cond.c revision 1.228 1 /* $NetBSD: cond.c,v 1.228 2020/12/20 13:03:48 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 directive, such as
76 * '.if <cond>', '.elifnmake <cond>', '.else', '.endif'.
77 *
78 * Cond_EvalCondition
79 * Evaluate the conditional, which is either the argument
80 * of one of the .if directives or the condition in a
81 * ':?then:else' variable modifier.
82 *
83 * Cond_save_depth
84 * Cond_restore_depth
85 * Save and restore the nesting of the conditions, at
86 * the start and end of including another makefile, to
87 * ensure that in each makefile the conditional
88 * directives are well-balanced.
89 */
90
91 #include <errno.h>
92
93 #include "make.h"
94 #include "dir.h"
95
96 /* "@(#)cond.c 8.2 (Berkeley) 1/2/94" */
97 MAKE_RCSID("$NetBSD: cond.c,v 1.228 2020/12/20 13:03:48 rillig Exp $");
98
99 /*
100 * The parsing of conditional expressions is based on this grammar:
101 * E -> F || E
102 * E -> F
103 * F -> T && F
104 * F -> T
105 * T -> defined(variable)
106 * T -> make(target)
107 * T -> exists(file)
108 * T -> empty(varspec)
109 * T -> target(name)
110 * T -> commands(name)
111 * T -> symbol
112 * T -> $(varspec) op value
113 * T -> $(varspec) == "string"
114 * T -> $(varspec) != "string"
115 * T -> "string"
116 * T -> ( E )
117 * T -> ! T
118 * op -> == | != | > | < | >= | <=
119 *
120 * 'symbol' is some other symbol to which the default function is applied.
121 *
122 * The tokens are scanned by CondToken, which returns:
123 * TOK_AND for '&' or '&&'
124 * TOK_OR for '|' or '||'
125 * TOK_NOT for '!'
126 * TOK_LPAREN for '('
127 * TOK_RPAREN for ')'
128 * Other terminal symbols are evaluated using either the default function or
129 * the function given in the terminal, they return either TOK_TRUE or
130 * TOK_FALSE.
131 *
132 * TOK_FALSE is 0 and TOK_TRUE 1 so we can directly assign C comparisons.
133 *
134 * All non-terminal functions (CondParser_Expr, CondParser_Factor and
135 * CondParser_Term) return either TOK_FALSE, TOK_TRUE, or TOK_ERROR on error.
136 */
137 typedef enum Token {
138 TOK_FALSE = 0, TOK_TRUE = 1, TOK_AND, TOK_OR, TOK_NOT,
139 TOK_LPAREN, TOK_RPAREN, TOK_EOF, TOK_NONE, TOK_ERROR
140 } Token;
141
142 typedef struct CondParser {
143 const struct If *if_info; /* Info for current statement */
144 const char *p; /* The remaining condition to parse */
145 Token curr; /* Single push-back token used in parsing */
146
147 /*
148 * Whether an error message has already been printed for this
149 * condition. The first available error message is usually the most
150 * specific one, therefore it makes sense to suppress the standard
151 * "Malformed conditional" message.
152 */
153 Boolean printedError;
154 } CondParser;
155
156 static Token CondParser_Expr(CondParser *par, Boolean);
157
158 static unsigned int cond_depth = 0; /* current .if nesting level */
159 static unsigned int cond_min_depth = 0; /* depth at makefile open */
160
161 /*
162 * Indicate when we should be strict about lhs of comparisons.
163 * In strict mode, the lhs must be a variable expression or a string literal
164 * in quotes. In non-strict mode it may also be an unquoted string literal.
165 *
166 * TRUE when CondEvalExpression is called from Cond_EvalLine (.if etc)
167 * FALSE when CondEvalExpression is called from ApplyModifier_IfElse
168 * since lhs is already expanded, and at that point we cannot tell if
169 * it was a variable reference or not.
170 */
171 static Boolean lhsStrict;
172
173 static int
174 is_token(const char *str, const char *tok, size_t len)
175 {
176 return strncmp(str, tok, len) == 0 && !ch_isalpha(str[len]);
177 }
178
179 static Token
180 ToToken(Boolean cond)
181 {
182 return cond ? TOK_TRUE : TOK_FALSE;
183 }
184
185 /* Push back the most recent token read. We only need one level of this. */
186 static void
187 CondParser_PushBack(CondParser *par, Token t)
188 {
189 assert(par->curr == TOK_NONE);
190 assert(t != TOK_NONE);
191
192 par->curr = t;
193 }
194
195 static void
196 CondParser_SkipWhitespace(CondParser *par)
197 {
198 cpp_skip_whitespace(&par->p);
199 }
200
201 /* Parse the argument of a built-in function.
202 *
203 * Arguments:
204 * *pp initially points at the '(',
205 * upon successful return it points right after the ')'.
206 *
207 * *out_arg receives the argument as string.
208 *
209 * func says whether the argument belongs to an actual function, or
210 * whether the parsed argument is passed to the default function.
211 *
212 * Return the length of the argument, or 0 on error. */
213 static size_t
214 ParseFuncArg(const char **pp, Boolean doEval, const char *func,
215 char **out_arg)
216 {
217 const char *p = *pp;
218 Buffer argBuf;
219 int paren_depth;
220 size_t argLen;
221
222 if (func != NULL)
223 p++; /* Skip opening '(' - verified by caller */
224
225 if (*p == '\0') {
226 *out_arg = NULL; /* Missing closing parenthesis: */
227 return 0; /* .if defined( */
228 }
229
230 cpp_skip_hspace(&p);
231
232 Buf_InitSize(&argBuf, 16);
233
234 paren_depth = 0;
235 for (;;) {
236 char ch = *p;
237 if (ch == '\0' || ch == ' ' || ch == '\t')
238 break;
239 if ((ch == '&' || ch == '|') && paren_depth == 0)
240 break;
241 if (*p == '$') {
242 /*
243 * Parse the variable expression and install it as
244 * part of the argument if it's valid. We tell
245 * Var_Parse to complain on an undefined variable,
246 * (XXX: but Var_Parse ignores that request)
247 * so we don't need to do it. Nor do we return an
248 * error, though perhaps we should.
249 */
250 void *nestedVal_freeIt;
251 VarEvalFlags eflags = doEval
252 ? VARE_WANTRES | VARE_UNDEFERR
253 : VARE_NONE;
254 const char *nestedVal;
255 (void)Var_Parse(&p, VAR_CMDLINE, eflags,
256 &nestedVal, &nestedVal_freeIt);
257 /* TODO: handle errors */
258 Buf_AddStr(&argBuf, nestedVal);
259 free(nestedVal_freeIt);
260 continue;
261 }
262 if (ch == '(')
263 paren_depth++;
264 else if (ch == ')' && --paren_depth < 0)
265 break;
266 Buf_AddByte(&argBuf, *p);
267 p++;
268 }
269
270 *out_arg = Buf_GetAll(&argBuf, &argLen);
271 Buf_Destroy(&argBuf, FALSE);
272
273 cpp_skip_hspace(&p);
274
275 if (func != NULL && *p++ != ')') {
276 Parse_Error(PARSE_WARNING,
277 "Missing closing parenthesis for %s()",
278 func);
279 /* The PARSE_FATAL follows in CondEvalExpression. */
280 return 0;
281 }
282
283 *pp = p;
284 return argLen;
285 }
286
287 /* Test whether the given variable is defined. */
288 static Boolean
289 FuncDefined(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
290 {
291 void *freeIt;
292 Boolean result = Var_Value(arg, VAR_CMDLINE, &freeIt) != NULL;
293 bmake_free(freeIt);
294 return result;
295 }
296
297 /* See if the given target is being made. */
298 static Boolean
299 FuncMake(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
300 {
301 StringListNode *ln;
302
303 for (ln = opts.create.first; ln != NULL; ln = ln->next)
304 if (Str_Match(ln->datum, arg))
305 return TRUE;
306 return FALSE;
307 }
308
309 /* See if the given file exists. */
310 static Boolean
311 FuncExists(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
312 {
313 Boolean result;
314 char *path;
315
316 path = Dir_FindFile(arg, &dirSearchPath);
317 DEBUG2(COND, "exists(%s) result is \"%s\"\n",
318 arg, path != NULL ? path : "");
319 result = path != NULL;
320 free(path);
321 return result;
322 }
323
324 /* See if the given node exists and is an actual target. */
325 static Boolean
326 FuncTarget(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
327 {
328 GNode *gn = Targ_FindNode(arg);
329 return gn != NULL && GNode_IsTarget(gn);
330 }
331
332 /* See if the given node exists and is an actual target with commands
333 * associated with it. */
334 static Boolean
335 FuncCommands(size_t argLen MAKE_ATTR_UNUSED, const char *arg)
336 {
337 GNode *gn = Targ_FindNode(arg);
338 return gn != NULL && GNode_IsTarget(gn) && !Lst_IsEmpty(&gn->commands);
339 }
340
341 /*
342 * Convert the given number into a double.
343 * We try a base 10 or 16 integer conversion first, if that fails
344 * then we try a floating point conversion instead.
345 *
346 * Results:
347 * Returns TRUE if the conversion succeeded.
348 * Sets 'out_value' to the converted number.
349 */
350 static Boolean
351 TryParseNumber(const char *str, double *out_value)
352 {
353 char *end;
354 unsigned long ul_val;
355 double dbl_val;
356
357 errno = 0;
358 if (str[0] == '\0') { /* XXX: why is an empty string a number? */
359 *out_value = 0.0;
360 return TRUE;
361 }
362
363 ul_val = strtoul(str, &end, str[1] == 'x' ? 16 : 10);
364 if (*end == '\0' && errno != ERANGE) {
365 *out_value = str[0] == '-' ? -(double)-ul_val : (double)ul_val;
366 return TRUE;
367 }
368
369 if (*end != '\0' && *end != '.' && *end != 'e' && *end != 'E')
370 return FALSE; /* skip the expensive strtod call */
371 dbl_val = strtod(str, &end);
372 if (*end != '\0')
373 return FALSE;
374
375 *out_value = dbl_val;
376 return TRUE;
377 }
378
379 static Boolean
380 is_separator(char ch)
381 {
382 return ch == '\0' || ch_isspace(ch) || strchr("!=><)", ch) != NULL;
383 }
384
385 /*-
386 * Parse a string from a variable reference or an optionally quoted
387 * string. This is called for the lhs and rhs of string comparisons.
388 *
389 * Results:
390 * Returns the string, absent any quotes, or NULL on error.
391 * Sets out_quoted if the string was quoted.
392 * Sets out_freeIt.
393 */
394 /* coverity:[+alloc : arg-*4] */
395 static void
396 CondParser_String(CondParser *par, Boolean doEval, Boolean strictLHS,
397 FStr *out_str, Boolean *out_quoted)
398 {
399 Buffer buf;
400 FStr str;
401 Boolean atStart;
402 const char *nested_p;
403 Boolean quoted;
404 const char *start;
405 VarEvalFlags eflags;
406 VarParseResult parseResult;
407
408 Buf_Init(&buf);
409 str = FStr_InitRefer(NULL);
410 *out_quoted = quoted = par->p[0] == '"';
411 start = par->p;
412 if (quoted)
413 par->p++;
414
415 while (par->p[0] != '\0' && str.str == NULL) {
416 switch (par->p[0]) {
417 case '\\':
418 par->p++;
419 if (par->p[0] != '\0') {
420 Buf_AddByte(&buf, par->p[0]);
421 par->p++;
422 }
423 continue;
424 case '"':
425 if (quoted) {
426 par->p++; /* skip the closing quote */
427 goto got_str;
428 }
429 Buf_AddByte(&buf, par->p[0]); /* likely? */
430 par->p++;
431 continue;
432 case ')': /* see is_separator */
433 case '!':
434 case '=':
435 case '>':
436 case '<':
437 case ' ':
438 case '\t':
439 if (!quoted)
440 goto got_str;
441 Buf_AddByte(&buf, par->p[0]);
442 par->p++;
443 continue;
444 case '$':
445 /* if we are in quotes, an undefined variable is ok */
446 eflags =
447 doEval && !quoted ? VARE_WANTRES | VARE_UNDEFERR :
448 doEval ? VARE_WANTRES :
449 VARE_NONE;
450
451 nested_p = par->p;
452 atStart = nested_p == start;
453 parseResult = Var_Parse(&nested_p, VAR_CMDLINE, eflags,
454 &str.str, &str.freeIt);
455 /* TODO: handle errors */
456 if (str.str == var_Error) {
457 if (parseResult & VPR_ANY_MSG)
458 par->printedError = TRUE;
459 /*
460 * XXX: Can there be any situation in which
461 * a returned var_Error requires freeIt?
462 */
463 FStr_Done(&str);
464 /*
465 * Even if !doEval, we still report syntax
466 * errors, which is what getting var_Error
467 * back with !doEval means.
468 */
469 str = FStr_InitRefer(NULL);
470 goto cleanup;
471 }
472 par->p = nested_p;
473
474 /*
475 * If the '$' started the string literal (which means
476 * no quotes), and the variable expression is followed
477 * by a space, looks like a comparison operator or is
478 * the end of the expression, we are done.
479 */
480 if (atStart && is_separator(par->p[0]))
481 goto cleanup;
482
483 Buf_AddStr(&buf, str.str);
484 FStr_Done(&str);
485 str = FStr_InitRefer(NULL); /* not finished yet */
486 continue;
487 default:
488 if (strictLHS && !quoted && *start != '$' &&
489 !ch_isdigit(*start)) {
490 /*
491 * The left-hand side must be quoted,
492 * a variable reference or a number.
493 */
494 str = FStr_InitRefer(NULL);
495 goto cleanup;
496 }
497 Buf_AddByte(&buf, par->p[0]);
498 par->p++;
499 continue;
500 }
501 }
502 got_str:
503 str = FStr_InitOwn(Buf_GetAll(&buf, NULL));
504 cleanup:
505 Buf_Destroy(&buf, FALSE);
506 *out_str = str;
507 }
508
509 struct If {
510 const char *form; /* Form of if */
511 size_t formlen; /* Length of form */
512 Boolean doNot; /* TRUE if default function should be negated */
513 /* The default function to apply on unquoted bare words. */
514 Boolean (*defProc)(size_t, const char *);
515 };
516
517 /* The different forms of .if directives. */
518 static const struct If ifs[] = {
519 { "def", 3, FALSE, FuncDefined },
520 { "ndef", 4, TRUE, FuncDefined },
521 { "make", 4, FALSE, FuncMake },
522 { "nmake", 5, TRUE, FuncMake },
523 { "", 0, FALSE, FuncDefined },
524 { NULL, 0, FALSE, NULL }
525 };
526 enum {
527 PLAIN_IF_INDEX = 4
528 };
529
530 static Boolean
531 If_Eval(const struct If *if_info, const char *arg, size_t arglen)
532 {
533 Boolean res = if_info->defProc(arglen, arg);
534 return if_info->doNot ? !res : res;
535 }
536
537 /* Evaluate a "comparison without operator", such as in ".if ${VAR}" or
538 * ".if 0". */
539 static Boolean
540 EvalNotEmpty(CondParser *par, const char *value, Boolean quoted)
541 {
542 double num;
543
544 /* For .ifxxx "...", check for non-empty string. */
545 if (quoted)
546 return value[0] != '\0';
547
548 /* For .ifxxx <number>, compare against zero */
549 if (TryParseNumber(value, &num))
550 return num != 0.0;
551
552 /* For .if ${...}, check for non-empty string. This is different from
553 * the evaluation function from that .if variant, which would test
554 * whether a variable of the given name were defined. */
555 /* XXX: Whitespace should count as empty, just as in ParseEmptyArg. */
556 if (par->if_info->form[0] == '\0')
557 return value[0] != '\0';
558
559 /* For the other variants of .ifxxx ${...}, use its default function. */
560 return If_Eval(par->if_info, value, strlen(value));
561 }
562
563 /* Evaluate a numerical comparison, such as in ".if ${VAR} >= 9". */
564 static Token
565 EvalCompareNum(double lhs, const char *op, double rhs)
566 {
567 DEBUG3(COND, "lhs = %f, rhs = %f, op = %.2s\n", lhs, rhs, op);
568
569 switch (op[0]) {
570 case '!':
571 if (op[1] != '=') {
572 Parse_Error(PARSE_WARNING, "Unknown operator");
573 /* The PARSE_FATAL follows in CondEvalExpression. */
574 return TOK_ERROR;
575 }
576 return ToToken(lhs != rhs);
577 case '=':
578 if (op[1] != '=') {
579 Parse_Error(PARSE_WARNING, "Unknown operator");
580 /* The PARSE_FATAL follows in CondEvalExpression. */
581 return TOK_ERROR;
582 }
583 return ToToken(lhs == rhs);
584 case '<':
585 return ToToken(op[1] == '=' ? lhs <= rhs : lhs < rhs);
586 case '>':
587 return ToToken(op[1] == '=' ? lhs >= rhs : lhs > rhs);
588 }
589 return TOK_ERROR;
590 }
591
592 static Token
593 EvalCompareStr(const char *lhs, const char *op, const char *rhs)
594 {
595 if (!((op[0] == '!' || op[0] == '=') && op[1] == '=')) {
596 Parse_Error(PARSE_WARNING,
597 "String comparison operator "
598 "must be either == or !=");
599 /* The PARSE_FATAL follows in CondEvalExpression. */
600 return TOK_ERROR;
601 }
602
603 DEBUG3(COND, "lhs = \"%s\", rhs = \"%s\", op = %.2s\n", lhs, rhs, op);
604 return ToToken((*op == '=') == (strcmp(lhs, rhs) == 0));
605 }
606
607 /* Evaluate a comparison, such as "${VAR} == 12345". */
608 static Token
609 EvalCompare(const char *lhs, Boolean lhsQuoted, const char *op,
610 const char *rhs, Boolean rhsQuoted)
611 {
612 double left, right;
613
614 if (!rhsQuoted && !lhsQuoted)
615 if (TryParseNumber(lhs, &left) && TryParseNumber(rhs, &right))
616 return EvalCompareNum(left, op, right);
617
618 return EvalCompareStr(lhs, op, rhs);
619 }
620
621 /* Parse a comparison condition such as:
622 *
623 * 0
624 * ${VAR:Mpattern}
625 * ${VAR} == value
626 * ${VAR:U0} < 12345
627 */
628 static Token
629 CondParser_Comparison(CondParser *par, Boolean doEval)
630 {
631 Token t = TOK_ERROR;
632 FStr lhs, rhs;
633 const char *op;
634 Boolean lhsQuoted, rhsQuoted;
635
636 /*
637 * Parse the variable spec and skip over it, saving its
638 * value in lhs.
639 */
640 CondParser_String(par, doEval, lhsStrict, &lhs, &lhsQuoted);
641 if (lhs.str == NULL)
642 goto done_lhs;
643
644 CondParser_SkipWhitespace(par);
645
646 op = par->p;
647 switch (par->p[0]) {
648 case '!':
649 case '=':
650 case '<':
651 case '>':
652 if (par->p[1] == '=')
653 par->p += 2;
654 else
655 par->p++;
656 break;
657 default:
658 /* Unknown operator, compare against an empty string or 0. */
659 t = ToToken(doEval && EvalNotEmpty(par, lhs.str, lhsQuoted));
660 goto done_lhs;
661 }
662
663 CondParser_SkipWhitespace(par);
664
665 if (par->p[0] == '\0') {
666 Parse_Error(PARSE_WARNING,
667 "Missing right-hand-side of operator");
668 /* The PARSE_FATAL follows in CondEvalExpression. */
669 goto done_lhs;
670 }
671
672 CondParser_String(par, doEval, FALSE, &rhs, &rhsQuoted);
673 if (rhs.str == NULL)
674 goto done_rhs;
675
676 if (!doEval) {
677 t = TOK_FALSE;
678 goto done_rhs;
679 }
680
681 t = EvalCompare(lhs.str, lhsQuoted, op, rhs.str, rhsQuoted);
682
683 done_rhs:
684 FStr_Done(&rhs);
685 done_lhs:
686 FStr_Done(&lhs);
687 return t;
688 }
689
690 /* The argument to empty() is a variable name, optionally followed by
691 * variable modifiers. */
692 static size_t
693 ParseEmptyArg(const char **pp, Boolean doEval,
694 const char *func MAKE_ATTR_UNUSED, char **out_arg)
695 {
696 void *val_freeIt;
697 const char *val;
698 size_t magic_res;
699
700 /* We do all the work here and return the result as the length */
701 *out_arg = NULL;
702
703 (*pp)--; /* Make (*pp)[1] point to the '('. */
704 (void)Var_Parse(pp, VAR_CMDLINE, doEval ? VARE_WANTRES : VARE_NONE,
705 &val, &val_freeIt);
706 /* TODO: handle errors */
707 /* If successful, *pp points beyond the closing ')' now. */
708
709 if (val == var_Error) {
710 free(val_freeIt);
711 return (size_t)-1;
712 }
713
714 /*
715 * A variable is empty when it just contains spaces...
716 * 4/15/92, christos
717 */
718 cpp_skip_whitespace(&val);
719
720 /*
721 * For consistency with the other functions we can't generate the
722 * true/false here.
723 */
724 magic_res = *val != '\0' ? 2 : 1;
725 free(val_freeIt);
726 return magic_res;
727 }
728
729 static Boolean
730 FuncEmpty(size_t arglen, const char *arg MAKE_ATTR_UNUSED)
731 {
732 /* Magic values ahead, see ParseEmptyArg. */
733 return arglen == 1;
734 }
735
736 static Boolean
737 CondParser_Func(CondParser *par, Boolean doEval, Token *out_token)
738 {
739 static const struct fn_def {
740 const char *fn_name;
741 size_t fn_name_len;
742 size_t (*fn_parse)(const char **, Boolean, const char *,
743 char **);
744 Boolean (*fn_eval)(size_t, const char *);
745 } fns[] = {
746 { "defined", 7, ParseFuncArg, FuncDefined },
747 { "make", 4, ParseFuncArg, FuncMake },
748 { "exists", 6, ParseFuncArg, FuncExists },
749 { "empty", 5, ParseEmptyArg, FuncEmpty },
750 { "target", 6, ParseFuncArg, FuncTarget },
751 { "commands", 8, ParseFuncArg, FuncCommands }
752 };
753 const struct fn_def *fn;
754 char *arg = NULL;
755 size_t arglen;
756 const char *cp = par->p;
757 const struct fn_def *fns_end = fns + sizeof fns / sizeof fns[0];
758
759 for (fn = fns; fn != fns_end; fn++) {
760 if (!is_token(cp, fn->fn_name, fn->fn_name_len))
761 continue;
762
763 cp += fn->fn_name_len;
764 cpp_skip_whitespace(&cp);
765 if (*cp != '(')
766 break;
767
768 arglen = fn->fn_parse(&cp, doEval, fn->fn_name, &arg);
769 if (arglen == 0 || arglen == (size_t)-1) {
770 par->p = cp;
771 *out_token = arglen == 0 ? TOK_FALSE : TOK_ERROR;
772 return TRUE;
773 }
774
775 /* Evaluate the argument using the required function. */
776 *out_token = ToToken(!doEval || fn->fn_eval(arglen, arg));
777 free(arg);
778 par->p = cp;
779 return TRUE;
780 }
781
782 return FALSE;
783 }
784
785 /* Parse a function call, a number, a variable expression or a string
786 * literal. */
787 static Token
788 CondParser_LeafToken(CondParser *par, Boolean doEval)
789 {
790 Token t;
791 char *arg = NULL;
792 size_t arglen;
793 const char *cp;
794 const char *cp1;
795
796 if (CondParser_Func(par, doEval, &t))
797 return t;
798
799 /* Push anything numeric through the compare expression */
800 cp = par->p;
801 if (ch_isdigit(cp[0]) || cp[0] == '-' || cp[0] == '+')
802 return CondParser_Comparison(par, doEval);
803
804 /*
805 * Most likely we have a naked token to apply the default function to.
806 * However ".if a == b" gets here when the "a" is unquoted and doesn't
807 * start with a '$'. This surprises people.
808 * If what follows the function argument is a '=' or '!' then the
809 * syntax would be invalid if we did "defined(a)" - so instead treat
810 * as an expression.
811 */
812 arglen = ParseFuncArg(&cp, doEval, NULL, &arg);
813 cp1 = cp;
814 cpp_skip_whitespace(&cp1);
815 if (*cp1 == '=' || *cp1 == '!')
816 return CondParser_Comparison(par, doEval);
817 par->p = cp;
818
819 /*
820 * Evaluate the argument using the default function.
821 * This path always treats .if as .ifdef. To get here, the character
822 * after .if must have been taken literally, so the argument cannot
823 * be empty - even if it contained a variable expansion.
824 */
825 t = ToToken(!doEval || If_Eval(par->if_info, arg, arglen));
826 free(arg);
827 return t;
828 }
829
830 /* Return the next token or comparison result from the parser. */
831 static Token
832 CondParser_Token(CondParser *par, Boolean doEval)
833 {
834 Token t;
835
836 t = par->curr;
837 if (t != TOK_NONE) {
838 par->curr = TOK_NONE;
839 return t;
840 }
841
842 cpp_skip_hspace(&par->p);
843
844 switch (par->p[0]) {
845
846 case '(':
847 par->p++;
848 return TOK_LPAREN;
849
850 case ')':
851 par->p++;
852 return TOK_RPAREN;
853
854 case '|':
855 par->p++;
856 if (par->p[0] == '|')
857 par->p++;
858 else if (opts.lint) {
859 Parse_Error(PARSE_FATAL, "Unknown operator '|'");
860 par->printedError = TRUE;
861 return TOK_ERROR;
862 }
863 return TOK_OR;
864
865 case '&':
866 par->p++;
867 if (par->p[0] == '&')
868 par->p++;
869 else if (opts.lint) {
870 Parse_Error(PARSE_FATAL, "Unknown operator '&'");
871 par->printedError = TRUE;
872 return TOK_ERROR;
873 }
874 return TOK_AND;
875
876 case '!':
877 par->p++;
878 return TOK_NOT;
879
880 case '#': /* XXX: see unit-tests/cond-token-plain.mk */
881 case '\n': /* XXX: why should this end the condition? */
882 /* Probably obsolete now, from 1993-03-21. */
883 case '\0':
884 return TOK_EOF;
885
886 case '"':
887 case '$':
888 return CondParser_Comparison(par, doEval);
889
890 default:
891 return CondParser_LeafToken(par, doEval);
892 }
893 }
894
895 /* Parse a single term in the expression. This consists of a terminal symbol
896 * or TOK_NOT and a term (not including the binary operators):
897 *
898 * T -> defined(variable) | make(target) | exists(file) | symbol
899 * T -> ! T | ( E )
900 *
901 * Results:
902 * TOK_TRUE, TOK_FALSE or TOK_ERROR.
903 */
904 static Token
905 CondParser_Term(CondParser *par, Boolean doEval)
906 {
907 Token t;
908
909 t = CondParser_Token(par, doEval);
910
911 if (t == TOK_EOF) {
912 /*
913 * If we reached the end of the expression, the expression
914 * is malformed...
915 */
916 t = TOK_ERROR;
917 } else if (t == TOK_LPAREN) {
918 /*
919 * T -> ( E )
920 */
921 t = CondParser_Expr(par, doEval);
922 if (t != TOK_ERROR) {
923 if (CondParser_Token(par, doEval) != TOK_RPAREN) {
924 t = TOK_ERROR;
925 }
926 }
927 } else if (t == TOK_NOT) {
928 t = CondParser_Term(par, doEval);
929 if (t == TOK_TRUE) {
930 t = TOK_FALSE;
931 } else if (t == TOK_FALSE) {
932 t = TOK_TRUE;
933 }
934 }
935 return t;
936 }
937
938 /* Parse a conjunctive factor (nice name, wot?)
939 *
940 * F -> T && F | T
941 *
942 * Results:
943 * TOK_TRUE, TOK_FALSE or TOK_ERROR
944 */
945 static Token
946 CondParser_Factor(CondParser *par, Boolean doEval)
947 {
948 Token l, o;
949
950 l = CondParser_Term(par, doEval);
951 if (l != TOK_ERROR) {
952 o = CondParser_Token(par, doEval);
953
954 if (o == TOK_AND) {
955 /*
956 * F -> T && F
957 *
958 * If T is TOK_FALSE, the whole thing will be
959 * TOK_FALSE, but we have to parse the r.h.s. anyway
960 * (to throw it away). If T is TOK_TRUE, the result
961 * is the r.h.s., be it a TOK_ERROR or not.
962 */
963 if (l == TOK_TRUE) {
964 l = CondParser_Factor(par, doEval);
965 } else {
966 (void)CondParser_Factor(par, FALSE);
967 }
968 } else {
969 /*
970 * F -> T
971 */
972 CondParser_PushBack(par, o);
973 }
974 }
975 return l;
976 }
977
978 /* Main expression production.
979 *
980 * E -> F || E | F
981 *
982 * Results:
983 * TOK_TRUE, TOK_FALSE or TOK_ERROR.
984 */
985 static Token
986 CondParser_Expr(CondParser *par, Boolean doEval)
987 {
988 Token l, o;
989
990 l = CondParser_Factor(par, doEval);
991 if (l != TOK_ERROR) {
992 o = CondParser_Token(par, doEval);
993
994 if (o == TOK_OR) {
995 /*
996 * E -> F || E
997 *
998 * A similar thing occurs for ||, except that here
999 * we make sure the l.h.s. is TOK_FALSE before we
1000 * bother to evaluate the r.h.s. Once again, if l
1001 * is TOK_FALSE, the result is the r.h.s. and once
1002 * again if l is TOK_TRUE, we parse the r.h.s. to
1003 * throw it away.
1004 */
1005 if (l == TOK_FALSE) {
1006 l = CondParser_Expr(par, doEval);
1007 } else {
1008 (void)CondParser_Expr(par, FALSE);
1009 }
1010 } else {
1011 /*
1012 * E -> F
1013 */
1014 CondParser_PushBack(par, o);
1015 }
1016 }
1017 return l;
1018 }
1019
1020 static CondEvalResult
1021 CondParser_Eval(CondParser *par, Boolean *value)
1022 {
1023 Token res;
1024
1025 DEBUG1(COND, "CondParser_Eval: %s\n", par->p);
1026
1027 res = CondParser_Expr(par, TRUE);
1028 if (res != TOK_FALSE && res != TOK_TRUE)
1029 return COND_INVALID;
1030
1031 if (CondParser_Token(par, FALSE) != TOK_EOF)
1032 return COND_INVALID;
1033
1034 *value = res == TOK_TRUE;
1035 return COND_PARSE;
1036 }
1037
1038 /* Evaluate the condition, including any side effects from the variable
1039 * expressions in the condition. The condition consists of &&, ||, !,
1040 * function(arg), comparisons and parenthetical groupings thereof.
1041 *
1042 * Results:
1043 * COND_PARSE if the condition was valid grammatically
1044 * COND_INVALID if not a valid conditional.
1045 *
1046 * (*value) is set to the boolean value of the condition
1047 */
1048 static CondEvalResult
1049 CondEvalExpression(const struct If *info, const char *cond, Boolean *value,
1050 Boolean eprint, Boolean strictLHS)
1051 {
1052 CondParser par;
1053 CondEvalResult rval;
1054
1055 lhsStrict = strictLHS;
1056
1057 cpp_skip_hspace(&cond);
1058
1059 par.if_info = info != NULL ? info : ifs + PLAIN_IF_INDEX;
1060 par.p = cond;
1061 par.curr = TOK_NONE;
1062 par.printedError = FALSE;
1063
1064 rval = CondParser_Eval(&par, value);
1065
1066 if (rval == COND_INVALID && eprint && !par.printedError)
1067 Parse_Error(PARSE_FATAL, "Malformed conditional (%s)", cond);
1068
1069 return rval;
1070 }
1071
1072 /* Evaluate a condition in a :? modifier, such as
1073 * ${"${VAR}" == value:?yes:no}. */
1074 CondEvalResult
1075 Cond_EvalCondition(const char *cond, Boolean *out_value)
1076 {
1077 return CondEvalExpression(NULL, cond, out_value, FALSE, FALSE);
1078 }
1079
1080 static Boolean
1081 IsEndif(const char *p)
1082 {
1083 return p[0] == 'e' && p[1] == 'n' && p[2] == 'd' &&
1084 p[3] == 'i' && p[4] == 'f' && !ch_isalpha(p[5]);
1085 }
1086
1087 /* Evaluate the conditional directive in the line, which is one of:
1088 *
1089 * .if <cond>
1090 * .ifmake <cond>
1091 * .ifnmake <cond>
1092 * .ifdef <cond>
1093 * .ifndef <cond>
1094 * .elif <cond>
1095 * .elifmake <cond>
1096 * .elifnmake <cond>
1097 * .elifdef <cond>
1098 * .elifndef <cond>
1099 * .else
1100 * .endif
1101 *
1102 * In these directives, <cond> consists of &&, ||, !, function(arg),
1103 * comparisons, expressions, bare words, numbers and strings, and
1104 * parenthetical groupings thereof.
1105 *
1106 * Results:
1107 * COND_PARSE to continue parsing the lines that follow the
1108 * conditional (when <cond> evaluates to TRUE)
1109 * COND_SKIP to skip the lines after the conditional
1110 * (when <cond> evaluates to FALSE, or when a previous
1111 * branch has already been taken)
1112 * COND_INVALID if the conditional was not valid, either because of
1113 * a syntax error or because some variable was undefined
1114 * or because the condition could not be evaluated
1115 */
1116 CondEvalResult
1117 Cond_EvalLine(const char *line)
1118 {
1119 typedef enum IfState {
1120
1121 /* None of the previous <cond> evaluated to TRUE. */
1122 IFS_INITIAL = 0,
1123
1124 /* The previous <cond> evaluated to TRUE.
1125 * The lines following this condition are interpreted. */
1126 IFS_ACTIVE = 1 << 0,
1127
1128 /* The previous directive was an '.else'. */
1129 IFS_SEEN_ELSE = 1 << 1,
1130
1131 /* One of the previous <cond> evaluated to TRUE. */
1132 IFS_WAS_ACTIVE = 1 << 2
1133
1134 } IfState;
1135
1136 static enum IfState *cond_states = NULL;
1137 static unsigned int cond_states_cap = 128;
1138
1139 const struct If *ifp;
1140 Boolean isElif;
1141 Boolean value;
1142 IfState state;
1143 const char *p = line;
1144
1145 if (cond_states == NULL) {
1146 cond_states = bmake_malloc(
1147 cond_states_cap * sizeof *cond_states);
1148 cond_states[0] = IFS_ACTIVE;
1149 }
1150
1151 p++; /* skip the leading '.' */
1152 cpp_skip_hspace(&p);
1153
1154 if (IsEndif(p)) { /* It is an '.endif'. */
1155 if (p[5] != '\0') {
1156 Parse_Error(PARSE_FATAL,
1157 "The .endif directive does not take arguments.");
1158 }
1159
1160 if (cond_depth == cond_min_depth) {
1161 Parse_Error(PARSE_FATAL, "if-less endif");
1162 return COND_PARSE;
1163 }
1164
1165 /* Return state for previous conditional */
1166 cond_depth--;
1167 return cond_states[cond_depth] & IFS_ACTIVE
1168 ? COND_PARSE : COND_SKIP;
1169 }
1170
1171 /* Parse the name of the directive, such as 'if', 'elif', 'endif'. */
1172 if (p[0] == 'e') {
1173 if (p[1] != 'l') {
1174 /*
1175 * Unknown directive. It might still be a
1176 * transformation rule like '.elisp.scm',
1177 * therefore no error message here.
1178 */
1179 return COND_INVALID;
1180 }
1181
1182 /* Quite likely this is 'else' or 'elif' */
1183 p += 2;
1184 if (is_token(p, "se", 2)) { /* It is an 'else'. */
1185
1186 if (p[2] != '\0')
1187 Parse_Error(PARSE_FATAL,
1188 "The .else directive "
1189 "does not take arguments.");
1190
1191 if (cond_depth == cond_min_depth) {
1192 Parse_Error(PARSE_FATAL, "if-less else");
1193 return COND_PARSE;
1194 }
1195
1196 state = cond_states[cond_depth];
1197 if (state == IFS_INITIAL) {
1198 state = IFS_ACTIVE | IFS_SEEN_ELSE;
1199 } else {
1200 if (state & IFS_SEEN_ELSE)
1201 Parse_Error(PARSE_WARNING,
1202 "extra else");
1203 state = IFS_WAS_ACTIVE | IFS_SEEN_ELSE;
1204 }
1205 cond_states[cond_depth] = state;
1206
1207 return state & IFS_ACTIVE ? COND_PARSE : COND_SKIP;
1208 }
1209 /* Assume for now it is an elif */
1210 isElif = TRUE;
1211 } else
1212 isElif = FALSE;
1213
1214 if (p[0] != 'i' || p[1] != 'f') {
1215 /*
1216 * Unknown directive. It might still be a transformation rule
1217 * like '.elisp.scm', therefore no error message here.
1218 */
1219 return COND_INVALID; /* Not an ifxxx or elifxxx line */
1220 }
1221
1222 /*
1223 * Figure out what sort of conditional it is -- what its default
1224 * function is, etc. -- by looking in the table of valid "ifs"
1225 */
1226 p += 2;
1227 for (ifp = ifs;; ifp++) {
1228 if (ifp->form == NULL) {
1229 /*
1230 * TODO: Add error message about unknown directive,
1231 * since there is no other known directive that starts
1232 * with 'el' or 'if'.
1233 *
1234 * Example: .elifx 123
1235 */
1236 return COND_INVALID;
1237 }
1238 if (is_token(p, ifp->form, ifp->formlen)) {
1239 p += ifp->formlen;
1240 break;
1241 }
1242 }
1243
1244 /* Now we know what sort of 'if' it is... */
1245
1246 if (isElif) {
1247 if (cond_depth == cond_min_depth) {
1248 Parse_Error(PARSE_FATAL, "if-less elif");
1249 return COND_PARSE;
1250 }
1251 state = cond_states[cond_depth];
1252 if (state & IFS_SEEN_ELSE) {
1253 Parse_Error(PARSE_WARNING, "extra elif");
1254 cond_states[cond_depth] =
1255 IFS_WAS_ACTIVE | IFS_SEEN_ELSE;
1256 return COND_SKIP;
1257 }
1258 if (state != IFS_INITIAL) {
1259 cond_states[cond_depth] = IFS_WAS_ACTIVE;
1260 return COND_SKIP;
1261 }
1262 } else {
1263 /* Normal .if */
1264 if (cond_depth + 1 >= cond_states_cap) {
1265 /*
1266 * This is rare, but not impossible.
1267 * In meta mode, dirdeps.mk (only runs at level 0)
1268 * can need more than the default.
1269 */
1270 cond_states_cap += 32;
1271 cond_states = bmake_realloc(cond_states,
1272 cond_states_cap *
1273 sizeof *cond_states);
1274 }
1275 state = cond_states[cond_depth];
1276 cond_depth++;
1277 if (!(state & IFS_ACTIVE)) {
1278 /*
1279 * If we aren't parsing the data,
1280 * treat as always false.
1281 */
1282 cond_states[cond_depth] = IFS_WAS_ACTIVE;
1283 return COND_SKIP;
1284 }
1285 }
1286
1287 /* And evaluate the conditional expression */
1288 if (CondEvalExpression(ifp, p, &value, TRUE, TRUE) == COND_INVALID) {
1289 /* Syntax error in conditional, error message already output. */
1290 /* Skip everything to matching .endif */
1291 /* XXX: An extra '.else' is not detected in this case. */
1292 cond_states[cond_depth] = IFS_WAS_ACTIVE;
1293 return COND_SKIP;
1294 }
1295
1296 if (!value) {
1297 cond_states[cond_depth] = IFS_INITIAL;
1298 return COND_SKIP;
1299 }
1300 cond_states[cond_depth] = IFS_ACTIVE;
1301 return COND_PARSE;
1302 }
1303
1304 void
1305 Cond_restore_depth(unsigned int saved_depth)
1306 {
1307 unsigned int open_conds = cond_depth - cond_min_depth;
1308
1309 if (open_conds != 0 || saved_depth > cond_depth) {
1310 Parse_Error(PARSE_FATAL, "%u open conditional%s",
1311 open_conds, open_conds == 1 ? "" : "s");
1312 cond_depth = cond_min_depth;
1313 }
1314
1315 cond_min_depth = saved_depth;
1316 }
1317
1318 unsigned int
1319 Cond_save_depth(void)
1320 {
1321 unsigned int depth = cond_min_depth;
1322
1323 cond_min_depth = cond_depth;
1324 return depth;
1325 }
1326