var.c revision 1.890 1 /* $NetBSD: var.c,v 1.890 2021/03/15 19:15:04 rillig Exp $ */
2
3 /*
4 * Copyright (c) 1988, 1989, 1990, 1993
5 * The Regents of the University of California. 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) 1989 by Berkeley Softworks
37 * All rights reserved.
38 *
39 * This code is derived from software contributed to Berkeley by
40 * Adam de Boor.
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. All advertising materials mentioning features or use of this software
51 * must display the following acknowledgement:
52 * This product includes software developed by the University of
53 * California, Berkeley and its contributors.
54 * 4. Neither the name of the University nor the names of its contributors
55 * may be used to endorse or promote products derived from this software
56 * without specific prior written permission.
57 *
58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68 * SUCH DAMAGE.
69 */
70
71 /*
72 * Handling of variables and the expressions formed from them.
73 *
74 * Variables are set using lines of the form VAR=value. Both the variable
75 * name and the value can contain references to other variables, by using
76 * expressions like ${VAR}, ${VAR:Modifiers}, ${${VARNAME}} or ${VAR:${MODS}}.
77 *
78 * Interface:
79 * Var_Init Initialize this module.
80 *
81 * Var_End Clean up the module.
82 *
83 * Var_Set
84 * Var_SetExpand
85 * Set the value of the variable, creating it if
86 * necessary.
87 *
88 * Var_Append
89 * Var_AppendExpand
90 * Append more characters to the variable, creating it if
91 * necessary. A space is placed between the old value and
92 * the new one.
93 *
94 * Var_Exists
95 * Var_ExistsExpand
96 * See if a variable exists.
97 *
98 * Var_Value Return the unexpanded value of a variable, or NULL if
99 * the variable is undefined.
100 *
101 * Var_Subst Substitute all variable expressions in a string.
102 *
103 * Var_Parse Parse a variable expression such as ${VAR:Mpattern}.
104 *
105 * Var_Delete
106 * Var_DeleteExpand
107 * Delete a variable.
108 *
109 * Var_ReexportVars
110 * Export some or even all variables to the environment
111 * of this process and its child processes.
112 *
113 * Var_Export Export the variable to the environment of this process
114 * and its child processes.
115 *
116 * Var_UnExport Don't export the variable anymore.
117 *
118 * Debugging:
119 * Var_Stats Print out hashing statistics if in -dh mode.
120 *
121 * Var_Dump Print out all variables defined in the given scope.
122 *
123 * XXX: There's a lot of almost duplicate code in these functions that only
124 * differs in subtle details that are not mentioned in the manual page.
125 */
126
127 #include <sys/stat.h>
128 #ifndef NO_REGEX
129 #include <sys/types.h>
130 #include <regex.h>
131 #endif
132 #include <errno.h>
133 #include <inttypes.h>
134 #include <limits.h>
135 #include <time.h>
136
137 #include "make.h"
138 #include "dir.h"
139 #include "job.h"
140 #include "metachar.h"
141
142 /* "@(#)var.c 8.3 (Berkeley) 3/19/94" */
143 MAKE_RCSID("$NetBSD: var.c,v 1.890 2021/03/15 19:15:04 rillig Exp $");
144
145 typedef enum VarFlags {
146 VFL_NONE = 0,
147
148 /*
149 * The variable's value is currently being used by Var_Parse or
150 * Var_Subst. This marker is used to avoid endless recursion.
151 */
152 VFL_IN_USE = 1 << 0,
153
154 /*
155 * The variable comes from the environment.
156 * These variables are not registered in any GNode, therefore they
157 * must be freed as soon as they are not used anymore.
158 */
159 VFL_FROM_ENV = 1 << 1,
160
161 /*
162 * The variable is exported to the environment, to be used by child
163 * processes.
164 */
165 VFL_EXPORTED = 1 << 2,
166
167 /*
168 * At the point where this variable was exported, it contained an
169 * unresolved reference to another variable. Before any child
170 * process is started, it needs to be exported again, in the hope
171 * that the referenced variable can then be resolved.
172 */
173 VFL_REEXPORT = 1 << 3,
174
175 /* The variable came from the command line. */
176 VFL_FROM_CMD = 1 << 4,
177
178 /*
179 * The variable value cannot be changed anymore, and the variable
180 * cannot be deleted. Any attempts to do so are silently ignored,
181 * they are logged with -dv though.
182 *
183 * See VAR_SET_READONLY.
184 */
185 VFL_READONLY = 1 << 5
186 } VarFlags;
187
188 /*
189 * Variables are defined using one of the VAR=value assignments. Their
190 * value can be queried by expressions such as $V, ${VAR}, or with modifiers
191 * such as ${VAR:S,from,to,g:Q}.
192 *
193 * There are 3 kinds of variables: scope variables, environment variables,
194 * undefined variables.
195 *
196 * Scope variables are stored in a GNode.scope. The only way to undefine
197 * a scope variable is using the .undef directive. In particular, it must
198 * not be possible to undefine a variable during the evaluation of an
199 * expression, or Var.name might point nowhere.
200 *
201 * Environment variables are temporary. They are returned by VarFind, and
202 * after using them, they must be freed using VarFreeEnv.
203 *
204 * Undefined variables occur during evaluation of variable expressions such
205 * as ${UNDEF:Ufallback} in Var_Parse and ApplyModifiers.
206 */
207 typedef struct Var {
208 /*
209 * The name of the variable, once set, doesn't change anymore.
210 * For scope variables, it aliases the corresponding HashEntry name.
211 * For environment and undefined variables, it is allocated.
212 */
213 FStr name;
214
215 /* The unexpanded value of the variable. */
216 Buffer val;
217 /* Miscellaneous status flags. */
218 VarFlags flags;
219 } Var;
220
221 /*
222 * Exporting variables is expensive and may leak memory, so skip it if we
223 * can.
224 *
225 * To avoid this, it might be worth encapsulating the environment variables
226 * in a separate data structure called EnvVars.
227 */
228 typedef enum VarExportedMode {
229 VAR_EXPORTED_NONE,
230 VAR_EXPORTED_SOME,
231 VAR_EXPORTED_ALL
232 } VarExportedMode;
233
234 typedef enum UnexportWhat {
235 /* Unexport the variables given by name. */
236 UNEXPORT_NAMED,
237 /*
238 * Unexport all globals previously exported, but keep the environment
239 * inherited from the parent.
240 */
241 UNEXPORT_ALL,
242 /*
243 * Unexport all globals previously exported and clear the environment
244 * inherited from the parent.
245 */
246 UNEXPORT_ENV
247 } UnexportWhat;
248
249 /* Flags for pattern matching in the :S and :C modifiers */
250 typedef struct VarPatternFlags {
251
252 /* Replace as often as possible ('g') */
253 Boolean subGlobal: 1;
254 /* Replace only once ('1') */
255 Boolean subOnce: 1;
256 /* Match at start of word ('^') */
257 Boolean anchorStart: 1;
258 /* Match at end of word ('$') */
259 Boolean anchorEnd: 1;
260 } VarPatternFlags;
261
262 /* SepBuf builds a string from words interleaved with separators. */
263 typedef struct SepBuf {
264 Buffer buf;
265 Boolean needSep;
266 /* Usually ' ', but see the ':ts' modifier. */
267 char sep;
268 } SepBuf;
269
270 static const char *
271 VarEvalFlags_ToString(VarEvalFlags eflags)
272 {
273 if (!eflags.wantRes) {
274 assert(!eflags.undefErr);
275 assert(!eflags.keepDollar && !eflags.keepUndef);
276 return "parse-only";
277 }
278 if (eflags.undefErr) {
279 assert(!eflags.keepDollar && !eflags.keepUndef);
280 return "eval-defined";
281 }
282 if (eflags.keepDollar && eflags.keepUndef)
283 return "eval-keep-dollar-and-undefined";
284 if (eflags.keepDollar)
285 return "eval-keep-dollar";
286 if (eflags.keepUndef)
287 return "eval-keep-undefined";
288 return "eval";
289 }
290
291 /*
292 * This lets us tell if we have replaced the original environ
293 * (which we cannot free).
294 */
295 char **savedEnv = NULL;
296
297 /*
298 * Special return value for Var_Parse, indicating a parse error. It may be
299 * caused by an undefined variable, a syntax error in a modifier or
300 * something entirely different.
301 */
302 char var_Error[] = "";
303
304 /*
305 * Special return value for Var_Parse, indicating an undefined variable in
306 * a case where VarEvalFlags.undefErr is not set. This undefined variable is
307 * typically a dynamic variable such as ${.TARGET}, whose expansion needs to
308 * be deferred until it is defined in an actual target.
309 *
310 * See VarEvalFlags.keepUndef.
311 */
312 static char varUndefined[] = "";
313
314 /*
315 * Traditionally this make consumed $$ during := like any other expansion.
316 * Other make's do not, and this make follows straight since 2016-01-09.
317 *
318 * This knob allows controlling the behavior.
319 * FALSE to consume $$ during := assignment.
320 * TRUE to preserve $$ during := assignment.
321 */
322 #define MAKE_SAVE_DOLLARS ".MAKE.SAVE_DOLLARS"
323 static Boolean save_dollars = TRUE;
324
325 /*
326 * A scope collects variable names and their values.
327 *
328 * The main scope is SCOPE_GLOBAL, which contains the variables that are set
329 * in the makefiles. SCOPE_INTERNAL acts as a fallback for SCOPE_GLOBAL and
330 * contains some internal make variables. These internal variables can thus
331 * be overridden, they can also be restored by undefining the overriding
332 * variable.
333 *
334 * SCOPE_CMDLINE contains variables from the command line arguments. These
335 * override variables from SCOPE_GLOBAL.
336 *
337 * There is no scope for environment variables, these are generated on-the-fly
338 * whenever they are referenced. If there were such a scope, each change to
339 * environment variables would have to be reflected in that scope, which may
340 * be simpler or more complex than the current implementation.
341 *
342 * Each target has its own scope, containing the 7 target-local variables
343 * .TARGET, .ALLSRC, etc. No other variables are in these scopes.
344 */
345
346 GNode *SCOPE_CMDLINE;
347 GNode *SCOPE_GLOBAL;
348 GNode *SCOPE_INTERNAL;
349
350 ENUM_FLAGS_RTTI_6(VarFlags,
351 VFL_IN_USE, VFL_FROM_ENV,
352 VFL_EXPORTED, VFL_REEXPORT, VFL_FROM_CMD, VFL_READONLY);
353
354 static VarExportedMode var_exportedVars = VAR_EXPORTED_NONE;
355
356
357 static Var *
358 VarNew(FStr name, const char *value, VarFlags flags)
359 {
360 size_t value_len = strlen(value);
361 Var *var = bmake_malloc(sizeof *var);
362 var->name = name;
363 Buf_InitSize(&var->val, value_len + 1);
364 Buf_AddBytes(&var->val, value, value_len);
365 var->flags = flags;
366 return var;
367 }
368
369 static const char *
370 CanonicalVarname(const char *name)
371 {
372 if (*name == '.' && ch_isupper(name[1])) {
373 switch (name[1]) {
374 case 'A':
375 if (strcmp(name, ".ALLSRC") == 0)
376 name = ALLSRC;
377 if (strcmp(name, ".ARCHIVE") == 0)
378 name = ARCHIVE;
379 break;
380 case 'I':
381 if (strcmp(name, ".IMPSRC") == 0)
382 name = IMPSRC;
383 break;
384 case 'M':
385 if (strcmp(name, ".MEMBER") == 0)
386 name = MEMBER;
387 break;
388 case 'O':
389 if (strcmp(name, ".OODATE") == 0)
390 name = OODATE;
391 break;
392 case 'P':
393 if (strcmp(name, ".PREFIX") == 0)
394 name = PREFIX;
395 break;
396 case 'S':
397 if (strcmp(name, ".SHELL") == 0) {
398 if (shellPath == NULL)
399 Shell_Init();
400 }
401 break;
402 case 'T':
403 if (strcmp(name, ".TARGET") == 0)
404 name = TARGET;
405 break;
406 }
407 }
408
409 /* GNU make has an additional alias $^ == ${.ALLSRC}. */
410
411 return name;
412 }
413
414 static Var *
415 GNode_FindVar(GNode *scope, const char *varname, unsigned int hash)
416 {
417 return HashTable_FindValueHash(&scope->vars, varname, hash);
418 }
419
420 /*
421 * Find the variable in the scope, and maybe in other scopes as well.
422 *
423 * Input:
424 * name name to find, is not expanded any further
425 * scope scope in which to look first
426 * elsewhere TRUE to look in other scopes as well
427 *
428 * Results:
429 * The found variable, or NULL if the variable does not exist.
430 * If the variable is an environment variable, it must be freed using
431 * VarFreeEnv after use.
432 */
433 static Var *
434 VarFind(const char *name, GNode *scope, Boolean elsewhere)
435 {
436 Var *var;
437 unsigned int nameHash;
438
439 /* Replace '.TARGET' with '@', likewise for other local variables. */
440 name = CanonicalVarname(name);
441 nameHash = Hash_Hash(name);
442
443 var = GNode_FindVar(scope, name, nameHash);
444 if (!elsewhere)
445 return var;
446
447 if (var == NULL && scope != SCOPE_CMDLINE)
448 var = GNode_FindVar(SCOPE_CMDLINE, name, nameHash);
449
450 if (!opts.checkEnvFirst && var == NULL && scope != SCOPE_GLOBAL) {
451 var = GNode_FindVar(SCOPE_GLOBAL, name, nameHash);
452 if (var == NULL && scope != SCOPE_INTERNAL) {
453 /* SCOPE_INTERNAL is subordinate to SCOPE_GLOBAL */
454 var = GNode_FindVar(SCOPE_INTERNAL, name, nameHash);
455 }
456 }
457
458 if (var == NULL) {
459 char *env;
460
461 if ((env = getenv(name)) != NULL) {
462 char *varname = bmake_strdup(name);
463 return VarNew(FStr_InitOwn(varname), env, VFL_FROM_ENV);
464 }
465
466 if (opts.checkEnvFirst && scope != SCOPE_GLOBAL) {
467 var = GNode_FindVar(SCOPE_GLOBAL, name, nameHash);
468 if (var == NULL && scope != SCOPE_INTERNAL)
469 var = GNode_FindVar(SCOPE_INTERNAL, name,
470 nameHash);
471 return var;
472 }
473
474 return NULL;
475 }
476
477 return var;
478 }
479
480 /* If the variable is an environment variable, free it, including its value. */
481 static void
482 VarFreeEnv(Var *v)
483 {
484 if (!(v->flags & VFL_FROM_ENV))
485 return;
486
487 FStr_Done(&v->name);
488 Buf_Done(&v->val);
489 free(v);
490 }
491
492 /* Add a new variable of the given name and value to the given scope. */
493 static Var *
494 VarAdd(const char *name, const char *value, GNode *scope, VarSetFlags flags)
495 {
496 HashEntry *he = HashTable_CreateEntry(&scope->vars, name, NULL);
497 Var *v = VarNew(FStr_InitRefer(/* aliased to */ he->key), value,
498 flags & VAR_SET_READONLY ? VFL_READONLY : VFL_NONE);
499 HashEntry_Set(he, v);
500 DEBUG3(VAR, "%s:%s = %s\n", scope->name, name, value);
501 return v;
502 }
503
504 /*
505 * Remove a variable from a scope, freeing all related memory as well.
506 * The variable name is kept as-is, it is not expanded.
507 */
508 void
509 Var_Delete(GNode *scope, const char *varname)
510 {
511 HashEntry *he = HashTable_FindEntry(&scope->vars, varname);
512 Var *v;
513
514 if (he == NULL) {
515 DEBUG2(VAR, "%s:delete %s (not found)\n", scope->name, varname);
516 return;
517 }
518
519 DEBUG2(VAR, "%s:delete %s\n", scope->name, varname);
520 v = he->value;
521 if (v->flags & VFL_EXPORTED)
522 unsetenv(v->name.str);
523 if (strcmp(v->name.str, MAKE_EXPORTED) == 0)
524 var_exportedVars = VAR_EXPORTED_NONE;
525 assert(v->name.freeIt == NULL);
526 HashTable_DeleteEntry(&scope->vars, he);
527 Buf_Done(&v->val);
528 free(v);
529 }
530
531 /*
532 * Remove a variable from a scope, freeing all related memory as well.
533 * The variable name is expanded once.
534 */
535 void
536 Var_DeleteExpand(GNode *scope, const char *name)
537 {
538 FStr varname = FStr_InitRefer(name);
539
540 if (strchr(varname.str, '$') != NULL) {
541 char *expanded;
542 (void)Var_Subst(varname.str, SCOPE_GLOBAL, VARE_WANTRES,
543 &expanded);
544 /* TODO: handle errors */
545 varname = FStr_InitOwn(expanded);
546 }
547
548 Var_Delete(scope, varname.str);
549 FStr_Done(&varname);
550 }
551
552 /*
553 * Undefine one or more variables from the global scope.
554 * The argument is expanded exactly once and then split into words.
555 */
556 void
557 Var_Undef(const char *arg)
558 {
559 VarParseResult vpr;
560 char *expanded;
561 Words varnames;
562 size_t i;
563
564 if (arg[0] == '\0') {
565 Parse_Error(PARSE_FATAL,
566 "The .undef directive requires an argument");
567 return;
568 }
569
570 vpr = Var_Subst(arg, SCOPE_GLOBAL, VARE_WANTRES, &expanded);
571 if (vpr != VPR_OK) {
572 Parse_Error(PARSE_FATAL,
573 "Error in variable names to be undefined");
574 return;
575 }
576
577 varnames = Str_Words(expanded, FALSE);
578 if (varnames.len == 1 && varnames.words[0][0] == '\0')
579 varnames.len = 0;
580
581 for (i = 0; i < varnames.len; i++) {
582 const char *varname = varnames.words[i];
583 Global_Delete(varname);
584 }
585
586 Words_Free(varnames);
587 free(expanded);
588 }
589
590 static Boolean
591 MayExport(const char *name)
592 {
593 if (name[0] == '.')
594 return FALSE; /* skip internals */
595 if (name[0] == '-')
596 return FALSE; /* skip misnamed variables */
597 if (name[1] == '\0') {
598 /*
599 * A single char.
600 * If it is one of the variables that should only appear in
601 * local scope, skip it, else we can get Var_Subst
602 * into a loop.
603 */
604 switch (name[0]) {
605 case '@':
606 case '%':
607 case '*':
608 case '!':
609 return FALSE;
610 }
611 }
612 return TRUE;
613 }
614
615 static Boolean
616 ExportVarEnv(Var *v)
617 {
618 const char *name = v->name.str;
619 char *val = v->val.data;
620 char *expr;
621
622 if ((v->flags & VFL_EXPORTED) && !(v->flags & VFL_REEXPORT))
623 return FALSE; /* nothing to do */
624
625 if (strchr(val, '$') == NULL) {
626 if (!(v->flags & VFL_EXPORTED))
627 setenv(name, val, 1);
628 return TRUE;
629 }
630
631 if (v->flags & VFL_IN_USE) {
632 /*
633 * We recursed while exporting in a child.
634 * This isn't going to end well, just skip it.
635 */
636 return FALSE;
637 }
638
639 /* XXX: name is injected without escaping it */
640 expr = str_concat3("${", name, "}");
641 (void)Var_Subst(expr, SCOPE_GLOBAL, VARE_WANTRES, &val);
642 /* TODO: handle errors */
643 setenv(name, val, 1);
644 free(val);
645 free(expr);
646 return TRUE;
647 }
648
649 static Boolean
650 ExportVarPlain(Var *v)
651 {
652 if (strchr(v->val.data, '$') == NULL) {
653 setenv(v->name.str, v->val.data, 1);
654 v->flags |= VFL_EXPORTED;
655 v->flags &= ~(unsigned)VFL_REEXPORT;
656 return TRUE;
657 }
658
659 /*
660 * Flag the variable as something we need to re-export.
661 * No point actually exporting it now though,
662 * the child process can do it at the last minute.
663 * Avoid calling setenv more often than necessary since it can leak.
664 */
665 v->flags |= VFL_EXPORTED | VFL_REEXPORT;
666 return TRUE;
667 }
668
669 static Boolean
670 ExportVarLiteral(Var *v)
671 {
672 if ((v->flags & VFL_EXPORTED) && !(v->flags & VFL_REEXPORT))
673 return FALSE;
674
675 if (!(v->flags & VFL_EXPORTED))
676 setenv(v->name.str, v->val.data, 1);
677
678 return TRUE;
679 }
680
681 /*
682 * Mark a single variable to be exported later for subprocesses.
683 *
684 * Internal variables (those starting with '.') are not exported.
685 */
686 static Boolean
687 ExportVar(const char *name, VarExportMode mode)
688 {
689 Var *v;
690
691 if (!MayExport(name))
692 return FALSE;
693
694 v = VarFind(name, SCOPE_GLOBAL, FALSE);
695 if (v == NULL)
696 return FALSE;
697
698 if (mode == VEM_ENV)
699 return ExportVarEnv(v);
700 else if (mode == VEM_PLAIN)
701 return ExportVarPlain(v);
702 else
703 return ExportVarLiteral(v);
704 }
705
706 /*
707 * Actually export the variables that have been marked as needing to be
708 * re-exported.
709 */
710 void
711 Var_ReexportVars(void)
712 {
713 char *xvarnames;
714
715 /*
716 * Several make implementations support this sort of mechanism for
717 * tracking recursion - but each uses a different name.
718 * We allow the makefiles to update MAKELEVEL and ensure
719 * children see a correctly incremented value.
720 */
721 char tmp[21];
722 snprintf(tmp, sizeof tmp, "%d", makelevel + 1);
723 setenv(MAKE_LEVEL_ENV, tmp, 1);
724
725 if (var_exportedVars == VAR_EXPORTED_NONE)
726 return;
727
728 if (var_exportedVars == VAR_EXPORTED_ALL) {
729 HashIter hi;
730
731 /* Ouch! Exporting all variables at once is crazy. */
732 HashIter_Init(&hi, &SCOPE_GLOBAL->vars);
733 while (HashIter_Next(&hi) != NULL) {
734 Var *var = hi.entry->value;
735 ExportVar(var->name.str, VEM_ENV);
736 }
737 return;
738 }
739
740 (void)Var_Subst("${" MAKE_EXPORTED ":O:u}", SCOPE_GLOBAL, VARE_WANTRES,
741 &xvarnames);
742 /* TODO: handle errors */
743 if (xvarnames[0] != '\0') {
744 Words varnames = Str_Words(xvarnames, FALSE);
745 size_t i;
746
747 for (i = 0; i < varnames.len; i++)
748 ExportVar(varnames.words[i], VEM_ENV);
749 Words_Free(varnames);
750 }
751 free(xvarnames);
752 }
753
754 static void
755 ExportVars(const char *varnames, Boolean isExport, VarExportMode mode)
756 /* TODO: try to combine the parameters 'isExport' and 'mode'. */
757 {
758 Words words = Str_Words(varnames, FALSE);
759 size_t i;
760
761 if (words.len == 1 && words.words[0][0] == '\0')
762 words.len = 0;
763
764 for (i = 0; i < words.len; i++) {
765 const char *varname = words.words[i];
766 if (!ExportVar(varname, mode))
767 continue;
768
769 if (var_exportedVars == VAR_EXPORTED_NONE)
770 var_exportedVars = VAR_EXPORTED_SOME;
771
772 if (isExport && mode == VEM_PLAIN)
773 Global_Append(MAKE_EXPORTED, varname);
774 }
775 Words_Free(words);
776 }
777
778 static void
779 ExportVarsExpand(const char *uvarnames, Boolean isExport, VarExportMode mode)
780 {
781 char *xvarnames;
782
783 (void)Var_Subst(uvarnames, SCOPE_GLOBAL, VARE_WANTRES, &xvarnames);
784 /* TODO: handle errors */
785 ExportVars(xvarnames, isExport, mode);
786 free(xvarnames);
787 }
788
789 /* Export the named variables, or all variables. */
790 void
791 Var_Export(VarExportMode mode, const char *varnames)
792 {
793 if (mode == VEM_PLAIN && varnames[0] == '\0') {
794 var_exportedVars = VAR_EXPORTED_ALL; /* use with caution! */
795 return;
796 }
797
798 ExportVarsExpand(varnames, TRUE, mode);
799 }
800
801 void
802 Var_ExportVars(const char *varnames)
803 {
804 ExportVarsExpand(varnames, FALSE, VEM_PLAIN);
805 }
806
807
808 extern char **environ;
809
810 static void
811 ClearEnv(void)
812 {
813 const char *cp;
814 char **newenv;
815
816 cp = getenv(MAKE_LEVEL_ENV); /* we should preserve this */
817 if (environ == savedEnv) {
818 /* we have been here before! */
819 newenv = bmake_realloc(environ, 2 * sizeof(char *));
820 } else {
821 if (savedEnv != NULL) {
822 free(savedEnv);
823 savedEnv = NULL;
824 }
825 newenv = bmake_malloc(2 * sizeof(char *));
826 }
827
828 /* Note: we cannot safely free() the original environ. */
829 environ = savedEnv = newenv;
830 newenv[0] = NULL;
831 newenv[1] = NULL;
832 if (cp != NULL && *cp != '\0')
833 setenv(MAKE_LEVEL_ENV, cp, 1);
834 }
835
836 static void
837 GetVarnamesToUnexport(Boolean isEnv, const char *arg,
838 FStr *out_varnames, UnexportWhat *out_what)
839 {
840 UnexportWhat what;
841 FStr varnames = FStr_InitRefer("");
842
843 if (isEnv) {
844 if (arg[0] != '\0') {
845 Parse_Error(PARSE_FATAL,
846 "The directive .unexport-env does not take "
847 "arguments");
848 /* continue anyway */
849 }
850 what = UNEXPORT_ENV;
851
852 } else {
853 what = arg[0] != '\0' ? UNEXPORT_NAMED : UNEXPORT_ALL;
854 if (what == UNEXPORT_NAMED)
855 varnames = FStr_InitRefer(arg);
856 }
857
858 if (what != UNEXPORT_NAMED) {
859 char *expanded;
860 /* Using .MAKE.EXPORTED */
861 (void)Var_Subst("${" MAKE_EXPORTED ":O:u}", SCOPE_GLOBAL,
862 VARE_WANTRES, &expanded);
863 /* TODO: handle errors */
864 varnames = FStr_InitOwn(expanded);
865 }
866
867 *out_varnames = varnames;
868 *out_what = what;
869 }
870
871 static void
872 UnexportVar(const char *varname, UnexportWhat what)
873 {
874 Var *v = VarFind(varname, SCOPE_GLOBAL, FALSE);
875 if (v == NULL) {
876 DEBUG1(VAR, "Not unexporting \"%s\" (not found)\n", varname);
877 return;
878 }
879
880 DEBUG1(VAR, "Unexporting \"%s\"\n", varname);
881 if (what != UNEXPORT_ENV &&
882 (v->flags & VFL_EXPORTED) && !(v->flags & VFL_REEXPORT))
883 unsetenv(v->name.str);
884 v->flags &= ~(unsigned)(VFL_EXPORTED | VFL_REEXPORT);
885
886 if (what == UNEXPORT_NAMED) {
887 /* Remove the variable names from .MAKE.EXPORTED. */
888 /* XXX: v->name is injected without escaping it */
889 char *expr = str_concat3("${" MAKE_EXPORTED ":N",
890 v->name.str, "}");
891 char *cp;
892 (void)Var_Subst(expr, SCOPE_GLOBAL, VARE_WANTRES, &cp);
893 /* TODO: handle errors */
894 Global_Set(MAKE_EXPORTED, cp);
895 free(cp);
896 free(expr);
897 }
898 }
899
900 static void
901 UnexportVars(FStr *varnames, UnexportWhat what)
902 {
903 size_t i;
904 Words words;
905
906 if (what == UNEXPORT_ENV)
907 ClearEnv();
908
909 words = Str_Words(varnames->str, FALSE);
910 for (i = 0; i < words.len; i++) {
911 const char *varname = words.words[i];
912 UnexportVar(varname, what);
913 }
914 Words_Free(words);
915
916 if (what != UNEXPORT_NAMED)
917 Global_Delete(MAKE_EXPORTED);
918 }
919
920 /*
921 * This is called when .unexport[-env] is seen.
922 *
923 * str must have the form "unexport[-env] varname...".
924 */
925 void
926 Var_UnExport(Boolean isEnv, const char *arg)
927 {
928 UnexportWhat what;
929 FStr varnames;
930
931 GetVarnamesToUnexport(isEnv, arg, &varnames, &what);
932 UnexportVars(&varnames, what);
933 FStr_Done(&varnames);
934 }
935
936 /*
937 * When there is a variable of the same name in the command line scope, the
938 * global variable would not be visible anywhere. Therefore there is no
939 * point in setting it at all.
940 *
941 * See 'scope == SCOPE_CMDLINE' in Var_SetWithFlags.
942 */
943 static Boolean
944 ExistsInCmdline(const char *name, const char *val)
945 {
946 Var *v;
947
948 v = VarFind(name, SCOPE_CMDLINE, FALSE);
949 if (v == NULL)
950 return FALSE;
951
952 if (v->flags & VFL_FROM_CMD) {
953 DEBUG3(VAR, "%s:%s = %s ignored!\n",
954 SCOPE_GLOBAL->name, name, val);
955 return TRUE;
956 }
957
958 VarFreeEnv(v);
959 return FALSE;
960 }
961
962 /* Set the variable to the value; the name is not expanded. */
963 void
964 Var_SetWithFlags(GNode *scope, const char *name, const char *val,
965 VarSetFlags flags)
966 {
967 Var *v;
968
969 assert(val != NULL);
970 if (name[0] == '\0') {
971 DEBUG0(VAR, "SetVar: variable name is empty - ignored\n");
972 return;
973 }
974
975 if (scope == SCOPE_GLOBAL && ExistsInCmdline(name, val))
976 return;
977
978 /*
979 * Only look for a variable in the given scope since anything set
980 * here will override anything in a lower scope, so there's not much
981 * point in searching them all.
982 */
983 v = VarFind(name, scope, FALSE);
984 if (v == NULL) {
985 if (scope == SCOPE_CMDLINE && !(flags & VAR_SET_NO_EXPORT)) {
986 /*
987 * This var would normally prevent the same name being
988 * added to SCOPE_GLOBAL, so delete it from there if
989 * needed. Otherwise -V name may show the wrong value.
990 *
991 * See ExistsInCmdline.
992 */
993 Var_Delete(SCOPE_GLOBAL, name);
994 }
995 v = VarAdd(name, val, scope, flags);
996 } else {
997 if ((v->flags & VFL_READONLY) && !(flags & VAR_SET_READONLY)) {
998 DEBUG3(VAR, "%s:%s = %s ignored (read-only)\n",
999 scope->name, name, val);
1000 return;
1001 }
1002 Buf_Empty(&v->val);
1003 Buf_AddStr(&v->val, val);
1004
1005 DEBUG3(VAR, "%s:%s = %s\n", scope->name, name, val);
1006 if (v->flags & VFL_EXPORTED)
1007 ExportVar(name, VEM_PLAIN);
1008 }
1009
1010 /*
1011 * Any variables given on the command line are automatically exported
1012 * to the environment (as per POSIX standard), except for internals.
1013 */
1014 if (scope == SCOPE_CMDLINE && !(flags & VAR_SET_NO_EXPORT) &&
1015 name[0] != '.') {
1016 v->flags |= VFL_FROM_CMD;
1017
1018 /*
1019 * If requested, don't export these in the environment
1020 * individually. We still put them in MAKEOVERRIDES so
1021 * that the command-line settings continue to override
1022 * Makefile settings.
1023 */
1024 if (!opts.varNoExportEnv)
1025 setenv(name, val, 1);
1026 /* XXX: What about .MAKE.EXPORTED? */
1027 /* XXX: Why not just mark the variable for needing export,
1028 * as in ExportVarPlain? */
1029
1030 Global_Append(MAKEOVERRIDES, name);
1031 }
1032
1033 if (name[0] == '.' && strcmp(name, MAKE_SAVE_DOLLARS) == 0)
1034 save_dollars = ParseBoolean(val, save_dollars);
1035
1036 if (v != NULL)
1037 VarFreeEnv(v);
1038 }
1039
1040 /* See Var_Set for documentation. */
1041 void
1042 Var_SetExpandWithFlags(GNode *scope, const char *name, const char *val,
1043 VarSetFlags flags)
1044 {
1045 const char *unexpanded_name = name;
1046 FStr varname = FStr_InitRefer(name);
1047
1048 assert(val != NULL);
1049
1050 if (strchr(varname.str, '$') != NULL) {
1051 char *expanded;
1052 (void)Var_Subst(varname.str, scope, VARE_WANTRES, &expanded);
1053 /* TODO: handle errors */
1054 varname = FStr_InitOwn(expanded);
1055 }
1056
1057 if (varname.str[0] == '\0') {
1058 DEBUG2(VAR, "Var_Set(\"%s\", \"%s\", ...) "
1059 "name expands to empty string - ignored\n",
1060 unexpanded_name, val);
1061 } else
1062 Var_SetWithFlags(scope, varname.str, val, flags);
1063
1064 FStr_Done(&varname);
1065 }
1066
1067 void
1068 Var_Set(GNode *scope, const char *name, const char *val)
1069 {
1070 Var_SetWithFlags(scope, name, val, VAR_SET_NONE);
1071 }
1072
1073 /*
1074 * Set the variable name to the value val in the given scope.
1075 *
1076 * If the variable doesn't yet exist, it is created.
1077 * Otherwise the new value overwrites and replaces the old value.
1078 *
1079 * Input:
1080 * name name of the variable to set, is expanded once
1081 * val value to give to the variable
1082 * scope scope in which to set it
1083 */
1084 void
1085 Var_SetExpand(GNode *scope, const char *name, const char *val)
1086 {
1087 Var_SetExpandWithFlags(scope, name, val, VAR_SET_NONE);
1088 }
1089
1090 void
1091 Global_Set(const char *name, const char *value)
1092 {
1093 Var_Set(SCOPE_GLOBAL, name, value);
1094 }
1095
1096 void
1097 Global_SetExpand(const char *name, const char *value)
1098 {
1099 Var_SetExpand(SCOPE_GLOBAL, name, value);
1100 }
1101
1102 void
1103 Global_Delete(const char *name)
1104 {
1105 Var_Delete(SCOPE_GLOBAL, name);
1106 }
1107
1108 /*
1109 * Append the value to the named variable.
1110 *
1111 * If the variable doesn't exist, it is created. Otherwise a single space
1112 * and the given value are appended.
1113 */
1114 void
1115 Var_Append(GNode *scope, const char *name, const char *val)
1116 {
1117 Var *v;
1118
1119 v = VarFind(name, scope, scope == SCOPE_GLOBAL);
1120
1121 if (v == NULL) {
1122 Var_SetWithFlags(scope, name, val, VAR_SET_NONE);
1123 } else if (v->flags & VFL_READONLY) {
1124 DEBUG1(VAR, "Ignoring append to %s since it is read-only\n",
1125 name);
1126 } else if (scope == SCOPE_CMDLINE || !(v->flags & VFL_FROM_CMD)) {
1127 Buf_AddByte(&v->val, ' ');
1128 Buf_AddStr(&v->val, val);
1129
1130 DEBUG3(VAR, "%s:%s = %s\n", scope->name, name, v->val.data);
1131
1132 if (v->flags & VFL_FROM_ENV) {
1133 /*
1134 * If the original variable came from the environment,
1135 * we have to install it in the global scope (we
1136 * could place it in the environment, but then we
1137 * should provide a way to export other variables...)
1138 */
1139 v->flags &= ~(unsigned)VFL_FROM_ENV;
1140 /*
1141 * This is the only place where a variable is
1142 * created whose v->name is not the same as
1143 * scope->vars->key.
1144 */
1145 HashTable_Set(&scope->vars, name, v);
1146 }
1147 }
1148 }
1149
1150 /*
1151 * The variable of the given name has the given value appended to it in the
1152 * given scope.
1153 *
1154 * If the variable doesn't exist, it is created. Otherwise the strings are
1155 * concatenated, with a space in between.
1156 *
1157 * Input:
1158 * name name of the variable to modify, is expanded once
1159 * val string to append to it
1160 * scope scope in which this should occur
1161 *
1162 * Notes:
1163 * Only if the variable is being sought in the global scope is the
1164 * environment searched.
1165 * XXX: Knows its calling circumstances in that if called with scope
1166 * an actual target, it will only search that scope since only
1167 * a local variable could be being appended to. This is actually
1168 * a big win and must be tolerated.
1169 */
1170 void
1171 Var_AppendExpand(GNode *scope, const char *name, const char *val)
1172 {
1173 FStr xname = FStr_InitRefer(name);
1174
1175 assert(val != NULL);
1176
1177 if (strchr(name, '$') != NULL) {
1178 char *expanded;
1179 (void)Var_Subst(name, scope, VARE_WANTRES, &expanded);
1180 /* TODO: handle errors */
1181 xname = FStr_InitOwn(expanded);
1182 if (expanded[0] == '\0') {
1183 /* TODO: update function name in the debug message */
1184 DEBUG2(VAR, "Var_Append(\"%s\", \"%s\", ...) "
1185 "name expands to empty string - ignored\n",
1186 name, val);
1187 FStr_Done(&xname);
1188 return;
1189 }
1190 }
1191
1192 Var_Append(scope, xname.str, val);
1193
1194 FStr_Done(&xname);
1195 }
1196
1197 void
1198 Global_Append(const char *name, const char *value)
1199 {
1200 Var_Append(SCOPE_GLOBAL, name, value);
1201 }
1202
1203 Boolean
1204 Var_Exists(GNode *scope, const char *name)
1205 {
1206 Var *v = VarFind(name, scope, TRUE);
1207 if (v == NULL)
1208 return FALSE;
1209
1210 VarFreeEnv(v);
1211 return TRUE;
1212 }
1213
1214 /*
1215 * See if the given variable exists, in the given scope or in other
1216 * fallback scopes.
1217 *
1218 * Input:
1219 * name Variable to find, is expanded once
1220 * scope Scope in which to start search
1221 */
1222 Boolean
1223 Var_ExistsExpand(GNode *scope, const char *name)
1224 {
1225 FStr varname = FStr_InitRefer(name);
1226 Boolean exists;
1227
1228 if (strchr(varname.str, '$') != NULL) {
1229 char *expanded;
1230 (void)Var_Subst(varname.str, scope, VARE_WANTRES, &expanded);
1231 /* TODO: handle errors */
1232 varname = FStr_InitOwn(expanded);
1233 }
1234
1235 exists = Var_Exists(scope, varname.str);
1236 FStr_Done(&varname);
1237 return exists;
1238 }
1239
1240 /*
1241 * Return the unexpanded value of the given variable in the given scope,
1242 * or the usual scopes.
1243 *
1244 * Input:
1245 * name name to find, is not expanded any further
1246 * scope scope in which to search for it
1247 *
1248 * Results:
1249 * The value if the variable exists, NULL if it doesn't.
1250 * The value is valid until the next modification to any variable.
1251 */
1252 FStr
1253 Var_Value(GNode *scope, const char *name)
1254 {
1255 Var *v = VarFind(name, scope, TRUE);
1256 char *value;
1257
1258 if (v == NULL)
1259 return FStr_InitRefer(NULL);
1260
1261 if (!(v->flags & VFL_FROM_ENV))
1262 return FStr_InitRefer(v->val.data);
1263
1264 /* Since environment variables are short-lived, free it now. */
1265 FStr_Done(&v->name);
1266 value = Buf_DoneData(&v->val);
1267 free(v);
1268 return FStr_InitOwn(value);
1269 }
1270
1271 /*
1272 * Return the unexpanded variable value from this node, without trying to look
1273 * up the variable in any other scope.
1274 */
1275 const char *
1276 GNode_ValueDirect(GNode *gn, const char *name)
1277 {
1278 Var *v = VarFind(name, gn, FALSE);
1279 return v != NULL ? v->val.data : NULL;
1280 }
1281
1282
1283 static void
1284 SepBuf_Init(SepBuf *buf, char sep)
1285 {
1286 Buf_InitSize(&buf->buf, 32);
1287 buf->needSep = FALSE;
1288 buf->sep = sep;
1289 }
1290
1291 static void
1292 SepBuf_Sep(SepBuf *buf)
1293 {
1294 buf->needSep = TRUE;
1295 }
1296
1297 static void
1298 SepBuf_AddBytes(SepBuf *buf, const char *mem, size_t mem_size)
1299 {
1300 if (mem_size == 0)
1301 return;
1302 if (buf->needSep && buf->sep != '\0') {
1303 Buf_AddByte(&buf->buf, buf->sep);
1304 buf->needSep = FALSE;
1305 }
1306 Buf_AddBytes(&buf->buf, mem, mem_size);
1307 }
1308
1309 static void
1310 SepBuf_AddBytesBetween(SepBuf *buf, const char *start, const char *end)
1311 {
1312 SepBuf_AddBytes(buf, start, (size_t)(end - start));
1313 }
1314
1315 static void
1316 SepBuf_AddStr(SepBuf *buf, const char *str)
1317 {
1318 SepBuf_AddBytes(buf, str, strlen(str));
1319 }
1320
1321 static char *
1322 SepBuf_DoneData(SepBuf *buf)
1323 {
1324 return Buf_DoneData(&buf->buf);
1325 }
1326
1327
1328 /*
1329 * This callback for ModifyWords gets a single word from a variable expression
1330 * and typically adds a modification of this word to the buffer. It may also
1331 * do nothing or add several words.
1332 *
1333 * For example, when evaluating the modifier ':M*b' in ${:Ua b c:M*b}, the
1334 * callback is called 3 times, once for "a", "b" and "c".
1335 */
1336 typedef void (*ModifyWordProc)(const char *word, SepBuf *buf, void *data);
1337
1338
1339 /*
1340 * Callback for ModifyWords to implement the :H modifier.
1341 * Add the dirname of the given word to the buffer.
1342 */
1343 /*ARGSUSED*/
1344 static void
1345 ModifyWord_Head(const char *word, SepBuf *buf, void *dummy MAKE_ATTR_UNUSED)
1346 {
1347 const char *slash = strrchr(word, '/');
1348 if (slash != NULL)
1349 SepBuf_AddBytesBetween(buf, word, slash);
1350 else
1351 SepBuf_AddStr(buf, ".");
1352 }
1353
1354 /*
1355 * Callback for ModifyWords to implement the :T modifier.
1356 * Add the basename of the given word to the buffer.
1357 */
1358 /*ARGSUSED*/
1359 static void
1360 ModifyWord_Tail(const char *word, SepBuf *buf, void *dummy MAKE_ATTR_UNUSED)
1361 {
1362 SepBuf_AddStr(buf, str_basename(word));
1363 }
1364
1365 /*
1366 * Callback for ModifyWords to implement the :E modifier.
1367 * Add the filename suffix of the given word to the buffer, if it exists.
1368 */
1369 /*ARGSUSED*/
1370 static void
1371 ModifyWord_Suffix(const char *word, SepBuf *buf, void *dummy MAKE_ATTR_UNUSED)
1372 {
1373 const char *lastDot = strrchr(word, '.');
1374 if (lastDot != NULL)
1375 SepBuf_AddStr(buf, lastDot + 1);
1376 }
1377
1378 /*
1379 * Callback for ModifyWords to implement the :R modifier.
1380 * Add the filename without extension of the given word to the buffer.
1381 */
1382 /*ARGSUSED*/
1383 static void
1384 ModifyWord_Root(const char *word, SepBuf *buf, void *dummy MAKE_ATTR_UNUSED)
1385 {
1386 const char *lastDot = strrchr(word, '.');
1387 size_t len = lastDot != NULL ? (size_t)(lastDot - word) : strlen(word);
1388 SepBuf_AddBytes(buf, word, len);
1389 }
1390
1391 /*
1392 * Callback for ModifyWords to implement the :M modifier.
1393 * Place the word in the buffer if it matches the given pattern.
1394 */
1395 static void
1396 ModifyWord_Match(const char *word, SepBuf *buf, void *data)
1397 {
1398 const char *pattern = data;
1399 DEBUG2(VAR, "VarMatch [%s] [%s]\n", word, pattern);
1400 if (Str_Match(word, pattern))
1401 SepBuf_AddStr(buf, word);
1402 }
1403
1404 /*
1405 * Callback for ModifyWords to implement the :N modifier.
1406 * Place the word in the buffer if it doesn't match the given pattern.
1407 */
1408 static void
1409 ModifyWord_NoMatch(const char *word, SepBuf *buf, void *data)
1410 {
1411 const char *pattern = data;
1412 if (!Str_Match(word, pattern))
1413 SepBuf_AddStr(buf, word);
1414 }
1415
1416 #ifdef SYSVVARSUB
1417
1418 /*
1419 * Check word against pattern for a match (% is a wildcard).
1420 *
1421 * Input:
1422 * word Word to examine
1423 * pattern Pattern to examine against
1424 *
1425 * Results:
1426 * Returns the start of the match, or NULL.
1427 * out_match_len returns the length of the match, if any.
1428 * out_hasPercent returns whether the pattern contains a percent.
1429 */
1430 static const char *
1431 SysVMatch(const char *word, const char *pattern,
1432 size_t *out_match_len, Boolean *out_hasPercent)
1433 {
1434 const char *p = pattern;
1435 const char *w = word;
1436 const char *percent;
1437 size_t w_len;
1438 size_t p_len;
1439 const char *w_tail;
1440
1441 *out_hasPercent = FALSE;
1442 percent = strchr(p, '%');
1443 if (percent != NULL) { /* ${VAR:...%...=...} */
1444 *out_hasPercent = TRUE;
1445 if (w[0] == '\0')
1446 return NULL; /* empty word does not match pattern */
1447
1448 /* check that the prefix matches */
1449 for (; p != percent && *w != '\0' && *w == *p; w++, p++)
1450 continue;
1451 if (p != percent)
1452 return NULL; /* No match */
1453
1454 p++; /* Skip the percent */
1455 if (*p == '\0') {
1456 /* No more pattern, return the rest of the string */
1457 *out_match_len = strlen(w);
1458 return w;
1459 }
1460 }
1461
1462 /* Test whether the tail matches */
1463 w_len = strlen(w);
1464 p_len = strlen(p);
1465 if (w_len < p_len)
1466 return NULL;
1467
1468 w_tail = w + w_len - p_len;
1469 if (memcmp(p, w_tail, p_len) != 0)
1470 return NULL;
1471
1472 *out_match_len = (size_t)(w_tail - w);
1473 return w;
1474 }
1475
1476 struct ModifyWord_SYSVSubstArgs {
1477 GNode *scope;
1478 const char *lhs;
1479 const char *rhs;
1480 };
1481
1482 /* Callback for ModifyWords to implement the :%.from=%.to modifier. */
1483 static void
1484 ModifyWord_SYSVSubst(const char *word, SepBuf *buf, void *data)
1485 {
1486 const struct ModifyWord_SYSVSubstArgs *args = data;
1487 char *rhs_expanded;
1488 const char *rhs;
1489 const char *percent;
1490
1491 size_t match_len;
1492 Boolean lhsPercent;
1493 const char *match = SysVMatch(word, args->lhs, &match_len, &lhsPercent);
1494 if (match == NULL) {
1495 SepBuf_AddStr(buf, word);
1496 return;
1497 }
1498
1499 /*
1500 * Append rhs to the buffer, substituting the first '%' with the
1501 * match, but only if the lhs had a '%' as well.
1502 */
1503
1504 (void)Var_Subst(args->rhs, args->scope, VARE_WANTRES, &rhs_expanded);
1505 /* TODO: handle errors */
1506
1507 rhs = rhs_expanded;
1508 percent = strchr(rhs, '%');
1509
1510 if (percent != NULL && lhsPercent) {
1511 /* Copy the prefix of the replacement pattern */
1512 SepBuf_AddBytesBetween(buf, rhs, percent);
1513 rhs = percent + 1;
1514 }
1515 if (percent != NULL || !lhsPercent)
1516 SepBuf_AddBytes(buf, match, match_len);
1517
1518 /* Append the suffix of the replacement pattern */
1519 SepBuf_AddStr(buf, rhs);
1520
1521 free(rhs_expanded);
1522 }
1523 #endif
1524
1525
1526 struct ModifyWord_SubstArgs {
1527 const char *lhs;
1528 size_t lhsLen;
1529 const char *rhs;
1530 size_t rhsLen;
1531 VarPatternFlags pflags;
1532 Boolean matched;
1533 };
1534
1535 /*
1536 * Callback for ModifyWords to implement the :S,from,to, modifier.
1537 * Perform a string substitution on the given word.
1538 */
1539 static void
1540 ModifyWord_Subst(const char *word, SepBuf *buf, void *data)
1541 {
1542 size_t wordLen = strlen(word);
1543 struct ModifyWord_SubstArgs *args = data;
1544 const char *match;
1545
1546 if (args->pflags.subOnce && args->matched)
1547 goto nosub;
1548
1549 if (args->pflags.anchorStart) {
1550 if (wordLen < args->lhsLen ||
1551 memcmp(word, args->lhs, args->lhsLen) != 0)
1552 goto nosub;
1553
1554 if (args->pflags.anchorEnd && wordLen != args->lhsLen)
1555 goto nosub;
1556
1557 /* :S,^prefix,replacement, or :S,^whole$,replacement, */
1558 SepBuf_AddBytes(buf, args->rhs, args->rhsLen);
1559 SepBuf_AddBytesBetween(buf,
1560 word + args->lhsLen, word + wordLen);
1561 args->matched = TRUE;
1562 return;
1563 }
1564
1565 if (args->pflags.anchorEnd) {
1566 const char *start;
1567
1568 if (wordLen < args->lhsLen)
1569 goto nosub;
1570
1571 start = word + (wordLen - args->lhsLen);
1572 if (memcmp(start, args->lhs, args->lhsLen) != 0)
1573 goto nosub;
1574
1575 /* :S,suffix$,replacement, */
1576 SepBuf_AddBytesBetween(buf, word, start);
1577 SepBuf_AddBytes(buf, args->rhs, args->rhsLen);
1578 args->matched = TRUE;
1579 return;
1580 }
1581
1582 if (args->lhs[0] == '\0')
1583 goto nosub;
1584
1585 /* unanchored case, may match more than once */
1586 while ((match = strstr(word, args->lhs)) != NULL) {
1587 SepBuf_AddBytesBetween(buf, word, match);
1588 SepBuf_AddBytes(buf, args->rhs, args->rhsLen);
1589 args->matched = TRUE;
1590 wordLen -= (size_t)(match - word) + args->lhsLen;
1591 word += (size_t)(match - word) + args->lhsLen;
1592 if (wordLen == 0 || !args->pflags.subGlobal)
1593 break;
1594 }
1595 nosub:
1596 SepBuf_AddBytes(buf, word, wordLen);
1597 }
1598
1599 #ifndef NO_REGEX
1600 /* Print the error caused by a regcomp or regexec call. */
1601 static void
1602 VarREError(int reerr, const regex_t *pat, const char *str)
1603 {
1604 size_t errlen = regerror(reerr, pat, NULL, 0);
1605 char *errbuf = bmake_malloc(errlen);
1606 regerror(reerr, pat, errbuf, errlen);
1607 Error("%s: %s", str, errbuf);
1608 free(errbuf);
1609 }
1610
1611 struct ModifyWord_SubstRegexArgs {
1612 regex_t re;
1613 size_t nsub;
1614 char *replace;
1615 VarPatternFlags pflags;
1616 Boolean matched;
1617 };
1618
1619 /*
1620 * Callback for ModifyWords to implement the :C/from/to/ modifier.
1621 * Perform a regex substitution on the given word.
1622 */
1623 static void
1624 ModifyWord_SubstRegex(const char *word, SepBuf *buf, void *data)
1625 {
1626 struct ModifyWord_SubstRegexArgs *args = data;
1627 int xrv;
1628 const char *wp = word;
1629 char *rp;
1630 int flags = 0;
1631 regmatch_t m[10];
1632
1633 if (args->pflags.subOnce && args->matched)
1634 goto nosub;
1635
1636 tryagain:
1637 xrv = regexec(&args->re, wp, args->nsub, m, flags);
1638
1639 switch (xrv) {
1640 case 0:
1641 args->matched = TRUE;
1642 SepBuf_AddBytes(buf, wp, (size_t)m[0].rm_so);
1643
1644 /*
1645 * Replacement of regular expressions is not specified by
1646 * POSIX, therefore implement it here.
1647 */
1648
1649 for (rp = args->replace; *rp != '\0'; rp++) {
1650 if (*rp == '\\' && (rp[1] == '&' || rp[1] == '\\')) {
1651 SepBuf_AddBytes(buf, rp + 1, 1);
1652 rp++;
1653 continue;
1654 }
1655
1656 if (*rp == '&') {
1657 SepBuf_AddBytesBetween(buf,
1658 wp + m[0].rm_so, wp + m[0].rm_eo);
1659 continue;
1660 }
1661
1662 if (*rp != '\\' || !ch_isdigit(rp[1])) {
1663 SepBuf_AddBytes(buf, rp, 1);
1664 continue;
1665 }
1666
1667 { /* \0 to \9 backreference */
1668 size_t n = (size_t)(rp[1] - '0');
1669 rp++;
1670
1671 if (n >= args->nsub) {
1672 Error("No subexpression \\%u",
1673 (unsigned)n);
1674 } else if (m[n].rm_so == -1) {
1675 Error(
1676 "No match for subexpression \\%u",
1677 (unsigned)n);
1678 } else {
1679 SepBuf_AddBytesBetween(buf,
1680 wp + m[n].rm_so, wp + m[n].rm_eo);
1681 }
1682 }
1683 }
1684
1685 wp += m[0].rm_eo;
1686 if (args->pflags.subGlobal) {
1687 flags |= REG_NOTBOL;
1688 if (m[0].rm_so == 0 && m[0].rm_eo == 0) {
1689 SepBuf_AddBytes(buf, wp, 1);
1690 wp++;
1691 }
1692 if (*wp != '\0')
1693 goto tryagain;
1694 }
1695 if (*wp != '\0')
1696 SepBuf_AddStr(buf, wp);
1697 break;
1698 default:
1699 VarREError(xrv, &args->re, "Unexpected regex error");
1700 /* FALLTHROUGH */
1701 case REG_NOMATCH:
1702 nosub:
1703 SepBuf_AddStr(buf, wp);
1704 break;
1705 }
1706 }
1707 #endif
1708
1709
1710 struct ModifyWord_LoopArgs {
1711 GNode *scope;
1712 char *tvar; /* name of temporary variable */
1713 char *str; /* string to expand */
1714 VarEvalFlags eflags;
1715 };
1716
1717 /* Callback for ModifyWords to implement the :@var (at) ...@ modifier of ODE make. */
1718 static void
1719 ModifyWord_Loop(const char *word, SepBuf *buf, void *data)
1720 {
1721 const struct ModifyWord_LoopArgs *args;
1722 char *s;
1723
1724 if (word[0] == '\0')
1725 return;
1726
1727 args = data;
1728 /* XXX: The variable name should not be expanded here. */
1729 Var_SetExpandWithFlags(args->scope, args->tvar, word,
1730 VAR_SET_NO_EXPORT);
1731 (void)Var_Subst(args->str, args->scope, args->eflags, &s);
1732 /* TODO: handle errors */
1733
1734 DEBUG4(VAR, "ModifyWord_Loop: "
1735 "in \"%s\", replace \"%s\" with \"%s\" to \"%s\"\n",
1736 word, args->tvar, args->str, s);
1737
1738 if (s[0] == '\n' || Buf_EndsWith(&buf->buf, '\n'))
1739 buf->needSep = FALSE;
1740 SepBuf_AddStr(buf, s);
1741 free(s);
1742 }
1743
1744
1745 /*
1746 * The :[first..last] modifier selects words from the expression.
1747 * It can also reverse the words.
1748 */
1749 static char *
1750 VarSelectWords(const char *str, int first, int last,
1751 char sep, Boolean oneBigWord)
1752 {
1753 Words words;
1754 int len, start, end, step;
1755 int i;
1756
1757 SepBuf buf;
1758 SepBuf_Init(&buf, sep);
1759
1760 if (oneBigWord) {
1761 /* fake what Str_Words() would do if there were only one word */
1762 words.len = 1;
1763 words.words = bmake_malloc(
1764 (words.len + 1) * sizeof(words.words[0]));
1765 words.freeIt = bmake_strdup(str);
1766 words.words[0] = words.freeIt;
1767 words.words[1] = NULL;
1768 } else {
1769 words = Str_Words(str, FALSE);
1770 }
1771
1772 /*
1773 * Now sanitize the given range. If first or last are negative,
1774 * convert them to the positive equivalents (-1 gets converted to len,
1775 * -2 gets converted to (len - 1), etc.).
1776 */
1777 len = (int)words.len;
1778 if (first < 0)
1779 first += len + 1;
1780 if (last < 0)
1781 last += len + 1;
1782
1783 /* We avoid scanning more of the list than we need to. */
1784 if (first > last) {
1785 start = (first > len ? len : first) - 1;
1786 end = last < 1 ? 0 : last - 1;
1787 step = -1;
1788 } else {
1789 start = first < 1 ? 0 : first - 1;
1790 end = last > len ? len : last;
1791 step = 1;
1792 }
1793
1794 for (i = start; (step < 0) == (i >= end); i += step) {
1795 SepBuf_AddStr(&buf, words.words[i]);
1796 SepBuf_Sep(&buf);
1797 }
1798
1799 Words_Free(words);
1800
1801 return SepBuf_DoneData(&buf);
1802 }
1803
1804
1805 /*
1806 * Callback for ModifyWords to implement the :tA modifier.
1807 * Replace each word with the result of realpath() if successful.
1808 */
1809 /*ARGSUSED*/
1810 static void
1811 ModifyWord_Realpath(const char *word, SepBuf *buf, void *data MAKE_ATTR_UNUSED)
1812 {
1813 struct stat st;
1814 char rbuf[MAXPATHLEN];
1815
1816 const char *rp = cached_realpath(word, rbuf);
1817 if (rp != NULL && *rp == '/' && stat(rp, &st) == 0)
1818 word = rp;
1819
1820 SepBuf_AddStr(buf, word);
1821 }
1822
1823
1824 static char *
1825 Words_JoinFree(Words words)
1826 {
1827 Buffer buf;
1828 size_t i;
1829
1830 Buf_Init(&buf);
1831
1832 for (i = 0; i < words.len; i++) {
1833 if (i != 0) {
1834 /* XXX: Use st->sep instead of ' ', for consistency. */
1835 Buf_AddByte(&buf, ' ');
1836 }
1837 Buf_AddStr(&buf, words.words[i]);
1838 }
1839
1840 Words_Free(words);
1841
1842 return Buf_DoneData(&buf);
1843 }
1844
1845 /* Remove adjacent duplicate words. */
1846 static char *
1847 VarUniq(const char *str)
1848 {
1849 Words words = Str_Words(str, FALSE);
1850
1851 if (words.len > 1) {
1852 size_t i, j;
1853 for (j = 0, i = 1; i < words.len; i++)
1854 if (strcmp(words.words[i], words.words[j]) != 0 &&
1855 (++j != i))
1856 words.words[j] = words.words[i];
1857 words.len = j + 1;
1858 }
1859
1860 return Words_JoinFree(words);
1861 }
1862
1863
1864 /*
1865 * Quote shell meta-characters and space characters in the string.
1866 * If quoteDollar is set, also quote and double any '$' characters.
1867 */
1868 static char *
1869 VarQuote(const char *str, Boolean quoteDollar)
1870 {
1871 Buffer buf;
1872 Buf_Init(&buf);
1873
1874 for (; *str != '\0'; str++) {
1875 if (*str == '\n') {
1876 const char *newline = Shell_GetNewline();
1877 if (newline == NULL)
1878 newline = "\\\n";
1879 Buf_AddStr(&buf, newline);
1880 continue;
1881 }
1882 if (ch_isspace(*str) || is_shell_metachar((unsigned char)*str))
1883 Buf_AddByte(&buf, '\\');
1884 Buf_AddByte(&buf, *str);
1885 if (quoteDollar && *str == '$')
1886 Buf_AddStr(&buf, "\\$");
1887 }
1888
1889 return Buf_DoneData(&buf);
1890 }
1891
1892 /*
1893 * Compute the 32-bit hash of the given string, using the MurmurHash3
1894 * algorithm. Output is encoded as 8 hex digits, in Little Endian order.
1895 */
1896 static char *
1897 VarHash(const char *str)
1898 {
1899 static const char hexdigits[16] = "0123456789abcdef";
1900 const unsigned char *ustr = (const unsigned char *)str;
1901
1902 uint32_t h = 0x971e137bU;
1903 uint32_t c1 = 0x95543787U;
1904 uint32_t c2 = 0x2ad7eb25U;
1905 size_t len2 = strlen(str);
1906
1907 char *buf;
1908 size_t i;
1909
1910 size_t len;
1911 for (len = len2; len != 0;) {
1912 uint32_t k = 0;
1913 switch (len) {
1914 default:
1915 k = ((uint32_t)ustr[3] << 24) |
1916 ((uint32_t)ustr[2] << 16) |
1917 ((uint32_t)ustr[1] << 8) |
1918 (uint32_t)ustr[0];
1919 len -= 4;
1920 ustr += 4;
1921 break;
1922 case 3:
1923 k |= (uint32_t)ustr[2] << 16;
1924 /* FALLTHROUGH */
1925 case 2:
1926 k |= (uint32_t)ustr[1] << 8;
1927 /* FALLTHROUGH */
1928 case 1:
1929 k |= (uint32_t)ustr[0];
1930 len = 0;
1931 }
1932 c1 = c1 * 5 + 0x7b7d159cU;
1933 c2 = c2 * 5 + 0x6bce6396U;
1934 k *= c1;
1935 k = (k << 11) ^ (k >> 21);
1936 k *= c2;
1937 h = (h << 13) ^ (h >> 19);
1938 h = h * 5 + 0x52dce729U;
1939 h ^= k;
1940 }
1941 h ^= (uint32_t)len2;
1942 h *= 0x85ebca6b;
1943 h ^= h >> 13;
1944 h *= 0xc2b2ae35;
1945 h ^= h >> 16;
1946
1947 buf = bmake_malloc(9);
1948 for (i = 0; i < 8; i++) {
1949 buf[i] = hexdigits[h & 0x0f];
1950 h >>= 4;
1951 }
1952 buf[8] = '\0';
1953 return buf;
1954 }
1955
1956 static char *
1957 VarStrftime(const char *fmt, Boolean zulu, time_t tim)
1958 {
1959 char buf[BUFSIZ];
1960
1961 if (tim == 0)
1962 time(&tim);
1963 if (*fmt == '\0')
1964 fmt = "%c";
1965 strftime(buf, sizeof buf, fmt, zulu ? gmtime(&tim) : localtime(&tim));
1966
1967 buf[sizeof buf - 1] = '\0';
1968 return bmake_strdup(buf);
1969 }
1970
1971 /*
1972 * The ApplyModifier functions take an expression that is being evaluated.
1973 * Their task is to apply a single modifier to the expression. This involves
1974 * parsing the modifier, evaluating it and finally updating the value of the
1975 * expression.
1976 *
1977 * Parsing the modifier
1978 *
1979 * If parsing succeeds, the parsing position *pp is updated to point to the
1980 * first character following the modifier, which typically is either ':' or
1981 * st->endc. The modifier doesn't have to check for this delimiter character,
1982 * this is done by ApplyModifiers.
1983 *
1984 * XXX: As of 2020-11-15, some modifiers such as :S, :C, :P, :L do not
1985 * need to be followed by a ':' or endc; this was an unintended mistake.
1986 *
1987 * If parsing fails because of a missing delimiter (as in the :S, :C or :@
1988 * modifiers), return AMR_CLEANUP.
1989 *
1990 * If parsing fails because the modifier is unknown, return AMR_UNKNOWN to
1991 * try the SysV modifier ${VAR:from=to} as fallback. This should only be
1992 * done as long as there have been no side effects from evaluating nested
1993 * variables, to avoid evaluating them more than once. In this case, the
1994 * parsing position may or may not be updated. (XXX: Why not? The original
1995 * parsing position is well-known in ApplyModifiers.)
1996 *
1997 * If parsing fails and the SysV modifier ${VAR:from=to} should not be used
1998 * as a fallback, either issue an error message using Error or Parse_Error
1999 * and then return AMR_CLEANUP, or return AMR_BAD for the default error
2000 * message. Both of these return values will stop processing the variable
2001 * expression. (XXX: As of 2020-08-23, evaluation of the whole string
2002 * continues nevertheless after skipping a few bytes, which essentially is
2003 * undefined behavior. Not in the sense of C, but still the resulting string
2004 * is garbage.)
2005 *
2006 * Evaluating the modifier
2007 *
2008 * After parsing, the modifier is evaluated. The side effects from evaluating
2009 * nested variable expressions in the modifier text often already happen
2010 * during parsing though. For most modifiers this doesn't matter since their
2011 * only noticeable effect is that the update the value of the expression.
2012 * Some modifiers such as ':sh' or '::=' have noticeable side effects though.
2013 *
2014 * Evaluating the modifier usually takes the current value of the variable
2015 * expression from st->expr->value, or the variable name from st->var->name
2016 * and stores the result back in expr->value via Expr_SetValueOwn or
2017 * Expr_SetValueRefer.
2018 *
2019 * If evaluating fails (as of 2020-08-23), an error message is printed using
2020 * Error. This function has no side-effects, it really just prints the error
2021 * message. Processing the expression continues as if everything were ok.
2022 * XXX: This should be fixed by adding proper error handling to Var_Subst,
2023 * Var_Parse, ApplyModifiers and ModifyWords.
2024 *
2025 * Housekeeping
2026 *
2027 * Some modifiers such as :D and :U turn undefined expressions into defined
2028 * expressions (see Expr_Define).
2029 *
2030 * Some modifiers need to free some memory.
2031 */
2032
2033 typedef enum ExprDefined {
2034 /* The variable expression is based on a regular, defined variable. */
2035 DEF_REGULAR,
2036 /* The variable expression is based on an undefined variable. */
2037 DEF_UNDEF,
2038 /*
2039 * The variable expression started as an undefined expression, but one
2040 * of the modifiers (such as ':D' or ':U') has turned the expression
2041 * from undefined to defined.
2042 */
2043 DEF_DEFINED
2044 } ExprDefined;
2045
2046 static const char *const ExprDefined_Name[] = {
2047 "regular",
2048 "undefined",
2049 "defined"
2050 };
2051
2052 /* A variable expression such as $@ or ${VAR:Mpattern:Q}. */
2053 typedef struct Expr {
2054 Var *var;
2055 FStr value;
2056 VarEvalFlags const eflags;
2057 GNode *const scope;
2058 ExprDefined defined;
2059 } Expr;
2060
2061 /*
2062 * The status of applying a chain of modifiers to an expression.
2063 *
2064 * The modifiers of an expression are broken into chains of modifiers,
2065 * starting a new chain whenever an indirect modifier starts or ends.
2066 *
2067 * For example, the expression ${VAR:M*:${IND1}:${IND2}:O:u} has 4 chains of
2068 * modifiers:
2069 *
2070 * Chain 1 is ':M*', consisting of the single modifier ':M*'.
2071 * Chain 2 is all modifiers from the value of the variable named 'IND1'.
2072 * Chain 3 is all modifiers from the value of the variable named 'IND2'.
2073 * Chain 4 is ':O:u', consisting of the 2 modifiers ':O' and ':u'.
2074 *
2075 * After such a chain has finished, its properties no longer have any effect.
2076 *
2077 * It may or may not have been intended that 'defined' has scope Expr while
2078 * 'sep' and 'oneBigWord' have smaller scope.
2079 *
2080 * See varmod-indirect.mk.
2081 */
2082 typedef struct ApplyModifiersState {
2083 Expr *expr;
2084 /* '\0' or '{' or '(' */
2085 const char startc;
2086 /* '\0' or '}' or ')' */
2087 const char endc;
2088 /* Word separator in expansions (see the :ts modifier). */
2089 char sep;
2090 /*
2091 * TRUE if some modifiers that otherwise split the variable value
2092 * into words, like :S and :C, treat the variable value as a single
2093 * big word, possibly containing spaces.
2094 */
2095 Boolean oneBigWord;
2096 } ApplyModifiersState;
2097
2098 static void
2099 Expr_Define(Expr *expr)
2100 {
2101 if (expr->defined == DEF_UNDEF)
2102 expr->defined = DEF_DEFINED;
2103 }
2104
2105 static void
2106 Expr_SetValueOwn(Expr *expr, char *value)
2107 {
2108 FStr_Done(&expr->value);
2109 expr->value = FStr_InitOwn(value);
2110 }
2111
2112 static void
2113 Expr_SetValueRefer(Expr *expr, const char *value)
2114 {
2115 FStr_Done(&expr->value);
2116 expr->value = FStr_InitRefer(value);
2117 }
2118
2119 typedef enum ApplyModifierResult {
2120 /* Continue parsing */
2121 AMR_OK,
2122 /* Not a match, try other modifiers as well. */
2123 AMR_UNKNOWN,
2124 /* Error out with "Bad modifier" message. */
2125 AMR_BAD,
2126 /* Error out without the standard error message. */
2127 AMR_CLEANUP
2128 } ApplyModifierResult;
2129
2130 /*
2131 * Allow backslashes to escape the delimiter, $, and \, but don't touch other
2132 * backslashes.
2133 */
2134 static Boolean
2135 IsEscapedModifierPart(const char *p, char delim,
2136 struct ModifyWord_SubstArgs *subst)
2137 {
2138 if (p[0] != '\\')
2139 return FALSE;
2140 if (p[1] == delim || p[1] == '\\' || p[1] == '$')
2141 return TRUE;
2142 return p[1] == '&' && subst != NULL;
2143 }
2144
2145 /* See ParseModifierPart */
2146 static VarParseResult
2147 ParseModifierPartSubst(
2148 const char **pp,
2149 char delim,
2150 VarEvalFlags eflags,
2151 ApplyModifiersState *st,
2152 char **out_part,
2153 /* Optionally stores the length of the returned string, just to save
2154 * another strlen call. */
2155 size_t *out_length,
2156 /* For the first part of the :S modifier, sets the VARP_ANCHOR_END flag
2157 * if the last character of the pattern is a $. */
2158 VarPatternFlags *out_pflags,
2159 /* For the second part of the :S modifier, allow ampersands to be
2160 * escaped and replace unescaped ampersands with subst->lhs. */
2161 struct ModifyWord_SubstArgs *subst
2162 )
2163 {
2164 Buffer buf;
2165 const char *p;
2166
2167 Buf_Init(&buf);
2168
2169 /*
2170 * Skim through until the matching delimiter is found; pick up
2171 * variable expressions on the way.
2172 */
2173 p = *pp;
2174 while (*p != '\0' && *p != delim) {
2175 const char *varstart;
2176
2177 if (IsEscapedModifierPart(p, delim, subst)) {
2178 Buf_AddByte(&buf, p[1]);
2179 p += 2;
2180 continue;
2181 }
2182
2183 if (*p != '$') { /* Unescaped, simple text */
2184 if (subst != NULL && *p == '&')
2185 Buf_AddBytes(&buf, subst->lhs, subst->lhsLen);
2186 else
2187 Buf_AddByte(&buf, *p);
2188 p++;
2189 continue;
2190 }
2191
2192 if (p[1] == delim) { /* Unescaped $ at end of pattern */
2193 if (out_pflags != NULL)
2194 out_pflags->anchorEnd = TRUE;
2195 else
2196 Buf_AddByte(&buf, *p);
2197 p++;
2198 continue;
2199 }
2200
2201 if (eflags.wantRes) { /* Nested variable, evaluated */
2202 const char *nested_p = p;
2203 FStr nested_val;
2204 VarEvalFlags nested_eflags = eflags;
2205 nested_eflags.keepDollar = FALSE;
2206
2207 (void)Var_Parse(&nested_p, st->expr->scope,
2208 nested_eflags, &nested_val);
2209 /* TODO: handle errors */
2210 Buf_AddStr(&buf, nested_val.str);
2211 FStr_Done(&nested_val);
2212 p += nested_p - p;
2213 continue;
2214 }
2215
2216 /*
2217 * XXX: This whole block is very similar to Var_Parse without
2218 * VarEvalFlags.wantRes. There may be subtle edge cases
2219 * though that are not yet covered in the unit tests and that
2220 * are parsed differently, depending on whether they are
2221 * evaluated or not.
2222 *
2223 * This subtle difference is not documented in the manual
2224 * page, neither is the difference between parsing :D and
2225 * :M documented. No code should ever depend on these
2226 * details, but who knows.
2227 */
2228
2229 varstart = p; /* Nested variable, only parsed */
2230 if (p[1] == '(' || p[1] == '{') {
2231 /*
2232 * Find the end of this variable reference
2233 * and suck it in without further ado.
2234 * It will be interpreted later.
2235 */
2236 char startc = p[1];
2237 int endc = startc == '(' ? ')' : '}';
2238 int depth = 1;
2239
2240 for (p += 2; *p != '\0' && depth > 0; p++) {
2241 if (p[-1] != '\\') {
2242 if (*p == startc)
2243 depth++;
2244 if (*p == endc)
2245 depth--;
2246 }
2247 }
2248 Buf_AddBytesBetween(&buf, varstart, p);
2249 } else {
2250 Buf_AddByte(&buf, *varstart);
2251 p++;
2252 }
2253 }
2254
2255 if (*p != delim) {
2256 *pp = p;
2257 Error("Unfinished modifier for \"%s\" ('%c' missing)",
2258 st->expr->var->name.str, delim);
2259 *out_part = NULL;
2260 return VPR_ERR;
2261 }
2262
2263 *pp = p + 1;
2264 if (out_length != NULL)
2265 *out_length = buf.len;
2266
2267 *out_part = Buf_DoneData(&buf);
2268 DEBUG1(VAR, "Modifier part: \"%s\"\n", *out_part);
2269 return VPR_OK;
2270 }
2271
2272 /*
2273 * Parse a part of a modifier such as the "from" and "to" in :S/from/to/ or
2274 * the "var" or "replacement ${var}" in :@var@replacement ${var}@, up to and
2275 * including the next unescaped delimiter. The delimiter, as well as the
2276 * backslash or the dollar, can be escaped with a backslash.
2277 *
2278 * Return the parsed (and possibly expanded) string, or NULL if no delimiter
2279 * was found. On successful return, the parsing position pp points right
2280 * after the delimiter. The delimiter is not included in the returned
2281 * value though.
2282 */
2283 static VarParseResult
2284 ParseModifierPart(
2285 /* The parsing position, updated upon return */
2286 const char **pp,
2287 /* Parsing stops at this delimiter */
2288 char delim,
2289 /* Flags for evaluating nested variables. */
2290 VarEvalFlags eflags,
2291 ApplyModifiersState *st,
2292 char **out_part
2293 )
2294 {
2295 return ParseModifierPartSubst(pp, delim, eflags, st, out_part,
2296 NULL, NULL, NULL);
2297 }
2298
2299 MAKE_INLINE Boolean
2300 IsDelimiter(char ch, const ApplyModifiersState *st)
2301 {
2302 return ch == ':' || ch == st->endc;
2303 }
2304
2305 /* Test whether mod starts with modname, followed by a delimiter. */
2306 MAKE_INLINE Boolean
2307 ModMatch(const char *mod, const char *modname, const ApplyModifiersState *st)
2308 {
2309 size_t n = strlen(modname);
2310 return strncmp(mod, modname, n) == 0 && IsDelimiter(mod[n], st);
2311 }
2312
2313 /* Test whether mod starts with modname, followed by a delimiter or '='. */
2314 MAKE_INLINE Boolean
2315 ModMatchEq(const char *mod, const char *modname, const ApplyModifiersState *st)
2316 {
2317 size_t n = strlen(modname);
2318 return strncmp(mod, modname, n) == 0 &&
2319 (IsDelimiter(mod[n], st) || mod[n] == '=');
2320 }
2321
2322 static Boolean
2323 TryParseIntBase0(const char **pp, int *out_num)
2324 {
2325 char *end;
2326 long n;
2327
2328 errno = 0;
2329 n = strtol(*pp, &end, 0);
2330
2331 if (end == *pp)
2332 return FALSE;
2333 if ((n == LONG_MIN || n == LONG_MAX) && errno == ERANGE)
2334 return FALSE;
2335 if (n < INT_MIN || n > INT_MAX)
2336 return FALSE;
2337
2338 *pp = end;
2339 *out_num = (int)n;
2340 return TRUE;
2341 }
2342
2343 static Boolean
2344 TryParseSize(const char **pp, size_t *out_num)
2345 {
2346 char *end;
2347 unsigned long n;
2348
2349 if (!ch_isdigit(**pp))
2350 return FALSE;
2351
2352 errno = 0;
2353 n = strtoul(*pp, &end, 10);
2354 if (n == ULONG_MAX && errno == ERANGE)
2355 return FALSE;
2356 if (n > SIZE_MAX)
2357 return FALSE;
2358
2359 *pp = end;
2360 *out_num = (size_t)n;
2361 return TRUE;
2362 }
2363
2364 static Boolean
2365 TryParseChar(const char **pp, int base, char *out_ch)
2366 {
2367 char *end;
2368 unsigned long n;
2369
2370 if (!ch_isalnum(**pp))
2371 return FALSE;
2372
2373 errno = 0;
2374 n = strtoul(*pp, &end, base);
2375 if (n == ULONG_MAX && errno == ERANGE)
2376 return FALSE;
2377 if (n > UCHAR_MAX)
2378 return FALSE;
2379
2380 *pp = end;
2381 *out_ch = (char)n;
2382 return TRUE;
2383 }
2384
2385 /*
2386 * Modify each word of the expression using the given function and place the
2387 * result back in the expression.
2388 */
2389 static void
2390 ModifyWords(ApplyModifiersState *st,
2391 ModifyWordProc modifyWord, void *modifyWord_args,
2392 Boolean oneBigWord)
2393 {
2394 Expr *expr = st->expr;
2395 const char *val = expr->value.str;
2396 SepBuf result;
2397 Words words;
2398 size_t i;
2399
2400 if (oneBigWord) {
2401 SepBuf_Init(&result, st->sep);
2402 modifyWord(val, &result, modifyWord_args);
2403 goto done;
2404 }
2405
2406 words = Str_Words(val, FALSE);
2407
2408 DEBUG2(VAR, "ModifyWords: split \"%s\" into %u words\n",
2409 val, (unsigned)words.len);
2410
2411 SepBuf_Init(&result, st->sep);
2412 for (i = 0; i < words.len; i++) {
2413 modifyWord(words.words[i], &result, modifyWord_args);
2414 if (result.buf.len > 0)
2415 SepBuf_Sep(&result);
2416 }
2417
2418 Words_Free(words);
2419
2420 done:
2421 Expr_SetValueOwn(expr, SepBuf_DoneData(&result));
2422 }
2423
2424 /* :@var (at) ...${var}...@ */
2425 static ApplyModifierResult
2426 ApplyModifier_Loop(const char **pp, ApplyModifiersState *st)
2427 {
2428 Expr *expr = st->expr;
2429 struct ModifyWord_LoopArgs args;
2430 char prev_sep;
2431 VarParseResult res;
2432
2433 args.scope = expr->scope;
2434
2435 (*pp)++; /* Skip the first '@' */
2436 res = ParseModifierPart(pp, '@', VARE_PARSE_ONLY, st, &args.tvar);
2437 if (res != VPR_OK)
2438 return AMR_CLEANUP;
2439 if (opts.strict && strchr(args.tvar, '$') != NULL) {
2440 Parse_Error(PARSE_FATAL,
2441 "In the :@ modifier of \"%s\", the variable name \"%s\" "
2442 "must not contain a dollar.",
2443 expr->var->name.str, args.tvar);
2444 return AMR_CLEANUP;
2445 }
2446
2447 res = ParseModifierPart(pp, '@', VARE_PARSE_ONLY, st, &args.str);
2448 if (res != VPR_OK)
2449 return AMR_CLEANUP;
2450
2451 if (!expr->eflags.wantRes)
2452 goto done;
2453
2454 args.eflags = expr->eflags;
2455 args.eflags.keepDollar = FALSE;
2456 prev_sep = st->sep;
2457 st->sep = ' '; /* XXX: should be st->sep for consistency */
2458 ModifyWords(st, ModifyWord_Loop, &args, st->oneBigWord);
2459 st->sep = prev_sep;
2460 /* XXX: Consider restoring the previous variable instead of deleting. */
2461 /*
2462 * XXX: The variable name should not be expanded here, see
2463 * ModifyWord_Loop.
2464 */
2465 Var_DeleteExpand(expr->scope, args.tvar);
2466
2467 done:
2468 free(args.tvar);
2469 free(args.str);
2470 return AMR_OK;
2471 }
2472
2473 /* :Ddefined or :Uundefined */
2474 static ApplyModifierResult
2475 ApplyModifier_Defined(const char **pp, ApplyModifiersState *st)
2476 {
2477 Expr *expr = st->expr;
2478 Buffer buf;
2479 const char *p;
2480
2481 VarEvalFlags eflags = VARE_PARSE_ONLY;
2482 if (expr->eflags.wantRes)
2483 if ((**pp == 'D') == (expr->defined == DEF_REGULAR))
2484 eflags = expr->eflags;
2485
2486 Buf_Init(&buf);
2487 p = *pp + 1;
2488 while (!IsDelimiter(*p, st) && *p != '\0') {
2489
2490 /* XXX: This code is similar to the one in Var_Parse.
2491 * See if the code can be merged.
2492 * See also ApplyModifier_Match and ParseModifierPart. */
2493
2494 /* Escaped delimiter or other special character */
2495 /* See Buf_AddEscaped in for.c. */
2496 if (*p == '\\') {
2497 char c = p[1];
2498 if (IsDelimiter(c, st) || c == '$' || c == '\\') {
2499 Buf_AddByte(&buf, c);
2500 p += 2;
2501 continue;
2502 }
2503 }
2504
2505 /* Nested variable expression */
2506 if (*p == '$') {
2507 FStr nested_val;
2508
2509 (void)Var_Parse(&p, expr->scope, eflags, &nested_val);
2510 /* TODO: handle errors */
2511 if (expr->eflags.wantRes)
2512 Buf_AddStr(&buf, nested_val.str);
2513 FStr_Done(&nested_val);
2514 continue;
2515 }
2516
2517 /* Ordinary text */
2518 Buf_AddByte(&buf, *p);
2519 p++;
2520 }
2521 *pp = p;
2522
2523 Expr_Define(expr);
2524
2525 if (eflags.wantRes)
2526 Expr_SetValueOwn(expr, Buf_DoneData(&buf));
2527 else
2528 Buf_Done(&buf);
2529
2530 return AMR_OK;
2531 }
2532
2533 /* :L */
2534 static ApplyModifierResult
2535 ApplyModifier_Literal(const char **pp, ApplyModifiersState *st)
2536 {
2537 Expr *expr = st->expr;
2538
2539 (*pp)++;
2540
2541 if (expr->eflags.wantRes) {
2542 Expr_Define(expr);
2543 Expr_SetValueOwn(expr, bmake_strdup(expr->var->name.str));
2544 }
2545
2546 return AMR_OK;
2547 }
2548
2549 static Boolean
2550 TryParseTime(const char **pp, time_t *out_time)
2551 {
2552 char *end;
2553 unsigned long n;
2554
2555 if (!ch_isdigit(**pp))
2556 return FALSE;
2557
2558 errno = 0;
2559 n = strtoul(*pp, &end, 10);
2560 if (n == ULONG_MAX && errno == ERANGE)
2561 return FALSE;
2562
2563 *pp = end;
2564 *out_time = (time_t)n; /* ignore possible truncation for now */
2565 return TRUE;
2566 }
2567
2568 /* :gmtime */
2569 static ApplyModifierResult
2570 ApplyModifier_Gmtime(const char **pp, ApplyModifiersState *st)
2571 {
2572 time_t utc;
2573
2574 const char *mod = *pp;
2575 if (!ModMatchEq(mod, "gmtime", st))
2576 return AMR_UNKNOWN;
2577
2578 if (mod[6] == '=') {
2579 const char *p = mod + 7;
2580 if (!TryParseTime(&p, &utc)) {
2581 Parse_Error(PARSE_FATAL,
2582 "Invalid time value: %s", mod + 7);
2583 return AMR_CLEANUP;
2584 }
2585 *pp = p;
2586 } else {
2587 utc = 0;
2588 *pp = mod + 6;
2589 }
2590
2591 if (st->expr->eflags.wantRes)
2592 Expr_SetValueOwn(st->expr,
2593 VarStrftime(st->expr->value.str, TRUE, utc));
2594
2595 return AMR_OK;
2596 }
2597
2598 /* :localtime */
2599 static ApplyModifierResult
2600 ApplyModifier_Localtime(const char **pp, ApplyModifiersState *st)
2601 {
2602 time_t utc;
2603
2604 const char *mod = *pp;
2605 if (!ModMatchEq(mod, "localtime", st))
2606 return AMR_UNKNOWN;
2607
2608 if (mod[9] == '=') {
2609 const char *p = mod + 10;
2610 if (!TryParseTime(&p, &utc)) {
2611 Parse_Error(PARSE_FATAL,
2612 "Invalid time value: %s", mod + 10);
2613 return AMR_CLEANUP;
2614 }
2615 *pp = p;
2616 } else {
2617 utc = 0;
2618 *pp = mod + 9;
2619 }
2620
2621 if (st->expr->eflags.wantRes)
2622 Expr_SetValueOwn(st->expr,
2623 VarStrftime(st->expr->value.str, FALSE, utc));
2624
2625 return AMR_OK;
2626 }
2627
2628 /* :hash */
2629 static ApplyModifierResult
2630 ApplyModifier_Hash(const char **pp, ApplyModifiersState *st)
2631 {
2632 if (!ModMatch(*pp, "hash", st))
2633 return AMR_UNKNOWN;
2634 *pp += 4;
2635
2636 if (st->expr->eflags.wantRes)
2637 Expr_SetValueOwn(st->expr, VarHash(st->expr->value.str));
2638
2639 return AMR_OK;
2640 }
2641
2642 /* :P */
2643 static ApplyModifierResult
2644 ApplyModifier_Path(const char **pp, ApplyModifiersState *st)
2645 {
2646 Expr *expr = st->expr;
2647 GNode *gn;
2648 char *path;
2649
2650 (*pp)++;
2651
2652 if (!st->expr->eflags.wantRes)
2653 return AMR_OK;
2654
2655 Expr_Define(expr);
2656
2657 gn = Targ_FindNode(expr->var->name.str);
2658 if (gn == NULL || gn->type & OP_NOPATH) {
2659 path = NULL;
2660 } else if (gn->path != NULL) {
2661 path = bmake_strdup(gn->path);
2662 } else {
2663 SearchPath *searchPath = Suff_FindPath(gn);
2664 path = Dir_FindFile(expr->var->name.str, searchPath);
2665 }
2666 if (path == NULL)
2667 path = bmake_strdup(expr->var->name.str);
2668 Expr_SetValueOwn(expr, path);
2669
2670 return AMR_OK;
2671 }
2672
2673 /* :!cmd! */
2674 static ApplyModifierResult
2675 ApplyModifier_ShellCommand(const char **pp, ApplyModifiersState *st)
2676 {
2677 Expr *expr = st->expr;
2678 char *cmd;
2679 const char *errfmt;
2680 VarParseResult res;
2681
2682 (*pp)++;
2683 res = ParseModifierPart(pp, '!', expr->eflags, st, &cmd);
2684 if (res != VPR_OK)
2685 return AMR_CLEANUP;
2686
2687 errfmt = NULL;
2688 if (expr->eflags.wantRes)
2689 Expr_SetValueOwn(expr, Cmd_Exec(cmd, &errfmt));
2690 else
2691 Expr_SetValueRefer(expr, "");
2692 if (errfmt != NULL)
2693 Error(errfmt, cmd); /* XXX: why still return AMR_OK? */
2694 free(cmd);
2695 Expr_Define(expr);
2696
2697 return AMR_OK;
2698 }
2699
2700 /*
2701 * The :range modifier generates an integer sequence as long as the words.
2702 * The :range=7 modifier generates an integer sequence from 1 to 7.
2703 */
2704 static ApplyModifierResult
2705 ApplyModifier_Range(const char **pp, ApplyModifiersState *st)
2706 {
2707 size_t n;
2708 Buffer buf;
2709 size_t i;
2710
2711 const char *mod = *pp;
2712 if (!ModMatchEq(mod, "range", st))
2713 return AMR_UNKNOWN;
2714
2715 if (mod[5] == '=') {
2716 const char *p = mod + 6;
2717 if (!TryParseSize(&p, &n)) {
2718 Parse_Error(PARSE_FATAL,
2719 "Invalid number \"%s\" for ':range' modifier",
2720 mod + 6);
2721 return AMR_CLEANUP;
2722 }
2723 *pp = p;
2724 } else {
2725 n = 0;
2726 *pp = mod + 5;
2727 }
2728
2729 if (!st->expr->eflags.wantRes)
2730 return AMR_OK;
2731
2732 if (n == 0) {
2733 Words words = Str_Words(st->expr->value.str, FALSE);
2734 n = words.len;
2735 Words_Free(words);
2736 }
2737
2738 Buf_Init(&buf);
2739
2740 for (i = 0; i < n; i++) {
2741 if (i != 0) {
2742 /* XXX: Use st->sep instead of ' ', for consistency. */
2743 Buf_AddByte(&buf, ' ');
2744 }
2745 Buf_AddInt(&buf, 1 + (int)i);
2746 }
2747
2748 Expr_SetValueOwn(st->expr, Buf_DoneData(&buf));
2749 return AMR_OK;
2750 }
2751
2752 /* Parse a ':M' or ':N' modifier. */
2753 static void
2754 ParseModifier_Match(const char **pp, const ApplyModifiersState *st,
2755 char **out_pattern)
2756 {
2757 const char *mod = *pp;
2758 Expr *expr = st->expr;
2759 Boolean copy = FALSE; /* pattern should be, or has been, copied */
2760 Boolean needSubst = FALSE;
2761 const char *endpat;
2762 char *pattern;
2763
2764 /*
2765 * In the loop below, ignore ':' unless we are at (or back to) the
2766 * original brace level.
2767 * XXX: This will likely not work right if $() and ${} are intermixed.
2768 */
2769 /*
2770 * XXX: This code is similar to the one in Var_Parse.
2771 * See if the code can be merged.
2772 * See also ApplyModifier_Defined.
2773 */
2774 int nest = 0;
2775 const char *p;
2776 for (p = mod + 1; *p != '\0' && !(*p == ':' && nest == 0); p++) {
2777 if (*p == '\\' &&
2778 (IsDelimiter(p[1], st) || p[1] == st->startc)) {
2779 if (!needSubst)
2780 copy = TRUE;
2781 p++;
2782 continue;
2783 }
2784 if (*p == '$')
2785 needSubst = TRUE;
2786 if (*p == '(' || *p == '{')
2787 nest++;
2788 if (*p == ')' || *p == '}') {
2789 nest--;
2790 if (nest < 0)
2791 break;
2792 }
2793 }
2794 *pp = p;
2795 endpat = p;
2796
2797 if (copy) {
2798 char *dst;
2799 const char *src;
2800
2801 /* Compress the \:'s out of the pattern. */
2802 pattern = bmake_malloc((size_t)(endpat - (mod + 1)) + 1);
2803 dst = pattern;
2804 src = mod + 1;
2805 for (; src < endpat; src++, dst++) {
2806 if (src[0] == '\\' && src + 1 < endpat &&
2807 /* XXX: st->startc is missing here; see above */
2808 IsDelimiter(src[1], st))
2809 src++;
2810 *dst = *src;
2811 }
2812 *dst = '\0';
2813 } else {
2814 pattern = bmake_strsedup(mod + 1, endpat);
2815 }
2816
2817 if (needSubst) {
2818 char *old_pattern = pattern;
2819 (void)Var_Subst(pattern, expr->scope, expr->eflags, &pattern);
2820 /* TODO: handle errors */
2821 free(old_pattern);
2822 }
2823
2824 DEBUG3(VAR, "Pattern[%s] for [%s] is [%s]\n",
2825 expr->var->name.str, expr->value.str, pattern);
2826
2827 *out_pattern = pattern;
2828 }
2829
2830 /* :Mpattern or :Npattern */
2831 static ApplyModifierResult
2832 ApplyModifier_Match(const char **pp, ApplyModifiersState *st)
2833 {
2834 const char mod = **pp;
2835 char *pattern;
2836
2837 ParseModifier_Match(pp, st, &pattern);
2838
2839 if (st->expr->eflags.wantRes) {
2840 ModifyWordProc modifyWord =
2841 mod == 'M' ? ModifyWord_Match : ModifyWord_NoMatch;
2842 ModifyWords(st, modifyWord, pattern, st->oneBigWord);
2843 }
2844
2845 free(pattern);
2846 return AMR_OK;
2847 }
2848
2849 static void
2850 ParsePatternFlags(const char **pp, VarPatternFlags *pflags, Boolean *oneBigWord)
2851 {
2852 for (;; (*pp)++) {
2853 if (**pp == 'g')
2854 pflags->subGlobal = TRUE;
2855 else if (**pp == '1')
2856 pflags->subOnce = TRUE;
2857 else if (**pp == 'W')
2858 *oneBigWord = TRUE;
2859 else
2860 break;
2861 }
2862 }
2863
2864 /* :S,from,to, */
2865 static ApplyModifierResult
2866 ApplyModifier_Subst(const char **pp, ApplyModifiersState *st)
2867 {
2868 struct ModifyWord_SubstArgs args;
2869 char *lhs, *rhs;
2870 Boolean oneBigWord;
2871 VarParseResult res;
2872
2873 char delim = (*pp)[1];
2874 if (delim == '\0') {
2875 Error("Missing delimiter for modifier ':S'");
2876 (*pp)++;
2877 return AMR_CLEANUP;
2878 }
2879
2880 *pp += 2;
2881
2882 args.pflags = (VarPatternFlags){ FALSE, FALSE, FALSE, FALSE };
2883 args.matched = FALSE;
2884
2885 if (**pp == '^') {
2886 args.pflags.anchorStart = TRUE;
2887 (*pp)++;
2888 }
2889
2890 res = ParseModifierPartSubst(pp, delim, st->expr->eflags, st, &lhs,
2891 &args.lhsLen, &args.pflags, NULL);
2892 if (res != VPR_OK)
2893 return AMR_CLEANUP;
2894 args.lhs = lhs;
2895
2896 res = ParseModifierPartSubst(pp, delim, st->expr->eflags, st, &rhs,
2897 &args.rhsLen, NULL, &args);
2898 if (res != VPR_OK)
2899 return AMR_CLEANUP;
2900 args.rhs = rhs;
2901
2902 oneBigWord = st->oneBigWord;
2903 ParsePatternFlags(pp, &args.pflags, &oneBigWord);
2904
2905 ModifyWords(st, ModifyWord_Subst, &args, oneBigWord);
2906
2907 free(lhs);
2908 free(rhs);
2909 return AMR_OK;
2910 }
2911
2912 #ifndef NO_REGEX
2913
2914 /* :C,from,to, */
2915 static ApplyModifierResult
2916 ApplyModifier_Regex(const char **pp, ApplyModifiersState *st)
2917 {
2918 char *re;
2919 struct ModifyWord_SubstRegexArgs args;
2920 Boolean oneBigWord;
2921 int error;
2922 VarParseResult res;
2923
2924 char delim = (*pp)[1];
2925 if (delim == '\0') {
2926 Error("Missing delimiter for :C modifier");
2927 (*pp)++;
2928 return AMR_CLEANUP;
2929 }
2930
2931 *pp += 2;
2932
2933 res = ParseModifierPart(pp, delim, st->expr->eflags, st, &re);
2934 if (res != VPR_OK)
2935 return AMR_CLEANUP;
2936
2937 res = ParseModifierPart(pp, delim, st->expr->eflags, st, &args.replace);
2938 if (args.replace == NULL) {
2939 free(re);
2940 return AMR_CLEANUP;
2941 }
2942
2943 args.pflags = (VarPatternFlags){ FALSE, FALSE, FALSE, FALSE };
2944 args.matched = FALSE;
2945 oneBigWord = st->oneBigWord;
2946 ParsePatternFlags(pp, &args.pflags, &oneBigWord);
2947
2948 if (!(st->expr->eflags.wantRes)) {
2949 free(args.replace);
2950 free(re);
2951 return AMR_OK;
2952 }
2953
2954 error = regcomp(&args.re, re, REG_EXTENDED);
2955 free(re);
2956 if (error != 0) {
2957 VarREError(error, &args.re, "Regex compilation error");
2958 free(args.replace);
2959 return AMR_CLEANUP;
2960 }
2961
2962 args.nsub = args.re.re_nsub + 1;
2963 if (args.nsub > 10)
2964 args.nsub = 10;
2965
2966 ModifyWords(st, ModifyWord_SubstRegex, &args, oneBigWord);
2967
2968 regfree(&args.re);
2969 free(args.replace);
2970 return AMR_OK;
2971 }
2972
2973 #endif
2974
2975 /* :Q, :q */
2976 static ApplyModifierResult
2977 ApplyModifier_Quote(const char **pp, ApplyModifiersState *st)
2978 {
2979 Boolean quoteDollar = **pp == 'q';
2980 if (!IsDelimiter((*pp)[1], st))
2981 return AMR_UNKNOWN;
2982 (*pp)++;
2983
2984 if (st->expr->eflags.wantRes)
2985 Expr_SetValueOwn(st->expr,
2986 VarQuote(st->expr->value.str, quoteDollar));
2987
2988 return AMR_OK;
2989 }
2990
2991 /*ARGSUSED*/
2992 static void
2993 ModifyWord_Copy(const char *word, SepBuf *buf, void *data MAKE_ATTR_UNUSED)
2994 {
2995 SepBuf_AddStr(buf, word);
2996 }
2997
2998 /* :ts<separator> */
2999 static ApplyModifierResult
3000 ApplyModifier_ToSep(const char **pp, ApplyModifiersState *st)
3001 {
3002 const char *sep = *pp + 2;
3003
3004 /*
3005 * Even in parse-only mode, proceed as normal since there is
3006 * neither any observable side effect nor a performance penalty.
3007 * Checking for wantRes for every single piece of code in here
3008 * would make the code in this function too hard to read.
3009 */
3010
3011 /* ":ts<any><endc>" or ":ts<any>:" */
3012 if (sep[0] != st->endc && IsDelimiter(sep[1], st)) {
3013 *pp = sep + 1;
3014 st->sep = sep[0];
3015 goto ok;
3016 }
3017
3018 /* ":ts<endc>" or ":ts:" */
3019 if (IsDelimiter(sep[0], st)) {
3020 *pp = sep;
3021 st->sep = '\0'; /* no separator */
3022 goto ok;
3023 }
3024
3025 /* ":ts<unrecognised><unrecognised>". */
3026 if (sep[0] != '\\') {
3027 (*pp)++; /* just for backwards compatibility */
3028 return AMR_BAD;
3029 }
3030
3031 /* ":ts\n" */
3032 if (sep[1] == 'n') {
3033 *pp = sep + 2;
3034 st->sep = '\n';
3035 goto ok;
3036 }
3037
3038 /* ":ts\t" */
3039 if (sep[1] == 't') {
3040 *pp = sep + 2;
3041 st->sep = '\t';
3042 goto ok;
3043 }
3044
3045 /* ":ts\x40" or ":ts\100" */
3046 {
3047 const char *p = sep + 1;
3048 int base = 8; /* assume octal */
3049
3050 if (sep[1] == 'x') {
3051 base = 16;
3052 p++;
3053 } else if (!ch_isdigit(sep[1])) {
3054 (*pp)++; /* just for backwards compatibility */
3055 return AMR_BAD; /* ":ts<backslash><unrecognised>". */
3056 }
3057
3058 if (!TryParseChar(&p, base, &st->sep)) {
3059 Parse_Error(PARSE_FATAL,
3060 "Invalid character number: %s", p);
3061 return AMR_CLEANUP;
3062 }
3063 if (!IsDelimiter(*p, st)) {
3064 (*pp)++; /* just for backwards compatibility */
3065 return AMR_BAD;
3066 }
3067
3068 *pp = p;
3069 }
3070
3071 ok:
3072 ModifyWords(st, ModifyWord_Copy, NULL, st->oneBigWord);
3073 return AMR_OK;
3074 }
3075
3076 static char *
3077 str_toupper(const char *str)
3078 {
3079 char *res;
3080 size_t i, len;
3081
3082 len = strlen(str);
3083 res = bmake_malloc(len + 1);
3084 for (i = 0; i < len + 1; i++)
3085 res[i] = ch_toupper(str[i]);
3086
3087 return res;
3088 }
3089
3090 static char *
3091 str_tolower(const char *str)
3092 {
3093 char *res;
3094 size_t i, len;
3095
3096 len = strlen(str);
3097 res = bmake_malloc(len + 1);
3098 for (i = 0; i < len + 1; i++)
3099 res[i] = ch_tolower(str[i]);
3100
3101 return res;
3102 }
3103
3104 /* :tA, :tu, :tl, :ts<separator>, etc. */
3105 static ApplyModifierResult
3106 ApplyModifier_To(const char **pp, ApplyModifiersState *st)
3107 {
3108 Expr *expr = st->expr;
3109 const char *mod = *pp;
3110 assert(mod[0] == 't');
3111
3112 if (IsDelimiter(mod[1], st) || mod[1] == '\0') {
3113 *pp = mod + 1;
3114 return AMR_BAD; /* Found ":t<endc>" or ":t:". */
3115 }
3116
3117 if (mod[1] == 's')
3118 return ApplyModifier_ToSep(pp, st);
3119
3120 if (!IsDelimiter(mod[2], st)) { /* :t<unrecognized> */
3121 *pp = mod + 1;
3122 return AMR_BAD;
3123 }
3124
3125 if (mod[1] == 'A') { /* :tA */
3126 *pp = mod + 2;
3127 ModifyWords(st, ModifyWord_Realpath, NULL, st->oneBigWord);
3128 return AMR_OK;
3129 }
3130
3131 if (mod[1] == 'u') { /* :tu */
3132 *pp = mod + 2;
3133 if (st->expr->eflags.wantRes)
3134 Expr_SetValueOwn(expr, str_toupper(expr->value.str));
3135 return AMR_OK;
3136 }
3137
3138 if (mod[1] == 'l') { /* :tl */
3139 *pp = mod + 2;
3140 if (st->expr->eflags.wantRes)
3141 Expr_SetValueOwn(expr, str_tolower(expr->value.str));
3142 return AMR_OK;
3143 }
3144
3145 if (mod[1] == 'W' || mod[1] == 'w') { /* :tW, :tw */
3146 *pp = mod + 2;
3147 st->oneBigWord = mod[1] == 'W';
3148 return AMR_OK;
3149 }
3150
3151 /* Found ":t<unrecognised>:" or ":t<unrecognised><endc>". */
3152 *pp = mod + 1; /* XXX: unnecessary but observable */
3153 return AMR_BAD;
3154 }
3155
3156 /* :[#], :[1], :[-1..1], etc. */
3157 static ApplyModifierResult
3158 ApplyModifier_Words(const char **pp, ApplyModifiersState *st)
3159 {
3160 Expr *expr = st->expr;
3161 char *estr;
3162 int first, last;
3163 VarParseResult res;
3164 const char *p;
3165
3166 (*pp)++; /* skip the '[' */
3167 res = ParseModifierPart(pp, ']', expr->eflags, st, &estr);
3168 if (res != VPR_OK)
3169 return AMR_CLEANUP;
3170
3171 if (!IsDelimiter(**pp, st))
3172 goto bad_modifier; /* Found junk after ']' */
3173
3174 if (!(expr->eflags.wantRes))
3175 goto ok;
3176
3177 if (estr[0] == '\0')
3178 goto bad_modifier; /* Found ":[]". */
3179
3180 if (estr[0] == '#' && estr[1] == '\0') { /* Found ":[#]" */
3181 if (st->oneBigWord) {
3182 Expr_SetValueRefer(expr, "1");
3183 } else {
3184 Buffer buf;
3185
3186 Words words = Str_Words(expr->value.str, FALSE);
3187 size_t ac = words.len;
3188 Words_Free(words);
3189
3190 /* 3 digits + '\0' is usually enough */
3191 Buf_InitSize(&buf, 4);
3192 Buf_AddInt(&buf, (int)ac);
3193 Expr_SetValueOwn(expr, Buf_DoneData(&buf));
3194 }
3195 goto ok;
3196 }
3197
3198 if (estr[0] == '*' && estr[1] == '\0') { /* Found ":[*]" */
3199 st->oneBigWord = TRUE;
3200 goto ok;
3201 }
3202
3203 if (estr[0] == '@' && estr[1] == '\0') { /* Found ":[@]" */
3204 st->oneBigWord = FALSE;
3205 goto ok;
3206 }
3207
3208 /*
3209 * We expect estr to contain a single integer for :[N], or two
3210 * integers separated by ".." for :[start..end].
3211 */
3212 p = estr;
3213 if (!TryParseIntBase0(&p, &first))
3214 goto bad_modifier; /* Found junk instead of a number */
3215
3216 if (p[0] == '\0') { /* Found only one integer in :[N] */
3217 last = first;
3218 } else if (p[0] == '.' && p[1] == '.' && p[2] != '\0') {
3219 /* Expecting another integer after ".." */
3220 p += 2;
3221 if (!TryParseIntBase0(&p, &last) || *p != '\0')
3222 goto bad_modifier; /* Found junk after ".." */
3223 } else
3224 goto bad_modifier; /* Found junk instead of ".." */
3225
3226 /*
3227 * Now first and last are properly filled in, but we still have to
3228 * check for 0 as a special case.
3229 */
3230 if (first == 0 && last == 0) {
3231 /* ":[0]" or perhaps ":[0..0]" */
3232 st->oneBigWord = TRUE;
3233 goto ok;
3234 }
3235
3236 /* ":[0..N]" or ":[N..0]" */
3237 if (first == 0 || last == 0)
3238 goto bad_modifier;
3239
3240 /* Normal case: select the words described by first and last. */
3241 Expr_SetValueOwn(expr,
3242 VarSelectWords(expr->value.str, first, last,
3243 st->sep, st->oneBigWord));
3244
3245 ok:
3246 free(estr);
3247 return AMR_OK;
3248
3249 bad_modifier:
3250 free(estr);
3251 return AMR_BAD;
3252 }
3253
3254 static int
3255 str_cmp_asc(const void *a, const void *b)
3256 {
3257 return strcmp(*(const char *const *)a, *(const char *const *)b);
3258 }
3259
3260 static int
3261 str_cmp_desc(const void *a, const void *b)
3262 {
3263 return strcmp(*(const char *const *)b, *(const char *const *)a);
3264 }
3265
3266 static void
3267 ShuffleStrings(char **strs, size_t n)
3268 {
3269 size_t i;
3270
3271 for (i = n - 1; i > 0; i--) {
3272 size_t rndidx = (size_t)random() % (i + 1);
3273 char *t = strs[i];
3274 strs[i] = strs[rndidx];
3275 strs[rndidx] = t;
3276 }
3277 }
3278
3279 /* :O (order ascending) or :Or (order descending) or :Ox (shuffle) */
3280 static ApplyModifierResult
3281 ApplyModifier_Order(const char **pp, ApplyModifiersState *st)
3282 {
3283 const char *mod = (*pp)++; /* skip past the 'O' in any case */
3284 Words words;
3285 enum SortMode {
3286 ASC, DESC, SHUFFLE
3287 } mode;
3288
3289 if (IsDelimiter(mod[1], st)) {
3290 mode = ASC;
3291 } else if ((mod[1] == 'r' || mod[1] == 'x') &&
3292 IsDelimiter(mod[2], st)) {
3293 (*pp)++;
3294 mode = mod[1] == 'r' ? DESC : SHUFFLE;
3295 } else
3296 return AMR_BAD;
3297
3298 if (!st->expr->eflags.wantRes)
3299 return AMR_OK;
3300
3301 words = Str_Words(st->expr->value.str, FALSE);
3302 if (mode == SHUFFLE)
3303 ShuffleStrings(words.words, words.len);
3304 else
3305 qsort(words.words, words.len, sizeof words.words[0],
3306 mode == ASC ? str_cmp_asc : str_cmp_desc);
3307 Expr_SetValueOwn(st->expr, Words_JoinFree(words));
3308
3309 return AMR_OK;
3310 }
3311
3312 /* :? then : else */
3313 static ApplyModifierResult
3314 ApplyModifier_IfElse(const char **pp, ApplyModifiersState *st)
3315 {
3316 Expr *expr = st->expr;
3317 char *then_expr, *else_expr;
3318 VarParseResult res;
3319
3320 Boolean value = FALSE;
3321 VarEvalFlags then_eflags = VARE_PARSE_ONLY;
3322 VarEvalFlags else_eflags = VARE_PARSE_ONLY;
3323
3324 int cond_rc = COND_PARSE; /* anything other than COND_INVALID */
3325 if (expr->eflags.wantRes) {
3326 cond_rc = Cond_EvalCondition(expr->var->name.str, &value);
3327 if (cond_rc != COND_INVALID && value)
3328 then_eflags = expr->eflags;
3329 if (cond_rc != COND_INVALID && !value)
3330 else_eflags = expr->eflags;
3331 }
3332
3333 (*pp)++; /* skip past the '?' */
3334 res = ParseModifierPart(pp, ':', then_eflags, st, &then_expr);
3335 if (res != VPR_OK)
3336 return AMR_CLEANUP;
3337
3338 res = ParseModifierPart(pp, st->endc, else_eflags, st, &else_expr);
3339 if (res != VPR_OK)
3340 return AMR_CLEANUP;
3341
3342 (*pp)--; /* Go back to the st->endc. */
3343
3344 if (cond_rc == COND_INVALID) {
3345 Error("Bad conditional expression `%s' in %s?%s:%s",
3346 expr->var->name.str, expr->var->name.str,
3347 then_expr, else_expr);
3348 return AMR_CLEANUP;
3349 }
3350
3351 if (!expr->eflags.wantRes) {
3352 free(then_expr);
3353 free(else_expr);
3354 } else if (value) {
3355 Expr_SetValueOwn(expr, then_expr);
3356 free(else_expr);
3357 } else {
3358 Expr_SetValueOwn(expr, else_expr);
3359 free(then_expr);
3360 }
3361 Expr_Define(expr);
3362 return AMR_OK;
3363 }
3364
3365 /*
3366 * The ::= modifiers are special in that they do not read the variable value
3367 * but instead assign to that variable. They always expand to an empty
3368 * string.
3369 *
3370 * Their main purpose is in supporting .for loops that generate shell commands
3371 * since an ordinary variable assignment at that point would terminate the
3372 * dependency group for these targets. For example:
3373 *
3374 * list-targets: .USE
3375 * .for i in ${.TARGET} ${.TARGET:R}.gz
3376 * @${t::=$i}
3377 * @echo 'The target is ${t:T}.'
3378 * .endfor
3379 *
3380 * ::=<str> Assigns <str> as the new value of variable.
3381 * ::?=<str> Assigns <str> as value of variable if
3382 * it was not already set.
3383 * ::+=<str> Appends <str> to variable.
3384 * ::!=<cmd> Assigns output of <cmd> as the new value of
3385 * variable.
3386 */
3387 static ApplyModifierResult
3388 ApplyModifier_Assign(const char **pp, ApplyModifiersState *st)
3389 {
3390 Expr *expr = st->expr;
3391 GNode *scope;
3392 char *val;
3393 VarParseResult res;
3394
3395 const char *mod = *pp;
3396 const char *op = mod + 1;
3397
3398 if (op[0] == '=')
3399 goto ok;
3400 if ((op[0] == '!' || op[0] == '+' || op[0] == '?') && op[1] == '=')
3401 goto ok;
3402 return AMR_UNKNOWN; /* "::<unrecognised>" */
3403
3404 ok:
3405 if (expr->var->name.str[0] == '\0') {
3406 *pp = mod + 1;
3407 return AMR_BAD;
3408 }
3409
3410 switch (op[0]) {
3411 case '+':
3412 case '?':
3413 case '!':
3414 *pp = mod + 3;
3415 break;
3416 default:
3417 *pp = mod + 2;
3418 break;
3419 }
3420
3421 res = ParseModifierPart(pp, st->endc, expr->eflags, st, &val);
3422 if (res != VPR_OK)
3423 return AMR_CLEANUP;
3424
3425 (*pp)--; /* Go back to the st->endc. */
3426
3427 if (!expr->eflags.wantRes)
3428 goto done;
3429
3430 scope = expr->scope; /* scope where v belongs */
3431 if (expr->defined == DEF_REGULAR && expr->scope != SCOPE_GLOBAL) {
3432 Var *gv = VarFind(expr->var->name.str, expr->scope, FALSE);
3433 if (gv == NULL)
3434 scope = SCOPE_GLOBAL;
3435 else
3436 VarFreeEnv(gv);
3437 }
3438
3439 switch (op[0]) {
3440 case '+':
3441 Var_Append(scope, expr->var->name.str, val);
3442 break;
3443 case '!': {
3444 const char *errfmt;
3445 char *cmd_output = Cmd_Exec(val, &errfmt);
3446 if (errfmt != NULL)
3447 Error(errfmt, val);
3448 else
3449 Var_Set(scope, expr->var->name.str, cmd_output);
3450 free(cmd_output);
3451 break;
3452 }
3453 case '?':
3454 if (expr->defined == DEF_REGULAR)
3455 break;
3456 /* FALLTHROUGH */
3457 default:
3458 Var_Set(scope, expr->var->name.str, val);
3459 break;
3460 }
3461 Expr_SetValueRefer(expr, "");
3462
3463 done:
3464 free(val);
3465 return AMR_OK;
3466 }
3467
3468 /*
3469 * :_=...
3470 * remember current value
3471 */
3472 static ApplyModifierResult
3473 ApplyModifier_Remember(const char **pp, ApplyModifiersState *st)
3474 {
3475 Expr *expr = st->expr;
3476 const char *mod = *pp;
3477 FStr name;
3478
3479 if (!ModMatchEq(mod, "_", st))
3480 return AMR_UNKNOWN;
3481
3482 name = FStr_InitRefer("_");
3483 if (mod[1] == '=') {
3484 /*
3485 * XXX: This ad-hoc call to strcspn deviates from the usual
3486 * behavior defined in ParseModifierPart. This creates an
3487 * unnecessary, undocumented inconsistency in make.
3488 */
3489 const char *arg = mod + 2;
3490 size_t argLen = strcspn(arg, ":)}");
3491 *pp = arg + argLen;
3492 name = FStr_InitOwn(bmake_strldup(arg, argLen));
3493 } else
3494 *pp = mod + 1;
3495
3496 if (expr->eflags.wantRes)
3497 Var_Set(expr->scope, name.str, expr->value.str);
3498 FStr_Done(&name);
3499
3500 return AMR_OK;
3501 }
3502
3503 /*
3504 * Apply the given function to each word of the variable value,
3505 * for a single-letter modifier such as :H, :T.
3506 */
3507 static ApplyModifierResult
3508 ApplyModifier_WordFunc(const char **pp, ApplyModifiersState *st,
3509 ModifyWordProc modifyWord)
3510 {
3511 if (!IsDelimiter((*pp)[1], st))
3512 return AMR_UNKNOWN;
3513 (*pp)++;
3514
3515 if (st->expr->eflags.wantRes)
3516 ModifyWords(st, modifyWord, NULL, st->oneBigWord);
3517
3518 return AMR_OK;
3519 }
3520
3521 static ApplyModifierResult
3522 ApplyModifier_Unique(const char **pp, ApplyModifiersState *st)
3523 {
3524 if (!IsDelimiter((*pp)[1], st))
3525 return AMR_UNKNOWN;
3526 (*pp)++;
3527
3528 if (st->expr->eflags.wantRes)
3529 Expr_SetValueOwn(st->expr, VarUniq(st->expr->value.str));
3530
3531 return AMR_OK;
3532 }
3533
3534 #ifdef SYSVVARSUB
3535 /* :from=to */
3536 static ApplyModifierResult
3537 ApplyModifier_SysV(const char **pp, ApplyModifiersState *st)
3538 {
3539 Expr *expr = st->expr;
3540 char *lhs, *rhs;
3541 VarParseResult res;
3542
3543 const char *mod = *pp;
3544 Boolean eqFound = FALSE;
3545
3546 /*
3547 * First we make a pass through the string trying to verify it is a
3548 * SysV-make-style translation. It must be: <lhs>=<rhs>
3549 */
3550 int depth = 1;
3551 const char *p = mod;
3552 while (*p != '\0' && depth > 0) {
3553 if (*p == '=') { /* XXX: should also test depth == 1 */
3554 eqFound = TRUE;
3555 /* continue looking for st->endc */
3556 } else if (*p == st->endc)
3557 depth--;
3558 else if (*p == st->startc)
3559 depth++;
3560 if (depth > 0)
3561 p++;
3562 }
3563 if (*p != st->endc || !eqFound)
3564 return AMR_UNKNOWN;
3565
3566 res = ParseModifierPart(pp, '=', expr->eflags, st, &lhs);
3567 if (res != VPR_OK)
3568 return AMR_CLEANUP;
3569
3570 /* The SysV modifier lasts until the end of the variable expression. */
3571 res = ParseModifierPart(pp, st->endc, expr->eflags, st, &rhs);
3572 if (res != VPR_OK)
3573 return AMR_CLEANUP;
3574
3575 (*pp)--; /* Go back to the st->endc. */
3576
3577 if (lhs[0] == '\0' && expr->value.str[0] == '\0') {
3578 /* Do not turn an empty expression into non-empty. */
3579 } else {
3580 struct ModifyWord_SYSVSubstArgs args = {
3581 expr->scope, lhs, rhs
3582 };
3583 ModifyWords(st, ModifyWord_SYSVSubst, &args, st->oneBigWord);
3584 }
3585 free(lhs);
3586 free(rhs);
3587 return AMR_OK;
3588 }
3589 #endif
3590
3591 #ifdef SUNSHCMD
3592 /* :sh */
3593 static ApplyModifierResult
3594 ApplyModifier_SunShell(const char **pp, ApplyModifiersState *st)
3595 {
3596 Expr *expr = st->expr;
3597 const char *p = *pp;
3598 if (!(p[1] == 'h' && IsDelimiter(p[2], st)))
3599 return AMR_UNKNOWN;
3600 *pp = p + 2;
3601
3602 if (expr->eflags.wantRes) {
3603 const char *errfmt;
3604 char *output = Cmd_Exec(expr->value.str, &errfmt);
3605 if (errfmt != NULL)
3606 Error(errfmt, expr->value.str);
3607 Expr_SetValueOwn(expr, output);
3608 }
3609
3610 return AMR_OK;
3611 }
3612 #endif
3613
3614 static void
3615 LogBeforeApply(const ApplyModifiersState *st, const char *mod)
3616 {
3617 const Expr *expr = st->expr;
3618 char vflags_str[VarFlags_ToStringSize];
3619 Boolean is_single_char = mod[0] != '\0' && IsDelimiter(mod[1], st);
3620
3621 /* At this point, only the first character of the modifier can
3622 * be used since the end of the modifier is not yet known. */
3623 debug_printf("Applying ${%s:%c%s} to \"%s\" (%s, %s, %s)\n",
3624 expr->var->name.str, mod[0], is_single_char ? "" : "...",
3625 expr->value.str,
3626 VarEvalFlags_ToString(expr->eflags),
3627 VarFlags_ToString(vflags_str, expr->var->flags),
3628 ExprDefined_Name[expr->defined]);
3629 }
3630
3631 static void
3632 LogAfterApply(const ApplyModifiersState *st, const char *p, const char *mod)
3633 {
3634 const Expr *expr = st->expr;
3635 const char *value = expr->value.str;
3636 char vflags_str[VarFlags_ToStringSize];
3637 const char *quot = value == var_Error ? "" : "\"";
3638
3639 debug_printf("Result of ${%s:%.*s} is %s%s%s (%s, %s, %s)\n",
3640 expr->var->name.str, (int)(p - mod), mod,
3641 quot, value == var_Error ? "error" : value, quot,
3642 VarEvalFlags_ToString(expr->eflags),
3643 VarFlags_ToString(vflags_str, expr->var->flags),
3644 ExprDefined_Name[expr->defined]);
3645 }
3646
3647 static ApplyModifierResult
3648 ApplyModifier(const char **pp, ApplyModifiersState *st)
3649 {
3650 switch (**pp) {
3651 case '!':
3652 return ApplyModifier_ShellCommand(pp, st);
3653 case ':':
3654 return ApplyModifier_Assign(pp, st);
3655 case '?':
3656 return ApplyModifier_IfElse(pp, st);
3657 case '@':
3658 return ApplyModifier_Loop(pp, st);
3659 case '[':
3660 return ApplyModifier_Words(pp, st);
3661 case '_':
3662 return ApplyModifier_Remember(pp, st);
3663 #ifndef NO_REGEX
3664 case 'C':
3665 return ApplyModifier_Regex(pp, st);
3666 #endif
3667 case 'D':
3668 return ApplyModifier_Defined(pp, st);
3669 case 'E':
3670 return ApplyModifier_WordFunc(pp, st, ModifyWord_Suffix);
3671 case 'g':
3672 return ApplyModifier_Gmtime(pp, st);
3673 case 'H':
3674 return ApplyModifier_WordFunc(pp, st, ModifyWord_Head);
3675 case 'h':
3676 return ApplyModifier_Hash(pp, st);
3677 case 'L':
3678 return ApplyModifier_Literal(pp, st);
3679 case 'l':
3680 return ApplyModifier_Localtime(pp, st);
3681 case 'M':
3682 case 'N':
3683 return ApplyModifier_Match(pp, st);
3684 case 'O':
3685 return ApplyModifier_Order(pp, st);
3686 case 'P':
3687 return ApplyModifier_Path(pp, st);
3688 case 'Q':
3689 case 'q':
3690 return ApplyModifier_Quote(pp, st);
3691 case 'R':
3692 return ApplyModifier_WordFunc(pp, st, ModifyWord_Root);
3693 case 'r':
3694 return ApplyModifier_Range(pp, st);
3695 case 'S':
3696 return ApplyModifier_Subst(pp, st);
3697 #ifdef SUNSHCMD
3698 case 's':
3699 return ApplyModifier_SunShell(pp, st);
3700 #endif
3701 case 'T':
3702 return ApplyModifier_WordFunc(pp, st, ModifyWord_Tail);
3703 case 't':
3704 return ApplyModifier_To(pp, st);
3705 case 'U':
3706 return ApplyModifier_Defined(pp, st);
3707 case 'u':
3708 return ApplyModifier_Unique(pp, st);
3709 default:
3710 return AMR_UNKNOWN;
3711 }
3712 }
3713
3714 static void ApplyModifiers(Expr *, const char **, char, char);
3715
3716 typedef enum ApplyModifiersIndirectResult {
3717 /* The indirect modifiers have been applied successfully. */
3718 AMIR_CONTINUE,
3719 /* Fall back to the SysV modifier. */
3720 AMIR_SYSV,
3721 /* Error out. */
3722 AMIR_OUT
3723 } ApplyModifiersIndirectResult;
3724
3725 /*
3726 * While expanding a variable expression, expand and apply indirect modifiers,
3727 * such as in ${VAR:${M_indirect}}.
3728 *
3729 * All indirect modifiers of a group must come from a single variable
3730 * expression. ${VAR:${M1}} is valid but ${VAR:${M1}${M2}} is not.
3731 *
3732 * Multiple groups of indirect modifiers can be chained by separating them
3733 * with colons. ${VAR:${M1}:${M2}} contains 2 indirect modifiers.
3734 *
3735 * If the variable expression is not followed by st->endc or ':', fall
3736 * back to trying the SysV modifier, such as in ${VAR:${FROM}=${TO}}.
3737 */
3738 static ApplyModifiersIndirectResult
3739 ApplyModifiersIndirect(ApplyModifiersState *st, const char **pp)
3740 {
3741 Expr *expr = st->expr;
3742 const char *p = *pp;
3743 FStr mods;
3744
3745 (void)Var_Parse(&p, expr->scope, expr->eflags, &mods);
3746 /* TODO: handle errors */
3747
3748 if (mods.str[0] != '\0' && *p != '\0' && !IsDelimiter(*p, st)) {
3749 FStr_Done(&mods);
3750 return AMIR_SYSV;
3751 }
3752
3753 DEBUG3(VAR, "Indirect modifier \"%s\" from \"%.*s\"\n",
3754 mods.str, (int)(p - *pp), *pp);
3755
3756 if (mods.str[0] != '\0') {
3757 const char *modsp = mods.str;
3758 ApplyModifiers(expr, &modsp, '\0', '\0');
3759 if (expr->value.str == var_Error || *modsp != '\0') {
3760 FStr_Done(&mods);
3761 *pp = p;
3762 return AMIR_OUT; /* error already reported */
3763 }
3764 }
3765 FStr_Done(&mods);
3766
3767 if (*p == ':')
3768 p++;
3769 else if (*p == '\0' && st->endc != '\0') {
3770 Error("Unclosed variable expression after indirect "
3771 "modifier, expecting '%c' for variable \"%s\"",
3772 st->endc, expr->var->name.str);
3773 *pp = p;
3774 return AMIR_OUT;
3775 }
3776
3777 *pp = p;
3778 return AMIR_CONTINUE;
3779 }
3780
3781 static ApplyModifierResult
3782 ApplySingleModifier(const char **pp, ApplyModifiersState *st)
3783 {
3784 ApplyModifierResult res;
3785 const char *mod = *pp;
3786 const char *p = *pp;
3787
3788 if (DEBUG(VAR))
3789 LogBeforeApply(st, mod);
3790
3791 res = ApplyModifier(&p, st);
3792
3793 #ifdef SYSVVARSUB
3794 if (res == AMR_UNKNOWN) {
3795 assert(p == mod);
3796 res = ApplyModifier_SysV(&p, st);
3797 }
3798 #endif
3799
3800 if (res == AMR_UNKNOWN) {
3801 /*
3802 * Guess the end of the current modifier.
3803 * XXX: Skipping the rest of the modifier hides
3804 * errors and leads to wrong results.
3805 * Parsing should rather stop here.
3806 */
3807 for (p++; !IsDelimiter(*p, st) && *p != '\0'; p++)
3808 continue;
3809 Parse_Error(PARSE_FATAL, "Unknown modifier \"%.*s\"",
3810 (int)(p - mod), mod);
3811 Expr_SetValueRefer(st->expr, var_Error);
3812 }
3813 if (res == AMR_CLEANUP || res == AMR_BAD) {
3814 *pp = p;
3815 return res;
3816 }
3817
3818 if (DEBUG(VAR))
3819 LogAfterApply(st, p, mod);
3820
3821 if (*p == '\0' && st->endc != '\0') {
3822 Error(
3823 "Unclosed variable expression, expecting '%c' for "
3824 "modifier \"%.*s\" of variable \"%s\" with value \"%s\"",
3825 st->endc,
3826 (int)(p - mod), mod,
3827 st->expr->var->name.str, st->expr->value.str);
3828 } else if (*p == ':') {
3829 p++;
3830 } else if (opts.strict && *p != '\0' && *p != st->endc) {
3831 Parse_Error(PARSE_FATAL,
3832 "Missing delimiter ':' after modifier \"%.*s\"",
3833 (int)(p - mod), mod);
3834 /*
3835 * TODO: propagate parse error to the enclosing
3836 * expression
3837 */
3838 }
3839 *pp = p;
3840 return AMR_OK;
3841 }
3842
3843 /* Apply any modifiers (such as :Mpattern or :@var@loop@ or :Q or ::=value). */
3844 static void
3845 ApplyModifiers(
3846 Expr *expr,
3847 const char **pp, /* the parsing position, updated upon return */
3848 char startc, /* '(' or '{'; or '\0' for indirect modifiers */
3849 char endc /* ')' or '}'; or '\0' for indirect modifiers */
3850 )
3851 {
3852 ApplyModifiersState st = {
3853 expr,
3854 startc,
3855 endc,
3856 ' ', /* .sep */
3857 FALSE /* .oneBigWord */
3858 };
3859 const char *p;
3860 const char *mod;
3861
3862 assert(startc == '(' || startc == '{' || startc == '\0');
3863 assert(endc == ')' || endc == '}' || endc == '\0');
3864 assert(expr->value.str != NULL);
3865
3866 p = *pp;
3867
3868 if (*p == '\0' && endc != '\0') {
3869 Error(
3870 "Unclosed variable expression (expecting '%c') for \"%s\"",
3871 st.endc, expr->var->name.str);
3872 goto cleanup;
3873 }
3874
3875 while (*p != '\0' && *p != endc) {
3876 ApplyModifierResult res;
3877
3878 if (*p == '$') {
3879 ApplyModifiersIndirectResult amir =
3880 ApplyModifiersIndirect(&st, &p);
3881 if (amir == AMIR_CONTINUE)
3882 continue;
3883 if (amir == AMIR_OUT)
3884 break;
3885 /*
3886 * It's neither '${VAR}:' nor '${VAR}}'. Try to parse
3887 * it as a SysV modifier, as that is the only modifier
3888 * that can start with '$'.
3889 */
3890 }
3891
3892 mod = p;
3893
3894 res = ApplySingleModifier(&p, &st);
3895 if (res == AMR_CLEANUP)
3896 goto cleanup;
3897 if (res == AMR_BAD)
3898 goto bad_modifier;
3899 }
3900
3901 *pp = p;
3902 assert(expr->value.str != NULL); /* Use var_Error or varUndefined. */
3903 return;
3904
3905 bad_modifier:
3906 /* XXX: The modifier end is only guessed. */
3907 Error("Bad modifier \":%.*s\" for variable \"%s\"",
3908 (int)strcspn(mod, ":)}"), mod, expr->var->name.str);
3909
3910 cleanup:
3911 /*
3912 * TODO: Use p + strlen(p) instead, to stop parsing immediately.
3913 *
3914 * In the unit tests, this generates a few unterminated strings in the
3915 * shell commands though. Instead of producing these unfinished
3916 * strings, commands with evaluation errors should not be run at all.
3917 *
3918 * To make that happen, Var_Subst must report the actual errors
3919 * instead of returning VPR_OK unconditionally.
3920 */
3921 *pp = p;
3922 Expr_SetValueRefer(expr, var_Error);
3923 }
3924
3925 /*
3926 * Only four of the local variables are treated specially as they are the
3927 * only four that will be set when dynamic sources are expanded.
3928 */
3929 static Boolean
3930 VarnameIsDynamic(const char *name, size_t len)
3931 {
3932 if (len == 1 || (len == 2 && (name[1] == 'F' || name[1] == 'D'))) {
3933 switch (name[0]) {
3934 case '@':
3935 case '%':
3936 case '*':
3937 case '!':
3938 return TRUE;
3939 }
3940 return FALSE;
3941 }
3942
3943 if ((len == 7 || len == 8) && name[0] == '.' && ch_isupper(name[1])) {
3944 return strcmp(name, ".TARGET") == 0 ||
3945 strcmp(name, ".ARCHIVE") == 0 ||
3946 strcmp(name, ".PREFIX") == 0 ||
3947 strcmp(name, ".MEMBER") == 0;
3948 }
3949
3950 return FALSE;
3951 }
3952
3953 static const char *
3954 UndefinedShortVarValue(char varname, const GNode *scope)
3955 {
3956 if (scope == SCOPE_CMDLINE || scope == SCOPE_GLOBAL) {
3957 /*
3958 * If substituting a local variable in a non-local scope,
3959 * assume it's for dynamic source stuff. We have to handle
3960 * this specially and return the longhand for the variable
3961 * with the dollar sign escaped so it makes it back to the
3962 * caller. Only four of the local variables are treated
3963 * specially as they are the only four that will be set
3964 * when dynamic sources are expanded.
3965 */
3966 switch (varname) {
3967 case '@':
3968 return "$(.TARGET)";
3969 case '%':
3970 return "$(.MEMBER)";
3971 case '*':
3972 return "$(.PREFIX)";
3973 case '!':
3974 return "$(.ARCHIVE)";
3975 }
3976 }
3977 return NULL;
3978 }
3979
3980 /*
3981 * Parse a variable name, until the end character or a colon, whichever
3982 * comes first.
3983 */
3984 static char *
3985 ParseVarname(const char **pp, char startc, char endc,
3986 GNode *scope, VarEvalFlags eflags,
3987 size_t *out_varname_len)
3988 {
3989 Buffer buf;
3990 const char *p = *pp;
3991 int depth = 0; /* Track depth so we can spot parse errors. */
3992
3993 Buf_Init(&buf);
3994
3995 while (*p != '\0') {
3996 if ((*p == endc || *p == ':') && depth == 0)
3997 break;
3998 if (*p == startc)
3999 depth++;
4000 if (*p == endc)
4001 depth--;
4002
4003 /* A variable inside a variable, expand. */
4004 if (*p == '$') {
4005 FStr nested_val;
4006 (void)Var_Parse(&p, scope, eflags, &nested_val);
4007 /* TODO: handle errors */
4008 Buf_AddStr(&buf, nested_val.str);
4009 FStr_Done(&nested_val);
4010 } else {
4011 Buf_AddByte(&buf, *p);
4012 p++;
4013 }
4014 }
4015 *pp = p;
4016 *out_varname_len = buf.len;
4017 return Buf_DoneData(&buf);
4018 }
4019
4020 static VarParseResult
4021 ValidShortVarname(char varname, const char *start)
4022 {
4023 if (varname != '$' && varname != ':' && varname != '}' &&
4024 varname != ')' && varname != '\0')
4025 return VPR_OK;
4026
4027 if (!opts.strict)
4028 return VPR_ERR; /* XXX: Missing error message */
4029
4030 if (varname == '$')
4031 Parse_Error(PARSE_FATAL,
4032 "To escape a dollar, use \\$, not $$, at \"%s\"", start);
4033 else if (varname == '\0')
4034 Parse_Error(PARSE_FATAL, "Dollar followed by nothing");
4035 else
4036 Parse_Error(PARSE_FATAL,
4037 "Invalid variable name '%c', at \"%s\"", varname, start);
4038
4039 return VPR_ERR;
4040 }
4041
4042 /*
4043 * Parse a single-character variable name such as in $V or $@.
4044 * Return whether to continue parsing.
4045 */
4046 static Boolean
4047 ParseVarnameShort(char varname, const char **pp, GNode *scope,
4048 VarEvalFlags eflags,
4049 VarParseResult *out_FALSE_res, const char **out_FALSE_val,
4050 Var **out_TRUE_var)
4051 {
4052 char name[2];
4053 Var *v;
4054 VarParseResult vpr;
4055
4056 vpr = ValidShortVarname(varname, *pp);
4057 if (vpr != VPR_OK) {
4058 (*pp)++;
4059 *out_FALSE_res = vpr;
4060 *out_FALSE_val = var_Error;
4061 return FALSE;
4062 }
4063
4064 name[0] = varname;
4065 name[1] = '\0';
4066 v = VarFind(name, scope, TRUE);
4067 if (v == NULL) {
4068 const char *val;
4069 *pp += 2;
4070
4071 val = UndefinedShortVarValue(varname, scope);
4072 if (val == NULL)
4073 val = eflags.undefErr ? var_Error : varUndefined;
4074
4075 if (opts.strict && val == var_Error) {
4076 Parse_Error(PARSE_FATAL,
4077 "Variable \"%s\" is undefined", name);
4078 *out_FALSE_res = VPR_ERR;
4079 *out_FALSE_val = val;
4080 return FALSE;
4081 }
4082
4083 /*
4084 * XXX: This looks completely wrong.
4085 *
4086 * If undefined expressions are not allowed, this should
4087 * rather be VPR_ERR instead of VPR_UNDEF, together with an
4088 * error message.
4089 *
4090 * If undefined expressions are allowed, this should rather
4091 * be VPR_UNDEF instead of VPR_OK.
4092 */
4093 *out_FALSE_res = eflags.undefErr ? VPR_UNDEF : VPR_OK;
4094 *out_FALSE_val = val;
4095 return FALSE;
4096 }
4097
4098 *out_TRUE_var = v;
4099 return TRUE;
4100 }
4101
4102 /* Find variables like @F or <D. */
4103 static Var *
4104 FindLocalLegacyVar(const char *varname, size_t namelen, GNode *scope,
4105 const char **out_extraModifiers)
4106 {
4107 /* Only resolve these variables if scope is a "real" target. */
4108 if (scope == SCOPE_CMDLINE || scope == SCOPE_GLOBAL)
4109 return NULL;
4110
4111 if (namelen != 2)
4112 return NULL;
4113 if (varname[1] != 'F' && varname[1] != 'D')
4114 return NULL;
4115 if (strchr("@%?*!<>", varname[0]) == NULL)
4116 return NULL;
4117
4118 {
4119 char name[] = { varname[0], '\0' };
4120 Var *v = VarFind(name, scope, FALSE);
4121
4122 if (v != NULL) {
4123 if (varname[1] == 'D') {
4124 *out_extraModifiers = "H:";
4125 } else { /* F */
4126 *out_extraModifiers = "T:";
4127 }
4128 }
4129 return v;
4130 }
4131 }
4132
4133 static VarParseResult
4134 EvalUndefined(Boolean dynamic, const char *start, const char *p, char *varname,
4135 VarEvalFlags eflags,
4136 FStr *out_val)
4137 {
4138 if (dynamic) {
4139 *out_val = FStr_InitOwn(bmake_strsedup(start, p));
4140 free(varname);
4141 return VPR_OK;
4142 }
4143
4144 if (eflags.undefErr && opts.strict) {
4145 Parse_Error(PARSE_FATAL,
4146 "Variable \"%s\" is undefined", varname);
4147 free(varname);
4148 *out_val = FStr_InitRefer(var_Error);
4149 return VPR_ERR;
4150 }
4151
4152 if (eflags.undefErr) {
4153 free(varname);
4154 *out_val = FStr_InitRefer(var_Error);
4155 return VPR_UNDEF; /* XXX: Should be VPR_ERR instead. */
4156 }
4157
4158 free(varname);
4159 *out_val = FStr_InitRefer(varUndefined);
4160 return VPR_OK;
4161 }
4162
4163 /*
4164 * Parse a long variable name enclosed in braces or parentheses such as $(VAR)
4165 * or ${VAR}, up to the closing brace or parenthesis, or in the case of
4166 * ${VAR:Modifiers}, up to the ':' that starts the modifiers.
4167 * Return whether to continue parsing.
4168 */
4169 static Boolean
4170 ParseVarnameLong(
4171 const char *p,
4172 char startc,
4173 GNode *scope,
4174 VarEvalFlags eflags,
4175
4176 const char **out_FALSE_pp,
4177 VarParseResult *out_FALSE_res,
4178 FStr *out_FALSE_val,
4179
4180 char *out_TRUE_endc,
4181 const char **out_TRUE_p,
4182 Var **out_TRUE_v,
4183 Boolean *out_TRUE_haveModifier,
4184 const char **out_TRUE_extraModifiers,
4185 Boolean *out_TRUE_dynamic,
4186 ExprDefined *out_TRUE_exprDefined
4187 )
4188 {
4189 size_t namelen;
4190 char *varname;
4191 Var *v;
4192 Boolean haveModifier;
4193 Boolean dynamic = FALSE;
4194
4195 const char *const start = p;
4196 char endc = startc == '(' ? ')' : '}';
4197
4198 p += 2; /* skip "${" or "$(" or "y(" */
4199 varname = ParseVarname(&p, startc, endc, scope, eflags, &namelen);
4200
4201 if (*p == ':') {
4202 haveModifier = TRUE;
4203 } else if (*p == endc) {
4204 haveModifier = FALSE;
4205 } else {
4206 Parse_Error(PARSE_FATAL, "Unclosed variable \"%s\"", varname);
4207 free(varname);
4208 *out_FALSE_pp = p;
4209 *out_FALSE_val = FStr_InitRefer(var_Error);
4210 *out_FALSE_res = VPR_ERR;
4211 return FALSE;
4212 }
4213
4214 v = VarFind(varname, scope, TRUE);
4215
4216 /* At this point, p points just after the variable name,
4217 * either at ':' or at endc. */
4218
4219 if (v == NULL) {
4220 v = FindLocalLegacyVar(varname, namelen, scope,
4221 out_TRUE_extraModifiers);
4222 }
4223
4224 if (v == NULL) {
4225 /*
4226 * Defer expansion of dynamic variables if they appear in
4227 * non-local scope since they are not defined there.
4228 */
4229 dynamic = VarnameIsDynamic(varname, namelen) &&
4230 (scope == SCOPE_CMDLINE || scope == SCOPE_GLOBAL);
4231
4232 if (!haveModifier) {
4233 p++; /* skip endc */
4234 *out_FALSE_pp = p;
4235 *out_FALSE_res = EvalUndefined(dynamic, start, p,
4236 varname, eflags, out_FALSE_val);
4237 return FALSE;
4238 }
4239
4240 /*
4241 * The variable expression is based on an undefined variable.
4242 * Nevertheless it needs a Var, for modifiers that access the
4243 * variable name, such as :L or :?.
4244 *
4245 * Most modifiers leave this expression in the "undefined"
4246 * state (VES_UNDEF), only a few modifiers like :D, :U, :L,
4247 * :P turn this undefined expression into a defined
4248 * expression (VES_DEF).
4249 *
4250 * In the end, after applying all modifiers, if the expression
4251 * is still undefined, Var_Parse will return an empty string
4252 * instead of the actually computed value.
4253 */
4254 v = VarNew(FStr_InitOwn(varname), "", VFL_NONE);
4255 *out_TRUE_exprDefined = DEF_UNDEF;
4256 } else
4257 free(varname);
4258
4259 *out_TRUE_endc = endc;
4260 *out_TRUE_p = p;
4261 *out_TRUE_v = v;
4262 *out_TRUE_haveModifier = haveModifier;
4263 *out_TRUE_dynamic = dynamic;
4264 return TRUE;
4265 }
4266
4267 /* Free the environment variable now since we own it. */
4268 static void
4269 FreeEnvVar(Var *v, FStr *inout_val)
4270 {
4271 char *varValue = Buf_DoneData(&v->val);
4272 if (inout_val->str == varValue)
4273 inout_val->freeIt = varValue;
4274 else
4275 free(varValue);
4276
4277 FStr_Done(&v->name);
4278 free(v);
4279 }
4280
4281 /*
4282 * Given the start of a variable expression (such as $v, $(VAR),
4283 * ${VAR:Mpattern}), extract the variable name and value, and the modifiers,
4284 * if any. While doing that, apply the modifiers to the value of the
4285 * expression, forming its final value. A few of the modifiers such as :!cmd!
4286 * or ::= have side effects.
4287 *
4288 * Input:
4289 * *pp The string to parse.
4290 * When parsing a condition in ParseEmptyArg, it may also
4291 * point to the "y" of "empty(VARNAME:Modifiers)", which
4292 * is syntactically the same.
4293 * scope The scope for finding variables
4294 * eflags Control the exact details of parsing
4295 *
4296 * Output:
4297 * *pp The position where to continue parsing.
4298 * TODO: After a parse error, the value of *pp is
4299 * unspecified. It may not have been updated at all,
4300 * point to some random character in the string, to the
4301 * location of the parse error, or at the end of the
4302 * string.
4303 * *out_val The value of the variable expression, never NULL.
4304 * *out_val var_Error if there was a parse error.
4305 * *out_val var_Error if the base variable of the expression was
4306 * undefined, eflags has undefErr set, and none of
4307 * the modifiers turned the undefined expression into a
4308 * defined expression.
4309 * XXX: It is not guaranteed that an error message has
4310 * been printed.
4311 * *out_val varUndefined if the base variable of the expression
4312 * was undefined, eflags did not have undefErr set,
4313 * and none of the modifiers turned the undefined
4314 * expression into a defined expression.
4315 * XXX: It is not guaranteed that an error message has
4316 * been printed.
4317 */
4318 VarParseResult
4319 Var_Parse(const char **pp, GNode *scope, VarEvalFlags eflags, FStr *out_val)
4320 {
4321 const char *p = *pp;
4322 const char *const start = p;
4323 /* TRUE if have modifiers for the variable. */
4324 Boolean haveModifier;
4325 /* Starting character if variable in parens or braces. */
4326 char startc;
4327 /* Ending character if variable in parens or braces. */
4328 char endc;
4329 /*
4330 * TRUE if the variable is local and we're expanding it in a
4331 * non-local scope. This is done to support dynamic sources.
4332 * The result is just the expression, unaltered.
4333 */
4334 Boolean dynamic;
4335 const char *extramodifiers;
4336 Var *v;
4337
4338 Expr expr = {
4339 NULL,
4340 #if defined(lint)
4341 /* NetBSD lint cannot fully parse C99 struct initializers. */
4342 { NULL, NULL },
4343 #else
4344 FStr_InitRefer(NULL),
4345 #endif
4346 eflags,
4347 scope,
4348 DEF_REGULAR
4349 };
4350
4351 DEBUG2(VAR, "Var_Parse: %s (%s)\n", start,
4352 VarEvalFlags_ToString(eflags));
4353
4354 *out_val = FStr_InitRefer(NULL);
4355 extramodifiers = NULL; /* extra modifiers to apply first */
4356 dynamic = FALSE;
4357
4358 /*
4359 * Appease GCC, which thinks that the variable might not be
4360 * initialized.
4361 */
4362 endc = '\0';
4363
4364 startc = p[1];
4365 if (startc != '(' && startc != '{') {
4366 VarParseResult res;
4367 if (!ParseVarnameShort(startc, pp, scope, eflags, &res,
4368 &out_val->str, &expr.var))
4369 return res;
4370 haveModifier = FALSE;
4371 p++;
4372 } else {
4373 VarParseResult res;
4374 if (!ParseVarnameLong(p, startc, scope, eflags,
4375 pp, &res, out_val,
4376 &endc, &p, &expr.var, &haveModifier, &extramodifiers,
4377 &dynamic, &expr.defined))
4378 return res;
4379 }
4380
4381 v = expr.var;
4382 if (v->flags & VFL_IN_USE)
4383 Fatal("Variable %s is recursive.", v->name.str);
4384
4385 /*
4386 * XXX: This assignment creates an alias to the current value of the
4387 * variable. This means that as long as the value of the expression
4388 * stays the same, the value of the variable must not change.
4389 * Using the '::=' modifier, it could be possible to do exactly this.
4390 * At the bottom of this function, the resulting value is compared to
4391 * the then-current value of the variable. This might also invoke
4392 * undefined behavior.
4393 */
4394 expr.value = FStr_InitRefer(v->val.data);
4395
4396 /*
4397 * Before applying any modifiers, expand any nested expressions from
4398 * the variable value.
4399 */
4400 if (strchr(expr.value.str, '$') != NULL && eflags.wantRes) {
4401 char *expanded;
4402 VarEvalFlags nested_eflags = eflags;
4403 if (opts.strict)
4404 nested_eflags.undefErr = FALSE;
4405 v->flags |= VFL_IN_USE;
4406 (void)Var_Subst(expr.value.str, scope, nested_eflags,
4407 &expanded);
4408 v->flags &= ~(unsigned)VFL_IN_USE;
4409 /* TODO: handle errors */
4410 Expr_SetValueOwn(&expr, expanded);
4411 }
4412
4413 if (extramodifiers != NULL) {
4414 const char *em = extramodifiers;
4415 ApplyModifiers(&expr, &em, '\0', '\0');
4416 }
4417
4418 if (haveModifier) {
4419 p++; /* Skip initial colon. */
4420 ApplyModifiers(&expr, &p, startc, endc);
4421 }
4422
4423 if (*p != '\0') /* Skip past endc if possible. */
4424 p++;
4425
4426 *pp = p;
4427
4428 if (v->flags & VFL_FROM_ENV) {
4429 FreeEnvVar(v, &expr.value);
4430
4431 } else if (expr.defined != DEF_REGULAR) {
4432 if (expr.defined == DEF_UNDEF) {
4433 if (dynamic) {
4434 Expr_SetValueOwn(&expr,
4435 bmake_strsedup(start, p));
4436 } else {
4437 /*
4438 * The expression is still undefined,
4439 * therefore discard the actual value and
4440 * return an error marker instead.
4441 */
4442 Expr_SetValueRefer(&expr,
4443 eflags.undefErr
4444 ? var_Error : varUndefined);
4445 }
4446 }
4447 /* XXX: This is not standard memory management. */
4448 if (expr.value.str != v->val.data)
4449 Buf_Done(&v->val);
4450 FStr_Done(&v->name);
4451 free(v);
4452 }
4453 *out_val = expr.value;
4454 return VPR_OK; /* XXX: Is not correct in all cases */
4455 }
4456
4457 static void
4458 VarSubstDollarDollar(const char **pp, Buffer *res, VarEvalFlags eflags)
4459 {
4460 /* A dollar sign may be escaped with another dollar sign. */
4461 if (save_dollars && eflags.keepDollar)
4462 Buf_AddByte(res, '$');
4463 Buf_AddByte(res, '$');
4464 *pp += 2;
4465 }
4466
4467 static void
4468 VarSubstExpr(const char **pp, Buffer *buf, GNode *scope,
4469 VarEvalFlags eflags, Boolean *inout_errorReported)
4470 {
4471 const char *p = *pp;
4472 const char *nested_p = p;
4473 FStr val;
4474
4475 (void)Var_Parse(&nested_p, scope, eflags, &val);
4476 /* TODO: handle errors */
4477
4478 if (val.str == var_Error || val.str == varUndefined) {
4479 if (!eflags.keepUndef) {
4480 p = nested_p;
4481 } else if (eflags.undefErr || val.str == var_Error) {
4482
4483 /*
4484 * XXX: This condition is wrong. If val == var_Error,
4485 * this doesn't necessarily mean there was an undefined
4486 * variable. It could equally well be a parse error;
4487 * see unit-tests/varmod-order.exp.
4488 */
4489
4490 /*
4491 * If variable is undefined, complain and skip the
4492 * variable. The complaint will stop us from doing
4493 * anything when the file is parsed.
4494 */
4495 if (!*inout_errorReported) {
4496 Parse_Error(PARSE_FATAL,
4497 "Undefined variable \"%.*s\"",
4498 (int)(size_t)(nested_p - p), p);
4499 }
4500 p = nested_p;
4501 *inout_errorReported = TRUE;
4502 } else {
4503 /* Copy the initial '$' of the undefined expression,
4504 * thereby deferring expansion of the expression, but
4505 * expand nested expressions if already possible.
4506 * See unit-tests/varparse-undef-partial.mk. */
4507 Buf_AddByte(buf, *p);
4508 p++;
4509 }
4510 } else {
4511 p = nested_p;
4512 Buf_AddStr(buf, val.str);
4513 }
4514
4515 FStr_Done(&val);
4516
4517 *pp = p;
4518 }
4519
4520 /*
4521 * Skip as many characters as possible -- either to the end of the string
4522 * or to the next dollar sign (variable expression).
4523 */
4524 static void
4525 VarSubstPlain(const char **pp, Buffer *res)
4526 {
4527 const char *p = *pp;
4528 const char *start = p;
4529
4530 for (p++; *p != '$' && *p != '\0'; p++)
4531 continue;
4532 Buf_AddBytesBetween(res, start, p);
4533 *pp = p;
4534 }
4535
4536 /*
4537 * Expand all variable expressions like $V, ${VAR}, $(VAR:Modifiers) in the
4538 * given string.
4539 *
4540 * Input:
4541 * str The string in which the variable expressions are
4542 * expanded.
4543 * scope The scope in which to start searching for
4544 * variables. The other scopes are searched as well.
4545 * eflags Special effects during expansion.
4546 */
4547 VarParseResult
4548 Var_Subst(const char *str, GNode *scope, VarEvalFlags eflags, char **out_res)
4549 {
4550 const char *p = str;
4551 Buffer res;
4552
4553 /* Set true if an error has already been reported,
4554 * to prevent a plethora of messages when recursing */
4555 /* XXX: Why is the 'static' necessary here? */
4556 static Boolean errorReported;
4557
4558 Buf_Init(&res);
4559 errorReported = FALSE;
4560
4561 while (*p != '\0') {
4562 if (p[0] == '$' && p[1] == '$')
4563 VarSubstDollarDollar(&p, &res, eflags);
4564 else if (p[0] == '$')
4565 VarSubstExpr(&p, &res, scope, eflags, &errorReported);
4566 else
4567 VarSubstPlain(&p, &res);
4568 }
4569
4570 *out_res = Buf_DoneDataCompact(&res);
4571 return VPR_OK;
4572 }
4573
4574 /* Initialize the variables module. */
4575 void
4576 Var_Init(void)
4577 {
4578 SCOPE_INTERNAL = GNode_New("Internal");
4579 SCOPE_GLOBAL = GNode_New("Global");
4580 SCOPE_CMDLINE = GNode_New("Command");
4581 }
4582
4583 /* Clean up the variables module. */
4584 void
4585 Var_End(void)
4586 {
4587 Var_Stats();
4588 }
4589
4590 void
4591 Var_Stats(void)
4592 {
4593 HashTable_DebugStats(&SCOPE_GLOBAL->vars, "Global variables");
4594 }
4595
4596 /* Print all variables in a scope, sorted by name. */
4597 void
4598 Var_Dump(GNode *scope)
4599 {
4600 Vector /* of const char * */ vec;
4601 HashIter hi;
4602 size_t i;
4603 const char **varnames;
4604
4605 Vector_Init(&vec, sizeof(const char *));
4606
4607 HashIter_Init(&hi, &scope->vars);
4608 while (HashIter_Next(&hi) != NULL)
4609 *(const char **)Vector_Push(&vec) = hi.entry->key;
4610 varnames = vec.items;
4611
4612 qsort(varnames, vec.len, sizeof varnames[0], str_cmp_asc);
4613
4614 for (i = 0; i < vec.len; i++) {
4615 const char *varname = varnames[i];
4616 Var *var = HashTable_FindValue(&scope->vars, varname);
4617 debug_printf("%-16s = %s\n", varname, var->val.data);
4618 }
4619
4620 Vector_Done(&vec);
4621 }
4622