    507 }
    508 
    509 /* The different forms of .if directives. */
    510 static const struct If {
    511     const char *form;		/* Form of if */
    512     size_t formlen;		/* Length of form */
    513     Boolean doNot;		/* TRUE if default function should be negated */
    514     Boolean (*defProc)(int, const char *); /* Default function to apply */
    515 } ifs[] = {
    516     { "def",   3, FALSE, FuncDefined },
    517     { "ndef",  4, TRUE,  FuncDefined },
    518     { "make",  4, FALSE, FuncMake },
    519     { "nmake", 5, TRUE,  FuncMake },
    520     { "",      0, FALSE, FuncDefined },
    521     { NULL,    0, FALSE, NULL }
    522 };
    523 
    524 /* Evaluate a "comparison without operator", such as in ".if ${VAR}" or
    525  * ".if 0". */
    526 static Token
    527 EvalNotEmpty(CondParser *par, const char *lhs, Boolean lhsQuoted)
    528 {
    529     double left;
    530 
    531     /* For .ifxxx "..." check for non-empty string. */
    532     if (lhsQuoted)
    533 	return lhs[0] != '\0';
    534 
    535     /* For .ifxxx <number> compare against zero */
    536     if (TryParseNumber(lhs, &left))
    537 	return left != 0.0;
    538 
    539     /* For .if ${...} check for non-empty string (defProc is ifdef). */
    540     if (par->if_info->form[0] == '\0')
    541 	return lhs[0] != 0;
    542 
    543     /* Otherwise action default test ... */
    544     return par->if_info->defProc(strlen(lhs), lhs) != par->if_info->doNot;
    545 }
    546 
    547 /* Evaluate a numerical comparison, such as in ".if ${VAR} >= 9". */
    548 static Token
    549 EvalCompareNum(double lhs, const char *op, double rhs)
    550 {
    551     if (DEBUG(COND))
    552 	fprintf(debug_file, "lhs = %f, rhs = %f, op = %.2s\n", lhs, rhs, op);
    553 
    554     switch (op[0]) {
    555     case '!':
    556 	if (op[1] != '=') {
    557 	    Parse_Error(PARSE_WARNING, "Unknown operator");
    558 	    /* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    559 	    return TOK_ERROR;
    560 	}
    561 	return lhs != rhs;
    562     case '=':
    563 	if (op[1] != '=') {
    564 	    Parse_Error(PARSE_WARNING, "Unknown operator");
    565 	    /* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    566 	    return TOK_ERROR;
    567 	}
    568 	return lhs == rhs;
    569     case '<':
    570 	return op[1] == '=' ? lhs <= rhs : lhs < rhs;
    571     case '>':
    572 	return op[1] == '=' ? lhs >= rhs : lhs > rhs;
    573     }
    574     return TOK_ERROR;
    575 }
    576 
    577 static Token
    578 EvalCompareStr(const char *lhs, const char *op, const char *rhs)
    579 {
    580     if (!((op[0] == '!' || op[0] == '=') && op[1] == '=')) {
    581 	Parse_Error(PARSE_WARNING,
    582 		    "String comparison operator must be either == or !=");
    583 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    584 	return TOK_ERROR;
    585     }
    586 
    587     if (DEBUG(COND)) {
    588 	fprintf(debug_file, "lhs = \"%s\", rhs = \"%s\", op = %.2s\n",
    589 		lhs, rhs, op);
    590     }
    591     return (*op == '=') == (strcmp(lhs, rhs) == 0);
    592 }
    593 
    594 /* Evaluate a comparison, such as "${VAR} == 12345". */
    595 static Token
    596 EvalCompare(const char *lhs, Boolean lhsQuoted, const char *op,
    597 	    const char *rhs, Boolean rhsQuoted)
    598 {
    599     double left, right;
    600 
    601     if (!rhsQuoted && !lhsQuoted)
    602 	if (TryParseNumber(lhs, &left) && TryParseNumber(rhs, &right))
    603 	    return EvalCompareNum(left, op, right);
    604 
    605     return EvalCompareStr(lhs, op, rhs);
    606 }
    607 
    608 /* Parse a comparison condition such as:
    609  *
    610  *	0
    611  *	${VAR:Mpattern}
    612  *	${VAR} == value
    613  *	${VAR:U0} < 12345
    614  */
    615 static Token
    616 CondParser_Comparison(CondParser *par, Boolean doEval)
    617 {
    618     Token t = TOK_ERROR;
    619     const char *lhs, *op, *rhs;
    620     void *lhsFree, *rhsFree;
    621     Boolean lhsQuoted, rhsQuoted;
    622 
    623     rhs = NULL;
    624     lhsFree = rhsFree = NULL;
    625     lhsQuoted = rhsQuoted = FALSE;
    626 
    627     /*
    628      * Parse the variable spec and skip over it, saving its
    629      * value in lhs.
    630      */
    631     lhs = CondParser_String(par, doEval, lhsStrict, &lhsQuoted, &lhsFree);
    632     if (!lhs)
    633 	goto done;
    634 
    635     CondParser_SkipWhitespace(par);
    636 
    637     /*
    638      * Make sure the operator is a valid one. If it isn't a
    639      * known relational operator, pretend we got a
    640      * != 0 comparison.
    641      */
    642     op = par->p;
    643     switch (par->p[0]) {
    644     case '!':
    645     case '=':
    646     case '<':
    647     case '>':
    648 	if (par->p[1] == '=') {
    649 	    par->p += 2;
    650 	} else {
    651 	    par->p += 1;
    652 	}
    653 	break;
    654     default:
    655 	t = doEval ? EvalNotEmpty(par, lhs, lhsQuoted) : TOK_FALSE;
    656 	goto done;
    657     }
    658 
    659     CondParser_SkipWhitespace(par);
    660 
    661     if (par->p[0] == '\0') {
    662 	Parse_Error(PARSE_WARNING, "Missing right-hand-side of operator");
    663 	/* The PARSE_FATAL is done as a follow-up by CondEvalExpression. */
    664 	goto done;
    665     }
    666 
    667     rhs = CondParser_String(par, doEval, FALSE, &rhsQuoted, &rhsFree);
    668     if (rhs == NULL)
    669 	goto done;
    670 
    671     if (!doEval) {
    672 	t = TOK_FALSE;
    673 	goto done;
    674     }
    675 
    676     t = EvalCompare(lhs, lhsQuoted, op, rhs, rhsQuoted);
    677 
    678 done:
    679     free(lhsFree);
    680     free(rhsFree);
    681     return t;
    682 }
    683 
    684 static int
    685 ParseEmptyArg(const char **linePtr, Boolean doEval,
    686 	      const char *func MAKE_ATTR_UNUSED, char **argPtr)
    687 {
    688     void *val_freeIt;
    689     const char *val;
    690     int magic_res;
    691 
    692     /* We do all the work here and return the result as the length */
    693     *argPtr = NULL;
    694 
    695     (*linePtr)--;		/* Make (*linePtr)[1] point to the '('. */
    696     val = Var_Parse(linePtr, VAR_CMD, doEval ? VARE_WANTRES : 0, &val_freeIt);
    697     /* If successful, *linePtr points beyond the closing ')' now. */
    698 
    699     if (val == var_Error) {
    700 	free(val_freeIt);
    701 	return -1;
    702     }
    703 
    704     /* A variable is empty when it just contains spaces... 4/15/92, christos */
    705     while (ch_isspace(val[0]))
    706 	val++;
    707 
    708     /*
    709      * For consistency with the other functions we can't generate the
    710      * true/false here.
    711      */
    712     magic_res = *val != '\0' ? 2 : 1;
    713     free(val_freeIt);
    714     return magic_res;
    715 }
    716 
    717 static Boolean
    718 FuncEmpty(int arglen, const char *arg MAKE_ATTR_UNUSED)
    719 {
    720     /* Magic values ahead, see ParseEmptyArg. */
    721     return arglen == 1;
    722 }
    723 
    724 static Token
    725 CondParser_Func(CondParser *par, Boolean doEval)
    726 {
    727     static const struct fn_def {
    728 	const char *fn_name;
    729 	size_t fn_name_len;
    730 	int (*fn_parse)(const char **, Boolean, const char *, char **);
    731 	Boolean (*fn_eval)(int, const char *);
    732     } fn_defs[] = {
    733 	{ "defined",  7, ParseFuncArg,  FuncDefined },
    734 	{ "make",     4, ParseFuncArg,  FuncMake },
    735 	{ "exists",   6, ParseFuncArg,  FuncExists },
    736 	{ "empty",    5, ParseEmptyArg, FuncEmpty },
    737 	{ "target",   6, ParseFuncArg,  FuncTarget },
    738 	{ "commands", 8, ParseFuncArg,  FuncCommands },
    739 	{ NULL,       0, NULL, NULL },
    740     };
    741     const struct fn_def *fn_def;
    742     Token t;
    743     char *arg = NULL;
    744     int arglen;
    745     const char *cp = par->p;
    746     const char *cp1;
    747 
    748     for (fn_def = fn_defs; fn_def->fn_name != NULL; fn_def++) {
    749 	if (!is_token(cp, fn_def->fn_name, fn_def->fn_name_len))
    750 	    continue;
    751 	cp += fn_def->fn_name_len;
    752 	/* There can only be whitespace before the '(' */
    753 	while (ch_isspace(*cp))
    754 	    cp++;
    755 	if (*cp != '(')
    756 	    break;
    757 
    758 	arglen = fn_def->fn_parse(&cp, doEval, fn_def->fn_name, &arg);
    759 	if (arglen <= 0) {
    760 	    par->p = cp;
    761 	    return arglen < 0 ? TOK_ERROR : TOK_FALSE;
    762 	}
    763 	/* Evaluate the argument using the required function. */
    764 	t = !doEval || fn_def->fn_eval(arglen, arg);
    765 	free(arg);
    766 	par->p = cp;
    767 	return t;
    768     }
    769 
    770     /* Push anything numeric through the compare expression */
    771     cp = par->p;
    772     if (ch_isdigit(cp[0]) || strchr("+-", cp[0]))
    773 	return CondParser_Comparison(par, doEval);
    774 
    775     /*
    776      * Most likely we have a naked token to apply the default function to.
    777      * However ".if a == b" gets here when the "a" is unquoted and doesn't
    778      * start with a '$'. This surprises people.
    779      * If what follows the function argument is a '=' or '!' then the syntax
    780      * would be invalid if we did "defined(a)" - so instead treat as an
    781      * expression.
    782      */
    783     arglen = ParseFuncArg(&cp, doEval, NULL, &arg);
    784     for (cp1 = cp; ch_isspace(*cp1); cp1++)
    785 	continue;
    786     if (*cp1 == '=' || *cp1 == '!')
    787 	return CondParser_Comparison(par, doEval);
    788     par->p = cp;
    789 
    790     /*
    791      * Evaluate the argument using the default function.
    792      * This path always treats .if as .ifdef. To get here, the character
    793      * after .if must have been taken literally, so the argument cannot
    794      * be empty - even if it contained a variable expansion.
    795      */
    796     t = !doEval || par->if_info->defProc(arglen, arg) != par->if_info->doNot;
    797     free(arg);
    798     return t;
    799 }
    800 
    801 /* Return the next token or comparison result from the parser. */
    802 static Token
    803 CondParser_Token(CondParser *par, Boolean doEval)
    804 {
    805     Token t;
    806 
    807     t = par->curr;
    808     if (t != TOK_NONE) {
    809 	par->curr = TOK_NONE;
    810 	return t;
    811     }
    812 
    813     while (par->p[0] == ' ' || par->p[0] == '\t') {
    814 	par->p++;
    815     }
    816 
    817     switch (par->p[0]) {
    818 
    819     case '(':
    820 	par->p++;
    821 	return TOK_LPAREN;
    822 
    823     case ')':
    824 	par->p++;
    825 	return TOK_RPAREN;
    826 
    827     case '|':
    828 	par->p++;
    829 	if (par->p[0] == '|') {
    830 	    par->p++;
    831 	}
    832 	return TOK_OR;
    833 
    834     case '&':
    835 	par->p++;
    836 	if (par->p[0] == '&') {
    837 	    par->p++;
    838 	}
    839 	return TOK_AND;
    840 
    841     case '!':
    842 	par->p++;
    843 	return TOK_NOT;
    844 
    845     case '#':
    846     case '\n':
    847     case '\0':
    848 	return TOK_EOF;
    849 
    850     case '"':
    851     case '$':
    852 	return CondParser_Comparison(par, doEval);
    853 
    854     default:
    855 	return CondParser_Func(par, doEval);
    856     }
    857 }
    858 
    859 /* Parse a single term in the expression. This consists of a terminal symbol
    860  * or TOK_NOT and a term (not including the binary operators):
    861  *
    862  *	T -> defined(variable) | make(target) | exists(file) | symbol
    863  *	T -> ! T | ( E )
    864  *
    865  * Results:
    866  *	TOK_TRUE, TOK_FALSE or TOK_ERROR.
    867  */
    868 static Token
    869 CondParser_Term(CondParser *par, Boolean doEval)
    870 {
    871     Token t;
    872 
    873     t = CondParser_Token(par, doEval);
    874 
    875     if (t == TOK_EOF) {
    876 	/*
    877 	 * If we reached the end of the expression, the expression
    878 	 * is malformed...
    879 	 */
    880 	t = TOK_ERROR;
    881     } else if (t == TOK_LPAREN) {
    882 	/*
    883 	 * T -> ( E )
    884 	 */
    885 	t = CondParser_Expr(par, doEval);
    886 	if (t != TOK_ERROR) {
    887 	    if (CondParser_Token(par, doEval) != TOK_RPAREN) {
    888 		t = TOK_ERROR;
    889 	    }
    890 	}
    891     } else if (t == TOK_NOT) {
    892 	t = CondParser_Term(par, doEval);
    893 	if (t == TOK_TRUE) {
    894 	    t = TOK_FALSE;
    895 	} else if (t == TOK_FALSE) {
    896 	    t = TOK_TRUE;
    897 	}
    898     }
    899     return t;
    900 }
    901 
    902 /* Parse a conjunctive factor (nice name, wot?)
    903  *
    904  *	F -> T && F | T
    905  *
    906  * Results:
    907  *	TOK_TRUE, TOK_FALSE or TOK_ERROR
    908  */
    909 static Token
    910 CondParser_Factor(CondParser *par, Boolean doEval)
    911 {
    912     Token l, o;
    913 
    914     l = CondParser_Term(par, doEval);
    915     if (l != TOK_ERROR) {
    916 	o = CondParser_Token(par, doEval);
    917 
    918 	if (o == TOK_AND) {
    919 	    /*
    920 	     * F -> T && F
    921 	     *
    922 	     * If T is TOK_FALSE, the whole thing will be TOK_FALSE, but we
    923 	     * have to parse the r.h.s. anyway (to throw it away).
    924 	     * If T is TOK_TRUE, the result is the r.h.s., be it a TOK_ERROR
    925 	     * or not.
    926 	     */
    927 	    if (l == TOK_TRUE) {
    928 		l = CondParser_Factor(par, doEval);
    929 	    } else {
    930 		(void)CondParser_Factor(par, FALSE);
    931 	    }
    932 	} else {
    933 	    /*
    934 	     * F -> T
    935 	     */
    936 	    CondParser_PushBack(par, o);
    937 	}
    938     }
    939     return l;
    940 }
    941 
    942 /* Main expression production.
    943  *
    944  *	E -> F || E | F
    945  *
    946  * Results:
    947  *	TOK_TRUE, TOK_FALSE or TOK_ERROR.
    948  */
    949 static Token
    950 CondParser_Expr(CondParser *par, Boolean doEval)
    951 {
    952     Token l, o;
    953 
    954     l = CondParser_Factor(par, doEval);
    955     if (l != TOK_ERROR) {
    956 	o = CondParser_Token(par, doEval);
    957 
    958 	if (o == TOK_OR) {
    959 	    /*
    960 	     * E -> F || E
    961 	     *
    962 	     * A similar thing occurs for ||, except that here we make sure
    963 	     * the l.h.s. is TOK_FALSE before we bother to evaluate the r.h.s.
    964 	     * Once again, if l is TOK_FALSE, the result is the r.h.s. and once
    965 	     * again if l is TOK_TRUE, we parse the r.h.s. to throw it away.
    966 	     */
    967 	    if (l == TOK_FALSE) {
    968 		l = CondParser_Expr(par, doEval);
    969 	    } else {
    970 		(void)CondParser_Expr(par, FALSE);
    971 	    }
    972 	} else {
    973 	    /*
    974 	     * E -> F
    975 	     */
    976 	    CondParser_PushBack(par, o);
    977 	}
    978     }
    979     return l;
    980 }
    981 
    982 static CondEvalResult
    983 CondParser_Eval(CondParser *par, Boolean *value)
    984 {
    985     Token res;
    986 
    987     if (DEBUG(COND))
    988 	fprintf(debug_file, "CondParser_Eval: %s\n", par->p);
    989 
    990     res = CondParser_Expr(par, TRUE);
    991     if (res != TOK_FALSE && res != TOK_TRUE)
    992 	return COND_INVALID;
    993 
    994     if (CondParser_Token(par, TRUE /* XXX: Why TRUE? */) != TOK_EOF)
    995 	return COND_INVALID;
    996 
    997     *value = res == TOK_TRUE;
    998     return COND_PARSE;
    999 }
   1000 
   1001 /* Evaluate the condition, including any side effects from the variable
   1002  * expressions in the condition. The condition consists of &&, ||, !,
   1003  * function(arg), comparisons and parenthetical groupings thereof.
   1004  *
   1005  * Results:
   1006  *	COND_PARSE	if the condition was valid grammatically
   1007  *	COND_INVALID  	if not a valid conditional.
   1008  *
   1009  *	(*value) is set to the boolean value of the condition
   1010  */
   1011 static CondEvalResult
   1012 CondEvalExpression(const struct If *info, const char *cond, Boolean *value,
   1013 		    Boolean eprint, Boolean strictLHS)
   1014 {
   1015     static const struct If *dflt_info;
   1016     CondParser par;
   1017     int rval;
   1018 
   1019     lhsStrict = strictLHS;
   1020 
   1021     while (*cond == ' ' || *cond == '\t')
   1022 	cond++;
   1023 
   1024     if (info == NULL && (info = dflt_info) == NULL) {
   1025 	/* Scan for the entry for .if - it can't be first */
   1026 	for (info = ifs;; info++)
   1027 	    if (info->form[0] == 0)
   1028 		break;
   1029 	dflt_info = info;
   1030     }
   1031     assert(info != NULL);
   1032 
   1033     par.if_info = info;
   1034     par.p = cond;
   1035     par.curr = TOK_NONE;
   1036 
   1037     rval = CondParser_Eval(&par, value);
   1038 
   1039     if (rval == COND_INVALID && eprint)
   1040 	Parse_Error(PARSE_FATAL, "Malformed conditional (%s)", cond);
   1041 
   1042     return rval;
   1043 }
   1044 
   1045 CondEvalResult
   1046 Cond_EvalCondition(const char *cond, Boolean *out_value)
   1047 {
   1048 	return CondEvalExpression(NULL, cond, out_value, FALSE, FALSE);
   1049 }
   1050 
   1051 /* Evaluate the conditional in the passed line. The line looks like this:
   1052  *	.<cond-type> <expr>
   1053  * In this line, <cond-type> is any of if, ifmake, ifnmake, ifdef, ifndef,
   1054  * elif, elifmake, elifnmake, elifdef, elifndef.
   1055  * In this line, <expr> consists of &&, ||, !, function(arg), comparisons
   1056  * and parenthetical groupings thereof.
   1057  *
   1058  * Note that the states IF_ACTIVE and ELSE_ACTIVE are only different in order
   1059  * to detect spurious .else lines (as are SKIP_TO_ELSE and SKIP_TO_ENDIF),
   1060  * otherwise .else could be treated as '.elif 1'.
   1061  *
   1062  * Results:
   1063  *	COND_PARSE	to continue parsing the lines after the conditional
   1064  *			(when .if or .else returns TRUE)
   1065  *	COND_SKIP	to skip the lines after the conditional
   1066  *			(when .if or .elif returns FALSE, or when a previous
   1067  *			branch has already been taken)
   1068  *	COND_INVALID  	if the conditional was not valid, either because of
   1069  *			a syntax error or because some variable was undefined
   1070  *			or because the condition could not be evaluated
   1071  */
   1072 CondEvalResult
   1073 Cond_EvalLine(const char *line)
   1074 {
   1075     enum { MAXIF = 128 };	/* maximum depth of .if'ing */
   1076     enum { MAXIF_BUMP = 32 };	/* how much to grow by */
   1077     enum if_states {
   1078 	IF_ACTIVE,		/* .if or .elif part active */
   1079 	ELSE_ACTIVE,		/* .else part active */
   1080 	SEARCH_FOR_ELIF,	/* searching for .elif/else to execute */
   1081 	SKIP_TO_ELSE,		/* has been true, but not seen '.else' */
   1082 	SKIP_TO_ENDIF		/* nothing else to execute */
   1083     };
   1084     static enum if_states *cond_state = NULL;
   1085     static unsigned int max_if_depth = MAXIF;
   1086 
   1087     const struct If *ifp;
   1088     Boolean isElif;
   1089     Boolean value;
   1090     enum if_states state;
   1091 
   1092     if (!cond_state) {
   1093 	cond_state = bmake_malloc(max_if_depth * sizeof(*cond_state));
   1094 	cond_state[0] = IF_ACTIVE;
   1095     }
   1096     /* skip leading character (the '.') and any whitespace */
   1097     for (line++; *line == ' ' || *line == '\t'; line++)
   1098 	continue;
   1099 
   1100     /* Find what type of if we're dealing with.  */
   1101     if (line[0] == 'e') {
   1102 	if (line[1] != 'l') {
   1103 	    if (!is_token(line + 1, "ndif", 4))
   1104 		return COND_INVALID;
   1105 	    /* End of conditional section */
   1106 	    if (cond_depth == cond_min_depth) {
   1107 		Parse_Error(PARSE_FATAL, "if-less endif");
   1108 		return COND_PARSE;
   1109 	    }
   1110 	    /* Return state for previous conditional */
   1111 	    cond_depth--;
   1112 	    return cond_state[cond_depth] <= ELSE_ACTIVE
   1113 		   ? COND_PARSE : COND_SKIP;
   1114 	}
   1115 
   1116 	/* Quite likely this is 'else' or 'elif' */
   1117 	line += 2;
   1118 	if (is_token(line, "se", 2)) {
   1119 	    /* It is else... */
   1120 	    if (cond_depth == cond_min_depth) {
   1121 		Parse_Error(PARSE_FATAL, "if-less else");
   1122 		return COND_PARSE;
   1123 	    }
   1124 
   1125 	    state = cond_state[cond_depth];
   1126 	    switch (state) {
   1127 	    case SEARCH_FOR_ELIF:
   1128 		state = ELSE_ACTIVE;
   1129 		break;
   1130 	    case ELSE_ACTIVE:
   1131 	    case SKIP_TO_ENDIF:
   1132 		Parse_Error(PARSE_WARNING, "extra else");
   1133 		/* FALLTHROUGH */
   1134 	    default:
   1135 	    case IF_ACTIVE:
   1136 	    case SKIP_TO_ELSE:
   1137 		state = SKIP_TO_ENDIF;
   1138 		break;
   1139 	    }
   1140 	    cond_state[cond_depth] = state;
   1141 	    return state <= ELSE_ACTIVE ? COND_PARSE : COND_SKIP;
   1142 	}
   1143 	/* Assume for now it is an elif */
   1144 	isElif = TRUE;
   1145     } else
   1146 	isElif = FALSE;
   1147 
   1148     if (line[0] != 'i' || line[1] != 'f')
   1149 	/* Not an ifxxx or elifxxx line */
   1150 	return COND_INVALID;
   1151 
   1152     /*
   1153      * Figure out what sort of conditional it is -- what its default
   1154      * function is, etc. -- by looking in the table of valid "ifs"
   1155      */
   1156     line += 2;
   1157     for (ifp = ifs;; ifp++) {
   1158 	if (ifp->form == NULL)
   1159 	    return COND_INVALID;
   1160 	if (is_token(ifp->form, line, ifp->formlen)) {
   1161 	    line += ifp->formlen;
   1162 	    break;
   1163 	}
   1164     }
   1165 
   1166     /* Now we know what sort of 'if' it is... */
   1167 
   1168     if (isElif) {
   1169 	if (cond_depth == cond_min_depth) {
   1170 	    Parse_Error(PARSE_FATAL, "if-less elif");
   1171 	    return COND_PARSE;
   1172 	}
   1173 	state = cond_state[cond_depth];
   1174 	if (state == SKIP_TO_ENDIF || state == ELSE_ACTIVE) {
   1175 	    Parse_Error(PARSE_WARNING, "extra elif");
   1176 	    cond_state[cond_depth] = SKIP_TO_ENDIF;
   1177 	    return COND_SKIP;
   1178 	}
   1179 	if (state != SEARCH_FOR_ELIF) {
   1180 	    /* Either just finished the 'true' block, or already SKIP_TO_ELSE */
   1181 	    cond_state[cond_depth] = SKIP_TO_ELSE;
   1182 	    return COND_SKIP;
   1183 	}
   1184     } else {
   1185 	/* Normal .if */
   1186 	if (cond_depth + 1 >= max_if_depth) {
   1187 	    /*
   1188 	     * This is rare, but not impossible.
   1189 	     * In meta mode, dirdeps.mk (only runs at level 0)
   1190 	     * can need more than the default.
   1191 	     */
   1192 	    max_if_depth += MAXIF_BUMP;
   1193 	    cond_state = bmake_realloc(cond_state,
   1194 				       max_if_depth * sizeof(*cond_state));
   1195 	}
   1196 	state = cond_state[cond_depth];
   1197 	cond_depth++;
   1198 	if (state > ELSE_ACTIVE) {
   1199 	    /* If we aren't parsing the data, treat as always false */
   1200 	    cond_state[cond_depth] = SKIP_TO_ELSE;
   1201 	    return COND_SKIP;
   1202 	}
   1203     }
   1204 
   1205     /* And evaluate the conditional expression */
   1206     if (CondEvalExpression(ifp, line, &value, TRUE, TRUE) == COND_INVALID) {
   1207 	/* Syntax error in conditional, error message already output. */
   1208 	/* Skip everything to matching .endif */
   1209 	cond_state[cond_depth] = SKIP_TO_ELSE;
   1210 	return COND_SKIP;
   1211     }
   1212 
   1213     if (!value) {
   1214 	cond_state[cond_depth] = SEARCH_FOR_ELIF;
   1215 	return COND_SKIP;
   1216     }
   1217     cond_state[cond_depth] = IF_ACTIVE;
   1218     return COND_PARSE;
   1219 }
   1220 
   1221 void
   1222 Cond_restore_depth(unsigned int saved_depth)
   1223 {
   1224     int open_conds = cond_depth - cond_min_depth;
   1225 
   1226     if (open_conds != 0 || saved_depth > cond_depth) {
   1227 	Parse_Error(PARSE_FATAL, "%d open conditional%s", open_conds,
   1228 		    open_conds == 1 ? "" : "s");
   1229 	cond_depth = cond_min_depth;
   1230     }
   1231 
   1232     cond_min_depth = saved_depth;
   1233 }
   1234 
   1235 unsigned int
   1236 Cond_save_depth(void)
   1237 {
   1238     int depth = cond_min_depth;
   1239 
   1240     cond_min_depth = cond_depth;
   1241     return depth;
   1242 }
   1243