1 2 3 4 FLEX(1) USER COMMANDS FLEX(1) 5 6 7 8 NAME 9 flex - fast lexical analyzer generator 10 11 SYNOPSIS 12 flex [-bcdfhilnpstvwBFILTV78+? -C[aefFmr] -ooutput -Pprefix 13 -Sskeleton] [--help --version] [filename ...] 14 15 OVERVIEW 16 This manual describes flex, a tool for generating programs 17 that perform pattern-matching on text. The manual includes 18 both tutorial and reference sections: 19 20 Description 21 a brief overview of the tool 22 23 Some Simple Examples 24 25 Format Of The Input File 26 27 Patterns 28 the extended regular expressions used by flex 29 30 How The Input Is Matched 31 the rules for determining what has been matched 32 33 Actions 34 how to specify what to do when a pattern is matched 35 36 The Generated Scanner 37 details regarding the scanner that flex produces; 38 how to control the input source 39 40 Start Conditions 41 introducing context into your scanners, and 42 managing "mini-scanners" 43 44 Multiple Input Buffers 45 how to manipulate multiple input sources; how to 46 scan from strings instead of files 47 48 End-of-file Rules 49 special rules for matching the end of the input 50 51 Miscellaneous Macros 52 a summary of macros available to the actions 53 54 Values Available To The User 55 a summary of values available to the actions 56 57 Interfacing With Yacc 58 connecting flex scanners together with yacc parsers 59 60 61 62 63 Version 2.5 Last change: April 1995 1 64 65 66 67 68 69 70 FLEX(1) USER COMMANDS FLEX(1) 71 72 73 74 Options 75 flex command-line options, and the "%option" 76 directive 77 78 Performance Considerations 79 how to make your scanner go as fast as possible 80 81 Generating C++ Scanners 82 the (experimental) facility for generating C++ 83 scanner classes 84 85 Incompatibilities With Lex And POSIX 86 how flex differs from AT&T lex and the POSIX lex 87 standard 88 89 Diagnostics 90 those error messages produced by flex (or scanners 91 it generates) whose meanings might not be apparent 92 93 Files 94 files used by flex 95 96 Deficiencies / Bugs 97 known problems with flex 98 99 See Also 100 other documentation, related tools 101 102 Author 103 includes contact information 104 105 106 DESCRIPTION 107 flex is a tool for generating scanners: programs which 108 recognized lexical patterns in text. flex reads the given 109 input files, or its standard input if no file names are 110 given, for a description of a scanner to generate. The 111 description is in the form of pairs of regular expressions 112 and C code, called rules. flex generates as output a C 113 source file, lex.yy.c, which defines a routine yylex(). This 114 file is compiled and linked with the -lfl library to produce 115 an executable. When the executable is run, it analyzes its 116 input for occurrences of the regular expressions. Whenever 117 it finds one, it executes the corresponding C code. 118 119 SOME SIMPLE EXAMPLES 120 First some simple examples to get the flavor of how one uses 121 flex. The following flex input specifies a scanner which 122 whenever it encounters the string "username" will replace it 123 with the user's login name: 124 125 %% 126 127 128 129 Version 2.5 Last change: April 1995 2 130 131 132 133 134 135 136 FLEX(1) USER COMMANDS FLEX(1) 137 138 139 140 username printf( "%s", getlogin() ); 141 142 By default, any text not matched by a flex scanner is copied 143 to the output, so the net effect of this scanner is to copy 144 its input file to its output with each occurrence of "user- 145 name" expanded. In this input, there is just one rule. 146 "username" is the pattern and the "printf" is the action. 147 The "%%" marks the beginning of the rules. 148 149 Here's another simple example: 150 151 int num_lines = 0, num_chars = 0; 152 153 %% 154 \n ++num_lines; ++num_chars; 155 . ++num_chars; 156 157 %% 158 main() 159 { 160 yylex(); 161 printf( "# of lines = %d, # of chars = %d\n", 162 num_lines, num_chars ); 163 } 164 165 This scanner counts the number of characters and the number 166 of lines in its input (it produces no output other than the 167 final report on the counts). The first line declares two 168 globals, "num_lines" and "num_chars", which are accessible 169 both inside yylex() and in the main() routine declared after 170 the second "%%". There are two rules, one which matches a 171 newline ("\n") and increments both the line count and the 172 character count, and one which matches any character other 173 than a newline (indicated by the "." regular expression). 174 175 A somewhat more complicated example: 176 177 /* scanner for a toy Pascal-like language */ 178 179 %{ 180 /* need this for the call to atof() below */ 181 #include <math.h> 182 %} 183 184 DIGIT [0-9] 185 ID [a-z][a-z0-9]* 186 187 %% 188 189 {DIGIT}+ { 190 printf( "An integer: %s (%d)\n", yytext, 191 atoi( yytext ) ); 192 193 194 195 Version 2.5 Last change: April 1995 3 196 197 198 199 200 201 202 FLEX(1) USER COMMANDS FLEX(1) 203 204 205 206 } 207 208 {DIGIT}+"."{DIGIT}* { 209 printf( "A float: %s (%g)\n", yytext, 210 atof( yytext ) ); 211 } 212 213 if|then|begin|end|procedure|function { 214 printf( "A keyword: %s\n", yytext ); 215 } 216 217 {ID} printf( "An identifier: %s\n", yytext ); 218 219 "+"|"-"|"*"|"/" printf( "An operator: %s\n", yytext ); 220 221 "{"[^}\n]*"}" /* eat up one-line comments */ 222 223 [ \t\n]+ /* eat up whitespace */ 224 225 . printf( "Unrecognized character: %s\n", yytext ); 226 227 %% 228 229 main( argc, argv ) 230 int argc; 231 char **argv; 232 { 233 ++argv, --argc; /* skip over program name */ 234 if ( argc > 0 ) 235 yyin = fopen( argv[0], "r" ); 236 else 237 yyin = stdin; 238 239 yylex(); 240 } 241 242 This is the beginnings of a simple scanner for a language 243 like Pascal. It identifies different types of tokens and 244 reports on what it has seen. 245 246 The details of this example will be explained in the follow- 247 ing sections. 248 249 FORMAT OF THE INPUT FILE 250 The flex input file consists of three sections, separated by 251 a line with just %% in it: 252 253 definitions 254 %% 255 rules 256 %% 257 user code 258 259 260 261 Version 2.5 Last change: April 1995 4 262 263 264 265 266 267 268 FLEX(1) USER COMMANDS FLEX(1) 269 270 271 272 The definitions section contains declarations of simple name 273 definitions to simplify the scanner specification, and 274 declarations of start conditions, which are explained in a 275 later section. 276 277 Name definitions have the form: 278 279 name definition 280 281 The "name" is a word beginning with a letter or an under- 282 score ('_') followed by zero or more letters, digits, '_', 283 or '-' (dash). The definition is taken to begin at the 284 first non-white-space character following the name and con- 285 tinuing to the end of the line. The definition can subse- 286 quently be referred to using "{name}", which will expand to 287 "(definition)". For example, 288 289 DIGIT [0-9] 290 ID [a-z][a-z0-9]* 291 292 defines "DIGIT" to be a regular expression which matches a 293 single digit, and "ID" to be a regular expression which 294 matches a letter followed by zero-or-more letters-or-digits. 295 A subsequent reference to 296 297 {DIGIT}+"."{DIGIT}* 298 299 is identical to 300 301 ([0-9])+"."([0-9])* 302 303 and matches one-or-more digits followed by a '.' followed by 304 zero-or-more digits. 305 306 The rules section of the flex input contains a series of 307 rules of the form: 308 309 pattern action 310 311 where the pattern must be unindented and the action must 312 begin on the same line. 313 314 See below for a further description of patterns and actions. 315 316 Finally, the user code section is simply copied to lex.yy.c 317 verbatim. It is used for companion routines which call or 318 are called by the scanner. The presence of this section is 319 optional; if it is missing, the second %% in the input file 320 may be skipped, too. 321 322 In the definitions and rules sections, any indented text or 323 text enclosed in %{ and %} is copied verbatim to the output 324 325 326 327 Version 2.5 Last change: April 1995 5 328 329 330 331 332 333 334 FLEX(1) USER COMMANDS FLEX(1) 335 336 337 338 (with the %{}'s removed). The %{}'s must appear unindented 339 on lines by themselves. 340 341 In the rules section, any indented or %{} text appearing 342 before the first rule may be used to declare variables which 343 are local to the scanning routine and (after the declara- 344 tions) code which is to be executed whenever the scanning 345 routine is entered. Other indented or %{} text in the rule 346 section is still copied to the output, but its meaning is 347 not well-defined and it may well cause compile-time errors 348 (this feature is present for POSIX compliance; see below for 349 other such features). 350 351 In the definitions section (but not in the rules section), 352 an unindented comment (i.e., a line beginning with "/*") is 353 also copied verbatim to the output up to the next "*/". 354 355 PATTERNS 356 The patterns in the input are written using an extended set 357 of regular expressions. These are: 358 359 x match the character 'x' 360 . any character (byte) except newline 361 [xyz] a "character class"; in this case, the pattern 362 matches either an 'x', a 'y', or a 'z' 363 [abj-oZ] a "character class" with a range in it; matches 364 an 'a', a 'b', any letter from 'j' through 'o', 365 or a 'Z' 366 [^A-Z] a "negated character class", i.e., any character 367 but those in the class. In this case, any 368 character EXCEPT an uppercase letter. 369 [^A-Z\n] any character EXCEPT an uppercase letter or 370 a newline 371 r* zero or more r's, where r is any regular expression 372 r+ one or more r's 373 r? zero or one r's (that is, "an optional r") 374 r{2,5} anywhere from two to five r's 375 r{2,} two or more r's 376 r{4} exactly 4 r's 377 {name} the expansion of the "name" definition 378 (see above) 379 "[xyz]\"foo" 380 the literal string: [xyz]"foo 381 \X if X is an 'a', 'b', 'f', 'n', 'r', 't', or 'v', 382 then the ANSI-C interpretation of \x. 383 Otherwise, a literal 'X' (used to escape 384 operators such as '*') 385 \0 a NUL character (ASCII code 0) 386 \123 the character with octal value 123 387 \x2a the character with hexadecimal value 2a 388 (r) match an r; parentheses are used to override 389 precedence (see below) 390 391 392 393 Version 2.5 Last change: April 1995 6 394 395 396 397 398 399 400 FLEX(1) USER COMMANDS FLEX(1) 401 402 403 404 rs the regular expression r followed by the 405 regular expression s; called "concatenation" 406 407 408 r|s either an r or an s 409 410 411 r/s an r but only if it is followed by an s. The 412 text matched by s is included when determining 413 whether this rule is the "longest match", 414 but is then returned to the input before 415 the action is executed. So the action only 416 sees the text matched by r. This type 417 of pattern is called trailing context". 418 (There are some combinations of r/s that flex 419 cannot match correctly; see notes in the 420 Deficiencies / Bugs section below regarding 421 "dangerous trailing context".) 422 ^r an r, but only at the beginning of a line (i.e., 423 which just starting to scan, or right after a 424 newline has been scanned). 425 r$ an r, but only at the end of a line (i.e., just 426 before a newline). Equivalent to "r/\n". 427 428 Note that flex's notion of "newline" is exactly 429 whatever the C compiler used to compile flex 430 interprets '\n' as; in particular, on some DOS 431 systems you must either filter out \r's in the 432 input yourself, or explicitly use r/\r\n for "r$". 433 434 435 <s>r an r, but only in start condition s (see 436 below for discussion of start conditions) 437 <s1,s2,s3>r 438 same, but in any of start conditions s1, 439 s2, or s3 440 <*>r an r in any start condition, even an exclusive one. 441 442 443 <<EOF>> an end-of-file 444 <s1,s2><<EOF>> 445 an end-of-file when in start condition s1 or s2 446 447 Note that inside of a character class, all regular expres- 448 sion operators lose their special meaning except escape 449 ('\') and the character class operators, '-', ']', and, at 450 the beginning of the class, '^'. 451 452 The regular expressions listed above are grouped according 453 to precedence, from highest precedence at the top to lowest 454 at the bottom. Those grouped together have equal pre- 455 cedence. For example, 456 457 458 459 Version 2.5 Last change: April 1995 7 460 461 462 463 464 465 466 FLEX(1) USER COMMANDS FLEX(1) 467 468 469 470 foo|bar* 471 472 is the same as 473 474 (foo)|(ba(r*)) 475 476 since the '*' operator has higher precedence than concatena- 477 tion, and concatenation higher than alternation ('|'). This 478 pattern therefore matches either the string "foo" or the 479 string "ba" followed by zero-or-more r's. To match "foo" or 480 zero-or-more "bar"'s, use: 481 482 foo|(bar)* 483 484 and to match zero-or-more "foo"'s-or-"bar"'s: 485 486 (foo|bar)* 487 488 489 In addition to characters and ranges of characters, charac- 490 ter classes can also contain character class expressions. 491 These are expressions enclosed inside [: and :] delimiters 492 (which themselves must appear between the '[' and ']' of the 493 character class; other elements may occur inside the charac- 494 ter class, too). The valid expressions are: 495 496 [:alnum:] [:alpha:] [:blank:] 497 [:cntrl:] [:digit:] [:graph:] 498 [:lower:] [:print:] [:punct:] 499 [:space:] [:upper:] [:xdigit:] 500 501 These expressions all designate a set of characters 502 equivalent to the corresponding standard C isXXX function. 503 For example, [:alnum:] designates those characters for which 504 isalnum() returns true - i.e., any alphabetic or numeric. 505 Some systems don't provide isblank(), so flex defines 506 [:blank:] as a blank or a tab. 507 508 For example, the following character classes are all 509 equivalent: 510 511 [[:alnum:]] 512 [[:alpha:][:digit:] 513 [[:alpha:]0-9] 514 [a-zA-Z0-9] 515 516 If your scanner is case-insensitive (the -i flag), then 517 [:upper:] and [:lower:] are equivalent to [:alpha:]. 518 519 Some notes on patterns: 520 521 - A negated character class such as the example "[^A-Z]" 522 523 524 525 Version 2.5 Last change: April 1995 8 526 527 528 529 530 531 532 FLEX(1) USER COMMANDS FLEX(1) 533 534 535 536 above will match a newline unless "\n" (or an 537 equivalent escape sequence) is one of the characters 538 explicitly present in the negated character class 539 (e.g., "[^A-Z\n]"). This is unlike how many other reg- 540 ular expression tools treat negated character classes, 541 but unfortunately the inconsistency is historically 542 entrenched. Matching newlines means that a pattern 543 like [^"]* can match the entire input unless there's 544 another quote in the input. 545 546 - A rule can have at most one instance of trailing con- 547 text (the '/' operator or the '$' operator). The start 548 condition, '^', and "<<EOF>>" patterns can only occur 549 at the beginning of a pattern, and, as well as with '/' 550 and '$', cannot be grouped inside parentheses. A '^' 551 which does not occur at the beginning of a rule or a 552 '$' which does not occur at the end of a rule loses its 553 special properties and is treated as a normal charac- 554 ter. 555 556 The following are illegal: 557 558 foo/bar$ 559 <sc1>foo<sc2>bar 560 561 Note that the first of these, can be written 562 "foo/bar\n". 563 564 The following will result in '$' or '^' being treated 565 as a normal character: 566 567 foo|(bar$) 568 foo|^bar 569 570 If what's wanted is a "foo" or a bar-followed-by-a- 571 newline, the following could be used (the special '|' 572 action is explained below): 573 574 foo | 575 bar$ /* action goes here */ 576 577 A similar trick will work for matching a foo or a bar- 578 at-the-beginning-of-a-line. 579 580 HOW THE INPUT IS MATCHED 581 When the generated scanner is run, it analyzes its input 582 looking for strings which match any of its patterns. If it 583 finds more than one match, it takes the one matching the 584 most text (for trailing context rules, this includes the 585 length of the trailing part, even though it will then be 586 returned to the input). If it finds two or more matches of 587 the same length, the rule listed first in the flex input 588 589 590 591 Version 2.5 Last change: April 1995 9 592 593 594 595 596 597 598 FLEX(1) USER COMMANDS FLEX(1) 599 600 601 602 file is chosen. 603 604 Once the match is determined, the text corresponding to the 605 match (called the token) is made available in the global 606 character pointer yytext, and its length in the global 607 integer yyleng. The action corresponding to the matched pat- 608 tern is then executed (a more detailed description of 609 actions follows), and then the remaining input is scanned 610 for another match. 611 612 If no match is found, then the default rule is executed: the 613 next character in the input is considered matched and copied 614 to the standard output. Thus, the simplest legal flex input 615 is: 616 617 %% 618 619 which generates a scanner that simply copies its input (one 620 character at a time) to its output. 621 622 Note that yytext can be defined in two different ways: 623 either as a character pointer or as a character array. You 624 can control which definition flex uses by including one of 625 the special directives %pointer or %array in the first 626 (definitions) section of your flex input. The default is 627 %pointer, unless you use the -l lex compatibility option, in 628 which case yytext will be an array. The advantage of using 629 %pointer is substantially faster scanning and no buffer 630 overflow when matching very large tokens (unless you run out 631 of dynamic memory). The disadvantage is that you are res- 632 tricted in how your actions can modify yytext (see the next 633 section), and calls to the unput() function destroys the 634 present contents of yytext, which can be a considerable 635 porting headache when moving between different lex versions. 636 637 The advantage of %array is that you can then modify yytext 638 to your heart's content, and calls to unput() do not destroy 639 yytext (see below). Furthermore, existing lex programs 640 sometimes access yytext externally using declarations of the 641 form: 642 extern char yytext[]; 643 This definition is erroneous when used with %pointer, but 644 correct for %array. 645 646 %array defines yytext to be an array of YYLMAX characters, 647 which defaults to a fairly large value. You can change the 648 size by simply #define'ing YYLMAX to a different value in 649 the first section of your flex input. As mentioned above, 650 with %pointer yytext grows dynamically to accommodate large 651 tokens. While this means your %pointer scanner can accommo- 652 date very large tokens (such as matching entire blocks of 653 comments), bear in mind that each time the scanner must 654 655 656 657 Version 2.5 Last change: April 1995 10 658 659 660 661 662 663 664 FLEX(1) USER COMMANDS FLEX(1) 665 666 667 668 resize yytext it also must rescan the entire token from the 669 beginning, so matching such tokens can prove slow. yytext 670 presently does not dynamically grow if a call to unput() 671 results in too much text being pushed back; instead, a run- 672 time error results. 673 674 Also note that you cannot use %array with C++ scanner 675 classes (the c++ option; see below). 676 677 ACTIONS 678 Each pattern in a rule has a corresponding action, which can 679 be any arbitrary C statement. The pattern ends at the first 680 non-escaped whitespace character; the remainder of the line 681 is its action. If the action is empty, then when the pat- 682 tern is matched the input token is simply discarded. For 683 example, here is the specification for a program which 684 deletes all occurrences of "zap me" from its input: 685 686 %% 687 "zap me" 688 689 (It will copy all other characters in the input to the out- 690 put since they will be matched by the default rule.) 691 692 Here is a program which compresses multiple blanks and tabs 693 down to a single blank, and throws away whitespace found at 694 the end of a line: 695 696 %% 697 [ \t]+ putchar( ' ' ); 698 [ \t]+$ /* ignore this token */ 699 700 701 If the action contains a '{', then the action spans till the 702 balancing '}' is found, and the action may cross multiple 703 lines. flex knows about C strings and comments and won't be 704 fooled by braces found within them, but also allows actions 705 to begin with %{ and will consider the action to be all the 706 text up to the next %} (regardless of ordinary braces inside 707 the action). 708 709 An action consisting solely of a vertical bar ('|') means 710 "same as the action for the next rule." See below for an 711 illustration. 712 713 Actions can include arbitrary C code, including return 714 statements to return a value to whatever routine called 715 yylex(). Each time yylex() is called it continues processing 716 tokens from where it last left off until it either reaches 717 the end of the file or executes a return. 718 719 720 721 722 723 Version 2.5 Last change: April 1995 11 724 725 726 727 728 729 730 FLEX(1) USER COMMANDS FLEX(1) 731 732 733 734 Actions are free to modify yytext except for lengthening it 735 (adding characters to its end--these will overwrite later 736 characters in the input stream). This however does not 737 apply when using %array (see above); in that case, yytext 738 may be freely modified in any way. 739 740 Actions are free to modify yyleng except they should not do 741 so if the action also includes use of yymore() (see below). 742 743 There are a number of special directives which can be 744 included within an action: 745 746 - ECHO copies yytext to the scanner's output. 747 748 - BEGIN followed by the name of a start condition places 749 the scanner in the corresponding start condition (see 750 below). 751 752 - REJECT directs the scanner to proceed on to the "second 753 best" rule which matched the input (or a prefix of the 754 input). The rule is chosen as described above in "How 755 the Input is Matched", and yytext and yyleng set up 756 appropriately. It may either be one which matched as 757 much text as the originally chosen rule but came later 758 in the flex input file, or one which matched less text. 759 For example, the following will both count the words in 760 the input and call the routine special() whenever 761 "frob" is seen: 762 763 int word_count = 0; 764 %% 765 766 frob special(); REJECT; 767 [^ \t\n]+ ++word_count; 768 769 Without the REJECT, any "frob"'s in the input would not 770 be counted as words, since the scanner normally exe- 771 cutes only one action per token. Multiple REJECT's are 772 allowed, each one finding the next best choice to the 773 currently active rule. For example, when the following 774 scanner scans the token "abcd", it will write "abcdab- 775 caba" to the output: 776 777 %% 778 a | 779 ab | 780 abc | 781 abcd ECHO; REJECT; 782 .|\n /* eat up any unmatched character */ 783 784 (The first three rules share the fourth's action since 785 they use the special '|' action.) REJECT is a 786 787 788 789 Version 2.5 Last change: April 1995 12 790 791 792 793 794 795 796 FLEX(1) USER COMMANDS FLEX(1) 797 798 799 800 particularly expensive feature in terms of scanner per- 801 formance; if it is used in any of the scanner's actions 802 it will slow down all of the scanner's matching. 803 Furthermore, REJECT cannot be used with the -Cf or -CF 804 options (see below). 805 806 Note also that unlike the other special actions, REJECT 807 is a branch; code immediately following it in the 808 action will not be executed. 809 810 - yymore() tells the scanner that the next time it 811 matches a rule, the corresponding token should be 812 appended onto the current value of yytext rather than 813 replacing it. For example, given the input "mega- 814 kludge" the following will write "mega-mega-kludge" to 815 the output: 816 817 %% 818 mega- ECHO; yymore(); 819 kludge ECHO; 820 821 First "mega-" is matched and echoed to the output. 822 Then "kludge" is matched, but the previous "mega-" is 823 still hanging around at the beginning of yytext so the 824 ECHO for the "kludge" rule will actually write "mega- 825 kludge". 826 827 Two notes regarding use of yymore(). First, yymore() depends 828 on the value of yyleng correctly reflecting the size of the 829 current token, so you must not modify yyleng if you are 830 using yymore(). Second, the presence of yymore() in the 831 scanner's action entails a minor performance penalty in the 832 scanner's matching speed. 833 834 - yyless(n) returns all but the first n characters of the 835 current token back to the input stream, where they will 836 be rescanned when the scanner looks for the next match. 837 yytext and yyleng are adjusted appropriately (e.g., 838 yyleng will now be equal to n ). For example, on the 839 input "foobar" the following will write out "foobar- 840 bar": 841 842 %% 843 foobar ECHO; yyless(3); 844 [a-z]+ ECHO; 845 846 An argument of 0 to yyless will cause the entire 847 current input string to be scanned again. Unless 848 you've changed how the scanner will subsequently pro- 849 cess its input (using BEGIN, for example), this will 850 result in an endless loop. 851 852 853 854 855 Version 2.5 Last change: April 1995 13 856 857 858 859 860 861 862 FLEX(1) USER COMMANDS FLEX(1) 863 864 865 866 Note that yyless is a macro and can only be used in the flex 867 input file, not from other source files. 868 869 - unput(c) puts the character c back onto the input 870 stream. It will be the next character scanned. The 871 following action will take the current token and cause 872 it to be rescanned enclosed in parentheses. 873 874 { 875 int i; 876 /* Copy yytext because unput() trashes yytext */ 877 char *yycopy = strdup( yytext ); 878 unput( ')' ); 879 for ( i = yyleng - 1; i >= 0; --i ) 880 unput( yycopy[i] ); 881 unput( '(' ); 882 free( yycopy ); 883 } 884 885 Note that since each unput() puts the given character 886 back at the beginning of the input stream, pushing back 887 strings must be done back-to-front. 888 889 An important potential problem when using unput() is that if 890 you are using %pointer (the default), a call to unput() des- 891 troys the contents of yytext, starting with its rightmost 892 character and devouring one character to the left with each 893 call. If you need the value of yytext preserved after a 894 call to unput() (as in the above example), you must either 895 first copy it elsewhere, or build your scanner using %array 896 instead (see How The Input Is Matched). 897 898 Finally, note that you cannot put back EOF to attempt to 899 mark the input stream with an end-of-file. 900 901 - input() reads the next character from the input stream. 902 For example, the following is one way to eat up C com- 903 ments: 904 905 %% 906 "/*" { 907 register int c; 908 909 for ( ; ; ) 910 { 911 while ( (c = input()) != '*' && 912 c != EOF ) 913 ; /* eat up text of comment */ 914 915 if ( c == '*' ) 916 { 917 while ( (c = input()) == '*' ) 918 919 920 921 Version 2.5 Last change: April 1995 14 922 923 924 925 926 927 928 FLEX(1) USER COMMANDS FLEX(1) 929 930 931 932 ; 933 if ( c == '/' ) 934 break; /* found the end */ 935 } 936 937 if ( c == EOF ) 938 { 939 error( "EOF in comment" ); 940 break; 941 } 942 } 943 } 944 945 (Note that if the scanner is compiled using C++, then 946 input() is instead referred to as yyinput(), in order 947 to avoid a name clash with the C++ stream by the name 948 of input.) 949 950 - YY_FLUSH_BUFFER flushes the scanner's internal buffer 951 so that the next time the scanner attempts to match a 952 token, it will first refill the buffer using YY_INPUT 953 (see The Generated Scanner, below). This action is a 954 special case of the more general yy_flush_buffer() 955 function, described below in the section Multiple Input 956 Buffers. 957 958 - yyterminate() can be used in lieu of a return statement 959 in an action. It terminates the scanner and returns a 960 0 to the scanner's caller, indicating "all done". By 961 default, yyterminate() is also called when an end-of- 962 file is encountered. It is a macro and may be rede- 963 fined. 964 965 THE GENERATED SCANNER 966 The output of flex is the file lex.yy.c, which contains the 967 scanning routine yylex(), a number of tables used by it for 968 matching tokens, and a number of auxiliary routines and mac- 969 ros. By default, yylex() is declared as follows: 970 971 int yylex() 972 { 973 ... various definitions and the actions in here ... 974 } 975 976 (If your environment supports function prototypes, then it 977 will be "int yylex( void )".) This definition may be 978 changed by defining the "YY_DECL" macro. For example, you 979 could use: 980 981 #define YY_DECL float lexscan( a, b ) float a, b; 982 983 to give the scanning routine the name lexscan, returning a 984 985 986 987 Version 2.5 Last change: April 1995 15 988 989 990 991 992 993 994 FLEX(1) USER COMMANDS FLEX(1) 995 996 997 998 float, and taking two floats as arguments. Note that if you 999 give arguments to the scanning routine using a K&R- 1000 style/non-prototyped function declaration, you must ter- 1001 minate the definition with a semi-colon (;). 1002 1003 Whenever yylex() is called, it scans tokens from the global 1004 input file yyin (which defaults to stdin). It continues 1005 until it either reaches an end-of-file (at which point it 1006 returns the value 0) or one of its actions executes a return 1007 statement. 1008 1009 If the scanner reaches an end-of-file, subsequent calls are 1010 undefined unless either yyin is pointed at a new input file 1011 (in which case scanning continues from that file), or yyres- 1012 tart() is called. yyrestart() takes one argument, a FILE * 1013 pointer (which can be nil, if you've set up YY_INPUT to scan 1014 from a source other than yyin), and initializes yyin for 1015 scanning from that file. Essentially there is no difference 1016 between just assigning yyin to a new input file or using 1017 yyrestart() to do so; the latter is available for compati- 1018 bility with previous versions of flex, and because it can be 1019 used to switch input files in the middle of scanning. It 1020 can also be used to throw away the current input buffer, by 1021 calling it with an argument of yyin; but better is to use 1022 YY_FLUSH_BUFFER (see above). Note that yyrestart() does not 1023 reset the start condition to INITIAL (see Start Conditions, 1024 below). 1025 1026 If yylex() stops scanning due to executing a return state- 1027 ment in one of the actions, the scanner may then be called 1028 again and it will resume scanning where it left off. 1029 1030 By default (and for purposes of efficiency), the scanner 1031 uses block-reads rather than simple getc() calls to read 1032 characters from yyin. The nature of how it gets its input 1033 can be controlled by defining the YY_INPUT macro. 1034 YY_INPUT's calling sequence is 1035 "YY_INPUT(buf,result,max_size)". Its action is to place up 1036 to max_size characters in the character array buf and return 1037 in the integer variable result either the number of charac- 1038 ters read or the constant YY_NULL (0 on Unix systems) to 1039 indicate EOF. The default YY_INPUT reads from the global 1040 file-pointer "yyin". 1041 1042 A sample definition of YY_INPUT (in the definitions section 1043 of the input file): 1044 1045 %{ 1046 #define YY_INPUT(buf,result,max_size) \ 1047 { \ 1048 int c = getchar(); \ 1049 result = (c == EOF) ? YY_NULL : (buf[0] = c, 1); \ 1050 1051 1052 1053 Version 2.5 Last change: April 1995 16 1054 1055 1056 1057 1058 1059 1060 FLEX(1) USER COMMANDS FLEX(1) 1061 1062 1063 1064 } 1065 %} 1066 1067 This definition will change the input processing to occur 1068 one character at a time. 1069 1070 When the scanner receives an end-of-file indication from 1071 YY_INPUT, it then checks the yywrap() function. If yywrap() 1072 returns false (zero), then it is assumed that the function 1073 has gone ahead and set up yyin to point to another input 1074 file, and scanning continues. If it returns true (non- 1075 zero), then the scanner terminates, returning 0 to its 1076 caller. Note that in either case, the start condition 1077 remains unchanged; it does not revert to INITIAL. 1078 1079 If you do not supply your own version of yywrap(), then you 1080 must either use %option noyywrap (in which case the scanner 1081 behaves as though yywrap() returned 1), or you must link 1082 with -lfl to obtain the default version of the routine, 1083 which always returns 1. 1084 1085 Three routines are available for scanning from in-memory 1086 buffers rather than files: yy_scan_string(), 1087 yy_scan_bytes(), and yy_scan_buffer(). See the discussion of 1088 them below in the section Multiple Input Buffers. 1089 1090 The scanner writes its ECHO output to the yyout global 1091 (default, stdout), which may be redefined by the user simply 1092 by assigning it to some other FILE pointer. 1093 1094 START CONDITIONS 1095 flex provides a mechanism for conditionally activating 1096 rules. Any rule whose pattern is prefixed with "<sc>" will 1097 only be active when the scanner is in the start condition 1098 named "sc". For example, 1099 1100 <STRING>[^"]* { /* eat up the string body ... */ 1101 ... 1102 } 1103 1104 will be active only when the scanner is in the "STRING" 1105 start condition, and 1106 1107 <INITIAL,STRING,QUOTE>\. { /* handle an escape ... */ 1108 ... 1109 } 1110 1111 will be active only when the current start condition is 1112 either "INITIAL", "STRING", or "QUOTE". 1113 1114 Start conditions are declared in the definitions (first) 1115 section of the input using unindented lines beginning with 1116 1117 1118 1119 Version 2.5 Last change: April 1995 17 1120 1121 1122 1123 1124 1125 1126 FLEX(1) USER COMMANDS FLEX(1) 1127 1128 1129 1130 either %s or %x followed by a list of names. The former 1131 declares inclusive start conditions, the latter exclusive 1132 start conditions. A start condition is activated using the 1133 BEGIN action. Until the next BEGIN action is executed, 1134 rules with the given start condition will be active and 1135 rules with other start conditions will be inactive. If the 1136 start condition is inclusive, then rules with no start con- 1137 ditions at all will also be active. If it is exclusive, 1138 then only rules qualified with the start condition will be 1139 active. A set of rules contingent on the same exclusive 1140 start condition describe a scanner which is independent of 1141 any of the other rules in the flex input. Because of this, 1142 exclusive start conditions make it easy to specify "mini- 1143 scanners" which scan portions of the input that are syntac- 1144 tically different from the rest (e.g., comments). 1145 1146 If the distinction between inclusive and exclusive start 1147 conditions is still a little vague, here's a simple example 1148 illustrating the connection between the two. The set of 1149 rules: 1150 1151 %s example 1152 %% 1153 1154 <example>foo do_something(); 1155 1156 bar something_else(); 1157 1158 is equivalent to 1159 1160 %x example 1161 %% 1162 1163 <example>foo do_something(); 1164 1165 <INITIAL,example>bar something_else(); 1166 1167 Without the <INITIAL,example> qualifier, the bar pattern in 1168 the second example wouldn't be active (i.e., couldn't match) 1169 when in start condition example. If we just used <example> 1170 to qualify bar, though, then it would only be active in 1171 example and not in INITIAL, while in the first example it's 1172 active in both, because in the first example the example 1173 startion condition is an inclusive (%s) start condition. 1174 1175 Also note that the special start-condition specifier <*> 1176 matches every start condition. Thus, the above example 1177 could also have been written; 1178 1179 %x example 1180 %% 1181 1182 1183 1184 1185 Version 2.5 Last change: April 1995 18 1186 1187 1188 1189 1190 1191 1192 FLEX(1) USER COMMANDS FLEX(1) 1193 1194 1195 1196 <example>foo do_something(); 1197 1198 <*>bar something_else(); 1199 1200 1201 The default rule (to ECHO any unmatched character) remains 1202 active in start conditions. It is equivalent to: 1203 1204 <*>.|\n ECHO; 1205 1206 1207 BEGIN(0) returns to the original state where only the rules 1208 with no start conditions are active. This state can also be 1209 referred to as the start-condition "INITIAL", so 1210 BEGIN(INITIAL) is equivalent to BEGIN(0). (The parentheses 1211 around the start condition name are not required but are 1212 considered good style.) 1213 1214 BEGIN actions can also be given as indented code at the 1215 beginning of the rules section. For example, the following 1216 will cause the scanner to enter the "SPECIAL" start condi- 1217 tion whenever yylex() is called and the global variable 1218 enter_special is true: 1219 1220 int enter_special; 1221 1222 %x SPECIAL 1223 %% 1224 if ( enter_special ) 1225 BEGIN(SPECIAL); 1226 1227 <SPECIAL>blahblahblah 1228 ...more rules follow... 1229 1230 1231 To illustrate the uses of start conditions, here is a 1232 scanner which provides two different interpretations of a 1233 string like "123.456". By default it will treat it as three 1234 tokens, the integer "123", a dot ('.'), and the integer 1235 "456". But if the string is preceded earlier in the line by 1236 the string "expect-floats" it will treat it as a single 1237 token, the floating-point number 123.456: 1238 1239 %{ 1240 #include <math.h> 1241 %} 1242 %s expect 1243 1244 %% 1245 expect-floats BEGIN(expect); 1246 1247 <expect>[0-9]+"."[0-9]+ { 1248 1249 1250 1251 Version 2.5 Last change: April 1995 19 1252 1253 1254 1255 1256 1257 1258 FLEX(1) USER COMMANDS FLEX(1) 1259 1260 1261 1262 printf( "found a float, = %f\n", 1263 atof( yytext ) ); 1264 } 1265 <expect>\n { 1266 /* that's the end of the line, so 1267 * we need another "expect-number" 1268 * before we'll recognize any more 1269 * numbers 1270 */ 1271 BEGIN(INITIAL); 1272 } 1273 1274 [0-9]+ { 1275 printf( "found an integer, = %d\n", 1276 atoi( yytext ) ); 1277 } 1278 1279 "." printf( "found a dot\n" ); 1280 1281 Here is a scanner which recognizes (and discards) C comments 1282 while maintaining a count of the current input line. 1283 1284 %x comment 1285 %% 1286 int line_num = 1; 1287 1288 "/*" BEGIN(comment); 1289 1290 <comment>[^*\n]* /* eat anything that's not a '*' */ 1291 <comment>"*"+[^*/\n]* /* eat up '*'s not followed by '/'s */ 1292 <comment>\n ++line_num; 1293 <comment>"*"+"/" BEGIN(INITIAL); 1294 1295 This scanner goes to a bit of trouble to match as much text 1296 as possible with each rule. In general, when attempting to 1297 write a high-speed scanner try to match as much possible in 1298 each rule, as it's a big win. 1299 1300 Note that start-conditions names are really integer values 1301 and can be stored as such. Thus, the above could be 1302 extended in the following fashion: 1303 1304 %x comment foo 1305 %% 1306 int line_num = 1; 1307 int comment_caller; 1308 1309 "/*" { 1310 comment_caller = INITIAL; 1311 BEGIN(comment); 1312 } 1313 1314 1315 1316 1317 Version 2.5 Last change: April 1995 20 1318 1319 1320 1321 1322 1323 1324 FLEX(1) USER COMMANDS FLEX(1) 1325 1326 1327 1328 ... 1329 1330 <foo>"/*" { 1331 comment_caller = foo; 1332 BEGIN(comment); 1333 } 1334 1335 <comment>[^*\n]* /* eat anything that's not a '*' */ 1336 <comment>"*"+[^*/\n]* /* eat up '*'s not followed by '/'s */ 1337 <comment>\n ++line_num; 1338 <comment>"*"+"/" BEGIN(comment_caller); 1339 1340 Furthermore, you can access the current start condition 1341 using the integer-valued YY_START macro. For example, the 1342 above assignments to comment_caller could instead be written 1343 1344 comment_caller = YY_START; 1345 1346 Flex provides YYSTATE as an alias for YY_START (since that 1347 is what's used by AT&T lex). 1348 1349 Note that start conditions do not have their own name-space; 1350 %s's and %x's declare names in the same fashion as 1351 #define's. 1352 1353 Finally, here's an example of how to match C-style quoted 1354 strings using exclusive start conditions, including expanded 1355 escape sequences (but not including checking for a string 1356 that's too long): 1357 1358 %x str 1359 1360 %% 1361 char string_buf[MAX_STR_CONST]; 1362 char *string_buf_ptr; 1363 1364 1365 \" string_buf_ptr = string_buf; BEGIN(str); 1366 1367 <str>\" { /* saw closing quote - all done */ 1368 BEGIN(INITIAL); 1369 *string_buf_ptr = '\0'; 1370 /* return string constant token type and 1371 * value to parser 1372 */ 1373 } 1374 1375 <str>\n { 1376 /* error - unterminated string constant */ 1377 /* generate error message */ 1378 } 1379 1380 1381 1382 1383 Version 2.5 Last change: April 1995 21 1384 1385 1386 1387 1388 1389 1390 FLEX(1) USER COMMANDS FLEX(1) 1391 1392 1393 1394 <str>\\[0-7]{1,3} { 1395 /* octal escape sequence */ 1396 int result; 1397 1398 (void) sscanf( yytext + 1, "%o", &result ); 1399 1400 if ( result > 0xff ) 1401 /* error, constant is out-of-bounds */ 1402 1403 *string_buf_ptr++ = result; 1404 } 1405 1406 <str>\\[0-9]+ { 1407 /* generate error - bad escape sequence; something 1408 * like '\48' or '\0777777' 1409 */ 1410 } 1411 1412 <str>\\n *string_buf_ptr++ = '\n'; 1413 <str>\\t *string_buf_ptr++ = '\t'; 1414 <str>\\r *string_buf_ptr++ = '\r'; 1415 <str>\\b *string_buf_ptr++ = '\b'; 1416 <str>\\f *string_buf_ptr++ = '\f'; 1417 1418 <str>\\(.|\n) *string_buf_ptr++ = yytext[1]; 1419 1420 <str>[^\\\n\"]+ { 1421 char *yptr = yytext; 1422 1423 while ( *yptr ) 1424 *string_buf_ptr++ = *yptr++; 1425 } 1426 1427 1428 Often, such as in some of the examples above, you wind up 1429 writing a whole bunch of rules all preceded by the same 1430 start condition(s). Flex makes this a little easier and 1431 cleaner by introducing a notion of start condition scope. A 1432 start condition scope is begun with: 1433 1434 <SCs>{ 1435 1436 where SCs is a list of one or more start conditions. Inside 1437 the start condition scope, every rule automatically has the 1438 prefix <SCs> applied to it, until a '}' which matches the 1439 initial '{'. So, for example, 1440 1441 <ESC>{ 1442 "\\n" return '\n'; 1443 "\\r" return '\r'; 1444 "\\f" return '\f'; 1445 "\\0" return '\0'; 1446 1447 1448 1449 Version 2.5 Last change: April 1995 22 1450 1451 1452 1453 1454 1455 1456 FLEX(1) USER COMMANDS FLEX(1) 1457 1458 1459 1460 } 1461 1462 is equivalent to: 1463 1464 <ESC>"\\n" return '\n'; 1465 <ESC>"\\r" return '\r'; 1466 <ESC>"\\f" return '\f'; 1467 <ESC>"\\0" return '\0'; 1468 1469 Start condition scopes may be nested. 1470 1471 Three routines are available for manipulating stacks of 1472 start conditions: 1473 1474 void yy_push_state(int new_state) 1475 pushes the current start condition onto the top of the 1476 start condition stack and switches to new_state as 1477 though you had used BEGIN new_state (recall that start 1478 condition names are also integers). 1479 1480 void yy_pop_state() 1481 pops the top of the stack and switches to it via BEGIN. 1482 1483 int yy_top_state() 1484 returns the top of the stack without altering the 1485 stack's contents. 1486 1487 The start condition stack grows dynamically and so has no 1488 built-in size limitation. If memory is exhausted, program 1489 execution aborts. 1490 1491 To use start condition stacks, your scanner must include a 1492 %option stack directive (see Options below). 1493 1494 MULTIPLE INPUT BUFFERS 1495 Some scanners (such as those which support "include" files) 1496 require reading from several input streams. As flex 1497 scanners do a large amount of buffering, one cannot control 1498 where the next input will be read from by simply writing a 1499 YY_INPUT which is sensitive to the scanning context. 1500 YY_INPUT is only called when the scanner reaches the end of 1501 its buffer, which may be a long time after scanning a state- 1502 ment such as an "include" which requires switching the input 1503 source. 1504 1505 To negotiate these sorts of problems, flex provides a 1506 mechanism for creating and switching between multiple input 1507 buffers. An input buffer is created by using: 1508 1509 YY_BUFFER_STATE yy_create_buffer( FILE *file, int size ) 1510 1511 which takes a FILE pointer and a size and creates a buffer 1512 1513 1514 1515 Version 2.5 Last change: April 1995 23 1516 1517 1518 1519 1520 1521 1522 FLEX(1) USER COMMANDS FLEX(1) 1523 1524 1525 1526 associated with the given file and large enough to hold size 1527 characters (when in doubt, use YY_BUF_SIZE for the size). 1528 It returns a YY_BUFFER_STATE handle, which may then be 1529 passed to other routines (see below). The YY_BUFFER_STATE 1530 type is a pointer to an opaque struct yy_buffer_state struc- 1531 ture, so you may safely initialize YY_BUFFER_STATE variables 1532 to ((YY_BUFFER_STATE) 0) if you wish, and also refer to the 1533 opaque structure in order to correctly declare input buffers 1534 in source files other than that of your scanner. Note that 1535 the FILE pointer in the call to yy_create_buffer is only 1536 used as the value of yyin seen by YY_INPUT; if you redefine 1537 YY_INPUT so it no longer uses yyin, then you can safely pass 1538 a nil FILE pointer to yy_create_buffer. You select a partic- 1539 ular buffer to scan from using: 1540 1541 void yy_switch_to_buffer( YY_BUFFER_STATE new_buffer ) 1542 1543 switches the scanner's input buffer so subsequent tokens 1544 will come from new_buffer. Note that yy_switch_to_buffer() 1545 may be used by yywrap() to set things up for continued scan- 1546 ning, instead of opening a new file and pointing yyin at it. 1547 Note also that switching input sources via either 1548 yy_switch_to_buffer() or yywrap() does not change the start 1549 condition. 1550 1551 void yy_delete_buffer( YY_BUFFER_STATE buffer ) 1552 1553 is used to reclaim the storage associated with a buffer. ( 1554 buffer can be nil, in which case the routine does nothing.) 1555 You can also clear the current contents of a buffer using: 1556 1557 void yy_flush_buffer( YY_BUFFER_STATE buffer ) 1558 1559 This function discards the buffer's contents, so the next 1560 time the scanner attempts to match a token from the buffer, 1561 it will first fill the buffer anew using YY_INPUT. 1562 1563 yy_new_buffer() is an alias for yy_create_buffer(), provided 1564 for compatibility with the C++ use of new and delete for 1565 creating and destroying dynamic objects. 1566 1567 Finally, the YY_CURRENT_BUFFER macro returns a 1568 YY_BUFFER_STATE handle to the current buffer. 1569 1570 Here is an example of using these features for writing a 1571 scanner which expands include files (the <<EOF>> feature is 1572 discussed below): 1573 1574 /* the "incl" state is used for picking up the name 1575 * of an include file 1576 */ 1577 %x incl 1578 1579 1580 1581 Version 2.5 Last change: April 1995 24 1582 1583 1584 1585 1586 1587 1588 FLEX(1) USER COMMANDS FLEX(1) 1589 1590 1591 1592 %{ 1593 #define MAX_INCLUDE_DEPTH 10 1594 YY_BUFFER_STATE include_stack[MAX_INCLUDE_DEPTH]; 1595 int include_stack_ptr = 0; 1596 %} 1597 1598 %% 1599 include BEGIN(incl); 1600 1601 [a-z]+ ECHO; 1602 [^a-z\n]*\n? ECHO; 1603 1604 <incl>[ \t]* /* eat the whitespace */ 1605 <incl>[^ \t\n]+ { /* got the include file name */ 1606 if ( include_stack_ptr >= MAX_INCLUDE_DEPTH ) 1607 { 1608 fprintf( stderr, "Includes nested too deeply" ); 1609 exit( 1 ); 1610 } 1611 1612 include_stack[include_stack_ptr++] = 1613 YY_CURRENT_BUFFER; 1614 1615 yyin = fopen( yytext, "r" ); 1616 1617 if ( ! yyin ) 1618 error( ... ); 1619 1620 yy_switch_to_buffer( 1621 yy_create_buffer( yyin, YY_BUF_SIZE ) ); 1622 1623 BEGIN(INITIAL); 1624 } 1625 1626 <<EOF>> { 1627 if ( --include_stack_ptr < 0 ) 1628 { 1629 yyterminate(); 1630 } 1631 1632 else 1633 { 1634 yy_delete_buffer( YY_CURRENT_BUFFER ); 1635 yy_switch_to_buffer( 1636 include_stack[include_stack_ptr] ); 1637 } 1638 } 1639 1640 Three routines are available for setting up input buffers 1641 for scanning in-memory strings instead of files. All of 1642 them create a new input buffer for scanning the string, and 1643 return a corresponding YY_BUFFER_STATE handle (which you 1644 1645 1646 1647 Version 2.5 Last change: April 1995 25 1648 1649 1650 1651 1652 1653 1654 FLEX(1) USER COMMANDS FLEX(1) 1655 1656 1657 1658 should delete with yy_delete_buffer() when done with it). 1659 They also switch to the new buffer using 1660 yy_switch_to_buffer(), so the next call to yylex() will 1661 start scanning the string. 1662 1663 yy_scan_string(const char *str) 1664 scans a NUL-terminated string. 1665 1666 yy_scan_bytes(const char *bytes, int len) 1667 scans len bytes (including possibly NUL's) starting at 1668 location bytes. 1669 1670 Note that both of these functions create and scan a copy of 1671 the string or bytes. (This may be desirable, since yylex() 1672 modifies the contents of the buffer it is scanning.) You 1673 can avoid the copy by using: 1674 1675 yy_scan_buffer(char *base, yy_size_t size) 1676 which scans in place the buffer starting at base, con- 1677 sisting of size bytes, the last two bytes of which must 1678 be YY_END_OF_BUFFER_CHAR (ASCII NUL). These last two 1679 bytes are not scanned; thus, scanning consists of 1680 base[0] through base[size-2], inclusive. 1681 1682 If you fail to set up base in this manner (i.e., forget 1683 the final two YY_END_OF_BUFFER_CHAR bytes), then 1684 yy_scan_buffer() returns a nil pointer instead of 1685 creating a new input buffer. 1686 1687 The type yy_size_t is an integral type to which you can 1688 cast an integer expression reflecting the size of the 1689 buffer. 1690 1691 END-OF-FILE RULES 1692 The special rule "<<EOF>>" indicates actions which are to be 1693 taken when an end-of-file is encountered and yywrap() 1694 returns non-zero (i.e., indicates no further files to pro- 1695 cess). The action must finish by doing one of four things: 1696 1697 - assigning yyin to a new input file (in previous ver- 1698 sions of flex, after doing the assignment you had to 1699 call the special action YY_NEW_FILE; this is no longer 1700 necessary); 1701 1702 - executing a return statement; 1703 1704 - executing the special yyterminate() action; 1705 1706 - or, switching to a new buffer using 1707 yy_switch_to_buffer() as shown in the example above. 1708 1709 1710 1711 1712 1713 Version 2.5 Last change: April 1995 26 1714 1715 1716 1717 1718 1719 1720 FLEX(1) USER COMMANDS FLEX(1) 1721 1722 1723 1724 <<EOF>> rules may not be used with other patterns; they may 1725 only be qualified with a list of start conditions. If an 1726 unqualified <<EOF>> rule is given, it applies to all start 1727 conditions which do not already have <<EOF>> actions. To 1728 specify an <<EOF>> rule for only the initial start condi- 1729 tion, use 1730 1731 <INITIAL><<EOF>> 1732 1733 1734 These rules are useful for catching things like unclosed 1735 comments. An example: 1736 1737 %x quote 1738 %% 1739 1740 ...other rules for dealing with quotes... 1741 1742 <quote><<EOF>> { 1743 error( "unterminated quote" ); 1744 yyterminate(); 1745 } 1746 <<EOF>> { 1747 if ( *++filelist ) 1748 yyin = fopen( *filelist, "r" ); 1749 else 1750 yyterminate(); 1751 } 1752 1753 1754 MISCELLANEOUS MACROS 1755 The macro YY_USER_ACTION can be defined to provide an action 1756 which is always executed prior to the matched rule's action. 1757 For example, it could be #define'd to call a routine to con- 1758 vert yytext to lower-case. When YY_USER_ACTION is invoked, 1759 the variable yy_act gives the number of the matched rule 1760 (rules are numbered starting with 1). Suppose you want to 1761 profile how often each of your rules is matched. The fol- 1762 lowing would do the trick: 1763 1764 #define YY_USER_ACTION ++ctr[yy_act] 1765 1766 where ctr is an array to hold the counts for the different 1767 rules. Note that the macro YY_NUM_RULES gives the total 1768 number of rules (including the default rule, even if you use 1769 -s), so a correct declaration for ctr is: 1770 1771 int ctr[YY_NUM_RULES]; 1772 1773 1774 The macro YY_USER_INIT may be defined to provide an action 1775 which is always executed before the first scan (and before 1776 1777 1778 1779 Version 2.5 Last change: April 1995 27 1780 1781 1782 1783 1784 1785 1786 FLEX(1) USER COMMANDS FLEX(1) 1787 1788 1789 1790 the scanner's internal initializations are done). For exam- 1791 ple, it could be used to call a routine to read in a data 1792 table or open a logging file. 1793 1794 The macro yy_set_interactive(is_interactive) can be used to 1795 control whether the current buffer is considered interac- 1796 tive. An interactive buffer is processed more slowly, but 1797 must be used when the scanner's input source is indeed 1798 interactive to avoid problems due to waiting to fill buffers 1799 (see the discussion of the -I flag below). A non-zero value 1800 in the macro invocation marks the buffer as interactive, a 1801 zero value as non-interactive. Note that use of this macro 1802 overrides %option always-interactive or %option never- 1803 interactive (see Options below). yy_set_interactive() must 1804 be invoked prior to beginning to scan the buffer that is (or 1805 is not) to be considered interactive. 1806 1807 The macro yy_set_bol(at_bol) can be used to control whether 1808 the current buffer's scanning context for the next token 1809 match is done as though at the beginning of a line. A non- 1810 zero macro argument makes rules anchored with 1811 1812 The macro YY_AT_BOL() returns true if the next token scanned 1813 from the current buffer will have '^' rules active, false 1814 otherwise. 1815 1816 In the generated scanner, the actions are all gathered in 1817 one large switch statement and separated using YY_BREAK, 1818 which may be redefined. By default, it is simply a "break", 1819 to separate each rule's action from the following rule's. 1820 Redefining YY_BREAK allows, for example, C++ users to 1821 #define YY_BREAK to do nothing (while being very careful 1822 that every rule ends with a "break" or a "return"!) to avoid 1823 suffering from unreachable statement warnings where because 1824 a rule's action ends with "return", the YY_BREAK is inacces- 1825 sible. 1826 1827 VALUES AVAILABLE TO THE USER 1828 This section summarizes the various values available to the 1829 user in the rule actions. 1830 1831 - char *yytext holds the text of the current token. It 1832 may be modified but not lengthened (you cannot append 1833 characters to the end). 1834 1835 If the special directive %array appears in the first 1836 section of the scanner description, then yytext is 1837 instead declared char yytext[YYLMAX], where YYLMAX is a 1838 macro definition that you can redefine in the first 1839 section if you don't like the default value (generally 1840 8KB). Using %array results in somewhat slower 1841 scanners, but the value of yytext becomes immune to 1842 1843 1844 1845 Version 2.5 Last change: April 1995 28 1846 1847 1848 1849 1850 1851 1852 FLEX(1) USER COMMANDS FLEX(1) 1853 1854 1855 1856 calls to input() and unput(), which potentially destroy 1857 its value when yytext is a character pointer. The 1858 opposite of %array is %pointer, which is the default. 1859 1860 You cannot use %array when generating C++ scanner 1861 classes (the -+ flag). 1862 1863 - int yyleng holds the length of the current token. 1864 1865 - FILE *yyin is the file which by default flex reads 1866 from. It may be redefined but doing so only makes 1867 sense before scanning begins or after an EOF has been 1868 encountered. Changing it in the midst of scanning will 1869 have unexpected results since flex buffers its input; 1870 use yyrestart() instead. Once scanning terminates 1871 because an end-of-file has been seen, you can assign 1872 yyin at the new input file and then call the scanner 1873 again to continue scanning. 1874 1875 - void yyrestart( FILE *new_file ) may be called to point 1876 yyin at the new input file. The switch-over to the new 1877 file is immediate (any previously buffered-up input is 1878 lost). Note that calling yyrestart() with yyin as an 1879 argument thus throws away the current input buffer and 1880 continues scanning the same input file. 1881 1882 - FILE *yyout is the file to which ECHO actions are done. 1883 It can be reassigned by the user. 1884 1885 - YY_CURRENT_BUFFER returns a YY_BUFFER_STATE handle to 1886 the current buffer. 1887 1888 - YY_START returns an integer value corresponding to the 1889 current start condition. You can subsequently use this 1890 value with BEGIN to return to that start condition. 1891 1892 INTERFACING WITH YACC 1893 One of the main uses of flex is as a companion to the yacc 1894 parser-generator. yacc parsers expect to call a routine 1895 named yylex() to find the next input token. The routine is 1896 supposed to return the type of the next token as well as 1897 putting any associated value in the global yylval. To use 1898 flex with yacc, one specifies the -d option to yacc to 1899 instruct it to generate the file y.tab.h containing defini- 1900 tions of all the %tokens appearing in the yacc input. This 1901 file is then included in the flex scanner. For example, if 1902 one of the tokens is "TOK_NUMBER", part of the scanner might 1903 look like: 1904 1905 %{ 1906 #include "y.tab.h" 1907 %} 1908 1909 1910 1911 Version 2.5 Last change: April 1995 29 1912 1913 1914 1915 1916 1917 1918 FLEX(1) USER COMMANDS FLEX(1) 1919 1920 1921 1922 %% 1923 1924 [0-9]+ yylval = atoi( yytext ); return TOK_NUMBER; 1925 1926 1927 OPTIONS 1928 flex has the following options: 1929 1930 -b Generate backing-up information to lex.backup. This is 1931 a list of scanner states which require backing up and 1932 the input characters on which they do so. By adding 1933 rules one can remove backing-up states. If all 1934 backing-up states are eliminated and -Cf or -CF is 1935 used, the generated scanner will run faster (see the -p 1936 flag). Only users who wish to squeeze every last cycle 1937 out of their scanners need worry about this option. 1938 (See the section on Performance Considerations below.) 1939 1940 -c is a do-nothing, deprecated option included for POSIX 1941 compliance. 1942 1943 -d makes the generated scanner run in debug mode. When- 1944 ever a pattern is recognized and the global 1945 yy_flex_debug is non-zero (which is the default), the 1946 scanner will write to stderr a line of the form: 1947 1948 --accepting rule at line 53 ("the matched text") 1949 1950 The line number refers to the location of the rule in 1951 the file defining the scanner (i.e., the file that was 1952 fed to flex). Messages are also generated when the 1953 scanner backs up, accepts the default rule, reaches the 1954 end of its input buffer (or encounters a NUL; at this 1955 point, the two look the same as far as the scanner's 1956 concerned), or reaches an end-of-file. 1957 1958 -f specifies fast scanner. No table compression is done 1959 and stdio is bypassed. The result is large but fast. 1960 This option is equivalent to -Cfr (see below). 1961 1962 -h generates a "help" summary of flex's options to stdout 1963 and then exits. -? and --help are synonyms for -h. 1964 1965 -i instructs flex to generate a case-insensitive scanner. 1966 The case of letters given in the flex input patterns 1967 will be ignored, and tokens in the input will be 1968 matched regardless of case. The matched text given in 1969 yytext will have the preserved case (i.e., it will not 1970 be folded). 1971 1972 -l turns on maximum compatibility with the original AT&T 1973 lex implementation. Note that this does not mean full 1974 1975 1976 1977 Version 2.5 Last change: April 1995 30 1978 1979 1980 1981 1982 1983 1984 FLEX(1) USER COMMANDS FLEX(1) 1985 1986 1987 1988 compatibility. Use of this option costs a considerable 1989 amount of performance, and it cannot be used with the 1990 -+, -f, -F, -Cf, or -CF options. For details on the 1991 compatibilities it provides, see the section "Incompa- 1992 tibilities With Lex And POSIX" below. This option also 1993 results in the name YY_FLEX_LEX_COMPAT being #define'd 1994 in the generated scanner. 1995 1996 -n is another do-nothing, deprecated option included only 1997 for POSIX compliance. 1998 1999 -p generates a performance report to stderr. The report 2000 consists of comments regarding features of the flex 2001 input file which will cause a serious loss of perfor- 2002 mance in the resulting scanner. If you give the flag 2003 twice, you will also get comments regarding features 2004 that lead to minor performance losses. 2005 2006 Note that the use of REJECT, %option yylineno, and 2007 variable trailing context (see the Deficiencies / Bugs 2008 section below) entails a substantial performance 2009 penalty; use of yymore(), the ^ operator, and the -I 2010 flag entail minor performance penalties. 2011 2012 -s causes the default rule (that unmatched scanner input 2013 is echoed to stdout) to be suppressed. If the scanner 2014 encounters input that does not match any of its rules, 2015 it aborts with an error. This option is useful for 2016 finding holes in a scanner's rule set. 2017 2018 -t instructs flex to write the scanner it generates to 2019 standard output instead of lex.yy.c. 2020 2021 -v specifies that flex should write to stderr a summary of 2022 statistics regarding the scanner it generates. Most of 2023 the statistics are meaningless to the casual flex user, 2024 but the first line identifies the version of flex (same 2025 as reported by -V), and the next line the flags used 2026 when generating the scanner, including those that are 2027 on by default. 2028 2029 -w suppresses warning messages. 2030 2031 -B instructs flex to generate a batch scanner, the oppo- 2032 site of interactive scanners generated by -I (see 2033 below). In general, you use -B when you are certain 2034 that your scanner will never be used interactively, and 2035 you want to squeeze a little more performance out of 2036 it. If your goal is instead to squeeze out a lot more 2037 performance, you should be using the -Cf or -CF 2038 options (discussed below), which turn on -B automati- 2039 cally anyway. 2040 2041 2042 2043 Version 2.5 Last change: April 1995 31 2044 2045 2046 2047 2048 2049 2050 FLEX(1) USER COMMANDS FLEX(1) 2051 2052 2053 2054 -F specifies that the fast scanner table representation 2055 should be used (and stdio bypassed). This representa- 2056 tion is about as fast as the full table representation 2057 (-f), and for some sets of patterns will be consider- 2058 ably smaller (and for others, larger). In general, if 2059 the pattern set contains both "keywords" and a catch- 2060 all, "identifier" rule, such as in the set: 2061 2062 "case" return TOK_CASE; 2063 "switch" return TOK_SWITCH; 2064 ... 2065 "default" return TOK_DEFAULT; 2066 [a-z]+ return TOK_ID; 2067 2068 then you're better off using the full table representa- 2069 tion. If only the "identifier" rule is present and you 2070 then use a hash table or some such to detect the key- 2071 words, you're better off using -F. 2072 2073 This option is equivalent to -CFr (see below). It can- 2074 not be used with -+. 2075 2076 -I instructs flex to generate an interactive scanner. An 2077 interactive scanner is one that only looks ahead to 2078 decide what token has been matched if it absolutely 2079 must. It turns out that always looking one extra char- 2080 acter ahead, even if the scanner has already seen 2081 enough text to disambiguate the current token, is a bit 2082 faster than only looking ahead when necessary. But 2083 scanners that always look ahead give dreadful interac- 2084 tive performance; for example, when a user types a new- 2085 line, it is not recognized as a newline token until 2086 they enter another token, which often means typing in 2087 another whole line. 2088 2089 Flex scanners default to interactive unless you use the 2090 -Cf or -CF table-compression options (see below). 2091 That's because if you're looking for high-performance 2092 you should be using one of these options, so if you 2093 didn't, flex assumes you'd rather trade off a bit of 2094 run-time performance for intuitive interactive 2095 behavior. Note also that you cannot use -I in conjunc- 2096 tion with -Cf or -CF. Thus, this option is not really 2097 needed; it is on by default for all those cases in 2098 which it is allowed. 2099 2100 You can force a scanner to not be interactive by using 2101 -B (see above). 2102 2103 -L instructs flex not to generate #line directives. 2104 Without this option, flex peppers the generated scanner 2105 with #line directives so error messages in the actions 2106 2107 2108 2109 Version 2.5 Last change: April 1995 32 2110 2111 2112 2113 2114 2115 2116 FLEX(1) USER COMMANDS FLEX(1) 2117 2118 2119 2120 will be correctly located with respect to either the 2121 original flex input file (if the errors are due to code 2122 in the input file), or lex.yy.c (if the errors are 2123 flex's fault -- you should report these sorts of errors 2124 to the email address given below). 2125 2126 -T makes flex run in trace mode. It will generate a lot 2127 of messages to stderr concerning the form of the input 2128 and the resultant non-deterministic and deterministic 2129 finite automata. This option is mostly for use in 2130 maintaining flex. 2131 2132 -V prints the version number to stdout and exits. --ver- 2133 sion is a synonym for -V. 2134 2135 -7 instructs flex to generate a 7-bit scanner, i.e., one 2136 which can only recognized 7-bit characters in its 2137 input. The advantage of using -7 is that the scanner's 2138 tables can be up to half the size of those generated 2139 using the -8 option (see below). The disadvantage is 2140 that such scanners often hang or crash if their input 2141 contains an 8-bit character. 2142 2143 Note, however, that unless you generate your scanner 2144 using the -Cf or -CF table compression options, use of 2145 -7 will save only a small amount of table space, and 2146 make your scanner considerably less portable. Flex's 2147 default behavior is to generate an 8-bit scanner unless 2148 you use the -Cf or -CF, in which case flex defaults to 2149 generating 7-bit scanners unless your site was always 2150 configured to generate 8-bit scanners (as will often be 2151 the case with non-USA sites). You can tell whether 2152 flex generated a 7-bit or an 8-bit scanner by inspect- 2153 ing the flag summary in the -v output as described 2154 above. 2155 2156 Note that if you use -Cfe or -CFe (those table compres- 2157 sion options, but also using equivalence classes as 2158 discussed see below), flex still defaults to generating 2159 an 8-bit scanner, since usually with these compression 2160 options full 8-bit tables are not much more expensive 2161 than 7-bit tables. 2162 2163 -8 instructs flex to generate an 8-bit scanner, i.e., one 2164 which can recognize 8-bit characters. This flag is 2165 only needed for scanners generated using -Cf or -CF, as 2166 otherwise flex defaults to generating an 8-bit scanner 2167 anyway. 2168 2169 See the discussion of -7 above for flex's default 2170 behavior and the tradeoffs between 7-bit and 8-bit 2171 scanners. 2172 2173 2174 2175 Version 2.5 Last change: April 1995 33 2176 2177 2178 2179 2180 2181 2182 FLEX(1) USER COMMANDS FLEX(1) 2183 2184 2185 2186 -+ specifies that you want flex to generate a C++ scanner 2187 class. See the section on Generating C++ Scanners 2188 below for details. 2189 2190 -C[aefFmr] 2191 controls the degree of table compression and, more gen- 2192 erally, trade-offs between small scanners and fast 2193 scanners. 2194 2195 -Ca ("align") instructs flex to trade off larger tables 2196 in the generated scanner for faster performance because 2197 the elements of the tables are better aligned for 2198 memory access and computation. On some RISC architec- 2199 tures, fetching and manipulating longwords is more 2200 efficient than with smaller-sized units such as short- 2201 words. This option can double the size of the tables 2202 used by your scanner. 2203 2204 -Ce directs flex to construct equivalence classes, 2205 i.e., sets of characters which have identical lexical 2206 properties (for example, if the only appearance of 2207 digits in the flex input is in the character class 2208 "[0-9]" then the digits '0', '1', ..., '9' will all be 2209 put in the same equivalence class). Equivalence 2210 classes usually give dramatic reductions in the final 2211 table/object file sizes (typically a factor of 2-5) and 2212 are pretty cheap performance-wise (one array look-up 2213 per character scanned). 2214 2215 -Cf specifies that the full scanner tables should be 2216 generated - flex should not compress the tables by tak- 2217 ing advantages of similar transition functions for dif- 2218 ferent states. 2219 2220 -CF specifies that the alternate fast scanner represen- 2221 tation (described above under the -F flag) should be 2222 used. This option cannot be used with -+. 2223 2224 -Cm directs flex to construct meta-equivalence classes, 2225 which are sets of equivalence classes (or characters, 2226 if equivalence classes are not being used) that are 2227 commonly used together. Meta-equivalence classes are 2228 often a big win when using compressed tables, but they 2229 have a moderate performance impact (one or two "if" 2230 tests and one array look-up per character scanned). 2231 2232 -Cr causes the generated scanner to bypass use of the 2233 standard I/O library (stdio) for input. Instead of 2234 calling fread() or getc(), the scanner will use the 2235 read() system call, resulting in a performance gain 2236 which varies from system to system, but in general is 2237 probably negligible unless you are also using -Cf or 2238 2239 2240 2241 Version 2.5 Last change: April 1995 34 2242 2243 2244 2245 2246 2247 2248 FLEX(1) USER COMMANDS FLEX(1) 2249 2250 2251 2252 -CF. Using -Cr can cause strange behavior if, for exam- 2253 ple, you read from yyin using stdio prior to calling 2254 the scanner (because the scanner will miss whatever 2255 text your previous reads left in the stdio input 2256 buffer). 2257 2258 -Cr has no effect if you define YY_INPUT (see The Gen- 2259 erated Scanner above). 2260 2261 A lone -C specifies that the scanner tables should be 2262 compressed but neither equivalence classes nor meta- 2263 equivalence classes should be used. 2264 2265 The options -Cf or -CF and -Cm do not make sense 2266 together - there is no opportunity for meta-equivalence 2267 classes if the table is not being compressed. Other- 2268 wise the options may be freely mixed, and are cumula- 2269 tive. 2270 2271 The default setting is -Cem, which specifies that flex 2272 should generate equivalence classes and meta- 2273 equivalence classes. This setting provides the highest 2274 degree of table compression. You can trade off 2275 faster-executing scanners at the cost of larger tables 2276 with the following generally being true: 2277 2278 slowest & smallest 2279 -Cem 2280 -Cm 2281 -Ce 2282 -C 2283 -C{f,F}e 2284 -C{f,F} 2285 -C{f,F}a 2286 fastest & largest 2287 2288 Note that scanners with the smallest tables are usually 2289 generated and compiled the quickest, so during develop- 2290 ment you will usually want to use the default, maximal 2291 compression. 2292 2293 -Cfe is often a good compromise between speed and size 2294 for production scanners. 2295 2296 -ooutput 2297 directs flex to write the scanner to the file output 2298 instead of lex.yy.c. If you combine -o with the -t 2299 option, then the scanner is written to stdout but its 2300 #line directives (see the -L option above) refer to the 2301 file output. 2302 2303 -Pprefix 2304 2305 2306 2307 Version 2.5 Last change: April 1995 35 2308 2309 2310 2311 2312 2313 2314 FLEX(1) USER COMMANDS FLEX(1) 2315 2316 2317 2318 changes the default yy prefix used by flex for all 2319 globally-visible variable and function names to instead 2320 be prefix. For example, -Pfoo changes the name of 2321 yytext to footext. It also changes the name of the 2322 default output file from lex.yy.c to lex.foo.c. Here 2323 are all of the names affected: 2324 2325 yy_create_buffer 2326 yy_delete_buffer 2327 yy_flex_debug 2328 yy_init_buffer 2329 yy_flush_buffer 2330 yy_load_buffer_state 2331 yy_switch_to_buffer 2332 yyin 2333 yyleng 2334 yylex 2335 yylineno 2336 yyout 2337 yyrestart 2338 yytext 2339 yywrap 2340 2341 (If you are using a C++ scanner, then only yywrap and 2342 yyFlexLexer are affected.) Within your scanner itself, 2343 you can still refer to the global variables and func- 2344 tions using either version of their name; but exter- 2345 nally, they have the modified name. 2346 2347 This option lets you easily link together multiple flex 2348 programs into the same executable. Note, though, that 2349 using this option also renames yywrap(), so you now 2350 must either provide your own (appropriately-named) ver- 2351 sion of the routine for your scanner, or use %option 2352 noyywrap, as linking with -lfl no longer provides one 2353 for you by default. 2354 2355 -Sskeleton_file 2356 overrides the default skeleton file from which flex 2357 constructs its scanners. You'll never need this option 2358 unless you are doing flex maintenance or development. 2359 2360 flex also provides a mechanism for controlling options 2361 within the scanner specification itself, rather than from 2362 the flex command-line. This is done by including %option 2363 directives in the first section of the scanner specifica- 2364 tion. You can specify multiple options with a single 2365 %option directive, and multiple directives in the first sec- 2366 tion of your flex input file. 2367 2368 Most options are given simply as names, optionally preceded 2369 by the word "no" (with no intervening whitespace) to negate 2370 2371 2372 2373 Version 2.5 Last change: April 1995 36 2374 2375 2376 2377 2378 2379 2380 FLEX(1) USER COMMANDS FLEX(1) 2381 2382 2383 2384 their meaning. A number are equivalent to flex flags or 2385 their negation: 2386 2387 7bit -7 option 2388 8bit -8 option 2389 align -Ca option 2390 backup -b option 2391 batch -B option 2392 c++ -+ option 2393 2394 caseful or 2395 case-sensitive opposite of -i (default) 2396 2397 case-insensitive or 2398 caseless -i option 2399 2400 debug -d option 2401 default opposite of -s option 2402 ecs -Ce option 2403 fast -F option 2404 full -f option 2405 interactive -I option 2406 lex-compat -l option 2407 meta-ecs -Cm option 2408 perf-report -p option 2409 read -Cr option 2410 stdout -t option 2411 verbose -v option 2412 warn opposite of -w option 2413 (use "%option nowarn" for -w) 2414 2415 array equivalent to "%array" 2416 pointer equivalent to "%pointer" (default) 2417 2418 Some %option's provide features otherwise not available: 2419 2420 always-interactive 2421 instructs flex to generate a scanner which always con- 2422 siders its input "interactive". Normally, on each new 2423 input file the scanner calls isatty() in an attempt to 2424 determine whether the scanner's input source is 2425 interactive and thus should be read a character at a 2426 time. When this option is used, however, then no such 2427 call is made. 2428 2429 main directs flex to provide a default main() program for 2430 the scanner, which simply calls yylex(). This option 2431 implies noyywrap (see below). 2432 2433 never-interactive 2434 instructs flex to generate a scanner which never con- 2435 siders its input "interactive" (again, no call made to 2436 2437 2438 2439 Version 2.5 Last change: April 1995 37 2440 2441 2442 2443 2444 2445 2446 FLEX(1) USER COMMANDS FLEX(1) 2447 2448 2449 2450 isatty()). This is the opposite of always-interactive. 2451 2452 stack 2453 enables the use of start condition stacks (see Start 2454 Conditions above). 2455 2456 stdinit 2457 if set (i.e., %option stdinit) initializes yyin and 2458 yyout to stdin and stdout, instead of the default of 2459 nil. Some existing lex programs depend on this 2460 behavior, even though it is not compliant with ANSI C, 2461 which does not require stdin and stdout to be compile- 2462 time constant. 2463 2464 yylineno 2465 directs flex to generate a scanner that maintains the 2466 number of the current line read from its input in the 2467 global variable yylineno. This option is implied by 2468 %option lex-compat. 2469 2470 yywrap 2471 if unset (i.e., %option noyywrap), makes the scanner 2472 not call yywrap() upon an end-of-file, but simply 2473 assume that there are no more files to scan (until the 2474 user points yyin at a new file and calls yylex() 2475 again). 2476 2477 flex scans your rule actions to determine whether you use 2478 the REJECT or yymore() features. The reject and yymore 2479 options are available to override its decision as to whether 2480 you use the options, either by setting them (e.g., %option 2481 reject) to indicate the feature is indeed used, or unsetting 2482 them to indicate it actually is not used (e.g., %option 2483 noyymore). 2484 2485 Three options take string-delimited values, offset with '=': 2486 2487 %option outfile="ABC" 2488 2489 is equivalent to -oABC, and 2490 2491 %option prefix="XYZ" 2492 2493 is equivalent to -PXYZ. Finally, 2494 2495 %option yyclass="foo" 2496 2497 only applies when generating a C++ scanner ( -+ option). It 2498 informs flex that you have derived foo as a subclass of 2499 yyFlexLexer, so flex will place your actions in the member 2500 function foo::yylex() instead of yyFlexLexer::yylex(). It 2501 also generates a yyFlexLexer::yylex() member function that 2502 2503 2504 2505 Version 2.5 Last change: April 1995 38 2506 2507 2508 2509 2510 2511 2512 FLEX(1) USER COMMANDS FLEX(1) 2513 2514 2515 2516 emits a run-time error (by invoking 2517 yyFlexLexer::LexerError()) if called. See Generating C++ 2518 Scanners, below, for additional information. 2519 2520 A number of options are available for lint purists who want 2521 to suppress the appearance of unneeded routines in the gen- 2522 erated scanner. Each of the following, if unset (e.g., 2523 %option nounput ), results in the corresponding routine not 2524 appearing in the generated scanner: 2525 2526 input, unput 2527 yy_push_state, yy_pop_state, yy_top_state 2528 yy_scan_buffer, yy_scan_bytes, yy_scan_string 2529 2530 (though yy_push_state() and friends won't appear anyway 2531 unless you use %option stack). 2532 2533 PERFORMANCE CONSIDERATIONS 2534 The main design goal of flex is that it generate high- 2535 performance scanners. It has been optimized for dealing 2536 well with large sets of rules. Aside from the effects on 2537 scanner speed of the table compression -C options outlined 2538 above, there are a number of options/actions which degrade 2539 performance. These are, from most expensive to least: 2540 2541 REJECT 2542 %option yylineno 2543 arbitrary trailing context 2544 2545 pattern sets that require backing up 2546 %array 2547 %option interactive 2548 %option always-interactive 2549 2550 '^' beginning-of-line operator 2551 yymore() 2552 2553 with the first three all being quite expensive and the last 2554 two being quite cheap. Note also that unput() is imple- 2555 mented as a routine call that potentially does quite a bit 2556 of work, while yyless() is a quite-cheap macro; so if just 2557 putting back some excess text you scanned, use yyless(). 2558 2559 REJECT should be avoided at all costs when performance is 2560 important. It is a particularly expensive option. 2561 2562 Getting rid of backing up is messy and often may be an enor- 2563 mous amount of work for a complicated scanner. In princi- 2564 pal, one begins by using the -b flag to generate a 2565 lex.backup file. For example, on the input 2566 2567 %% 2568 2569 2570 2571 Version 2.5 Last change: April 1995 39 2572 2573 2574 2575 2576 2577 2578 FLEX(1) USER COMMANDS FLEX(1) 2579 2580 2581 2582 foo return TOK_KEYWORD; 2583 foobar return TOK_KEYWORD; 2584 2585 the file looks like: 2586 2587 State #6 is non-accepting - 2588 associated rule line numbers: 2589 2 3 2590 out-transitions: [ o ] 2591 jam-transitions: EOF [ \001-n p-\177 ] 2592 2593 State #8 is non-accepting - 2594 associated rule line numbers: 2595 3 2596 out-transitions: [ a ] 2597 jam-transitions: EOF [ \001-` b-\177 ] 2598 2599 State #9 is non-accepting - 2600 associated rule line numbers: 2601 3 2602 out-transitions: [ r ] 2603 jam-transitions: EOF [ \001-q s-\177 ] 2604 2605 Compressed tables always back up. 2606 2607 The first few lines tell us that there's a scanner state in 2608 which it can make a transition on an 'o' but not on any 2609 other character, and that in that state the currently 2610 scanned text does not match any rule. The state occurs when 2611 trying to match the rules found at lines 2 and 3 in the 2612 input file. If the scanner is in that state and then reads 2613 something other than an 'o', it will have to back up to find 2614 a rule which is matched. With a bit of headscratching one 2615 can see that this must be the state it's in when it has seen 2616 "fo". When this has happened, if anything other than 2617 another 'o' is seen, the scanner will have to back up to 2618 simply match the 'f' (by the default rule). 2619 2620 The comment regarding State #8 indicates there's a problem 2621 when "foob" has been scanned. Indeed, on any character 2622 other than an 'a', the scanner will have to back up to 2623 accept "foo". Similarly, the comment for State #9 concerns 2624 when "fooba" has been scanned and an 'r' does not follow. 2625 2626 The final comment reminds us that there's no point going to 2627 all the trouble of removing backing up from the rules unless 2628 we're using -Cf or -CF, since there's no performance gain 2629 doing so with compressed scanners. 2630 2631 The way to remove the backing up is to add "error" rules: 2632 2633 %% 2634 2635 2636 2637 Version 2.5 Last change: April 1995 40 2638 2639 2640 2641 2642 2643 2644 FLEX(1) USER COMMANDS FLEX(1) 2645 2646 2647 2648 foo return TOK_KEYWORD; 2649 foobar return TOK_KEYWORD; 2650 2651 fooba | 2652 foob | 2653 fo { 2654 /* false alarm, not really a keyword */ 2655 return TOK_ID; 2656 } 2657 2658 2659 Eliminating backing up among a list of keywords can also be 2660 done using a "catch-all" rule: 2661 2662 %% 2663 foo return TOK_KEYWORD; 2664 foobar return TOK_KEYWORD; 2665 2666 [a-z]+ return TOK_ID; 2667 2668 This is usually the best solution when appropriate. 2669 2670 Backing up messages tend to cascade. With a complicated set 2671 of rules it's not uncommon to get hundreds of messages. If 2672 one can decipher them, though, it often only takes a dozen 2673 or so rules to eliminate the backing up (though it's easy to 2674 make a mistake and have an error rule accidentally match a 2675 valid token. A possible future flex feature will be to 2676 automatically add rules to eliminate backing up). 2677 2678 It's important to keep in mind that you gain the benefits of 2679 eliminating backing up only if you eliminate every instance 2680 of backing up. Leaving just one means you gain nothing. 2681 2682 Variable trailing context (where both the leading and trail- 2683 ing parts do not have a fixed length) entails almost the 2684 same performance loss as REJECT (i.e., substantial). So 2685 when possible a rule like: 2686 2687 %% 2688 mouse|rat/(cat|dog) run(); 2689 2690 is better written: 2691 2692 %% 2693 mouse/cat|dog run(); 2694 rat/cat|dog run(); 2695 2696 or as 2697 2698 %% 2699 mouse|rat/cat run(); 2700 2701 2702 2703 Version 2.5 Last change: April 1995 41 2704 2705 2706 2707 2708 2709 2710 FLEX(1) USER COMMANDS FLEX(1) 2711 2712 2713 2714 mouse|rat/dog run(); 2715 2716 Note that here the special '|' action does not provide any 2717 savings, and can even make things worse (see Deficiencies / 2718 Bugs below). 2719 2720 Another area where the user can increase a scanner's perfor- 2721 mance (and one that's easier to implement) arises from the 2722 fact that the longer the tokens matched, the faster the 2723 scanner will run. This is because with long tokens the pro- 2724 cessing of most input characters takes place in the (short) 2725 inner scanning loop, and does not often have to go through 2726 the additional work of setting up the scanning environment 2727 (e.g., yytext) for the action. Recall the scanner for C 2728 comments: 2729 2730 %x comment 2731 %% 2732 int line_num = 1; 2733 2734 "/*" BEGIN(comment); 2735 2736 <comment>[^*\n]* 2737 <comment>"*"+[^*/\n]* 2738 <comment>\n ++line_num; 2739 <comment>"*"+"/" BEGIN(INITIAL); 2740 2741 This could be sped up by writing it as: 2742 2743 %x comment 2744 %% 2745 int line_num = 1; 2746 2747 "/*" BEGIN(comment); 2748 2749 <comment>[^*\n]* 2750 <comment>[^*\n]*\n ++line_num; 2751 <comment>"*"+[^*/\n]* 2752 <comment>"*"+[^*/\n]*\n ++line_num; 2753 <comment>"*"+"/" BEGIN(INITIAL); 2754 2755 Now instead of each newline requiring the processing of 2756 another action, recognizing the newlines is "distributed" 2757 over the other rules to keep the matched text as long as 2758 possible. Note that adding rules does not slow down the 2759 scanner! The speed of the scanner is independent of the 2760 number of rules or (modulo the considerations given at the 2761 beginning of this section) how complicated the rules are 2762 with regard to operators such as '*' and '|'. 2763 2764 A final example in speeding up a scanner: suppose you want 2765 to scan through a file containing identifiers and keywords, 2766 2767 2768 2769 Version 2.5 Last change: April 1995 42 2770 2771 2772 2773 2774 2775 2776 FLEX(1) USER COMMANDS FLEX(1) 2777 2778 2779 2780 one per line and with no other extraneous characters, and 2781 recognize all the keywords. A natural first approach is: 2782 2783 %% 2784 asm | 2785 auto | 2786 break | 2787 ... etc ... 2788 volatile | 2789 while /* it's a keyword */ 2790 2791 .|\n /* it's not a keyword */ 2792 2793 To eliminate the back-tracking, introduce a catch-all rule: 2794 2795 %% 2796 asm | 2797 auto | 2798 break | 2799 ... etc ... 2800 volatile | 2801 while /* it's a keyword */ 2802 2803 [a-z]+ | 2804 .|\n /* it's not a keyword */ 2805 2806 Now, if it's guaranteed that there's exactly one word per 2807 line, then we can reduce the total number of matches by a 2808 half by merging in the recognition of newlines with that of 2809 the other tokens: 2810 2811 %% 2812 asm\n | 2813 auto\n | 2814 break\n | 2815 ... etc ... 2816 volatile\n | 2817 while\n /* it's a keyword */ 2818 2819 [a-z]+\n | 2820 .|\n /* it's not a keyword */ 2821 2822 One has to be careful here, as we have now reintroduced 2823 backing up into the scanner. In particular, while we know 2824 that there will never be any characters in the input stream 2825 other than letters or newlines, flex can't figure this out, 2826 and it will plan for possibly needing to back up when it has 2827 scanned a token like "auto" and then the next character is 2828 something other than a newline or a letter. Previously it 2829 would then just match the "auto" rule and be done, but now 2830 it has no "auto" rule, only a "auto\n" rule. To eliminate 2831 the possibility of backing up, we could either duplicate all 2832 2833 2834 2835 Version 2.5 Last change: April 1995 43 2836 2837 2838 2839 2840 2841 2842 FLEX(1) USER COMMANDS FLEX(1) 2843 2844 2845 2846 rules but without final newlines, or, since we never expect 2847 to encounter such an input and therefore don't how it's 2848 classified, we can introduce one more catch-all rule, this 2849 one which doesn't include a newline: 2850 2851 %% 2852 asm\n | 2853 auto\n | 2854 break\n | 2855 ... etc ... 2856 volatile\n | 2857 while\n /* it's a keyword */ 2858 2859 [a-z]+\n | 2860 [a-z]+ | 2861 .|\n /* it's not a keyword */ 2862 2863 Compiled with -Cf, this is about as fast as one can get a 2864 flex scanner to go for this particular problem. 2865 2866 A final note: flex is slow when matching NUL's, particularly 2867 when a token contains multiple NUL's. It's best to write 2868 rules which match short amounts of text if it's anticipated 2869 that the text will often include NUL's. 2870 2871 Another final note regarding performance: as mentioned above 2872 in the section How the Input is Matched, dynamically resiz- 2873 ing yytext to accommodate huge tokens is a slow process 2874 because it presently requires that the (huge) token be res- 2875 canned from the beginning. Thus if performance is vital, 2876 you should attempt to match "large" quantities of text but 2877 not "huge" quantities, where the cutoff between the two is 2878 at about 8K characters/token. 2879 2880 GENERATING C++ SCANNERS 2881 flex provides two different ways to generate scanners for 2882 use with C++. The first way is to simply compile a scanner 2883 generated by flex using a C++ compiler instead of a C com- 2884 piler. You should not encounter any compilations errors 2885 (please report any you find to the email address given in 2886 the Author section below). You can then use C++ code in 2887 your rule actions instead of C code. Note that the default 2888 input source for your scanner remains yyin, and default 2889 echoing is still done to yyout. Both of these remain FILE * 2890 variables and not C++ streams. 2891 2892 You can also use flex to generate a C++ scanner class, using 2893 the -+ option (or, equivalently, %option c++), which is 2894 automatically specified if the name of the flex executable 2895 ends in a '+', such as flex++. When using this option, flex 2896 defaults to generating the scanner to the file lex.yy.cc 2897 instead of lex.yy.c. The generated scanner includes the 2898 2899 2900 2901 Version 2.5 Last change: April 1995 44 2902 2903 2904 2905 2906 2907 2908 FLEX(1) USER COMMANDS FLEX(1) 2909 2910 2911 2912 header file FlexLexer.h, which defines the interface to two 2913 C++ classes. 2914 2915 The first class, FlexLexer, provides an abstract base class 2916 defining the general scanner class interface. It provides 2917 the following member functions: 2918 2919 const char* YYText() 2920 returns the text of the most recently matched token, 2921 the equivalent of yytext. 2922 2923 int YYLeng() 2924 returns the length of the most recently matched token, 2925 the equivalent of yyleng. 2926 2927 int lineno() const 2928 returns the current input line number (see %option 2929 yylineno), or 1 if %option yylineno was not used. 2930 2931 void set_debug( int flag ) 2932 sets the debugging flag for the scanner, equivalent to 2933 assigning to yy_flex_debug (see the Options section 2934 above). Note that you must build the scanner using 2935 %option debug to include debugging information in it. 2936 2937 int debug() const 2938 returns the current setting of the debugging flag. 2939 2940 Also provided are member functions equivalent to 2941 yy_switch_to_buffer(), yy_create_buffer() (though the first 2942 argument is an istream* object pointer and not a FILE*), 2943 yy_flush_buffer(), yy_delete_buffer(), and yyrestart() 2944 (again, the first argument is a istream* object pointer). 2945 2946 The second class defined in FlexLexer.h is yyFlexLexer, 2947 which is derived from FlexLexer. It defines the following 2948 additional member functions: 2949 2950 yyFlexLexer( istream* arg_yyin = 0, ostream* arg_yyout = 0 ) 2951 constructs a yyFlexLexer object using the given streams 2952 for input and output. If not specified, the streams 2953 default to cin and cout, respectively. 2954 2955 virtual int yylex() 2956 performs the same role is yylex() does for ordinary 2957 flex scanners: it scans the input stream, consuming 2958 tokens, until a rule's action returns a value. If you 2959 derive a subclass S from yyFlexLexer and want to access 2960 the member functions and variables of S inside yylex(), 2961 then you need to use %option yyclass="S" to inform flex 2962 that you will be using that subclass instead of yyFlex- 2963 Lexer. In this case, rather than generating 2964 2965 2966 2967 Version 2.5 Last change: April 1995 45 2968 2969 2970 2971 2972 2973 2974 FLEX(1) USER COMMANDS FLEX(1) 2975 2976 2977 2978 yyFlexLexer::yylex(), flex generates S::yylex() (and 2979 also generates a dummy yyFlexLexer::yylex() that calls 2980 yyFlexLexer::LexerError() if called). 2981 2982 virtual void switch_streams(istream* new_in = 0, 2983 ostream* new_out = 0) reassigns yyin to new_in (if 2984 non-nil) and yyout to new_out (ditto), deleting the 2985 previous input buffer if yyin is reassigned. 2986 2987 int yylex( istream* new_in, ostream* new_out = 0 ) 2988 first switches the input streams via switch_streams( 2989 new_in, new_out ) and then returns the value of 2990 yylex(). 2991 2992 In addition, yyFlexLexer defines the following protected 2993 virtual functions which you can redefine in derived classes 2994 to tailor the scanner: 2995 2996 virtual int LexerInput( char* buf, int max_size ) 2997 reads up to max_size characters into buf and returns 2998 the number of characters read. To indicate end-of- 2999 input, return 0 characters. Note that "interactive" 3000 scanners (see the -B and -I flags) define the macro 3001 YY_INTERACTIVE. If you redefine LexerInput() and need 3002 to take different actions depending on whether or not 3003 the scanner might be scanning an interactive input 3004 source, you can test for the presence of this name via 3005 #ifdef. 3006 3007 virtual void LexerOutput( const char* buf, int size ) 3008 writes out size characters from the buffer buf, which, 3009 while NUL-terminated, may also contain "internal" NUL's 3010 if the scanner's rules can match text with NUL's in 3011 them. 3012 3013 virtual void LexerError( const char* msg ) 3014 reports a fatal error message. The default version of 3015 this function writes the message to the stream cerr and 3016 exits. 3017 3018 Note that a yyFlexLexer object contains its entire scanning 3019 state. Thus you can use such objects to create reentrant 3020 scanners. You can instantiate multiple instances of the 3021 same yyFlexLexer class, and you can also combine multiple 3022 C++ scanner classes together in the same program using the 3023 -P option discussed above. 3024 3025 Finally, note that the %array feature is not available to 3026 C++ scanner classes; you must use %pointer (the default). 3027 3028 Here is an example of a simple C++ scanner: 3029 3030 3031 3032 3033 Version 2.5 Last change: April 1995 46 3034 3035 3036 3037 3038 3039 3040 FLEX(1) USER COMMANDS FLEX(1) 3041 3042 3043 3044 // An example of using the flex C++ scanner class. 3045 3046 %{ 3047 int mylineno = 0; 3048 %} 3049 3050 string \"[^\n"]+\" 3051 3052 ws [ \t]+ 3053 3054 alpha [A-Za-z] 3055 dig [0-9] 3056 name ({alpha}|{dig}|\$)({alpha}|{dig}|[_.\-/$])* 3057 num1 [-+]?{dig}+\.?([eE][-+]?{dig}+)? 3058 num2 [-+]?{dig}*\.{dig}+([eE][-+]?{dig}+)? 3059 number {num1}|{num2} 3060 3061 %% 3062 3063 {ws} /* skip blanks and tabs */ 3064 3065 "/*" { 3066 int c; 3067 3068 while((c = yyinput()) != 0) 3069 { 3070 if(c == '\n') 3071 ++mylineno; 3072 3073 else if(c == '*') 3074 { 3075 if((c = yyinput()) == '/') 3076 break; 3077 else 3078 unput(c); 3079 } 3080 } 3081 } 3082 3083 {number} cout << "number " << YYText() << '\n'; 3084 3085 \n mylineno++; 3086 3087 {name} cout << "name " << YYText() << '\n'; 3088 3089 {string} cout << "string " << YYText() << '\n'; 3090 3091 %% 3092 3093 int main( int /* argc */, char** /* argv */ ) 3094 { 3095 FlexLexer* lexer = new yyFlexLexer; 3096 3097 3098 3099 Version 2.5 Last change: April 1995 47 3100 3101 3102 3103 3104 3105 3106 FLEX(1) USER COMMANDS FLEX(1) 3107 3108 3109 3110 while(lexer->yylex() != 0) 3111 ; 3112 return 0; 3113 } 3114 If you want to create multiple (different) lexer classes, 3115 you use the -P flag (or the prefix= option) to rename each 3116 yyFlexLexer to some other xxFlexLexer. You then can include 3117 <FlexLexer.h> in your other sources once per lexer class, 3118 first renaming yyFlexLexer as follows: 3119 3120 #undef yyFlexLexer 3121 #define yyFlexLexer xxFlexLexer 3122 #include <FlexLexer.h> 3123 3124 #undef yyFlexLexer 3125 #define yyFlexLexer zzFlexLexer 3126 #include <FlexLexer.h> 3127 3128 if, for example, you used %option prefix="xx" for one of 3129 your scanners and %option prefix="zz" for the other. 3130 3131 IMPORTANT: the present form of the scanning class is experi- 3132 mental and may change considerably between major releases. 3133 3134 INCOMPATIBILITIES WITH LEX AND POSIX 3135 flex is a rewrite of the AT&T Unix lex tool (the two imple- 3136 mentations do not share any code, though), with some exten- 3137 sions and incompatibilities, both of which are of concern to 3138 those who wish to write scanners acceptable to either imple- 3139 mentation. Flex is fully compliant with the POSIX lex 3140 specification, except that when using %pointer (the 3141 default), a call to unput() destroys the contents of yytext, 3142 which is counter to the POSIX specification. 3143 3144 In this section we discuss all of the known areas of incom- 3145 patibility between flex, AT&T lex, and the POSIX specifica- 3146 tion. 3147 3148 flex's -l option turns on maximum compatibility with the 3149 original AT&T lex implementation, at the cost of a major 3150 loss in the generated scanner's performance. We note below 3151 which incompatibilities can be overcome using the -l option. 3152 3153 flex is fully compatible with lex with the following excep- 3154 tions: 3155 3156 - The undocumented lex scanner internal variable yylineno 3157 is not supported unless -l or %option yylineno is used. 3158 3159 yylineno should be maintained on a per-buffer basis, 3160 rather than a per-scanner (single global variable) 3161 basis. 3162 3163 3164 3165 Version 2.5 Last change: April 1995 48 3166 3167 3168 3169 3170 3171 3172 FLEX(1) USER COMMANDS FLEX(1) 3173 3174 3175 3176 yylineno is not part of the POSIX specification. 3177 3178 - The input() routine is not redefinable, though it may 3179 be called to read characters following whatever has 3180 been matched by a rule. If input() encounters an end- 3181 of-file the normal yywrap() processing is done. A 3182 ``real'' end-of-file is returned by input() as EOF. 3183 3184 Input is instead controlled by defining the YY_INPUT 3185 macro. 3186 3187 The flex restriction that input() cannot be redefined 3188 is in accordance with the POSIX specification, which 3189 simply does not specify any way of controlling the 3190 scanner's input other than by making an initial assign- 3191 ment to yyin. 3192 3193 - The unput() routine is not redefinable. This restric- 3194 tion is in accordance with POSIX. 3195 3196 - flex scanners are not as reentrant as lex scanners. In 3197 particular, if you have an interactive scanner and an 3198 interrupt handler which long-jumps out of the scanner, 3199 and the scanner is subsequently called again, you may 3200 get the following message: 3201 3202 fatal flex scanner internal error--end of buffer missed 3203 3204 To reenter the scanner, first use 3205 3206 yyrestart( yyin ); 3207 3208 Note that this call will throw away any buffered input; 3209 usually this isn't a problem with an interactive 3210 scanner. 3211 3212 Also note that flex C++ scanner classes are reentrant, 3213 so if using C++ is an option for you, you should use 3214 them instead. See "Generating C++ Scanners" above for 3215 details. 3216 3217 - output() is not supported. Output from the ECHO macro 3218 is done to the file-pointer yyout (default stdout). 3219 3220 output() is not part of the POSIX specification. 3221 3222 - lex does not support exclusive start conditions (%x), 3223 though they are in the POSIX specification. 3224 3225 - When definitions are expanded, flex encloses them in 3226 parentheses. With lex, the following: 3227 3228 3229 3230 3231 Version 2.5 Last change: April 1995 49 3232 3233 3234 3235 3236 3237 3238 FLEX(1) USER COMMANDS FLEX(1) 3239 3240 3241 3242 NAME [A-Z][A-Z0-9]* 3243 %% 3244 foo{NAME}? printf( "Found it\n" ); 3245 %% 3246 3247 will not match the string "foo" because when the macro 3248 is expanded the rule is equivalent to "foo[A-Z][A-Z0- 3249 9]*?" and the precedence is such that the '?' is asso- 3250 ciated with "[A-Z0-9]*". With flex, the rule will be 3251 expanded to "foo([A-Z][A-Z0-9]*)?" and so the string 3252 "foo" will match. 3253 3254 Note that if the definition begins with ^ or ends with 3255 $ then it is not expanded with parentheses, to allow 3256 these operators to appear in definitions without losing 3257 their special meanings. But the <s>, /, and <<EOF>> 3258 operators cannot be used in a flex definition. 3259 3260 Using -l results in the lex behavior of no parentheses 3261 around the definition. 3262 3263 The POSIX specification is that the definition be 3264 enclosed in parentheses. 3265 3266 - Some implementations of lex allow a rule's action to 3267 begin on a separate line, if the rule's pattern has 3268 trailing whitespace: 3269 3270 %% 3271 foo|bar<space here> 3272 { foobar_action(); } 3273 3274 flex does not support this feature. 3275 3276 - The lex %r (generate a Ratfor scanner) option is not 3277 supported. It is not part of the POSIX specification. 3278 3279 - After a call to unput(), yytext is undefined until the 3280 next token is matched, unless the scanner was built 3281 using %array. This is not the case with lex or the 3282 POSIX specification. The -l option does away with this 3283 incompatibility. 3284 3285 - The precedence of the {} (numeric range) operator is 3286 different. lex interprets "abc{1,3}" as "match one, 3287 two, or three occurrences of 'abc'", whereas flex 3288 interprets it as "match 'ab' followed by one, two, or 3289 three occurrences of 'c'". The latter is in agreement 3290 with the POSIX specification. 3291 3292 - The precedence of the ^ operator is different. lex 3293 interprets "^foo|bar" as "match either 'foo' at the 3294 3295 3296 3297 Version 2.5 Last change: April 1995 50 3298 3299 3300 3301 3302 3303 3304 FLEX(1) USER COMMANDS FLEX(1) 3305 3306 3307 3308 beginning of a line, or 'bar' anywhere", whereas flex 3309 interprets it as "match either 'foo' or 'bar' if they 3310 come at the beginning of a line". The latter is in 3311 agreement with the POSIX specification. 3312 3313 - The special table-size declarations such as %a sup- 3314 ported by lex are not required by flex scanners; flex 3315 ignores them. 3316 3317 - The name FLEX_SCANNER is #define'd so scanners may be 3318 written for use with either flex or lex. Scanners also 3319 include YY_FLEX_MAJOR_VERSION and YY_FLEX_MINOR_VERSION 3320 indicating which version of flex generated the scanner 3321 (for example, for the 2.5 release, these defines would 3322 be 2 and 5 respectively). 3323 3324 The following flex features are not included in lex or the 3325 POSIX specification: 3326 3327 C++ scanners 3328 %option 3329 start condition scopes 3330 start condition stacks 3331 interactive/non-interactive scanners 3332 yy_scan_string() and friends 3333 yyterminate() 3334 yy_set_interactive() 3335 yy_set_bol() 3336 YY_AT_BOL() 3337 <<EOF>> 3338 <*> 3339 YY_DECL 3340 YY_START 3341 YY_USER_ACTION 3342 YY_USER_INIT 3343 #line directives 3344 %{}'s around actions 3345 multiple actions on a line 3346 3347 plus almost all of the flex flags. The last feature in the 3348 list refers to the fact that with flex you can put multiple 3349 actions on the same line, separated with semi-colons, while 3350 with lex, the following 3351 3352 foo handle_foo(); ++num_foos_seen; 3353 3354 is (rather surprisingly) truncated to 3355 3356 foo handle_foo(); 3357 3358 flex does not truncate the action. Actions that are not 3359 enclosed in braces are simply terminated at the end of the 3360 3361 3362 3363 Version 2.5 Last change: April 1995 51 3364 3365 3366 3367 3368 3369 3370 FLEX(1) USER COMMANDS FLEX(1) 3371 3372 3373 3374 line. 3375 3376 DIAGNOSTICS 3377 warning, rule cannot be matched indicates that the given 3378 rule cannot be matched because it follows other rules that 3379 will always match the same text as it. For example, in the 3380 following "foo" cannot be matched because it comes after an 3381 identifier "catch-all" rule: 3382 3383 [a-z]+ got_identifier(); 3384 foo got_foo(); 3385 3386 Using REJECT in a scanner suppresses this warning. 3387 3388 warning, -s option given but default rule can be matched 3389 means that it is possible (perhaps only in a particular 3390 start condition) that the default rule (match any single 3391 character) is the only one that will match a particular 3392 input. Since -s was given, presumably this is not intended. 3393 3394 reject_used_but_not_detected undefined or 3395 yymore_used_but_not_detected undefined - These errors can 3396 occur at compile time. They indicate that the scanner uses 3397 REJECT or yymore() but that flex failed to notice the fact, 3398 meaning that flex scanned the first two sections looking for 3399 occurrences of these actions and failed to find any, but 3400 somehow you snuck some in (via a #include file, for exam- 3401 ple). Use %option reject or %option yymore to indicate to 3402 flex that you really do use these features. 3403 3404 flex scanner jammed - a scanner compiled with -s has encoun- 3405 tered an input string which wasn't matched by any of its 3406 rules. This error can also occur due to internal problems. 3407 3408 token too large, exceeds YYLMAX - your scanner uses %array 3409 and one of its rules matched a string longer than the YYLMAX 3410 constant (8K bytes by default). You can increase the value 3411 by #define'ing YYLMAX in the definitions section of your 3412 flex input. 3413 3414 scanner requires -8 flag to use the character 'x' - Your 3415 scanner specification includes recognizing the 8-bit charac- 3416 ter 'x' and you did not specify the -8 flag, and your 3417 scanner defaulted to 7-bit because you used the -Cf or -CF 3418 table compression options. See the discussion of the -7 3419 flag for details. 3420 3421 flex scanner push-back overflow - you used unput() to push 3422 back so much text that the scanner's buffer could not hold 3423 both the pushed-back text and the current token in yytext. 3424 Ideally the scanner should dynamically resize the buffer in 3425 this case, but at present it does not. 3426 3427 3428 3429 Version 2.5 Last change: April 1995 52 3430 3431 3432 3433 3434 3435 3436 FLEX(1) USER COMMANDS FLEX(1) 3437 3438 3439 3440 input buffer overflow, can't enlarge buffer because scanner 3441 uses REJECT - the scanner was working on matching an 3442 extremely large token and needed to expand the input buffer. 3443 This doesn't work with scanners that use REJECT. 3444 3445 fatal flex scanner internal error--end of buffer missed - 3446 This can occur in an scanner which is reentered after a 3447 long-jump has jumped out (or over) the scanner's activation 3448 frame. Before reentering the scanner, use: 3449 3450 yyrestart( yyin ); 3451 3452 or, as noted above, switch to using the C++ scanner class. 3453 3454 too many start conditions in <> you listed more start condi- 3455 tions in a <> construct than exist (so you must have listed 3456 at least one of them twice). 3457 3458 FILES 3459 -lfl library with which scanners must be linked. 3460 3461 lex.yy.c 3462 generated scanner (called lexyy.c on some systems). 3463 3464 lex.yy.cc 3465 generated C++ scanner class, when using -+. 3466 3467 <FlexLexer.h> 3468 header file defining the C++ scanner base class, Flex- 3469 Lexer, and its derived class, yyFlexLexer. 3470 3471 flex.skl 3472 skeleton scanner. This file is only used when building 3473 flex, not when flex executes. 3474 3475 lex.backup 3476 backing-up information for -b flag (called lex.bck on 3477 some systems). 3478 3479 DEFICIENCIES / BUGS 3480 Some trailing context patterns cannot be properly matched 3481 and generate warning messages ("dangerous trailing con- 3482 text"). These are patterns where the ending of the first 3483 part of the rule matches the beginning of the second part, 3484 such as "zx*/xy*", where the 'x*' matches the 'x' at the 3485 beginning of the trailing context. (Note that the POSIX 3486 draft states that the text matched by such patterns is unde- 3487 fined.) 3488 3489 For some trailing context rules, parts which are actually 3490 fixed-length are not recognized as such, leading to the 3491 abovementioned performance loss. In particular, parts using 3492 3493 3494 3495 Version 2.5 Last change: April 1995 53 3496 3497 3498 3499 3500 3501 3502 FLEX(1) USER COMMANDS FLEX(1) 3503 3504 3505 3506 '|' or {n} (such as "foo{3}") are always considered 3507 variable-length. 3508 3509 Combining trailing context with the special '|' action can 3510 result in fixed trailing context being turned into the more 3511 expensive variable trailing context. For example, in the 3512 following: 3513 3514 %% 3515 abc | 3516 xyz/def 3517 3518 3519 Use of unput() invalidates yytext and yyleng, unless the 3520 %array directive or the -l option has been used. 3521 3522 Pattern-matching of NUL's is substantially slower than 3523 matching other characters. 3524 3525 Dynamic resizing of the input buffer is slow, as it entails 3526 rescanning all the text matched so far by the current (gen- 3527 erally huge) token. 3528 3529 Due to both buffering of input and read-ahead, you cannot 3530 intermix calls to <stdio.h> routines, such as, for example, 3531 getchar(), with flex rules and expect it to work. Call 3532 input() instead. 3533 3534 The total table entries listed by the -v flag excludes the 3535 number of table entries needed to determine what rule has 3536 been matched. The number of entries is equal to the number 3537 of DFA states if the scanner does not use REJECT, and some- 3538 what greater than the number of states if it does. 3539 3540 REJECT cannot be used with the -f or -F options. 3541 3542 The flex internal algorithms need documentation. 3543 3544 SEE ALSO 3545 lex(1), yacc(1), sed(1), awk(1). 3546 3547 John Levine, Tony Mason, and Doug Brown, Lex & Yacc, 3548 O'Reilly and Associates. Be sure to get the 2nd edition. 3549 3550 M. E. Lesk and E. Schmidt, LEX - Lexical Analyzer Generator 3551 3552 Alfred Aho, Ravi Sethi and Jeffrey Ullman, Compilers: Prin- 3553 ciples, Techniques and Tools, Addison-Wesley (1986). 3554 Describes the pattern-matching techniques used by flex 3555 (deterministic finite automata). 3556 3557 3558 3559 3560 3561 Version 2.5 Last change: April 1995 54 3562 3563 3564 3565 3566 3567 3568 FLEX(1) USER COMMANDS FLEX(1) 3569 3570 3571 3572 AUTHOR 3573 Vern Paxson, with the help of many ideas and much inspira- 3574 tion from Van Jacobson. Original version by Jef Poskanzer. 3575 The fast table representation is a partial implementation of 3576 a design done by Van Jacobson. The implementation was done 3577 by Kevin Gong and Vern Paxson. 3578 3579 Thanks to the many flex beta-testers, feedbackers, and con- 3580 tributors, especially Francois Pinard, Casey Leedom, Robert 3581 Abramovitz, Stan Adermann, Terry Allen, David Barker- 3582 Plummer, John Basrai, Neal Becker, Nelson H.F. Beebe, 3583 benson (a] odi.com, Karl Berry, Peter A. Bigot, Simon Blanchard, 3584 Keith Bostic, Frederic Brehm, Ian Brockbank, Kin Cho, Nick 3585 Christopher, Brian Clapper, J.T. Conklin, Jason Coughlin, 3586 Bill Cox, Nick Cropper, Dave Curtis, Scott David Daniels, 3587 Chris G. Demetriou, Theo Deraadt, Mike Donahue, Chuck 3588 Doucette, Tom Epperly, Leo Eskin, Chris Faylor, Chris 3589 Flatters, Jon Forrest, Jeffrey Friedl, Joe Gayda, Kaveh R. 3590 Ghazi, Wolfgang Glunz, Eric Goldman, Christopher M. Gould, 3591 Ulrich Grepel, Peer Griebel, Jan Hajic, Charles Hemphill, 3592 NORO Hideo, Jarkko Hietaniemi, Scott Hofmann, Jeff Honig, 3593 Dana Hudes, Eric Hughes, John Interrante, Ceriel Jacobs, 3594 Michal Jaegermann, Sakari Jalovaara, Jeffrey R. Jones, Henry 3595 Juengst, Klaus Kaempf, Jonathan I. Kamens, Terrence O Kane, 3596 Amir Katz, ken (a] ken.hilco.com, Kevin B. Kenny, Steve Kirsch, 3597 Winfried Koenig, Marq Kole, Ronald Lamprecht, Greg Lee, 3598 Rohan Lenard, Craig Leres, John Levine, Steve Liddle, David 3599 Loffredo, Mike Long, Mohamed el Lozy, Brian Madsen, Malte, 3600 Joe Marshall, Bengt Martensson, Chris Metcalf, Luke Mewburn, 3601 Jim Meyering, R. Alexander Milowski, Erik Naggum, G.T. 3602 Nicol, Landon Noll, James Nordby, Marc Nozell, Richard 3603 Ohnemus, Karsten Pahnke, Sven Panne, Roland Pesch, Walter 3604 Pelissero, Gaumond Pierre, Esmond Pitt, Jef Poskanzer, Joe 3605 Rahmeh, Jarmo Raiha, Frederic Raimbault, Pat Rankin, Rick 3606 Richardson, Kevin Rodgers, Kai Uwe Rommel, Jim Roskind, 3607 Alberto Santini, Andreas Scherer, Darrell Schiebel, Raf 3608 Schietekat, Doug Schmidt, Philippe Schnoebelen, Andreas 3609 Schwab, Larry Schwimmer, Alex Siegel, Eckehard Stolz, Jan- 3610 Erik Strvmquist, Mike Stump, Paul Stuart, Dave Tallman, Ian 3611 Lance Taylor, Chris Thewalt, Richard M. Timoney, Jodi Tsai, 3612 Paul Tuinenga, Gary Weik, Frank Whaley, Gerhard Wilhelms, 3613 Kent Williams, Ken Yap, Ron Zellar, Nathan Zelle, David 3614 Zuhn, and those whose names have slipped my marginal mail- 3615 archiving skills but whose contributions are appreciated all 3616 the same. 3617 3618 Thanks to Keith Bostic, Jon Forrest, Noah Friedman, John 3619 Gilmore, Craig Leres, John Levine, Bob Mulcahy, G.T. Nicol, 3620 Francois Pinard, Rich Salz, and Richard Stallman for help 3621 with various distribution headaches. 3622 3623 3624 3625 3626 3627 Version 2.5 Last change: April 1995 55 3628 3629 3630 3631 3632 3633 3634 FLEX(1) USER COMMANDS FLEX(1) 3635 3636 3637 3638 Thanks to Esmond Pitt and Earle Horton for 8-bit character 3639 support; to Benson Margulies and Fred Burke for C++ support; 3640 to Kent Williams and Tom Epperly for C++ class support; to 3641 Ove Ewerlid for support of NUL's; and to Eric Hughes for 3642 support of multiple buffers. 3643 3644 This work was primarily done when I was with the Real Time 3645 Systems Group at the Lawrence Berkeley Laboratory in Berke- 3646 ley, CA. Many thanks to all there for the support I 3647 received. 3648 3649 Send comments to vern (a] ee.lbl.gov. 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 Version 2.5 Last change: April 1995 56 3694 3695 3696 3697
