1 /* args.c - Command line argument parsing. 2 * 3 * Copyright 2006 Rob Landley <rob (at) landley.net> 4 */ 5 6 #include "toys.h" 7 8 // Design goals: 9 // Don't use getopt() out of libc. 10 // Don't permute original arguments (screwing up ps/top output). 11 // Integrated --long options "(noshort)a(along)b(blong1)(blong2)" 12 13 /* This uses a getopt-like option string, but not getopt() itself. We call 14 * it the get_opt string. 15 * 16 * Each option in the get_opt string corresponds to a bit position in the 17 * return value. The rightmost argument is (1<<0), the next to last is (1<<1) 18 * and so on. If the option isn't seen in argv[], its bit remains 0. 19 * 20 * Options which have an argument fill in the corresponding slot in the global 21 * union "this" (see generated/globals.h), which it treats as an array of longs 22 * (note that sizeof(long)==sizeof(pointer) is guaranteed by LP64). 23 * 24 * You don't have to free the option strings, which point into the environment 25 * space. List objects should be freed by main() when command_main() returns. 26 * 27 * Example: 28 * Calling get_optflags() when toys.which->options="ab:c:d" and 29 * argv = ["command", "-b", "fruit", "-d", "walrus"] results in: 30 * 31 * Changes to struct toys: 32 * toys.optflags = 5 (I.E. 0101 so -b = 4 | -d = 1) 33 * toys.optargs[0] = "walrus" (leftover argument) 34 * toys.optargs[1] = NULL (end of list) 35 * toys.optc = 1 (there was 1 leftover argument) 36 * 37 * Changes to union this: 38 * this[0]=NULL (because -c didn't get an argument this time) 39 * this[1]="fruit" (argument to -b) 40 */ 41 42 // Enabling TOYBOX_DEBUG in .config adds syntax checks to option string parsing 43 // which aren't needed in the final code (your option string is hardwired and 44 // should be correct when you ship), but are useful for development. 45 46 // What you can put in a get_opt string: 47 // Any otherwise unused character (all letters, unprefixed numbers) specify 48 // an option that sets a flag. The bit value is the same as the binary digit 49 // if you string the option characters together in order. 50 // So in "abcdefgh" a = 128, h = 1 51 // 52 // Suffixes specify that this option takes an argument (stored in GLOBALS): 53 // Note that pointer and long are always the same size, even on 64 bit. 54 // : plus a string argument, keep most recent if more than one 55 // * plus a string argument, appended to a list 56 // # plus a signed long argument 57 // <LOW - die if less than LOW 58 // >HIGH - die if greater than HIGH 59 // =DEFAULT - value if not specified 60 // - plus a signed long argument defaulting to negative (say + for positive) 61 // . plus a double precision floating point argument (with CFG_TOYBOX_FLOAT) 62 // Chop this option out with USE_TOYBOX_FLOAT() in option string 63 // Same <LOW>HIGH=DEFAULT as # 64 // @ plus an occurrence counter (which is a long) 65 // (longopt) 66 // | this is required. If more than one marked, only one required. 67 // ; long option's argument is optional (can only be supplied with --opt=) 68 // ^ Stop parsing after encountering this argument 69 // " " (space char) the "plus an argument" must be separate 70 // I.E. "-j 3" not "-j3". So "kill -stop" != "kill -s top" 71 // 72 // At the beginning of the get_opt string (before any options): 73 // ^ stop at first nonoption argument 74 // <0 die if less than # leftover arguments (default 0) 75 // >9 die if > # leftover arguments (default MAX_INT) 76 // ? Allow unknown arguments (pass them through to command). 77 // & first argument has imaginary dash (ala tar/ps) 78 // If given twice, all arguments have imaginary dash 79 // 80 // At the end: [groups] of previously seen options 81 // - Only one in group (switch off) [-abc] means -ab=-b, -ba=-a, -abc=-c 82 // + Synonyms (switch on all) [+abc] means -ab=-abc, -c=-abc 83 // ! More than one in group is error [!abc] means -ab calls error_exit() 84 // primarily useful if you can switch things back off again. 85 86 // Notes from getopt man page 87 // - and -- cannot be arguments. 88 // -- force end of arguments 89 // - is a synonym for stdin in file arguments 90 // -abcd means -a -b -c -d (but if -b takes an argument, then it's -a -b cd) 91 92 // Linked list of all known options (option string parsed into this). 93 // Hangs off getoptflagstate, freed at end of option parsing. 94 struct opts { 95 struct opts *next; 96 long *arg; // Pointer into union "this" to store arguments at. 97 int c; // Argument character to match 98 int flags; // |=1, ^=2 99 unsigned dex[3]; // which bits to disable/enable/exclude in toys.optflags 100 char type; // Type of arguments to store union "this" 101 union { 102 long l; 103 FLOAT f; 104 } val[3]; // low, high, default - range of allowed values 105 }; 106 107 // linked list of long options. (Hangs off getoptflagstate, free at end of 108 // option parsing, details about flag to set and global slot to fill out 109 // stored in related short option struct, but if opt->c = -1 the long option 110 // is "bare" (has no corresponding short option). 111 struct longopts { 112 struct longopts *next; 113 struct opts *opt; 114 char *str; 115 int len; 116 }; 117 118 // State during argument parsing. 119 struct getoptflagstate 120 { 121 int argc, minargs, maxargs, nodash; 122 char *arg; 123 struct opts *opts; 124 struct longopts *longopts; 125 int noerror, nodash_now, stopearly; 126 unsigned excludes, requires; 127 }; 128 129 // Use getoptflagstate to parse parse one command line option from argv 130 static int gotflag(struct getoptflagstate *gof, struct opts *opt) 131 { 132 int type; 133 134 // Did we recognize this option? 135 if (!opt) { 136 if (gof->noerror) return 1; 137 error_exit("Unknown option %s", gof->arg); 138 } 139 140 // Might enabling this switch off something else? 141 if (toys.optflags & opt->dex[0]) { 142 struct opts *clr; 143 unsigned i = 1; 144 145 // Forget saved argument for flag we switch back off 146 for (clr=gof->opts, i=1; clr; clr = clr->next, i<<=1) 147 if (clr->arg && (i & toys.optflags & opt->dex[0])) *clr->arg = 0; 148 toys.optflags &= ~opt->dex[0]; 149 } 150 151 // Set flags 152 toys.optflags |= opt->dex[1]; 153 gof->excludes |= opt->dex[2]; 154 if (opt->flags&2) gof->stopearly=2; 155 156 if (toys.optflags & gof->excludes) { 157 struct opts *bad; 158 unsigned i = 1; 159 160 for (bad=gof->opts, i=1; ;bad = bad->next, i<<=1) { 161 if (opt == bad || !(i & toys.optflags)) continue; 162 if (toys.optflags & bad->dex[2]) break; 163 } 164 error_exit("No '%c' with '%c'", opt->c, bad->c); 165 } 166 167 // Does this option take an argument? 168 if (!gof->arg) { 169 if (opt->flags & 8) return 0; 170 gof->arg = ""; 171 } else gof->arg++; 172 type = opt->type; 173 174 if (type == '@') ++*(opt->arg); 175 else if (type) { 176 char *arg = gof->arg; 177 178 // Handle "-xblah" and "-x blah", but also a third case: "abxc blah" 179 // to make "tar xCjfv blah1 blah2 thingy" work like 180 // "tar -x -C blah1 -j -f blah2 -v thingy" 181 182 if (gof->nodash_now || (!arg[0] && !(opt->flags & 8))) 183 arg = toys.argv[++gof->argc]; 184 if (!arg) { 185 char *s = "Missing argument to "; 186 struct longopts *lo; 187 188 if (opt->c != -1) error_exit("%s-%c", s, opt->c); 189 190 for (lo = gof->longopts; lo->opt != opt; lo = lo->next); 191 error_exit("%s--%.*s", s, lo->len, lo->str); 192 } 193 194 if (type == ':') *(opt->arg) = (long)arg; 195 else if (type == '*') { 196 struct arg_list **list; 197 198 list = (struct arg_list **)opt->arg; 199 while (*list) list=&((*list)->next); 200 *list = xzalloc(sizeof(struct arg_list)); 201 (*list)->arg = arg; 202 } else if (type == '#' || type == '-') { 203 long l = atolx(arg); 204 if (type == '-' && !ispunct(*arg)) l*=-1; 205 if (l < opt->val[0].l) error_exit("-%c < %ld", opt->c, opt->val[0].l); 206 if (l > opt->val[1].l) error_exit("-%c > %ld", opt->c, opt->val[1].l); 207 208 *(opt->arg) = l; 209 } else if (CFG_TOYBOX_FLOAT && type == '.') { 210 FLOAT *f = (FLOAT *)(opt->arg); 211 212 *f = strtod(arg, &arg); 213 if (opt->val[0].l != LONG_MIN && *f < opt->val[0].f) 214 error_exit("-%c < %lf", opt->c, (double)opt->val[0].f); 215 if (opt->val[1].l != LONG_MAX && *f > opt->val[1].f) 216 error_exit("-%c > %lf", opt->c, (double)opt->val[1].f); 217 } 218 219 if (!gof->nodash_now) gof->arg = ""; 220 } 221 222 return 0; 223 } 224 225 // Parse this command's options string into struct getoptflagstate, which 226 // includes a struct opts linked list in reverse order (I.E. right-to-left) 227 void parse_optflaglist(struct getoptflagstate *gof) 228 { 229 char *options = toys.which->options; 230 long *nextarg = (long *)&this; 231 struct opts *new = 0; 232 int idx; 233 234 // Parse option format string 235 memset(gof, 0, sizeof(struct getoptflagstate)); 236 gof->maxargs = INT_MAX; 237 if (!options) return; 238 239 // Parse leading special behavior indicators 240 for (;;) { 241 if (*options == '^') gof->stopearly++; 242 else if (*options == '<') gof->minargs=*(++options)-'0'; 243 else if (*options == '>') gof->maxargs=*(++options)-'0'; 244 else if (*options == '?') gof->noerror++; 245 else if (*options == '&') gof->nodash++; 246 else break; 247 options++; 248 } 249 250 // Parse option string into a linked list of options with attributes. 251 252 if (!*options) gof->stopearly++; 253 while (*options) { 254 char *temp; 255 256 // Option groups come after all options are defined 257 if (*options == '[') break; 258 259 // Allocate a new list entry when necessary 260 if (!new) { 261 new = xzalloc(sizeof(struct opts)); 262 new->next = gof->opts; 263 gof->opts = new; 264 new->val[0].l = LONG_MIN; 265 new->val[1].l = LONG_MAX; 266 } 267 // Each option must start with "(" or an option character. (Bare 268 // longopts only come at the start of the string.) 269 if (*options == '(' && new->c != -1) { 270 char *end; 271 struct longopts *lo; 272 273 // Find the end of the longopt 274 for (end = ++options; *end && *end != ')'; end++); 275 if (CFG_TOYBOX_DEBUG && !*end) error_exit("(longopt) didn't end"); 276 277 // init a new struct longopts 278 lo = xmalloc(sizeof(struct longopts)); 279 lo->next = gof->longopts; 280 lo->opt = new; 281 lo->str = options; 282 lo->len = end-options; 283 gof->longopts = lo; 284 options = ++end; 285 286 // Mark this struct opt as used, even when no short opt. 287 if (!new->c) new->c = -1; 288 289 continue; 290 291 // If this is the start of a new option that wasn't a longopt, 292 293 } else if (strchr(":*#@.-", *options)) { 294 if (CFG_TOYBOX_DEBUG && new->type) 295 error_exit("multiple types %c:%c%c", new->c, new->type, *options); 296 new->type = *options; 297 } else if (-1 != (idx = stridx("|^ ;", *options))) new->flags |= 1<<idx; 298 // bounds checking 299 else if (-1 != (idx = stridx("<>=", *options))) { 300 if (new->type == '#') { 301 long l = strtol(++options, &temp, 10); 302 if (temp != options) new->val[idx].l = l; 303 } else if (CFG_TOYBOX_FLOAT && new->type == '.') { 304 FLOAT f = strtod(++options, &temp); 305 if (temp != options) new->val[idx].f = f; 306 } else if (CFG_TOYBOX_DEBUG) error_exit("<>= only after .#"); 307 options = --temp; 308 309 // At this point, we've hit the end of the previous option. The 310 // current character is the start of a new option. If we've already 311 // assigned an option to this struct, loop to allocate a new one. 312 // (It'll get back here afterwards and fall through to next else.) 313 } else if (new->c) { 314 new = 0; 315 continue; 316 317 // Claim this option, loop to see what's after it. 318 } else new->c = *options; 319 320 options++; 321 } 322 323 // Initialize enable/disable/exclude masks and pointers to store arguments. 324 // (This goes right to left so we need the whole list before we can start.) 325 idx = 0; 326 for (new = gof->opts; new; new = new->next) { 327 unsigned u = 1<<idx++; 328 329 if (new->c == 1) new->c = 0; 330 new->dex[1] = u; 331 if (new->flags & 1) gof->requires |= u; 332 if (new->type) { 333 new->arg = (void *)nextarg; 334 *(nextarg++) = new->val[2].l; 335 } 336 } 337 338 // Parse trailing group indicators 339 while (*options) { 340 unsigned bits = 0; 341 342 if (CFG_TOYBOX_DEBUG && *options != '[') error_exit("trailing %s", options); 343 344 idx = stridx("-+!", *++options); 345 if (CFG_TOYBOX_DEBUG && idx == -1) error_exit("[ needs +-!"); 346 if (CFG_TOYBOX_DEBUG && (options[1] == ']' || !options[1])) 347 error_exit("empty []"); 348 349 // Don't advance past ] but do process it once in loop. 350 while (*options++ != ']') { 351 struct opts *opt; 352 int i; 353 354 if (CFG_TOYBOX_DEBUG && !*options) error_exit("[ without ]"); 355 // Find this option flag (in previously parsed struct opt) 356 for (i=0, opt = gof->opts; ; i++, opt = opt->next) { 357 if (*options == ']') { 358 if (!opt) break; 359 if (bits&(1<<i)) opt->dex[idx] |= bits&~(1<<i); 360 } else { 361 if (CFG_TOYBOX_DEBUG && !opt) 362 error_exit("[] unknown target %c", *options); 363 if (opt->c == *options) { 364 bits |= 1<<i; 365 break; 366 } 367 } 368 } 369 } 370 } 371 } 372 373 // Fill out toys.optflags, toys.optargs, and this[] from toys.argv 374 375 void get_optflags(void) 376 { 377 struct getoptflagstate gof; 378 struct opts *catch; 379 long saveflags; 380 char *letters[]={"s",""}; 381 382 // Option parsing is a two stage process: parse the option string into 383 // a struct opts list, then use that list to process argv[]; 384 385 toys.exithelp++; 386 // Allocate memory for optargs 387 saveflags = 0; 388 while (toys.argv[saveflags++]); 389 toys.optargs = xzalloc(sizeof(char *)*saveflags); 390 391 parse_optflaglist(&gof); 392 393 // Iterate through command line arguments, skipping argv[0] 394 for (gof.argc=1; toys.argv[gof.argc]; gof.argc++) { 395 gof.arg = toys.argv[gof.argc]; 396 catch = NULL; 397 398 // Parse this argument 399 if (gof.stopearly>1) goto notflag; 400 401 gof.nodash_now = 0; 402 403 // Various things with dashes 404 if (*gof.arg == '-') { 405 406 // Handle - 407 if (!gof.arg[1]) goto notflag; 408 gof.arg++; 409 if (*gof.arg=='-') { 410 struct longopts *lo; 411 412 gof.arg++; 413 // Handle -- 414 if (!*gof.arg) { 415 gof.stopearly += 2; 416 continue; 417 } 418 419 // do we match a known --longopt? 420 for (lo = gof.longopts; lo; lo = lo->next) { 421 if (!strncmp(gof.arg, lo->str, lo->len)) { 422 if (!gof.arg[lo->len]) gof.arg = 0; 423 else if (gof.arg[lo->len] == '=' && lo->opt->type) 424 gof.arg += lo->len; 425 else continue; 426 // It's a match. 427 catch = lo->opt; 428 break; 429 } 430 } 431 432 // Should we handle this --longopt as a non-option argument? 433 if (!lo && gof.noerror) { 434 gof.arg -= 2; 435 goto notflag; 436 } 437 438 // Long option parsed, handle option. 439 gotflag(&gof, catch); 440 continue; 441 } 442 443 // Handle things that don't start with a dash. 444 } else { 445 if (gof.nodash && (gof.nodash>1 || gof.argc == 1)) gof.nodash_now = 1; 446 else goto notflag; 447 } 448 449 // At this point, we have the args part of -args. Loop through 450 // each entry (could be -abc meaning -a -b -c) 451 saveflags = toys.optflags; 452 while (*gof.arg) { 453 454 // Identify next option char. 455 for (catch = gof.opts; catch; catch = catch->next) 456 if (*gof.arg == catch->c) 457 if (!((catch->flags&4) && gof.arg[1])) break; 458 459 // Handle option char (advancing past what was used) 460 if (gotflag(&gof, catch) ) { 461 toys.optflags = saveflags; 462 gof.arg = toys.argv[gof.argc]; 463 goto notflag; 464 } 465 } 466 continue; 467 468 // Not a flag, save value in toys.optargs[] 469 notflag: 470 if (gof.stopearly) gof.stopearly++; 471 toys.optargs[toys.optc++] = toys.argv[gof.argc]; 472 } 473 474 // Sanity check 475 if (toys.optc<gof.minargs) 476 error_exit("Need%s %d argument%s", letters[!!(gof.minargs-1)], 477 gof.minargs, letters[!(gof.minargs-1)]); 478 if (toys.optc>gof.maxargs) 479 error_exit("Max %d argument%s", gof.maxargs, letters[!(gof.maxargs-1)]); 480 if (gof.requires && !(gof.requires & toys.optflags)) { 481 struct opts *req; 482 char needs[32], *s = needs; 483 484 for (req = gof.opts; req; req = req->next) 485 if (req->flags & 1) *(s++) = req->c; 486 *s = 0; 487 488 error_exit("Needs %s-%s", s[1] ? "one of " : "", needs); 489 } 490 toys.exithelp = 0; 491 492 if (CFG_TOYBOX_FREE) { 493 llist_traverse(gof.opts, free); 494 llist_traverse(gof.longopts, free); 495 } 496 } 497