1 \input texinfo.tex @c -*-texinfo-*-
5 @settitle Lexical Analysis With Flex, for Flex @value{VERSION}
6 @set authors Vern Paxson, Will Estes and John Millaway
11 @dircategory Programming
13 * flex: (flex). Fast lexical analyzer generator (lex replacement).
19 The flex manual is placed under the same licensing conditions as the
22 Copyright @copyright{} 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2012
25 Copyright @copyright{} 1990, 1997 The Regents of the University of California.
28 This code is derived from software contributed to Berkeley by
31 The United States Government has rights in this work pursuant
32 to contract no. DE-AC03-76SF00098 between the United States
33 Department of Energy and the University of California.
35 Redistribution and use in source and binary forms, with or without
36 modification, are permitted provided that the following conditions
41 Redistributions of source code must retain the above copyright
42 notice, this list of conditions and the following disclaimer.
45 Redistributions in binary form must reproduce the above copyright
46 notice, this list of conditions and the following disclaimer in the
47 documentation and/or other materials provided with the distribution.
50 Neither the name of the University nor the names of its contributors
51 may be used to endorse or promote products derived from this software
52 without specific prior written permission.
54 THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
55 IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
56 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
61 @title Lexical Analysis with Flex
62 @subtitle Edition @value{EDITION}, @value{UPDATED}
63 @author @value{authors}
65 @vskip 0pt plus 1filll
70 @node Top, Copyright, (dir), (dir)
73 This manual describes @code{flex}, a tool for generating programs that
74 perform pattern-matching on text. The manual includes both tutorial and
77 This edition of @cite{The flex Manual} documents @code{flex} version
78 @value{VERSION}. It was last updated on @value{UPDATED}.
80 This manual was written by @value{authors}.
93 * Multiple Input Buffers::
103 * Memory Management::
104 * Serialized Tables::
113 --- The Detailed Node Listing ---
115 Format of the Input File
117 * Definitions Section::
119 * User Code Section::
120 * Comments in the Input::
124 * Options for Specifying Filenames::
125 * Options Affecting Scanner Behavior::
126 * Code-Level And API Options::
127 * Options for Scanner Speed and Size::
128 * Debugging Options::
129 * Miscellaneous Options::
134 * Reentrant Overview::
135 * Reentrant Example::
137 * Reentrant Functions::
139 The Reentrant API in Detail
141 * Specify Reentrant::
142 * Extra Reentrant Argument::
143 * Global Replacement::
144 * Init and Destroy Functions::
151 * The Default Memory Management::
152 * Overriding The Default Memory Management::
153 * A Note About yytext And Memory::
157 * Creating Serialized Tables::
158 * Loading and Unloading Serialized Tables::
159 * Tables File Format::
163 * When was flex born?::
164 * How do I expand backslash-escape sequences in C-style quoted strings?::
165 * Why do flex scanners call fileno if it is not ANSI compatible?::
166 * Does flex support recursive pattern definitions?::
167 * How do I skip huge chunks of input (tens of megabytes) while using flex?::
168 * Flex is not matching my patterns in the same order that I defined them.::
169 * My actions are executing out of order or sometimes not at all.::
170 * How can I have multiple input sources feed into the same scanner at the same time?::
171 * Can I build nested parsers that work with the same input file?::
172 * How can I match text only at the end of a file?::
173 * How can I make REJECT cascade across start condition boundaries?::
174 * Why cant I use fast or full tables with interactive mode?::
175 * How much faster is -F or -f than -C?::
176 * If I have a simple grammar cant I just parse it with flex?::
177 * Why doesn't yyrestart() set the start state back to INITIAL?::
178 * How can I match C-style comments?::
179 * The period isn't working the way I expected.::
180 * Can I get the flex manual in another format?::
181 * Does there exist a "faster" NDFA->DFA algorithm?::
182 * How does flex compile the DFA so quickly?::
183 * How can I use more than 8192 rules?::
184 * How do I abandon a file in the middle of a scan and switch to a new file?::
185 * How do I execute code only during initialization (only before the first scan)?::
186 * How do I execute code at termination?::
187 * Where else can I find help?::
188 * Can I include comments in the "rules" section of the file?::
189 * I get an error about undefined yywrap().::
190 * How can I change the matching pattern at run time?::
191 * How can I expand macros in the input?::
192 * How can I build a two-pass scanner?::
193 * How do I match any string not matched in the preceding rules?::
194 * I am trying to port code from AT&T lex that uses yysptr and yysbuf.::
195 * Is there a way to make flex treat NULL like a regular character?::
196 * Whenever flex can not match the input it says "flex scanner jammed".::
197 * Why doesn't flex have non-greedy operators like perl does?::
198 * Memory leak - 16386 bytes allocated by malloc.::
199 * How do I track the byte offset for lseek()?::
200 * How do I use my own I/O classes in a C++ scanner?::
201 * How do I skip as many chars as possible?::
203 * Are certain equivalent patterns faster than others?::
204 * Is backing up a big deal?::
205 * Can I fake multi-byte character support?::
207 * Can you discuss some flex internals?::
208 * unput() messes up yy_at_bol::
209 * The | operator is not doing what I want::
210 * Why can't flex understand this variable trailing context pattern?::
211 * The ^ operator isn't working::
212 * Trailing context is getting confused with trailing optional patterns::
213 * Is flex GNU or not?::
215 * I need to scan if-then-else blocks and while loops::
219 * Is there a repository for flex scanners?::
220 * How can I conditionally compile or preprocess my flex input file?::
221 * Where can I find grammars for lex and yacc?::
222 * I get an end-of-buffer message for each character scanned.::
262 * What is the difference between YYLEX_PARAM and YY_DECL?::
263 * Why do I get "conflicting types for yylex" error?::
264 * How do I access the values set in a Flex action from within a Bison action?::
268 * Makefiles and Flex::
276 * Index of Functions and Macros::
277 * Index of Variables::
278 * Index of Data Types::
280 * Index of Scanner Options::
285 @node Copyright, Reporting Bugs, Top, Top
288 @cindex copyright of flex
289 @cindex distributing flex
292 @node Reporting Bugs, Introduction, Copyright, Top
293 @chapter Reporting Bugs
295 @cindex bugs, reporting
296 @cindex reporting bugs
298 If you find a bug in @code{flex}, please report it using
299 GitHub's issue tracking facility at @url{https://github.com/westes/flex/issues/}
301 @node Introduction, Simple Examples, Reporting Bugs, Top
302 @chapter Introduction
304 @cindex scanner, definition of
305 @code{flex} is a tool for generating @dfn{scanners}. A scanner is a
306 program which recognizes lexical patterns in text. The @code{flex}
307 program reads the given input files, or its standard input if no file
308 names are given, for a description of a scanner to generate. The
309 description is in the form of pairs of regular expressions and C code,
310 called @dfn{rules}. @code{flex} generates as output a C source file,
311 @file{lex.yy.c} by default, which defines a routine @code{yylex()}.
312 This file can be compiled and linked with the flex runtime library to
313 produce an executable. When the executable is run, it analyzes its
314 input for occurrences of the regular expressions. Whenever it finds
315 one, it executes the corresponding C code.
317 @node Simple Examples, Format, Introduction, Top
318 @chapter Some Simple Examples
320 First some simple examples to get the flavor of how one uses
323 @cindex username expansion
324 The following @code{flex} input specifies a scanner which, when it
325 encounters the string @samp{username} will replace it with the user's
331 username printf( "%s", getlogin() );
336 @cindex rules, default
337 By default, any text not matched by a @code{flex} scanner is copied to
338 the output, so the net effect of this scanner is to copy its input file
339 to its output with each occurrence of @samp{username} expanded. In this
340 input, there is just one rule. @samp{username} is the @dfn{pattern} and
341 the @samp{printf} is the @dfn{action}. The @samp{%%} symbol marks the
342 beginning of the rules.
344 Here's another simple example:
346 @cindex counting characters and lines
349 int num_lines = 0, num_chars = 0;
352 \n ++num_lines; ++num_chars;
360 printf( "# of lines = %d, # of chars = %d\n",
361 num_lines, num_chars );
366 This scanner counts the number of characters and the number of lines in
367 its input. It produces no output other than the final report on the
368 character and line counts. The first line declares two globals,
369 @code{num_lines} and @code{num_chars}, which are accessible both inside
370 @code{yylex()} and in the @code{main()} routine declared after the
371 second @samp{%%}. There are two rules, one which matches a newline
372 (@samp{\n}) and increments both the line count and the character count,
373 and one which matches any character other than a newline (indicated by
374 the @samp{.} regular expression).
376 A somewhat more complicated example:
378 @cindex Pascal-like language
381 /* scanner for a toy Pascal-like language */
384 /* need this for the call to atof() below */
394 printf( "An integer: %s (%d)\n", yytext,
398 {DIGIT}+"."{DIGIT}* {
399 printf( "A float: %s (%g)\n", yytext,
403 if|then|begin|end|procedure|function {
404 printf( "A keyword: %s\n", yytext );
407 {ID} printf( "An identifier: %s\n", yytext );
409 "+"|"-"|"*"|"/" printf( "An operator: %s\n", yytext );
411 "{"[^{}\n]*"}" /* eat up one-line comments */
413 [ \t\n]+ /* eat up whitespace */
415 . printf( "Unrecognized character: %s\n", yytext );
419 int main( int argc, char **argv )
421 ++argv, --argc; /* skip over program name */
423 yyin = fopen( argv[0], "r" );
432 This is the beginnings of a simple scanner for a language like Pascal.
433 It identifies different types of @dfn{tokens} and reports on what it has
436 The details of this example will be explained in the following
439 @node Format, Patterns, Simple Examples, Top
440 @chapter Format of the Input File
443 @cindex format of flex input
444 @cindex input, format of
446 @cindex sections of flex input
448 The @code{flex} input file consists of three sections, separated by a
449 line containing only @samp{%%}.
451 @cindex format of input file
463 * Definitions Section::
465 * User Code Section::
466 * Comments in the Input::
469 @node Definitions Section, Rules Section, Format, Format
470 @section Format of the Definitions Section
472 @cindex input file, Definitions section
473 @cindex Definitions, in flex input
474 The @dfn{definitions section} contains declarations of simple @dfn{name}
475 definitions to simplify the scanner specification, and declarations of
476 @dfn{start conditions}, which are explained in a later section.
478 @cindex aliases, how to define
479 @cindex pattern aliases, how to define
480 Name definitions have the form:
488 The @samp{name} is a word beginning with a letter or an underscore
489 (@samp{_}) followed by zero or more letters, digits, @samp{_}, or
490 @samp{-} (dash). The definition is taken to begin at the first
491 non-whitespace character following the name and continuing to the end of
492 the line. The definition can subsequently be referred to using
493 @samp{@{name@}}, which will expand to @samp{(definition)}. For example,
495 @cindex pattern aliases, defining
496 @cindex defining pattern aliases
504 Defines @samp{DIGIT} to be a regular expression which matches a single
505 digit, and @samp{ID} to be a regular expression which matches a letter
506 followed by zero-or-more letters-or-digits. A subsequent reference to
508 @cindex pattern aliases, use of
523 and matches one-or-more digits followed by a @samp{.} followed by
526 @cindex comments in flex input
527 An unindented comment (i.e., a line
528 beginning with @samp{/*}) is copied verbatim to the output up
529 to the next @samp{*/}.
531 @cindex %@{ and %@}, in Definitions Section
532 @cindex embedding C code in flex input
533 @cindex C code in flex input
534 Any @emph{indented} text or text enclosed in @samp{%@{} and @samp{%@}}
535 is also copied verbatim to the output (with the %@{ and %@} symbols
536 removed). The %@{ and %@} symbols must appear unindented on lines by
541 A @code{%top} block is similar to a @samp{%@{} ... @samp{%@}} block, except
542 that the code in a @code{%top} block is relocated to the @emph{top} of the
543 generated file, before any flex definitions @footnote{Actually,
544 @code{yyIN_HEADER} is defined before the @samp{%top} block.}.
545 The @code{%top} block is useful when you want certain preprocessor macros to be
546 defined or certain files to be included before the generated code.
547 The single characters, @samp{@{} and @samp{@}} are used to delimit the
548 @code{%top} block, as show in the example below:
553 /* This code goes at the "top" of the generated file. */
555 #include <inttypes.h>
560 Multiple @code{%top} blocks are allowed, and their order is preserved.
562 @node Rules Section, User Code Section, Definitions Section, Format
563 @section Format of the Rules Section
565 @cindex input file, Rules Section
566 @cindex rules, in flex input
567 The @dfn{rules} section of the @code{flex} input contains a series of
576 where the pattern must be unindented and the action must begin
578 @xref{Patterns}, for a further description of patterns and actions.
580 In the rules section, any indented or %@{ %@} enclosed text appearing
581 before the first rule may be used to declare variables which are local
582 to the scanning routine and (after the declarations) code which is to be
583 executed whenever the scanning routine is entered. Other indented or
584 %@{ %@} text in the rule section is still copied to the output, but its
585 meaning is not well-defined and it may well cause compile-time errors
586 (this feature is present for @acronym{POSIX} compliance. @xref{Lex and
587 Posix}, for other such features).
589 Any @emph{indented} text or text enclosed in @samp{%@{} and @samp{%@}}
590 is copied verbatim to the output (with the %@{ and %@} symbols removed).
591 The %@{ and %@} symbols must appear unindented on lines by themselves.
593 @node User Code Section, Comments in the Input, Rules Section, Format
594 @section Format of the User Code Section
596 @cindex input file, user code Section
597 @cindex user code, in flex input
598 The user code section is simply copied to @file{lex.yy.c} verbatim. It
599 is used for companion routines which call or are called by the scanner.
600 The presence of this section is optional; if it is missing, the second
601 @samp{%%} in the input file may be skipped, too.
603 @node Comments in the Input, , User Code Section, Format
604 @section Comments in the Input
606 @cindex comments, syntax of
607 Flex supports C-style comments, that is, anything between @samp{/*} and
609 considered a comment. Whenever flex encounters a comment, it copies the
610 entire comment verbatim to the generated source code. Comments may
611 appear just about anywhere, but with the following exceptions:
614 @cindex comments, in rules section
616 Comments may not appear in the Rules Section wherever flex is expecting
617 a regular expression. This means comments may not appear at the
618 beginning of a line, or immediately following a list of scanner states.
620 Comments may not appear on an @samp{%option} line in the Definitions
624 If you want to follow a simple rule, then always begin a comment on a
625 new line, with one or more whitespace characters before the initial
626 @samp{/*}). This rule will work anywhere in the input file.
628 All the comments in the following example are valid:
630 @cindex comments, valid uses of
631 @cindex comments in the input
638 /* Definitions Section */
643 ruleA /* after regex */ { /* code block */ } /* after code block */
644 /* Rules Section (indented) */
653 /* User Code Section */
658 @node Patterns, Matching, Format, Top
661 @cindex patterns, in rules section
662 @cindex regular expressions, in patterns
663 The patterns in the input (see @ref{Rules Section}) are written using an
664 extended set of regular expressions. These are:
666 @cindex patterns, syntax
667 @cindex patterns, syntax
670 match the character 'x'
673 any character (byte) except newline
675 @cindex [] in patterns
676 @cindex character classes in patterns, syntax of
677 @cindex POSIX, character classes in patterns, syntax of
679 a @dfn{character class}; in this case, the pattern
680 matches either an 'x', a 'y', or a 'z'
682 @cindex ranges in patterns
684 a "character class" with a range in it; matches
685 an 'a', a 'b', any letter from 'j' through 'o',
688 @cindex ranges in patterns, negating
689 @cindex negating ranges in patterns
691 a "negated character class", i.e., any character
692 but those in the class. In this case, any
693 character EXCEPT an uppercase letter.
696 any character EXCEPT an uppercase letter or
699 @item [a-z]@{-@}[aeiou]
700 the lowercase consonants
703 zero or more r's, where r is any regular expression
709 zero or one r's (that is, ``an optional r'')
711 @cindex braces in patterns
713 anywhere from two to five r's
721 @cindex pattern aliases, expansion of
723 the expansion of the @samp{name} definition
726 @cindex literal text in patterns, syntax of
727 @cindex verbatim text in patterns, syntax of
729 the literal string: @samp{[xyz]"foo}
731 @cindex escape sequences in patterns, syntax of
733 if X is @samp{a}, @samp{b}, @samp{f}, @samp{n}, @samp{r}, @samp{t}, or
734 @samp{v}, then the ANSI-C interpretation of @samp{\x}. Otherwise, a
735 literal @samp{X} (used to escape operators such as @samp{*})
737 @cindex NULL character in patterns, syntax of
739 a NUL character (ASCII code 0)
741 @cindex octal characters in patterns
743 the character with octal value 123
746 the character with hexadecimal value 2a
749 match an @samp{r}; parentheses are used to override precedence (see below)
752 apply option @samp{r} and omit option @samp{s} while interpreting pattern.
753 Options may be zero or more of the characters @samp{i}, @samp{s}, or @samp{x}.
755 @samp{i} means case-insensitive. @samp{-i} means case-sensitive.
757 @samp{s} alters the meaning of the @samp{.} syntax to match any single byte whatsoever.
758 @samp{-s} alters the meaning of @samp{.} to match any byte except @samp{\n}.
760 @samp{x} ignores comments and whitespace in patterns. Whitespace is ignored unless
761 it is backslash-escaped, contained within @samp{""}s, or appears inside a
764 The following are all valid:
767 (?:foo) same as (foo)
768 (?i:ab7) same as ([aA][bB]7)
769 (?-i:ab) same as (ab)
770 (?s:.) same as [\x00-\xFF]
771 (?-s:.) same as [^\n]
772 (?ix-s: a . b) same as ([Aa][^\n][bB])
773 (?x:a b) same as ("ab")
774 (?x:a\ b) same as ("a b")
775 (?x:a" "b) same as ("a b")
776 (?x:a[ ]b) same as ("a b")
784 omit everything within @samp{()}. The first @samp{)}
785 character encountered ends the pattern. It is not possible to for the comment
786 to contain a @samp{)} character. The comment may span lines.
788 @cindex concatenation, in patterns
790 the regular expression @samp{r} followed by the regular expression @samp{s}; called
794 either an @samp{r} or an @samp{s}
796 @cindex trailing context, in patterns
798 an @samp{r} but only if it is followed by an @samp{s}. The text matched by @samp{s} is
799 included when determining whether this rule is the longest match, but is
800 then returned to the input before the action is executed. So the action
801 only sees the text matched by @samp{r}. This type of pattern is called
802 @dfn{trailing context}. (There are some combinations of @samp{r/s} that flex
803 cannot match correctly. @xref{Limitations}, regarding dangerous trailing
806 @cindex beginning of line, in patterns
807 @cindex BOL, in patterns
809 an @samp{r}, but only at the beginning of a line (i.e.,
810 when just starting to scan, or right after a
811 newline has been scanned).
813 @cindex end of line, in patterns
814 @cindex EOL, in patterns
816 an @samp{r}, but only at the end of a line (i.e., just before a
817 newline). Equivalent to @samp{r/\n}.
819 @cindex newline, matching in patterns
820 Note that @code{flex}'s notion of ``newline'' is exactly
821 whatever the C compiler used to compile @code{flex}
822 interprets @samp{\n} as; in particular, on some DOS
823 systems you must either filter out @samp{\r}s in the
824 input yourself, or explicitly use @samp{r/\r\n} for @samp{r$}.
826 @cindex start conditions, in patterns
828 an @samp{r}, but only in start condition @code{s} (see @ref{Start
829 Conditions} for discussion of start conditions).
832 same, but in any of start conditions @code{s1}, @code{s2}, or @code{s3}.
835 an @samp{r} in any start condition, even an exclusive one.
837 @cindex end of file, in patterns
838 @cindex EOF in patterns, syntax of
843 an end-of-file when in start condition @code{s1} or @code{s2}
846 Note that inside of a character class, all regular expression operators
847 lose their special meaning except escape (@samp{\}) and the character class
848 operators, @samp{-}, @samp{]]}, and, at the beginning of the class, @samp{^}.
850 @cindex patterns, precedence of operators
851 The regular expressions listed above are grouped according to
852 precedence, from highest precedence at the top to lowest at the bottom.
853 Those grouped together have equal precedence (see special note on the
854 precedence of the repeat operator, @samp{@{@}}, under the documentation
855 for the @samp{--posix} POSIX compliance option). For example,
857 @cindex patterns, grouping and precedence
872 since the @samp{*} operator has higher precedence than concatenation,
873 and concatenation higher than alternation (@samp{|}). This pattern
874 therefore matches @emph{either} the string @samp{foo} @emph{or} the
875 string @samp{ba} followed by zero-or-more @samp{r}'s. To match
876 @samp{foo} or zero-or-more repetitions of the string @samp{bar}, use:
884 And to match a sequence of zero or more repetitions of @samp{foo} and
887 @cindex patterns, repetitions with grouping
894 @cindex character classes in patterns
895 In addition to characters and ranges of characters, character classes
896 can also contain @dfn{character class expressions}. These are
897 expressions enclosed inside @samp{[:} and @samp{:]} delimiters (which
898 themselves must appear between the @samp{[} and @samp{]} of the
899 character class. Other elements may occur inside the character class,
900 too). The valid expressions are:
902 @cindex patterns, valid character classes
905 [:alnum:] [:alpha:] [:blank:]
906 [:cntrl:] [:digit:] [:graph:]
907 [:lower:] [:print:] [:punct:]
908 [:space:] [:upper:] [:xdigit:]
912 These expressions all designate a set of characters equivalent to the
913 corresponding standard C @code{isXXX} function. For example,
914 @samp{[:alnum:]} designates those characters for which @code{isalnum()}
915 returns true - i.e., any alphabetic or numeric character. Some systems
916 don't provide @code{isblank()}, so flex defines @samp{[:blank:]} as a
919 For example, the following character classes are all equivalent:
921 @cindex character classes, equivalence of
922 @cindex patterns, character class equivalence
932 A word of caution. Character classes are expanded immediately when seen in the @code{flex} input.
933 This means the character classes are sensitive to the locale in which @code{flex}
934 is executed, and the resulting scanner will not be sensitive to the runtime locale.
935 This may or may not be desirable.
939 @cindex case-insensitive, effect on character classes
940 @item If your scanner is case-insensitive (the @samp{-i} flag), then
941 @samp{[:upper:]} and @samp{[:lower:]} are equivalent to
944 @anchor{case and character ranges}
945 @item Character classes with ranges, such as @samp{[a-Z]}, should be used with
946 caution in a case-insensitive scanner if the range spans upper or lowercase
947 characters. Flex does not know if you want to fold all upper and lowercase
948 characters together, or if you want the literal numeric range specified (with
949 no case folding). When in doubt, flex will assume that you meant the literal
950 numeric range, and will issue a warning. The exception to this rule is a
951 character range such as @samp{[a-z]} or @samp{[S-W]} where it is obvious that you
952 want case-folding to occur. Here are some examples with the @samp{-i} flag
955 @multitable {@samp{[a-zA-Z]}} {ambiguous} {@samp{[A-Z\[\\\]_`a-t]}} {@samp{[@@A-Z\[\\\]_`abc]}}
956 @item Range @tab Result @tab Literal Range @tab Alternate Range
957 @item @samp{[a-t]} @tab ok @tab @samp{[a-tA-T]} @tab
958 @item @samp{[A-T]} @tab ok @tab @samp{[a-tA-T]} @tab
959 @item @samp{[A-t]} @tab ambiguous @tab @samp{[A-Z\[\\\]_`a-t]} @tab @samp{[a-tA-T]}
960 @item @samp{[_-@{]} @tab ambiguous @tab @samp{[_`a-z@{]} @tab @samp{[_`a-zA-Z@{]}
961 @item @samp{[@@-C]} @tab ambiguous @tab @samp{[@@ABC]} @tab @samp{[@@A-Z\[\\\]_`abc]}
964 @cindex end of line, in negated character classes
965 @cindex EOL, in negated character classes
967 A negated character class such as the example @samp{[^A-Z]} above
968 @emph{will} match a newline unless @samp{\n} (or an equivalent escape
969 sequence) is one of the characters explicitly present in the negated
970 character class (e.g., @samp{[^A-Z\n]}). This is unlike how many other
971 regular expression tools treat negated character classes, but
972 unfortunately the inconsistency is historically entrenched. Matching
973 newlines means that a pattern like @samp{[^"]*} can match the entire
974 input unless there's another quote in the input.
976 Flex allows negation of character class expressions by prepending @samp{^} to
977 the POSIX character class name.
981 [:^alnum:] [:^alpha:] [:^blank:]
982 [:^cntrl:] [:^digit:] [:^graph:]
983 [:^lower:] [:^print:] [:^punct:]
984 [:^space:] [:^upper:] [:^xdigit:]
988 Flex will issue a warning if the expressions @samp{[:^upper:]} and
989 @samp{[:^lower:]} appear in a case-insensitive scanner, since their meaning is
990 unclear. The current behavior is to skip them entirely, but this may change
991 without notice in future revisions of flex.
995 The @samp{@{-@}} operator computes the difference of two character classes. For
996 example, @samp{[a-c]@{-@}[b-z]} represents all the characters in the class
997 @samp{[a-c]} that are not in the class @samp{[b-z]} (which in this case, is
998 just the single character @samp{a}). The @samp{@{-@}} operator is left
999 associative, so @samp{[abc]@{-@}[b]@{-@}[c]} is the same as @samp{[a]}. Be careful
1000 not to accidentally create an empty set, which will never match.
1004 The @samp{@{+@}} operator computes the union of two character classes. For
1005 example, @samp{[a-z]@{+@}[0-9]} is the same as @samp{[a-z0-9]}. This operator
1006 is useful when preceded by the result of a difference operation, as in,
1007 @samp{[[:alpha:]]@{-@}[[:lower:]]@{+@}[q]}, which is equivalent to
1008 @samp{[A-Zq]} in the "C" locale.
1010 @cindex trailing context, limits of
1011 @cindex ^ as non-special character in patterns
1012 @cindex $ as normal character in patterns
1014 A rule can have at most one instance of trailing context (the @samp{/} operator
1015 or the @samp{$} operator). The start condition, @samp{^}, and @samp{<<EOF>>} patterns
1016 can only occur at the beginning of a pattern, and, as well as with @samp{/} and @samp{$},
1017 cannot be grouped inside parentheses. A @samp{^} which does not occur at
1018 the beginning of a rule or a @samp{$} which does not occur at the end of
1019 a rule loses its special properties and is treated as a normal character.
1022 The following are invalid:
1024 @cindex patterns, invalid trailing context
1032 Note that the first of these can be written @samp{foo/bar\n}.
1035 The following will result in @samp{$} or @samp{^} being treated as a normal character:
1037 @cindex patterns, special characters treated as non-special
1045 If the desired meaning is a @samp{foo} or a
1046 @samp{bar}-followed-by-a-newline, the following could be used (the
1047 special @code{|} action is explained below, @pxref{Actions}):
1049 @cindex patterns, end of line
1053 bar$ /* action goes here */
1057 A similar trick will work for matching a @samp{foo} or a
1058 @samp{bar}-at-the-beginning-of-a-line.
1061 @node Matching, Actions, Patterns, Top
1062 @chapter How the Input Is Matched
1064 @cindex patterns, matching
1065 @cindex input, matching
1066 @cindex trailing context, matching
1067 @cindex matching, and trailing context
1068 @cindex matching, length of
1069 @cindex matching, multiple matches
1070 When the generated scanner is run, it analyzes its input looking for
1071 strings which match any of its patterns. If it finds more than one
1072 match, it takes the one matching the most text (for trailing context
1073 rules, this includes the length of the trailing part, even though it
1074 will then be returned to the input). If it finds two or more matches of
1075 the same length, the rule listed first in the @code{flex} input file is
1081 Once the match is determined, the text corresponding to the match
1082 (called the @dfn{token}) is made available in the global character
1083 pointer @code{yytext}, and its length in the global integer
1084 @code{yyleng}. The @dfn{action} corresponding to the matched pattern is
1085 then executed (@pxref{Actions}), and then the remaining input is scanned
1088 @cindex default rule
1089 If no match is found, then the @dfn{default rule} is executed: the next
1090 character in the input is considered matched and copied to the standard
1091 output. Thus, the simplest valid @code{flex} input is:
1093 @cindex minimal scanner
1100 which generates a scanner that simply copies its input (one character at
1101 a time) to its output.
1103 @cindex yytext, two types of
1104 @cindex %array, use of
1105 @cindex %pointer, use of
1107 Note that @code{yytext} can be defined in two different ways: either as
1108 a character @emph{pointer} or as a character @emph{array}. You can
1109 control which definition @code{flex} uses by including one of the
1110 special directives @code{%pointer} or @code{%array} in the first
1111 (definitions) section of your flex input. The default is
1112 @code{%pointer}, unless you use the @samp{-l} lex compatibility option,
1113 in which case @code{yytext} will be an array. The advantage of using
1114 @code{%pointer} is substantially faster scanning and no buffer overflow
1115 when matching very large tokens (unless you run out of dynamic memory).
1116 The disadvantage is that you are restricted in how your actions can
1117 modify @code{yytext} (@pxref{Actions}), and calls to the @code{unput()}
1118 function destroys the present contents of @code{yytext}, which can be a
1119 considerable porting headache when moving between different @code{lex}
1122 @cindex %array, advantages of
1123 The advantage of @code{%array} is that you can then modify @code{yytext}
1124 to your heart's content, and calls to @code{unput()} do not destroy
1125 @code{yytext} (@pxref{Actions}). Furthermore, existing @code{lex}
1126 programs sometimes access @code{yytext} externally using declarations of
1131 extern char yytext[];
1135 This definition is erroneous when used with @code{%pointer}, but correct
1138 The @code{%array} declaration defines @code{yytext} to be an array of
1139 @code{YYLMAX} characters, which defaults to a fairly large value. You
1140 can change the size by simply #define'ing @code{YYLMAX} to a different
1141 value in the first section of your @code{flex} input. As mentioned
1142 above, with @code{%pointer} yytext grows dynamically to accommodate
1143 large tokens. While this means your @code{%pointer} scanner can
1144 accommodate very large tokens (such as matching entire blocks of
1145 comments), bear in mind that each time the scanner must resize
1146 @code{yytext} it also must rescan the entire token from the beginning,
1147 so matching such tokens can prove slow. @code{yytext} presently does
1148 @emph{not} dynamically grow if a call to @code{unput()} results in too
1149 much text being pushed back; instead, a run-time error results.
1151 @cindex %array, with C++
1152 Also note that you cannot use @code{%array} with C++ scanner classes
1155 @node Actions, Generated Scanner, Matching, Top
1159 Each pattern in a rule has a corresponding @dfn{action}, which can be
1160 any arbitrary C statement. The pattern ends at the first non-escaped
1161 whitespace character; the remainder of the line is its action. If the
1162 action is empty, then when the pattern is matched the input token is
1163 simply discarded. For example, here is the specification for a program
1164 which deletes all occurrences of @samp{zap me} from its input:
1166 @cindex deleting lines from input
1174 This example will copy all other characters in the input to the output
1175 since they will be matched by the default rule.
1177 Here is a program which compresses multiple blanks and tabs down to a
1178 single blank, and throws away whitespace found at the end of a line:
1180 @cindex whitespace, compressing
1181 @cindex compressing whitespace
1185 [ \t]+ putchar( ' ' );
1186 [ \t]+$ /* ignore this token */
1190 @cindex %@{ and %@}, in Rules Section
1191 @cindex actions, use of @{ and @}
1192 @cindex actions, embedded C strings
1193 @cindex C-strings, in actions
1194 @cindex comments, in actions
1195 If the action contains a @samp{@{}, then the action spans till the
1196 balancing @samp{@}} is found, and the action may cross multiple lines.
1197 @code{flex} knows about C strings and comments and won't be fooled by
1198 braces found within them, but also allows actions to begin with
1199 @samp{%@{} and will consider the action to be all the text up to the
1200 next @samp{%@}} (regardless of ordinary braces inside the action).
1202 @cindex |, in actions
1203 An action consisting solely of a vertical bar (@samp{|}) means ``same as the
1204 action for the next rule''. See below for an illustration.
1206 Actions can include arbitrary C code, including @code{return} statements
1207 to return a value to whatever routine called @code{yylex()}. Each time
1208 @code{yylex()} is called it continues processing tokens from where it
1209 last left off until it either reaches the end of the file or executes a
1212 @cindex yytext, modification of
1213 Actions are free to modify @code{yytext} except for lengthening it
1214 (adding characters to its end--these will overwrite later characters in
1215 the input stream). This however does not apply when using @code{%array}
1216 (@pxref{Matching}). In that case, @code{yytext} may be freely modified
1219 @cindex yyleng, modification of
1220 @cindex yymore, and yyleng
1221 Actions are free to modify @code{yyleng} except they should not do so if
1222 the action also includes use of @code{yymore()} (see below).
1224 @cindex preprocessor macros, for use in actions
1225 There are a number of special directives which can be included within an
1231 copies yytext to the scanner's output.
1235 followed by the name of a start condition places the scanner in the
1236 corresponding start condition (see below).
1240 directs the scanner to proceed on to the ``second best'' rule which
1241 matched the input (or a prefix of the input). The rule is chosen as
1242 described above in @ref{Matching}, and @code{yytext} and @code{yyleng}
1243 set up appropriately. It may either be one which matched as much text
1244 as the originally chosen rule but came later in the @code{flex} input
1245 file, or one which matched less text. For example, the following will
1246 both count the words in the input and call the routine @code{special()}
1247 whenever @samp{frob} is seen:
1254 frob special(); REJECT;
1255 [^ \t\n]+ ++word_count;
1259 Without the @code{REJECT}, any occurrences of @samp{frob} in the input
1260 would not be counted as words, since the scanner normally executes only
1261 one action per token. Multiple uses of @code{REJECT} are allowed, each
1262 one finding the next best choice to the currently active rule. For
1263 example, when the following scanner scans the token @samp{abcd}, it will
1264 write @samp{abcdabcaba} to the output:
1266 @cindex REJECT, calling multiple times
1275 .|\n /* eat up any unmatched character */
1279 The first three rules share the fourth's action since they use the
1280 special @samp{|} action.
1282 @code{REJECT} is a particularly expensive feature in terms of scanner
1283 performance; if it is used in @emph{any} of the scanner's actions it
1284 will slow down @emph{all} of the scanner's matching. Furthermore,
1285 @code{REJECT} cannot be used with the @samp{-Cf} or @samp{-CF} options
1286 (@pxref{Scanner Options}).
1288 Note also that unlike the other special actions, @code{REJECT} is a
1289 @emph{branch}. Code immediately following it in the action will
1290 @emph{not} be executed.
1294 tells the scanner that the next time it matches a rule, the
1295 corresponding token should be @emph{appended} onto the current value of
1296 @code{yytext} rather than replacing it. For example, given the input
1297 @samp{mega-kludge} the following will write @samp{mega-mega-kludge} to
1300 @cindex yymore(), mega-kludge
1301 @cindex yymore() to append token to previous token
1305 mega- ECHO; yymore();
1310 First @samp{mega-} is matched and echoed to the output. Then @samp{kludge}
1311 is matched, but the previous @samp{mega-} is still hanging around at the
1316 for the @samp{kludge} rule will actually write @samp{mega-kludge}.
1319 @cindex yymore, performance penalty of
1320 Two notes regarding use of @code{yymore()}. First, @code{yymore()}
1321 depends on the value of @code{yyleng} correctly reflecting the size of
1322 the current token, so you must not modify @code{yyleng} if you are using
1323 @code{yymore()}. Second, the presence of @code{yymore()} in the
1324 scanner's action entails a minor performance penalty in the scanner's
1328 @code{yyless(n)} returns all but the first @code{n} characters of the
1329 current token back to the input stream, where they will be rescanned
1330 when the scanner looks for the next match. @code{yytext} and
1331 @code{yyleng} are adjusted appropriately (e.g., @code{yyleng} will now
1332 be equal to @code{n}). For example, on the input @samp{foobar} the
1333 following will write out @samp{foobarbar}:
1335 @cindex yyless(), pushing back characters
1336 @cindex pushing back characters with yyless
1340 foobar ECHO; yyless(3);
1345 An argument of 0 to @code{yyless()} will cause the entire current input
1346 string to be scanned again. Unless you've changed how the scanner will
1347 subsequently process its input (using @code{BEGIN}, for example), this
1348 will result in an endless loop.
1350 Note that @code{yyless()} is a macro and can only be used in the flex
1351 input file, not from other source files.
1354 @cindex pushing back characters with unput
1355 @code{unput(c)} puts the character @code{c} back onto the input stream.
1356 It will be the next character scanned. The following action will take
1357 the current token and cause it to be rescanned enclosed in parentheses.
1359 @cindex unput(), pushing back characters
1360 @cindex pushing back characters with unput()
1365 /* Copy yytext because unput() trashes yytext */
1366 char *yycopy = strdup( yytext );
1368 for ( i = yyleng - 1; i >= 0; --i )
1376 Note that since each @code{unput()} puts the given character back at the
1377 @emph{beginning} of the input stream, pushing back strings must be done
1380 @cindex %pointer, and unput()
1381 @cindex unput(), and %pointer
1382 An important potential problem when using @code{unput()} is that if you
1383 are using @code{%pointer} (the default), a call to @code{unput()}
1384 @emph{destroys} the contents of @code{yytext}, starting with its
1385 rightmost character and devouring one character to the left with each
1386 call. If you need the value of @code{yytext} preserved after a call to
1387 @code{unput()} (as in the above example), you must either first copy it
1388 elsewhere, or build your scanner using @code{%array} instead
1391 @cindex pushing back EOF
1392 @cindex EOF, pushing back
1393 Finally, note that you cannot put back @samp{EOF} to attempt to mark the
1394 input stream with an end-of-file.
1397 @code{input()} reads the next character from the input stream. For
1398 example, the following is one way to eat up C comments:
1400 @cindex comments, discarding
1401 @cindex discarding C comments
1410 while ( (c = input()) != '*' &&
1412 ; /* eat up text of comment */
1416 while ( (c = input()) == '*' )
1419 break; /* found the end */
1424 error( "EOF in comment" );
1432 @cindex input(), and C++
1434 (Note that if the scanner is compiled using @code{C++}, then
1435 @code{input()} is instead referred to as @b{yyinput()}, in order to
1436 avoid a name clash with the @code{C++} stream by the name of
1439 @cindex flushing the internal buffer
1440 @cindex YY_FLUSH_BUFFER
1441 @code{YY_FLUSH_BUFFER;} flushes the scanner's internal buffer so that
1442 the next time the scanner attempts to match a token, it will first
1443 refill the buffer using @code{YY_INPUT()} (@pxref{Generated Scanner}).
1444 This action is a special case of the more general
1445 @code{yy_flush_buffer;} function, described below (@pxref{Multiple
1448 @cindex yyterminate()
1449 @cindex terminating with yyterminate()
1450 @cindex exiting with yyterminate()
1451 @cindex halting with yyterminate()
1452 @code{yyterminate()} can be used in lieu of a return statement in an
1453 action. It terminates the scanner and returns a 0 to the scanner's
1454 caller, indicating ``all done''. By default, @code{yyterminate()} is
1455 also called when an end-of-file is encountered. It is a macro and may
1458 @node Generated Scanner, Start Conditions, Actions, Top
1459 @chapter The Generated Scanner
1461 @cindex yylex(), in generated scanner
1462 The output of @code{flex} is the file @file{lex.yy.c}, which contains
1463 the scanning routine @code{yylex()}, a number of tables used by it for
1464 matching tokens, and a number of auxiliary routines and macros. By
1465 default, @code{yylex()} is declared as follows:
1471 ... various definitions and the actions in here ...
1476 @cindex yylex(), overriding
1477 (If your environment supports function prototypes, then it will be
1478 @code{int yylex( void )}.) This definition may be changed by defining
1479 the @code{YY_DECL} macro. For example, you could use:
1481 @cindex yylex, overriding the prototype of
1484 #define YY_DECL float lexscan( a, b ) float a, b;
1488 to give the scanning routine the name @code{lexscan}, returning a float,
1489 and taking two floats as arguments. Note that if you give arguments to
1490 the scanning routine using a K&R-style/non-prototyped function
1491 declaration, you must terminate the definition with a semi-colon (;).
1493 @code{flex} generates @samp{C99} function definitions by
1494 default. Flex used to have the ability to generate obsolete, er,
1495 @samp{traditional}, function definitions. This was to support
1496 bootstrapping gcc on old systems. Unfortunately, traditional
1497 definitions prevent us from using any standard data types smaller than
1498 int (such as short, char, or bool) as function arguments. Furthermore,
1499 traditional definitions support added extra complexity in the skeleton file.
1500 For this reason, current versions of @code{flex} generate standard C99 code
1501 only, leaving K&R-style functions to the historians.
1503 @cindex stdin, default for yyin
1505 Whenever @code{yylex()} is called, it scans tokens from the global input
1506 file @file{yyin} (which defaults to stdin). It continues until it
1507 either reaches an end-of-file (at which point it returns the value 0) or
1508 one of its actions executes a @code{return} statement.
1510 @cindex EOF and yyrestart()
1511 @cindex end-of-file, and yyrestart()
1513 If the scanner reaches an end-of-file, subsequent calls are undefined
1514 unless either @file{yyin} is pointed at a new input file (in which case
1515 scanning continues from that file), or @code{yyrestart()} is called.
1516 @code{yyrestart()} takes one argument, a @code{FILE *} pointer (which
1517 can be NULL, if you've set up @code{YY_INPUT} to scan from a source other
1518 than @code{yyin}), and initializes @file{yyin} for scanning from that
1519 file. Essentially there is no difference between just assigning
1520 @file{yyin} to a new input file or using @code{yyrestart()} to do so;
1521 the latter is available for compatibility with previous versions of
1522 @code{flex}, and because it can be used to switch input files in the
1523 middle of scanning. It can also be used to throw away the current input
1524 buffer, by calling it with an argument of @file{yyin}; but it would be
1525 better to use @code{YY_FLUSH_BUFFER} (@pxref{Actions}). Note that
1526 @code{yyrestart()} does @emph{not} reset the start condition to
1527 @code{INITIAL} (@pxref{Start Conditions}).
1529 @cindex RETURN, within actions
1530 If @code{yylex()} stops scanning due to executing a @code{return}
1531 statement in one of the actions, the scanner may then be called again
1532 and it will resume scanning where it left off.
1535 By default (and for purposes of efficiency), the scanner uses
1536 block-reads rather than simple @code{getc()} calls to read characters
1537 from @file{yyin}. The nature of how it gets its input can be controlled
1538 by defining the @code{YY_INPUT} macro. The calling sequence for
1539 @code{YY_INPUT()} is @code{YY_INPUT(buf,result,max_size)}. Its action
1540 is to place up to @code{max_size} characters in the character array
1541 @code{buf} and return in the integer variable @code{result} either the
1542 number of characters read or the constant @code{YY_NULL} (0 on Unix
1543 systems) to indicate @samp{EOF}. The default @code{YY_INPUT} reads from
1544 the global file-pointer @file{yyin}.
1546 @cindex YY_INPUT, overriding
1547 Here is a sample definition of @code{YY_INPUT} (in the definitions
1548 section of the input file):
1553 #define YY_INPUT(buf,result,max_size) \
1555 int c = getchar(); \
1556 result = (c == EOF) ? YY_NULL : (buf[0] = c, 1); \
1562 This definition will change the input processing to occur one character
1566 When the scanner receives an end-of-file indication from YY_INPUT, it
1567 then checks the @code{yywrap()} function. If @code{yywrap()} returns
1568 false (zero), then it is assumed that the function has gone ahead and
1569 set up @file{yyin} to point to another input file, and scanning
1570 continues. If it returns true (non-zero), then the scanner terminates,
1571 returning 0 to its caller. Note that in either case, the start
1572 condition remains unchanged; it does @emph{not} revert to
1575 @cindex yywrap, default for
1576 @cindex noyywrap, %option
1577 @cindex %option noyywrapp
1578 If you do not supply your own version of @code{yywrap()}, then you must
1579 either use @code{%option noyywrap} (in which case the scanner behaves as
1580 though @code{yywrap()} returned 1), or you must link with @samp{-lfl} to
1581 obtain the default version of the routine, which always returns 1.
1583 For scanning from in-memory buffers (e.g., scanning strings), see
1584 @ref{Scanning Strings}. @xref{Multiple Input Buffers}.
1586 @cindex ECHO, and yyout
1588 @cindex stdout, as default for yyout
1589 The scanner writes its @code{ECHO} output to the @file{yyout} global
1590 (default, @file{stdout}), which may be redefined by the user simply by
1591 assigning it to some other @code{FILE} pointer.
1593 @node Start Conditions, Multiple Input Buffers, Generated Scanner, Top
1594 @chapter Start Conditions
1596 @cindex start conditions
1597 @code{flex} provides a mechanism for conditionally activating rules.
1598 Any rule whose pattern is prefixed with @samp{<sc>} will only be active
1599 when the scanner is in the @dfn{start condition} named @code{sc}. For
1604 <STRING>[^"]* { /* eat up the string body ... */
1610 will be active only when the scanner is in the @code{STRING} start
1613 @cindex start conditions, multiple
1616 <INITIAL,STRING,QUOTE>\. { /* handle an escape ... */
1622 will be active only when the current start condition is either
1623 @code{INITIAL}, @code{STRING}, or @code{QUOTE}.
1625 @cindex start conditions, inclusive v.s.@: exclusive
1626 Start conditions are declared in the definitions (first) section of the
1627 input using unindented lines beginning with either @samp{%s} or
1628 @samp{%x} followed by a list of names. The former declares
1629 @dfn{inclusive} start conditions, the latter @dfn{exclusive} start
1630 conditions. A start condition is activated using the @code{BEGIN}
1631 action. Until the next @code{BEGIN} action is executed, rules with the
1632 given start condition will be active and rules with other start
1633 conditions will be inactive. If the start condition is inclusive, then
1634 rules with no start conditions at all will also be active. If it is
1635 exclusive, then @emph{only} rules qualified with the start condition
1636 will be active. A set of rules contingent on the same exclusive start
1637 condition describe a scanner which is independent of any of the other
1638 rules in the @code{flex} input. Because of this, exclusive start
1639 conditions make it easy to specify ``mini-scanners'' which scan portions
1640 of the input that are syntactically different from the rest (e.g.,
1643 If the distinction between inclusive and exclusive start conditions
1644 is still a little vague, here's a simple example illustrating the
1645 connection between the two. The set of rules:
1647 @cindex start conditions, inclusive
1653 <example>foo do_something();
1655 bar something_else();
1661 @cindex start conditions, exclusive
1667 <example>foo do_something();
1669 <INITIAL,example>bar something_else();
1673 Without the @code{<INITIAL,example>} qualifier, the @code{bar} pattern in
1674 the second example wouldn't be active (i.e., couldn't match) when in
1675 start condition @code{example}. If we just used @code{<example>} to
1676 qualify @code{bar}, though, then it would only be active in
1677 @code{example} and not in @code{INITIAL}, while in the first example
1678 it's active in both, because in the first example the @code{example}
1679 start condition is an inclusive @code{(%s)} start condition.
1681 @cindex start conditions, special wildcard condition
1682 Also note that the special start-condition specifier
1684 matches every start condition. Thus, the above example could also
1687 @cindex start conditions, use of wildcard condition (<*>)
1693 <example>foo do_something();
1695 <*>bar something_else();
1699 The default rule (to @code{ECHO} any unmatched character) remains active
1700 in start conditions. It is equivalent to:
1702 @cindex start conditions, behavior of default rule
1709 @cindex BEGIN, explanation
1712 @code{BEGIN(0)} returns to the original state where only the rules with
1713 no start conditions are active. This state can also be referred to as
1714 the start-condition @code{INITIAL}, so @code{BEGIN(INITIAL)} is
1715 equivalent to @code{BEGIN(0)}. (The parentheses around the start
1716 condition name are not required but are considered good style.)
1718 @code{BEGIN} actions can also be given as indented code at the beginning
1719 of the rules section. For example, the following will cause the scanner
1720 to enter the @code{SPECIAL} start condition whenever @code{yylex()} is
1721 called and the global variable @code{enter_special} is true:
1723 @cindex start conditions, using BEGIN
1730 if ( enter_special )
1733 <SPECIAL>blahblahblah
1734 ...more rules follow...
1738 To illustrate the uses of start conditions, here is a scanner which
1739 provides two different interpretations of a string like @samp{123.456}.
1740 By default it will treat it as three tokens, the integer @samp{123}, a
1741 dot (@samp{.}), and the integer @samp{456}. But if the string is
1742 preceded earlier in the line by the string @samp{expect-floats} it will
1743 treat it as a single token, the floating-point number @samp{123.456}:
1745 @cindex start conditions, for different interpretations of same input
1754 expect-floats BEGIN(expect);
1756 <expect>[0-9]+.[0-9]+ {
1757 printf( "found a float, = %f\n",
1761 /* that's the end of the line, so
1762 * we need another "expect-number"
1763 * before we'll recognize any more
1770 printf( "found an integer, = %d\n",
1774 "." printf( "found a dot\n" );
1778 @cindex comments, example of scanning C comments
1779 Here is a scanner which recognizes (and discards) C comments while
1780 maintaining a count of the current input line.
1782 @cindex recognizing C comments
1789 "/*" BEGIN(comment);
1791 <comment>[^*\n]* /* eat anything that's not a '*' */
1792 <comment>"*"+[^*/\n]* /* eat up '*'s not followed by '/'s */
1793 <comment>\n ++line_num;
1794 <comment>"*"+"/" BEGIN(INITIAL);
1798 This scanner goes to a bit of trouble to match as much
1799 text as possible with each rule. In general, when attempting to write
1800 a high-speed scanner try to match as much possible in each rule, as
1803 Note that start-conditions names are really integer values and
1804 can be stored as such. Thus, the above could be extended in the
1807 @cindex start conditions, integer values
1808 @cindex using integer values of start condition names
1817 comment_caller = INITIAL;
1824 comment_caller = foo;
1828 <comment>[^*\n]* /* eat anything that's not a '*' */
1829 <comment>"*"+[^*/\n]* /* eat up '*'s not followed by '/'s */
1830 <comment>\n ++line_num;
1831 <comment>"*"+"/" BEGIN(comment_caller);
1835 @cindex YY_START, example
1836 Furthermore, you can access the current start condition using the
1837 integer-valued @code{YY_START} macro. For example, the above
1838 assignments to @code{comment_caller} could instead be written
1840 @cindex getting current start state with YY_START
1843 comment_caller = YY_START;
1848 Flex provides @code{YYSTATE} as an alias for @code{YY_START} (since that
1849 is what's used by AT&T @code{lex}).
1851 For historical reasons, start conditions do not have their own
1852 name-space within the generated scanner. The start condition names are
1853 unmodified in the generated scanner and generated header.
1854 @xref{option-header}. @xref{option-prefix}.
1858 Finally, here's an example of how to match C-style quoted strings using
1859 exclusive start conditions, including expanded escape sequences (but
1860 not including checking for a string that's too long):
1862 @cindex matching C-style double-quoted strings
1868 char string_buf[MAX_STR_CONST];
1869 char *string_buf_ptr;
1872 \" string_buf_ptr = string_buf; BEGIN(str);
1874 <str>\" { /* saw closing quote - all done */
1876 *string_buf_ptr = '\0';
1877 /* return string constant token type and
1883 /* error - unterminated string constant */
1884 /* generate error message */
1888 /* octal escape sequence */
1891 (void) sscanf( yytext + 1, "%o", &result );
1893 if ( result > 0xff )
1894 /* error, constant is out-of-bounds */
1896 *string_buf_ptr++ = result;
1900 /* generate error - bad escape sequence; something
1901 * like '\48' or '\0777777'
1905 <str>\\n *string_buf_ptr++ = '\n';
1906 <str>\\t *string_buf_ptr++ = '\t';
1907 <str>\\r *string_buf_ptr++ = '\r';
1908 <str>\\b *string_buf_ptr++ = '\b';
1909 <str>\\f *string_buf_ptr++ = '\f';
1911 <str>\\(.|\n) *string_buf_ptr++ = yytext[1];
1914 char *yptr = yytext;
1917 *string_buf_ptr++ = *yptr++;
1922 @cindex start condition, applying to multiple patterns
1923 Often, such as in some of the examples above, you wind up writing a
1924 whole bunch of rules all preceded by the same start condition(s). Flex
1925 makes this a little easier and cleaner by introducing a notion of start
1926 condition @dfn{scope}. A start condition scope is begun with:
1934 where @code{<SCs>} is a list of one or more start conditions. Inside the
1935 start condition scope, every rule automatically has the prefix
1936 @code{<SCs>} applied to it, until a @samp{@}} which matches the initial
1937 @samp{@{}. So, for example,
1939 @cindex extended scope of start conditions
1955 <ESC>"\\n" return '\n';
1956 <ESC>"\\r" return '\r';
1957 <ESC>"\\f" return '\f';
1958 <ESC>"\\0" return '\0';
1962 Start condition scopes may be nested.
1964 @cindex stacks, routines for manipulating
1965 @cindex start conditions, use of a stack
1967 The following routines are available for manipulating stacks of start conditions:
1969 @deftypefun void yy_push_state ( int @code{new_state} )
1970 pushes the current start condition onto the top of the start condition
1971 stack and switches to
1973 as though you had used
1974 @code{BEGIN new_state}
1975 (recall that start condition names are also integers).
1978 @deftypefun void yy_pop_state ()
1979 pops the top of the stack and switches to it via
1983 @deftypefun int yy_top_state ()
1984 returns the top of the stack without altering the stack's contents.
1987 @cindex memory, for start condition stacks
1988 The start condition stack grows dynamically and so has no built-in size
1989 limitation. If memory is exhausted, program execution aborts.
1991 To use start condition stacks, your scanner must include a @code{%option
1992 stack} directive (@pxref{Scanner Options}).
1994 @node Multiple Input Buffers, EOF, Start Conditions, Top
1995 @chapter Multiple Input Buffers
1997 @cindex multiple input streams
1998 Some scanners (such as those which support ``include'' files) require
1999 reading from several input streams. As @code{flex} scanners do a large
2000 amount of buffering, one cannot control where the next input will be
2001 read from by simply writing a @code{YY_INPUT()} which is sensitive to
2002 the scanning context. @code{YY_INPUT()} is only called when the scanner
2003 reaches the end of its buffer, which may be a long time after scanning a
2004 statement such as an @code{include} statement which requires switching
2007 To negotiate these sorts of problems, @code{flex} provides a mechanism
2008 for creating and switching between multiple input buffers. An input
2009 buffer is created by using:
2011 @cindex memory, allocating input buffers
2012 @deftypefun YY_BUFFER_STATE yy_create_buffer ( FILE *file, int size )
2015 which takes a @code{FILE} pointer and a size and creates a buffer
2016 associated with the given file and large enough to hold @code{size}
2017 characters (when in doubt, use @code{YY_BUF_SIZE} for the size). It
2018 returns a @code{YY_BUFFER_STATE} handle, which may then be passed to
2019 other routines (see below).
2020 @tindex YY_BUFFER_STATE
2021 The @code{YY_BUFFER_STATE} type is a
2022 pointer to an opaque @code{struct yy_buffer_state} structure, so you may
2023 safely initialize @code{YY_BUFFER_STATE} variables to @code{((YY_BUFFER_STATE)
2024 0)} if you wish, and also refer to the opaque structure in order to
2025 correctly declare input buffers in source files other than that of your
2026 scanner. Note that the @code{FILE} pointer in the call to
2027 @code{yy_create_buffer} is only used as the value of @file{yyin} seen by
2028 @code{YY_INPUT}. If you redefine @code{YY_INPUT()} so it no longer uses
2029 @file{yyin}, then you can safely pass a NULL @code{FILE} pointer to
2030 @code{yy_create_buffer}. You select a particular buffer to scan from
2033 @deftypefun void yy_switch_to_buffer ( YY_BUFFER_STATE new_buffer )
2036 The above function switches the scanner's input buffer so subsequent tokens
2037 will come from @code{new_buffer}. Note that @code{yy_switch_to_buffer()} may
2038 be used by @code{yywrap()} to set things up for continued scanning, instead of
2039 opening a new file and pointing @file{yyin} at it. If you are looking for a
2040 stack of input buffers, then you want to use @code{yypush_buffer_state()}
2041 instead of this function. Note also that switching input sources via either
2042 @code{yy_switch_to_buffer()} or @code{yywrap()} does @emph{not} change the
2045 @cindex memory, deleting input buffers
2046 @deftypefun void yy_delete_buffer ( YY_BUFFER_STATE buffer )
2049 is used to reclaim the storage associated with a buffer. (@code{buffer}
2050 can be NULL, in which case the routine does nothing.) You can also clear
2051 the current contents of a buffer using:
2053 @cindex pushing an input buffer
2054 @cindex stack, input buffer push
2055 @deftypefun void yypush_buffer_state ( YY_BUFFER_STATE buffer )
2058 This function pushes the new buffer state onto an internal stack. The pushed
2059 state becomes the new current state. The stack is maintained by flex and will
2060 grow as required. This function is intended to be used instead of
2061 @code{yy_switch_to_buffer}, when you want to change states, but preserve the
2062 current state for later use.
2064 @cindex popping an input buffer
2065 @cindex stack, input buffer pop
2066 @deftypefun void yypop_buffer_state ( )
2069 This function removes the current state from the top of the stack, and deletes
2070 it by calling @code{yy_delete_buffer}. The next state on the stack, if any,
2071 becomes the new current state.
2073 @cindex clearing an input buffer
2074 @cindex flushing an input buffer
2075 @deftypefun void yy_flush_buffer ( YY_BUFFER_STATE buffer )
2078 This function discards the buffer's contents,
2079 so the next time the scanner attempts to match a token from the
2080 buffer, it will first fill the buffer anew using
2083 @deftypefun YY_BUFFER_STATE yy_new_buffer ( FILE *file, int size )
2086 is an alias for @code{yy_create_buffer()},
2087 provided for compatibility with the C++ use of @code{new} and
2088 @code{delete} for creating and destroying dynamic objects.
2090 @cindex YY_CURRENT_BUFFER, and multiple buffers Finally, the macro
2091 @code{YY_CURRENT_BUFFER} macro returns a @code{YY_BUFFER_STATE} handle to the
2092 current buffer. It should not be used as an lvalue.
2094 @cindex EOF, example using multiple input buffers
2095 Here are two examples of using these features for writing a scanner
2096 which expands include files (the
2098 feature is discussed below).
2100 This first example uses yypush_buffer_state and yypop_buffer_state. Flex
2101 maintains the stack internally.
2103 @cindex handling include files with multiple input buffers
2106 /* the "incl" state is used for picking up the name
2107 * of an include file
2111 include BEGIN(incl);
2116 <incl>[ \t]* /* eat the whitespace */
2117 <incl>[^ \t\n]+ { /* got the include file name */
2118 yyin = fopen( yytext, "r" );
2123 yypush_buffer_state(yy_create_buffer( yyin, YY_BUF_SIZE ));
2129 yypop_buffer_state();
2131 if ( !YY_CURRENT_BUFFER )
2139 The second example, below, does the same thing as the previous example did, but
2140 manages its own input buffer stack manually (instead of letting flex do it).
2142 @cindex handling include files with multiple input buffers
2145 /* the "incl" state is used for picking up the name
2146 * of an include file
2151 #define MAX_INCLUDE_DEPTH 10
2152 YY_BUFFER_STATE include_stack[MAX_INCLUDE_DEPTH];
2153 int include_stack_ptr = 0;
2157 include BEGIN(incl);
2162 <incl>[ \t]* /* eat the whitespace */
2163 <incl>[^ \t\n]+ { /* got the include file name */
2164 if ( include_stack_ptr >= MAX_INCLUDE_DEPTH )
2166 fprintf( stderr, "Includes nested too deeply" );
2170 include_stack[include_stack_ptr++] =
2173 yyin = fopen( yytext, "r" );
2178 yy_switch_to_buffer(
2179 yy_create_buffer( yyin, YY_BUF_SIZE ) );
2185 if ( --include_stack_ptr == 0 )
2192 yy_delete_buffer( YY_CURRENT_BUFFER );
2193 yy_switch_to_buffer(
2194 include_stack[include_stack_ptr] );
2200 @anchor{Scanning Strings}
2201 @cindex strings, scanning strings instead of files
2202 The following routines are available for setting up input buffers for
2203 scanning in-memory strings instead of files. All of them create a new
2204 input buffer for scanning the string, and return a corresponding
2205 @code{YY_BUFFER_STATE} handle (which you should delete with
2206 @code{yy_delete_buffer()} when done with it). They also switch to the
2207 new buffer using @code{yy_switch_to_buffer()}, so the next call to
2208 @code{yylex()} will start scanning the string.
2210 @deftypefun YY_BUFFER_STATE yy_scan_string ( const char *str )
2211 scans a NUL-terminated string.
2214 @deftypefun YY_BUFFER_STATE yy_scan_bytes ( const char *bytes, int len )
2215 scans @code{len} bytes (including possibly @code{NUL}s) starting at location
2219 Note that both of these functions create and scan a @emph{copy} of the
2220 string or bytes. (This may be desirable, since @code{yylex()} modifies
2221 the contents of the buffer it is scanning.) You can avoid the copy by
2224 @vindex YY_END_OF_BUFFER_CHAR
2225 @deftypefun YY_BUFFER_STATE yy_scan_buffer (char *base, yy_size_t size)
2226 which scans in place the buffer starting at @code{base}, consisting of
2227 @code{size} bytes, the last two bytes of which @emph{must} be
2228 @code{YY_END_OF_BUFFER_CHAR} (ASCII NUL). These last two bytes are not
2229 scanned; thus, scanning consists of @code{base[0]} through
2230 @code{base[size-2]}, inclusive.
2233 If you fail to set up @code{base} in this manner (i.e., forget the final
2234 two @code{YY_END_OF_BUFFER_CHAR} bytes), then @code{yy_scan_buffer()}
2235 returns a NULL pointer instead of creating a new input buffer.
2237 @deftp {Data type} yy_size_t
2238 is an integral type to which you can cast an integer expression
2239 reflecting the size of the buffer.
2242 @node EOF, Misc Macros, Multiple Input Buffers, Top
2243 @chapter End-of-File Rules
2245 @cindex EOF, explanation
2246 The special rule @code{<<EOF>>} indicates
2247 actions which are to be taken when an end-of-file is
2248 encountered and @code{yywrap()} returns non-zero (i.e., indicates
2249 no further files to process). The action must finish
2250 by doing one of the following things:
2254 @findex YY_NEW_FILE (now obsolete)
2255 assigning @file{yyin} to a new input file (in previous versions of
2256 @code{flex}, after doing the assignment you had to call the special
2257 action @code{YY_NEW_FILE}. This is no longer necessary.)
2260 executing a @code{return} statement;
2263 executing the special @code{yyterminate()} action.
2266 or, switching to a new buffer using @code{yy_switch_to_buffer()} as
2267 shown in the example above.
2270 <<EOF>> rules may not be used with other patterns; they may only be
2271 qualified with a list of start conditions. If an unqualified <<EOF>>
2272 rule is given, it applies to @emph{all} start conditions which do not
2273 already have <<EOF>> actions. To specify an <<EOF>> rule for only the
2274 initial start condition, use:
2282 These rules are useful for catching things like unclosed comments. An
2285 @cindex <<EOF>>, use of
2291 ...other rules for dealing with quotes...
2294 error( "unterminated quote" );
2299 yyin = fopen( *filelist, "r" );
2306 @node Misc Macros, User Values, EOF, Top
2307 @chapter Miscellaneous Macros
2309 @hkindex YY_USER_ACTION
2310 The macro @code{YY_USER_ACTION} can be defined to provide an action
2311 which is always executed prior to the matched rule's action. For
2312 example, it could be #define'd to call a routine to convert yytext to
2313 lower-case. When @code{YY_USER_ACTION} is invoked, the variable
2314 @code{yy_act} gives the number of the matched rule (rules are numbered
2315 starting with 1). Suppose you want to profile how often each of your
2316 rules is matched. The following would do the trick:
2318 @cindex YY_USER_ACTION to track each time a rule is matched
2321 #define YY_USER_ACTION ++ctr[yy_act]
2325 @vindex YY_NUM_RULES
2326 where @code{ctr} is an array to hold the counts for the different rules.
2327 Note that the macro @code{YY_NUM_RULES} gives the total number of rules
2328 (including the default rule), even if you use @samp{-s)}, so a correct
2329 declaration for @code{ctr} is:
2333 int ctr[YY_NUM_RULES];
2337 @hkindex YY_USER_INIT
2338 The macro @code{YY_USER_INIT} may be defined to provide an action which
2339 is always executed before the first scan (and before the scanner's
2340 internal initializations are done). For example, it could be used to
2341 call a routine to read in a data table or open a logging file.
2343 @findex yy_set_interactive
2344 The macro @code{yy_set_interactive(is_interactive)} can be used to
2345 control whether the current buffer is considered @dfn{interactive}. An
2346 interactive buffer is processed more slowly, but must be used when the
2347 scanner's input source is indeed interactive to avoid problems due to
2348 waiting to fill buffers (see the discussion of the @samp{-I} flag in
2349 @ref{Scanner Options}). A non-zero value in the macro invocation marks
2350 the buffer as interactive, a zero value as non-interactive. Note that
2351 use of this macro overrides @code{%option always-interactive} or
2352 @code{%option never-interactive} (@pxref{Scanner Options}).
2353 @code{yy_set_interactive()} must be invoked prior to beginning to scan
2354 the buffer that is (or is not) to be considered interactive.
2356 @cindex BOL, setting it
2358 The macro @code{yy_set_bol(at_bol)} can be used to control whether the
2359 current buffer's scanning context for the next token match is done as
2360 though at the beginning of a line. A non-zero macro argument makes
2361 rules anchored with @samp{^} active, while a zero argument makes
2362 @samp{^} rules inactive.
2364 @cindex BOL, checking the BOL flag
2366 The macro @code{YY_AT_BOL()} returns true if the next token scanned from
2367 the current buffer will have @samp{^} rules active, false otherwise.
2369 @cindex actions, redefining YY_BREAK
2371 In the generated scanner, the actions are all gathered in one large
2372 switch statement and separated using @code{YY_BREAK}, which may be
2373 redefined. By default, it is simply a @code{break}, to separate each
2374 rule's action from the following rule's. Redefining @code{YY_BREAK}
2375 allows, for example, C++ users to #define YY_BREAK to do nothing (while
2376 being very careful that every rule ends with a @code{break} or a
2377 @code{return}!) to avoid suffering from unreachable statement warnings
2378 where because a rule's action ends with @code{return}, the
2379 @code{YY_BREAK} is inaccessible.
2381 @node User Values, Yacc, Misc Macros, Top
2382 @chapter Values Available To the User
2384 This chapter summarizes the various values available to the user in the
2390 holds the text of the current token. It may be modified but not
2391 lengthened (you cannot append characters to the end).
2393 @cindex yytext, default array size
2394 @cindex array, default size for yytext
2396 If the special directive @code{%array} appears in the first section of
2397 the scanner description, then @code{yytext} is instead declared
2398 @code{char yytext[YYLMAX]}, where @code{YYLMAX} is a macro definition
2399 that you can redefine in the first section if you don't like the default
2400 value (generally 8KB). Using @code{%array} results in somewhat slower
2401 scanners, but the value of @code{yytext} becomes immune to calls to
2402 @code{unput()}, which potentially destroy its value when @code{yytext} is
2403 a character pointer. The opposite of @code{%array} is @code{%pointer},
2404 which is the default.
2406 @cindex C++ and %array
2407 You cannot use @code{%array} when generating C++ scanner classes (the
2412 holds the length of the current token.
2416 is the file which by default @code{flex} reads from. It may be
2417 redefined but doing so only makes sense before scanning begins or after
2418 an EOF has been encountered. Changing it in the midst of scanning will
2419 have unexpected results since @code{flex} buffers its input; use
2420 @code{yyrestart()} instead. Once scanning terminates because an
2421 end-of-file has been seen, you can assign @file{yyin} at the new input
2422 file and then call the scanner again to continue scanning.
2425 @item void yyrestart( FILE *new_file )
2426 may be called to point @file{yyin} at the new input file. The
2427 switch-over to the new file is immediate (any previously buffered-up
2428 input is lost). Note that calling @code{yyrestart()} with @file{yyin}
2429 as an argument thus throws away the current input buffer and continues
2430 scanning the same input file.
2434 is the file to which @code{ECHO} actions are done. It can be reassigned
2437 @vindex YY_CURRENT_BUFFER
2438 @item YY_CURRENT_BUFFER
2439 returns a @code{YY_BUFFER_STATE} handle to the current buffer.
2443 returns an integer value corresponding to the current start condition.
2444 You can subsequently use this value with @code{BEGIN} to return to that
2448 @node Yacc, Scanner Options, User Values, Top
2449 @chapter Interfacing with Yacc
2451 @cindex yacc, interface
2453 @vindex yylval, with yacc
2454 One of the main uses of @code{flex} is as a companion to the @code{yacc}
2455 parser-generator. @code{yacc} parsers expect to call a routine named
2456 @code{yylex()} to find the next input token. The routine is supposed to
2457 return the type of the next token as well as putting any associated
2458 value in the global @code{yylval}. To use @code{flex} with @code{yacc},
2459 one specifies the @samp{-d} option to @code{yacc} to instruct it to
2460 generate the file @file{y.tab.h} containing definitions of all the
2461 @code{%tokens} appearing in the @code{yacc} input. This file is then
2462 included in the @code{flex} scanner. For example, if one of the tokens
2463 is @code{TOK_NUMBER}, part of the scanner might look like:
2465 @cindex yacc interface
2474 [0-9]+ yylval = atoi( yytext ); return TOK_NUMBER;
2478 @node Scanner Options, Performance, Yacc, Top
2479 @chapter Scanner Options
2481 @cindex command-line options
2482 @cindex options, command-line
2483 @cindex arguments, command-line
2485 The various @code{flex} options are categorized by function in the following
2486 menu. If you want to lookup a particular option by name, @xref{Index of Scanner Options}.
2489 * Options for Specifying Filenames::
2490 * Options Affecting Scanner Behavior::
2491 * Code-Level And API Options::
2492 * Options for Scanner Speed and Size::
2493 * Debugging Options::
2494 * Miscellaneous Options::
2497 Even though there are many scanner options, a typical scanner might only
2498 specify the following options:
2502 %option 8bit reentrant bison-bridge
2503 %option warn nodefault
2505 %option outfile="scanner.c" header-file="scanner.h"
2509 The first line specifies the general type of scanner we want. The second line
2510 specifies that we are being careful. The third line asks flex to track line
2511 numbers. The last line tells flex what to name the files. (The options can be
2512 specified in any order. We just divided them.)
2514 @code{flex} also provides a mechanism for controlling options within the
2515 scanner specification itself, rather than from the flex command-line.
2516 This is done by including @code{%option} directives in the first section
2517 of the scanner specification. You can specify multiple options with a
2518 single @code{%option} directive, and multiple directives in the first
2519 section of your flex input file.
2521 Most options are given simply as names, optionally preceded by the
2522 word @samp{no} (with no intervening whitespace) to negate their meaning.
2523 The names are the same as their long-option equivalents (but without the
2524 leading @samp{--} ).
2526 @code{flex} scans your rule actions to determine whether you use the
2527 @code{REJECT} or @code{yymore()} features. The @code{REJECT} and
2528 @code{yymore} options are available to override its decision as to
2529 whether you use the options, either by setting them (e.g., @code{%option
2530 reject)} to indicate the feature is indeed used, or unsetting them to
2531 indicate it actually is not used (e.g., @code{%option noyymore)}.
2534 A number of options are available for lint purists who want to suppress
2535 the appearance of unneeded routines in the generated scanner. Each of
2536 the following, if unset (e.g., @code{%option nounput}), results in the
2537 corresponding routine not appearing in the generated scanner:
2542 yy_push_state, yy_pop_state, yy_top_state
2543 yy_scan_buffer, yy_scan_bytes, yy_scan_string
2545 yyget_extra, yyset_extra, yyget_leng, yyget_text,
2546 yyget_lineno, yyset_lineno, yyget_in, yyset_in,
2547 yyget_out, yyset_out, yyget_lval, yyset_lval,
2548 yyget_lloc, yyset_lloc, yyget_debug, yyset_debug
2552 (though @code{yy_push_state()} and friends won't appear anyway unless
2553 you use @code{%option stack)}.
2555 @node Options for Specifying Filenames, Options Affecting Scanner Behavior, Scanner Options, Scanner Options
2556 @section Options for Specifying Filenames
2560 @anchor{option-header}
2561 @opindex ---header-file
2562 @opindex header-file
2563 @item --header-file=FILE, @code{%option header-file="FILE"}
2564 instructs flex to write a C header to @file{FILE}. This file contains
2565 function prototypes, extern variables, and types used by the scanner.
2566 Only the external API is exported by the header file. Many macros that
2567 are usable from within scanner actions are not exported to the header
2568 file. This is due to namespace problems and the goal of a clean
2571 While in the header, the macro @code{yyIN_HEADER} is defined, where @samp{yy}
2572 is substituted with the appropriate prefix.
2574 The @samp{--header-file} option is not compatible with the @samp{--c++} option,
2575 since the C++ scanner provides its own header in @file{yyFlexLexer.h}.
2579 @anchor{option-outfile}
2583 @item -oFILE, --outfile=FILE, @code{%option outfile="FILE"}
2584 directs flex to write the scanner to the file @file{FILE} instead of
2585 @file{lex.yy.c}. If you combine @samp{--outfile} with the @samp{--stdout} option,
2586 then the scanner is written to @file{stdout} but its @code{#line}
2587 directives (see the @samp{-l} option above) refer to the file
2592 @anchor{option-stdout}
2596 @item -t, --stdout, @code{%option stdout}
2597 instructs @code{flex} to write the scanner it generates to standard
2598 output instead of @file{lex.yy.c}.
2603 @item -SFILE, --skel=FILE
2604 overrides the default skeleton file from which
2606 constructs its scanners. You'll never need this option unless you are doing
2608 maintenance or development.
2610 @opindex ---tables-file
2611 @opindex tables-file
2612 @item --tables-file=FILE
2613 Write serialized scanner dfa tables to FILE. The generated scanner will not
2614 contain the tables, and requires them to be loaded at runtime.
2615 @xref{serialization}.
2617 @opindex ---tables-verify
2618 @opindex tables-verify
2619 @item --tables-verify
2620 This option is for flex development. We document it here in case you stumble
2621 upon it by accident or in case you suspect some inconsistency in the serialized
2622 tables. Flex will serialize the scanner dfa tables but will also generate the
2623 in-code tables as it normally does. At runtime, the scanner will verify that
2624 the serialized tables match the in-code tables, instead of loading them.
2628 @node Options Affecting Scanner Behavior, Code-Level And API Options, Options for Specifying Filenames, Scanner Options
2629 @section Options Affecting Scanner Behavior
2632 @anchor{option-case-insensitive}
2634 @opindex ---case-insensitive
2635 @opindex case-insensitive
2636 @item -i, --case-insensitive, @code{%option case-insensitive}
2637 instructs @code{flex} to generate a @dfn{case-insensitive} scanner. The
2638 case of letters given in the @code{flex} input patterns will be ignored,
2639 and tokens in the input will be matched regardless of case. The matched
2640 text given in @code{yytext} will have the preserved case (i.e., it will
2641 not be folded). For tricky behavior, see @ref{case and character ranges}.
2645 @anchor{option-lex-compat}
2647 @opindex ---lex-compat
2649 @item -l, --lex-compat, @code{%option lex-compat}
2650 turns on maximum compatibility with the original AT&T @code{lex}
2651 implementation. Note that this does not mean @emph{full} compatibility.
2652 Use of this option costs a considerable amount of performance, and it
2653 cannot be used with the @samp{--c++}, @samp{--full}, @samp{--fast}, @samp{-Cf}, or
2654 @samp{-CF} options. For details on the compatibilities it provides, see
2655 @ref{Lex and Posix}. This option also results in the name
2656 @code{YY_FLEX_LEX_COMPAT} being @code{#define}'d in the generated scanner.
2660 @anchor{option-batch}
2664 @item -B, --batch, @code{%option batch}
2665 instructs @code{flex} to generate a @dfn{batch} scanner, the opposite of
2666 @emph{interactive} scanners generated by @samp{--interactive} (see below). In
2667 general, you use @samp{-B} when you are @emph{certain} that your scanner
2668 will never be used interactively, and you want to squeeze a
2669 @emph{little} more performance out of it. If your goal is instead to
2670 squeeze out a @emph{lot} more performance, you should be using the
2671 @samp{-Cf} or @samp{-CF} options, which turn on @samp{--batch} automatically
2676 @anchor{option-interactive}
2678 @opindex ---interactive
2679 @opindex interactive
2680 @item -I, --interactive, @code{%option interactive}
2681 instructs @code{flex} to generate an @i{interactive} scanner. An
2682 interactive scanner is one that only looks ahead to decide what token
2683 has been matched if it absolutely must. It turns out that always
2684 looking one extra character ahead, even if the scanner has already seen
2685 enough text to disambiguate the current token, is a bit faster than only
2686 looking ahead when necessary. But scanners that always look ahead give
2687 dreadful interactive performance; for example, when a user types a
2688 newline, it is not recognized as a newline token until they enter
2689 @emph{another} token, which often means typing in another whole line.
2691 @code{flex} scanners default to @code{interactive} unless you use the
2692 @samp{-Cf} or @samp{-CF} table-compression options
2693 (@pxref{Performance}). That's because if you're looking for
2694 high-performance you should be using one of these options, so if you
2695 didn't, @code{flex} assumes you'd rather trade off a bit of run-time
2696 performance for intuitive interactive behavior. Note also that you
2697 @emph{cannot} use @samp{--interactive} in conjunction with @samp{-Cf} or
2698 @samp{-CF}. Thus, this option is not really needed; it is on by default
2699 for all those cases in which it is allowed.
2701 You can force a scanner to
2703 be interactive by using
2708 @anchor{option-7bit}
2712 @item -7, --7bit, @code{%option 7bit}
2713 instructs @code{flex} to generate a 7-bit scanner, i.e., one which can
2714 only recognize 7-bit characters in its input. The advantage of using
2715 @samp{--7bit} is that the scanner's tables can be up to half the size of
2716 those generated using the @samp{--8bit}. The disadvantage is that such
2717 scanners often hang or crash if their input contains an 8-bit character.
2719 Note, however, that unless you generate your scanner using the
2720 @samp{-Cf} or @samp{-CF} table compression options, use of @samp{--7bit}
2721 will save only a small amount of table space, and make your scanner
2722 considerably less portable. @code{Flex}'s default behavior is to
2723 generate an 8-bit scanner unless you use the @samp{-Cf} or @samp{-CF},
2724 in which case @code{flex} defaults to generating 7-bit scanners unless
2725 your site was always configured to generate 8-bit scanners (as will
2726 often be the case with non-USA sites). You can tell whether flex
2727 generated a 7-bit or an 8-bit scanner by inspecting the flag summary in
2728 the @samp{--verbose} output as described above.
2730 Note that if you use @samp{-Cfe} or @samp{-CFe} @code{flex} still
2731 defaults to generating an 8-bit scanner, since usually with these
2732 compression options full 8-bit tables are not much more expensive than
2737 @anchor{option-8bit}
2741 @item -8, --8bit, @code{%option 8bit}
2742 instructs @code{flex} to generate an 8-bit scanner, i.e., one which can
2743 recognize 8-bit characters. This flag is only needed for scanners
2744 generated using @samp{-Cf} or @samp{-CF}, as otherwise flex defaults to
2745 generating an 8-bit scanner anyway.
2747 See the discussion of
2749 above for @code{flex}'s default behavior and the tradeoffs between 7-bit
2754 @anchor{option-default}
2757 @item --default, @code{%option default}
2758 generate the default rule.
2762 @anchor{option-always-interactive}
2763 @opindex ---always-interactive
2764 @opindex always-interactive
2765 @item --always-interactive, @code{%option always-interactive}
2766 instructs flex to generate a scanner which always considers its input
2767 @emph{interactive}. Normally, on each new input file the scanner calls
2768 @code{isatty()} in an attempt to determine whether the scanner's input
2769 source is interactive and thus should be read a character at a time.
2770 When this option is used, however, then no such call is made.
2774 @opindex ---never-interactive
2775 @item --never-interactive, @code{--never-interactive}
2776 instructs flex to generate a scanner which never considers its input
2777 interactive. This is the opposite of @code{always-interactive}.
2780 @anchor{option-posix}
2784 @item -X, --posix, @code{%option posix}
2785 turns on maximum compatibility with the POSIX 1003.2-1992 definition of
2786 @code{lex}. Since @code{flex} was originally designed to implement the
2787 POSIX definition of @code{lex} this generally involves very few changes
2788 in behavior. At the current writing the known differences between
2789 @code{flex} and the POSIX standard are:
2793 In POSIX and AT&T @code{lex}, the repeat operator, @samp{@{@}}, has lower
2794 precedence than concatenation (thus @samp{ab@{3@}} yields @samp{ababab}).
2795 Most POSIX utilities use an Extended Regular Expression (ERE) precedence
2796 that has the precedence of the repeat operator higher than concatenation
2797 (which causes @samp{ab@{3@}} to yield @samp{abbb}). By default, @code{flex}
2798 places the precedence of the repeat operator higher than concatenation
2799 which matches the ERE processing of other POSIX utilities. When either
2800 @samp{--posix} or @samp{-l} are specified, @code{flex} will use the
2801 traditional AT&T and POSIX-compliant precedence for the repeat operator
2802 where concatenation has higher precedence than the repeat operator.
2806 @anchor{option-stack}
2809 @item --stack, @code{%option stack}
2811 start condition stacks (@pxref{Start Conditions}).
2815 @anchor{option-stdinit}
2818 @item --stdinit, @code{%option stdinit}
2819 if set (i.e., @b{%option stdinit)} initializes @code{yyin} and
2820 @code{yyout} to @file{stdin} and @file{stdout}, instead of the default of
2821 @file{NULL}. Some existing @code{lex} programs depend on this behavior,
2822 even though it is not compliant with ANSI C, which does not require
2823 @file{stdin} and @file{stdout} to be compile-time constant. In a
2824 reentrant scanner, however, this is not a problem since initialization
2825 is performed in @code{yylex_init} at runtime.
2829 @anchor{option-yylineno}
2830 @opindex ---yylineno
2832 @item --yylineno, @code{%option yylineno}
2833 directs @code{flex} to generate a scanner
2834 that maintains the number of the current line read from its input in the
2835 global variable @code{yylineno}. This option is implied by @code{%option
2836 lex-compat}. In a reentrant C scanner, the macro @code{yylineno} is
2837 accessible regardless of the value of @code{%option yylineno}, however, its
2838 value is not modified by @code{flex} unless @code{%option yylineno} is enabled.
2842 @anchor{option-yywrap}
2845 @item --yywrap, @code{%option yywrap}
2846 if unset (i.e., @code{--noyywrap)}, makes the scanner not call
2847 @code{yywrap()} upon an end-of-file, but simply assume that there are no
2848 more files to scan (until the user points @file{yyin} at a new file and
2849 calls @code{yylex()} again).
2853 @node Code-Level And API Options, Options for Scanner Speed and Size, Options Affecting Scanner Behavior, Scanner Options
2854 @section Code-Level And API Options
2858 @anchor{option-ansi-definitions}
2859 @opindex ---option-ansi-definitions
2860 @opindex ansi-definitions
2861 @item --ansi-definitions, @code{%option ansi-definitions}
2864 @anchor{option-ansi-prototypes}
2865 @opindex ---option-ansi-prototypes
2866 @opindex ansi-prototypes
2867 @item --ansi-prototypes, @code{%option ansi-prototypes}
2870 @anchor{option-bison-bridge}
2871 @opindex ---bison-bridge
2872 @opindex bison-bridge
2873 @item --bison-bridge, @code{%option bison-bridge}
2874 instructs flex to generate a C scanner that is
2875 meant to be called by a
2877 parser. The scanner has minor API changes for
2879 compatibility. In particular, the declaration of
2881 is modified to take an additional parameter,
2883 @xref{Bison Bridge}.
2885 @anchor{option-bison-locations}
2886 @opindex ---bison-locations
2887 @opindex bison-locations
2888 @item --bison-locations, @code{%option bison-locations}
2890 @code{GNU bison} @code{%locations} are being used.
2891 This means @code{yylex} will be passed
2892 an additional parameter, @code{yylloc}. This option
2893 implies @code{%option bison-bridge}.
2894 @xref{Bison Bridge}.
2896 @anchor{option-noline}
2900 @item -L, --noline, @code{%option noline}
2905 directives. Without this option,
2907 peppers the generated scanner
2908 with @code{#line} directives so error messages in the actions will be correctly
2909 located with respect to either the original
2911 input file (if the errors are due to code in the input file), or
2915 fault -- you should report these sorts of errors to the email address
2916 given in @ref{Reporting Bugs}).
2920 @anchor{option-reentrant}
2922 @opindex ---reentrant
2924 @item -R, --reentrant, @code{%option reentrant}
2925 instructs flex to generate a reentrant C scanner. The generated scanner
2926 may safely be used in a multi-threaded environment. The API for a
2927 reentrant scanner is different than for a non-reentrant scanner
2928 @pxref{Reentrant}). Because of the API difference between
2929 reentrant and non-reentrant @code{flex} scanners, non-reentrant flex
2930 code must be modified before it is suitable for use with this option.
2931 This option is not compatible with the @samp{--c++} option.
2933 The option @samp{--reentrant} does not affect the performance of
2942 @item -+, --c++, @code{%option c++}
2943 specifies that you want flex to generate a C++
2944 scanner class. @xref{Cxx}, for
2949 @anchor{option-array}
2952 @item --array, @code{%option array}
2953 specifies that you want yytext to be an array instead of a char*
2957 @anchor{option-pointer}
2960 @item --pointer, @code{%option pointer}
2961 specify that @code{yytext} should be a @code{char *}, not an array.
2962 This default is @code{char *}.
2966 @anchor{option-prefix}
2970 @item -PPREFIX, --prefix=PREFIX, @code{%option prefix="PREFIX"}
2971 changes the default @samp{yy} prefix used by @code{flex} for all
2972 globally-visible variable and function names to instead be
2973 @samp{PREFIX}. For example, @samp{--prefix=foo} changes the name of
2974 @code{yytext} to @code{footext}. It also changes the name of the default
2975 output file from @file{lex.yy.c} to @file{lex.foo.c}. Here is a partial
2976 list of the names affected:
2985 yy_load_buffer_state
3001 (If you are using a C++ scanner, then only @code{yywrap} and
3002 @code{yyFlexLexer} are affected.) Within your scanner itself, you can
3003 still refer to the global variables and functions using either version
3004 of their name; but externally, they have the modified name.
3006 This option lets you easily link together multiple
3008 programs into the same executable. Note, though, that using this
3014 provide your own (appropriately-named) version of the routine for your
3016 @code{%option noyywrap},
3019 no longer provides one for you by default.
3023 @anchor{option-main}
3026 @item --main, @code{%option main}
3027 directs flex to provide a default @code{main()} program for the
3028 scanner, which simply calls @code{yylex()}. This option implies
3029 @code{noyywrap} (see below).
3033 @anchor{option-nounistd}
3034 @opindex ---nounistd
3036 @item --nounistd, @code{%option nounistd}
3037 suppresses inclusion of the non-ANSI header file @file{unistd.h}. This option
3038 is meant to target environments in which @file{unistd.h} does not exist. Be aware
3039 that certain options may cause flex to generate code that relies on functions
3040 normally found in @file{unistd.h}, (e.g. @code{isatty()}, @code{read()}.)
3041 If you wish to use these functions, you will have to inform your compiler where
3043 @xref{option-always-interactive}. @xref{option-read}.
3047 @anchor{option-yyclass}
3050 @item --yyclass=NAME, @code{%option yyclass="NAME"}
3051 only applies when generating a C++ scanner (the @samp{--c++} option). It
3052 informs @code{flex} that you have derived @code{NAME} as a subclass of
3053 @code{yyFlexLexer}, so @code{flex} will place your actions in the member
3054 function @code{foo::yylex()} instead of @code{yyFlexLexer::yylex()}. It
3055 also generates a @code{yyFlexLexer::yylex()} member function that emits
3056 a run-time error (by invoking @code{yyFlexLexer::LexerError())} if
3061 @node Options for Scanner Speed and Size, Debugging Options, Code-Level And API Options, Scanner Options
3062 @section Options for Scanner Speed and Size
3067 controls the degree of table compression and, more generally, trade-offs
3068 between small scanners and fast scanners.
3073 A lone @samp{-C} specifies that the scanner tables should be compressed
3074 but neither equivalence classes nor meta-equivalence classes should be
3077 @anchor{option-align}
3081 @item -Ca, --align, @code{%option align}
3082 (``align'') instructs flex to trade off larger tables in the
3083 generated scanner for faster performance because the elements of
3084 the tables are better aligned for memory access and computation. On some
3085 RISC architectures, fetching and manipulating longwords is more efficient
3086 than with smaller-sized units such as shortwords. This option can
3087 quadruple the size of the tables used by your scanner.
3093 @item -Ce, --ecs, @code{%option ecs}
3094 directs @code{flex} to construct @dfn{equivalence classes}, i.e., sets
3095 of characters which have identical lexical properties (for example, if
3096 the only appearance of digits in the @code{flex} input is in the
3097 character class ``[0-9]'' then the digits '0', '1', ..., '9' will all be
3098 put in the same equivalence class). Equivalence classes usually give
3099 dramatic reductions in the final table/object file sizes (typically a
3100 factor of 2-5) and are pretty cheap performance-wise (one array look-up
3101 per character scanned).
3105 specifies that the @dfn{full} scanner tables should be generated -
3106 @code{flex} should not compress the tables by taking advantages of
3107 similar transition functions for different states.
3111 specifies that the alternate fast scanner representation (described
3112 above under the @samp{--fast} flag) should be used. This option cannot be
3113 used with @samp{--c++}.
3115 @anchor{option-meta-ecs}
3117 @opindex ---meta-ecs
3119 @item -Cm, --meta-ecs, @code{%option meta-ecs}
3123 @dfn{meta-equivalence classes},
3124 which are sets of equivalence classes (or characters, if equivalence
3125 classes are not being used) that are commonly used together. Meta-equivalence
3126 classes are often a big win when using compressed tables, but they
3127 have a moderate performance impact (one or two @code{if} tests and one
3128 array look-up per character scanned).
3130 @anchor{option-read}
3134 @item -Cr, --read, @code{%option read}
3135 causes the generated scanner to @emph{bypass} use of the standard I/O
3136 library (@code{stdio}) for input. Instead of calling @code{fread()} or
3137 @code{getc()}, the scanner will use the @code{read()} system call,
3138 resulting in a performance gain which varies from system to system, but
3139 in general is probably negligible unless you are also using @samp{-Cf}
3140 or @samp{-CF}. Using @samp{-Cr} can cause strange behavior if, for
3141 example, you read from @file{yyin} using @code{stdio} prior to calling
3142 the scanner (because the scanner will miss whatever text your previous
3143 reads left in the @code{stdio} input buffer). @samp{-Cr} has no effect
3144 if you define @code{YY_INPUT()} (@pxref{Generated Scanner}).
3147 The options @samp{-Cf} or @samp{-CF} and @samp{-Cm} do not make sense
3148 together - there is no opportunity for meta-equivalence classes if the
3149 table is not being compressed. Otherwise the options may be freely
3150 mixed, and are cumulative.
3152 The default setting is @samp{-Cem}, which specifies that @code{flex}
3153 should generate equivalence classes and meta-equivalence classes. This
3154 setting provides the highest degree of table compression. You can trade
3155 off faster-executing scanners at the cost of larger tables with the
3156 following generally being true:
3172 Note that scanners with the smallest tables are usually generated and
3173 compiled the quickest, so during development you will usually want to
3174 use the default, maximal compression.
3176 @samp{-Cfe} is often a good compromise between speed and size for
3177 production scanners.
3179 @anchor{option-full}
3183 @item -f, --full, @code{%option full}
3186 No table compression is done and @code{stdio} is bypassed.
3187 The result is large but fast. This option is equivalent to
3191 @anchor{option-fast}
3195 @item -F, --fast, @code{%option fast}
3196 specifies that the @emph{fast} scanner table representation should be
3197 used (and @code{stdio} bypassed). This representation is about as fast
3198 as the full table representation @samp{--full}, and for some sets of
3199 patterns will be considerably smaller (and for others, larger). In
3200 general, if the pattern set contains both @emph{keywords} and a
3201 catch-all, @emph{identifier} rule, such as in the set:
3205 "case" return TOK_CASE;
3206 "switch" return TOK_SWITCH;
3208 "default" return TOK_DEFAULT;
3209 [a-z]+ return TOK_ID;
3213 then you're better off using the full table representation. If only
3214 the @emph{identifier} rule is present and you then use a hash table or some such
3215 to detect the keywords, you're better off using
3218 This option is equivalent to @samp{-CFr}. It cannot be used
3223 @node Debugging Options, Miscellaneous Options, Options for Scanner Speed and Size, Scanner Options
3224 @section Debugging Options
3228 @anchor{option-backup}
3232 @item -b, --backup, @code{%option backup}
3233 Generate backing-up information to @file{lex.backup}. This is a list of
3234 scanner states which require backing up and the input characters on
3235 which they do so. By adding rules one can remove backing-up states. If
3236 @emph{all} backing-up states are eliminated and @samp{-Cf} or @code{-CF}
3237 is used, the generated scanner will run faster (see the @samp{--perf-report} flag).
3238 Only users who wish to squeeze every last cycle out of their scanners
3239 need worry about this option. (@pxref{Performance}).
3243 @anchor{option-debug}
3247 @item -d, --debug, @code{%option debug}
3248 makes the generated scanner run in @dfn{debug} mode. Whenever a pattern
3249 is recognized and the global variable @code{yy_flex_debug} is non-zero
3250 (which is the default), the scanner will write to @file{stderr} a line
3255 -accepting rule at line 53 ("the matched text")
3259 The line number refers to the location of the rule in the file defining
3260 the scanner (i.e., the file that was fed to flex). Messages are also
3261 generated when the scanner backs up, accepts the default rule, reaches
3262 the end of its input buffer (or encounters a NUL; at this point, the two
3263 look the same as far as the scanner's concerned), or reaches an
3268 @anchor{option-perf-report}
3270 @opindex ---perf-report
3271 @opindex perf-report
3272 @item -p, --perf-report, @code{%option perf-report}
3273 generates a performance report to @file{stderr}. The report consists of
3274 comments regarding features of the @code{flex} input file which will
3275 cause a serious loss of performance in the resulting scanner. If you
3276 give the flag twice, you will also get comments regarding features that
3277 lead to minor performance losses.
3279 Note that the use of @code{REJECT}, and
3280 variable trailing context (@pxref{Limitations}) entails a substantial
3281 performance penalty; use of @code{yymore()}, the @samp{^} operator, and
3282 the @samp{--interactive} flag entail minor performance penalties.
3286 @anchor{option-nodefault}
3288 @opindex ---nodefault
3290 @item -s, --nodefault, @code{%option nodefault}
3291 causes the @emph{default rule} (that unmatched scanner input is echoed
3292 to @file{stdout)} to be suppressed. If the scanner encounters input
3293 that does not match any of its rules, it aborts with an error. This
3294 option is useful for finding holes in a scanner's rule set.
3298 @anchor{option-trace}
3302 @item -T, --trace, @code{%option trace}
3303 makes @code{flex} run in @dfn{trace} mode. It will generate a lot of
3304 messages to @file{stderr} concerning the form of the input and the
3305 resultant non-deterministic and deterministic finite automata. This
3306 option is mostly for use in maintaining @code{flex}.
3310 @anchor{option-nowarn}
3314 @item -w, --nowarn, @code{%option nowarn}
3315 suppresses warning messages.
3319 @anchor{option-verbose}
3323 @item -v, --verbose, @code{%option verbose}
3324 specifies that @code{flex} should write to @file{stderr} a summary of
3325 statistics regarding the scanner it generates. Most of the statistics
3326 are meaningless to the casual @code{flex} user, but the first line
3327 identifies the version of @code{flex} (same as reported by @samp{--version}),
3328 and the next line the flags used when generating the scanner, including
3329 those that are on by default.
3333 @anchor{option-warn}
3336 @item --warn, @code{%option warn}
3337 warn about certain things. In particular, if the default rule can be
3338 matched but no default rule has been given, the flex will warn you.
3339 We recommend using this option always.
3343 @node Miscellaneous Options, , Debugging Options, Scanner Options
3344 @section Miscellaneous Options
3349 A do-nothing option included for POSIX compliance.
3353 @item -h, -?, --help
3354 generates a ``help'' summary of @code{flex}'s options to @file{stdout}
3359 Another do-nothing option included for
3365 prints the version number to @file{stdout} and exits.
3370 @node Performance, Cxx, Scanner Options, Top
3371 @chapter Performance Considerations
3373 @cindex performance, considerations
3374 The main design goal of @code{flex} is that it generate high-performance
3375 scanners. It has been optimized for dealing well with large sets of
3376 rules. Aside from the effects on scanner speed of the table compression
3377 @samp{-C} options outlined above, there are a number of options/actions
3378 which degrade performance. These are, from most expensive to least:
3380 @cindex REJECT, performance costs
3381 @cindex yylineno, performance costs
3382 @cindex trailing context, performance costs
3386 arbitrary trailing context
3388 pattern sets that require backing up
3393 %option always-interactive
3395 ^ beginning-of-line operator
3400 with the first two all being quite expensive and the last two being
3401 quite cheap. Note also that @code{unput()} is implemented as a routine
3402 call that potentially does quite a bit of work, while @code{yyless()} is
3403 a quite-cheap macro. So if you are just putting back some excess text
3404 you scanned, use @code{yyless()}.
3406 @code{REJECT} should be avoided at all costs when performance is
3407 important. It is a particularly expensive option.
3409 There is one case when @code{%option yylineno} can be expensive. That is when
3410 your patterns match long tokens that could @emph{possibly} contain a newline
3411 character. There is no performance penalty for rules that can not possibly
3412 match newlines, since flex does not need to check them for newlines. In
3413 general, you should avoid rules such as @code{[^f]+}, which match very long
3414 tokens, including newlines, and may possibly match your entire file! A better
3415 approach is to separate @code{[^f]+} into two rules:
3426 The above scanner does not incur a performance penalty.
3428 @cindex patterns, tuning for performance
3429 @cindex performance, backing up
3430 @cindex backing up, example of eliminating
3431 Getting rid of backing up is messy and often may be an enormous amount
3432 of work for a complicated scanner. In principal, one begins by using
3433 the @samp{-b} flag to generate a @file{lex.backup} file. For example,
3436 @cindex backing up, eliminating
3440 foo return TOK_KEYWORD;
3441 foobar return TOK_KEYWORD;
3445 the file looks like:
3449 State #6 is non-accepting -
3450 associated rule line numbers:
3452 out-transitions: [ o ]
3453 jam-transitions: EOF [ \001-n p-\177 ]
3455 State #8 is non-accepting -
3456 associated rule line numbers:
3458 out-transitions: [ a ]
3459 jam-transitions: EOF [ \001-` b-\177 ]
3461 State #9 is non-accepting -
3462 associated rule line numbers:
3464 out-transitions: [ r ]
3465 jam-transitions: EOF [ \001-q s-\177 ]
3467 Compressed tables always back up.
3471 The first few lines tell us that there's a scanner state in which it can
3472 make a transition on an 'o' but not on any other character, and that in
3473 that state the currently scanned text does not match any rule. The
3474 state occurs when trying to match the rules found at lines 2 and 3 in
3475 the input file. If the scanner is in that state and then reads
3476 something other than an 'o', it will have to back up to find a rule
3477 which is matched. With a bit of headscratching one can see that this
3478 must be the state it's in when it has seen @samp{fo}. When this has
3479 happened, if anything other than another @samp{o} is seen, the scanner
3480 will have to back up to simply match the @samp{f} (by the default rule).
3482 The comment regarding State #8 indicates there's a problem when
3483 @samp{foob} has been scanned. Indeed, on any character other than an
3484 @samp{a}, the scanner will have to back up to accept "foo". Similarly,
3485 the comment for State #9 concerns when @samp{fooba} has been scanned and
3486 an @samp{r} does not follow.
3488 The final comment reminds us that there's no point going to all the
3489 trouble of removing backing up from the rules unless we're using
3490 @samp{-Cf} or @samp{-CF}, since there's no performance gain doing so
3491 with compressed scanners.
3493 @cindex error rules, to eliminate backing up
3494 The way to remove the backing up is to add ``error'' rules:
3496 @cindex backing up, eliminating by adding error rules
3500 foo return TOK_KEYWORD;
3501 foobar return TOK_KEYWORD;
3506 /* false alarm, not really a keyword */
3512 Eliminating backing up among a list of keywords can also be done using a
3515 @cindex backing up, eliminating with catch-all rule
3519 foo return TOK_KEYWORD;
3520 foobar return TOK_KEYWORD;
3522 [a-z]+ return TOK_ID;
3526 This is usually the best solution when appropriate.
3528 Backing up messages tend to cascade. With a complicated set of rules
3529 it's not uncommon to get hundreds of messages. If one can decipher
3530 them, though, it often only takes a dozen or so rules to eliminate the
3531 backing up (though it's easy to make a mistake and have an error rule
3532 accidentally match a valid token. A possible future @code{flex} feature
3533 will be to automatically add rules to eliminate backing up).
3535 It's important to keep in mind that you gain the benefits of eliminating
3536 backing up only if you eliminate @emph{every} instance of backing up.
3537 Leaving just one means you gain nothing.
3539 @emph{Variable} trailing context (where both the leading and trailing
3540 parts do not have a fixed length) entails almost the same performance
3541 loss as @code{REJECT} (i.e., substantial). So when possible a rule
3544 @cindex trailing context, variable length
3548 mouse|rat/(cat|dog) run();
3557 mouse/cat|dog run();
3567 mouse|rat/cat run();
3568 mouse|rat/dog run();
3572 Note that here the special '|' action does @emph{not} provide any
3573 savings, and can even make things worse (@pxref{Limitations}).
3575 Another area where the user can increase a scanner's performance (and
3576 one that's easier to implement) arises from the fact that the longer the
3577 tokens matched, the faster the scanner will run. This is because with
3578 long tokens the processing of most input characters takes place in the
3579 (short) inner scanning loop, and does not often have to go through the
3580 additional work of setting up the scanning environment (e.g.,
3581 @code{yytext}) for the action. Recall the scanner for C comments:
3583 @cindex performance optimization, matching longer tokens
3590 "/*" BEGIN(comment);
3593 <comment>"*"+[^*/\n]*
3594 <comment>\n ++line_num;
3595 <comment>"*"+"/" BEGIN(INITIAL);
3599 This could be sped up by writing it as:
3607 "/*" BEGIN(comment);
3610 <comment>[^*\n]*\n ++line_num;
3611 <comment>"*"+[^*/\n]*
3612 <comment>"*"+[^*/\n]*\n ++line_num;
3613 <comment>"*"+"/" BEGIN(INITIAL);
3617 Now instead of each newline requiring the processing of another action,
3618 recognizing the newlines is distributed over the other rules to keep the
3619 matched text as long as possible. Note that @emph{adding} rules does
3620 @emph{not} slow down the scanner! The speed of the scanner is
3621 independent of the number of rules or (modulo the considerations given
3622 at the beginning of this section) how complicated the rules are with
3623 regard to operators such as @samp{*} and @samp{|}.
3625 @cindex keywords, for performance
3626 @cindex performance, using keywords
3627 A final example in speeding up a scanner: suppose you want to scan
3628 through a file containing identifiers and keywords, one per line
3629 and with no other extraneous characters, and recognize all the
3630 keywords. A natural first approach is:
3632 @cindex performance optimization, recognizing keywords
3641 while /* it's a keyword */
3643 .|\n /* it's not a keyword */
3647 To eliminate the back-tracking, introduce a catch-all rule:
3657 while /* it's a keyword */
3660 .|\n /* it's not a keyword */
3664 Now, if it's guaranteed that there's exactly one word per line, then we
3665 can reduce the total number of matches by a half by merging in the
3666 recognition of newlines with that of the other tokens:
3676 while\n /* it's a keyword */
3679 .|\n /* it's not a keyword */
3683 One has to be careful here, as we have now reintroduced backing up
3684 into the scanner. In particular, while
3686 know that there will never be any characters in the input stream
3687 other than letters or newlines,
3689 can't figure this out, and it will plan for possibly needing to back up
3690 when it has scanned a token like @samp{auto} and then the next character
3691 is something other than a newline or a letter. Previously it would
3692 then just match the @samp{auto} rule and be done, but now it has no @samp{auto}
3693 rule, only a @samp{auto\n} rule. To eliminate the possibility of backing up,
3694 we could either duplicate all rules but without final newlines, or,
3695 since we never expect to encounter such an input and therefore don't
3696 how it's classified, we can introduce one more catch-all rule, this
3697 one which doesn't include a newline:
3707 while\n /* it's a keyword */
3711 .|\n /* it's not a keyword */
3715 Compiled with @samp{-Cf}, this is about as fast as one can get a
3716 @code{flex} scanner to go for this particular problem.
3718 A final note: @code{flex} is slow when matching @code{NUL}s,
3719 particularly when a token contains multiple @code{NUL}s. It's best to
3720 write rules which match @emph{short} amounts of text if it's anticipated
3721 that the text will often include @code{NUL}s.
3723 Another final note regarding performance: as mentioned in
3724 @ref{Matching}, dynamically resizing @code{yytext} to accommodate huge
3725 tokens is a slow process because it presently requires that the (huge)
3726 token be rescanned from the beginning. Thus if performance is vital,
3727 you should attempt to match ``large'' quantities of text but not
3728 ``huge'' quantities, where the cutoff between the two is at about 8K
3729 characters per token.
3731 @node Cxx, Reentrant, Performance, Top
3732 @chapter Generating C++ Scanners
3734 @cindex c++, experimental form of scanner class
3735 @cindex experimental form of c++ scanner class
3736 @strong{IMPORTANT}: the present form of the scanning class is @emph{experimental}
3737 and may change considerably between major releases.
3740 @cindex member functions, C++
3741 @cindex methods, c++
3742 @code{flex} provides two different ways to generate scanners for use
3743 with C++. The first way is to simply compile a scanner generated by
3744 @code{flex} using a C++ compiler instead of a C compiler. You should
3745 not encounter any compilation errors (@pxref{Reporting Bugs}). You can
3746 then use C++ code in your rule actions instead of C code. Note that the
3747 default input source for your scanner remains @file{yyin}, and default
3748 echoing is still done to @file{yyout}. Both of these remain @code{FILE
3749 *} variables and not C++ @emph{streams}.
3751 You can also use @code{flex} to generate a C++ scanner class, using the
3752 @samp{-+} option (or, equivalently, @code{%option c++)}, which is
3753 automatically specified if the name of the @code{flex} executable ends
3754 in a '+', such as @code{flex++}. When using this option, @code{flex}
3755 defaults to generating the scanner to the file @file{lex.yy.cc} instead
3756 of @file{lex.yy.c}. The generated scanner includes the header file
3757 @file{FlexLexer.h}, which defines the interface to two C++ classes.
3759 The first class in @file{FlexLexer.h}, @code{FlexLexer},
3760 provides an abstract base class defining the general scanner class
3761 interface. It provides the following member functions:
3764 @findex YYText (C++ only)
3765 @item const char* YYText()
3766 returns the text of the most recently matched token, the equivalent of
3769 @findex YYLeng (C++ only)
3771 returns the length of the most recently matched token, the equivalent of
3774 @findex lineno (C++ only)
3775 @item int lineno() const
3776 returns the current input line number (see @code{%option yylineno)}, or
3777 @code{1} if @code{%option yylineno} was not used.
3779 @findex set_debug (C++ only)
3780 @item void set_debug( int flag )
3781 sets the debugging flag for the scanner, equivalent to assigning to
3782 @code{yy_flex_debug} (@pxref{Scanner Options}). Note that you must build
3783 the scanner using @code{%option debug} to include debugging information
3786 @findex debug (C++ only)
3787 @item int debug() const
3788 returns the current setting of the debugging flag.
3791 Also provided are member functions equivalent to
3792 @code{yy_switch_to_buffer()}, @code{yy_create_buffer()} (though the
3793 first argument is an @code{istream&} object reference and not a
3794 @code{FILE*)}, @code{yy_flush_buffer()}, @code{yy_delete_buffer()}, and
3795 @code{yyrestart()} (again, the first argument is a @code{istream&}
3798 @tindex yyFlexLexer (C++ only)
3799 @tindex FlexLexer (C++ only)
3800 The second class defined in @file{FlexLexer.h} is @code{yyFlexLexer},
3801 which is derived from @code{FlexLexer}. It defines the following
3802 additional member functions:
3805 @findex yyFlexLexer constructor (C++ only)
3806 @item yyFlexLexer( istream* arg_yyin = 0, ostream* arg_yyout = 0 )
3807 @item yyFlexLexer( istream& arg_yyin, ostream& arg_yyout )
3808 constructs a @code{yyFlexLexer} object using the given streams for input
3809 and output. If not specified, the streams default to @code{cin} and
3810 @code{cout}, respectively. @code{yyFlexLexer} does not take ownership of
3811 its stream arguments. It's up to the user to ensure the streams pointed
3812 to remain alive at least as long as the @code{yyFlexLexer} instance.
3814 @findex yylex (C++ version)
3815 @item virtual int yylex()
3816 performs the same role is @code{yylex()} does for ordinary @code{flex}
3817 scanners: it scans the input stream, consuming tokens, until a rule's
3818 action returns a value. If you derive a subclass @code{S} from
3819 @code{yyFlexLexer} and want to access the member functions and variables
3820 of @code{S} inside @code{yylex()}, then you need to use @code{%option
3821 yyclass="S"} to inform @code{flex} that you will be using that subclass
3822 instead of @code{yyFlexLexer}. In this case, rather than generating
3823 @code{yyFlexLexer::yylex()}, @code{flex} generates @code{S::yylex()}
3824 (and also generates a dummy @code{yyFlexLexer::yylex()} that calls
3825 @code{yyFlexLexer::LexerError()} if called).
3827 @findex switch_streams (C++ only)
3828 @item virtual void switch_streams(istream* new_in = 0, ostream* new_out = 0)
3829 @item virtual void switch_streams(istream& new_in, ostream& new_out)
3830 reassigns @code{yyin} to @code{new_in} (if non-null) and @code{yyout} to
3831 @code{new_out} (if non-null), deleting the previous input buffer if
3832 @code{yyin} is reassigned.
3834 @item int yylex( istream* new_in, ostream* new_out = 0 )
3835 @item int yylex( istream& new_in, ostream& new_out )
3836 first switches the input streams via @code{switch_streams( new_in,
3837 new_out )} and then returns the value of @code{yylex()}.
3840 In addition, @code{yyFlexLexer} defines the following protected virtual
3841 functions which you can redefine in derived classes to tailor the
3845 @findex LexerInput (C++ only)
3846 @item virtual int LexerInput( char* buf, int max_size )
3847 reads up to @code{max_size} characters into @code{buf} and returns the
3848 number of characters read. To indicate end-of-input, return 0
3849 characters. Note that @code{interactive} scanners (see the @samp{-B}
3850 and @samp{-I} flags in @ref{Scanner Options}) define the macro
3851 @code{YY_INTERACTIVE}. If you redefine @code{LexerInput()} and need to
3852 take different actions depending on whether or not the scanner might be
3853 scanning an interactive input source, you can test for the presence of
3854 this name via @code{#ifdef} statements.
3856 @findex LexerOutput (C++ only)
3857 @item virtual void LexerOutput( const char* buf, int size )
3858 writes out @code{size} characters from the buffer @code{buf}, which, while
3859 @code{NUL}-terminated, may also contain internal @code{NUL}s if the
3860 scanner's rules can match text with @code{NUL}s in them.
3862 @cindex error reporting, in C++
3863 @findex LexerError (C++ only)
3864 @item virtual void LexerError( const char* msg )
3865 reports a fatal error message. The default version of this function
3866 writes the message to the stream @code{cerr} and exits.
3869 Note that a @code{yyFlexLexer} object contains its @emph{entire}
3870 scanning state. Thus you can use such objects to create reentrant
3871 scanners, but see also @ref{Reentrant}. You can instantiate multiple
3872 instances of the same @code{yyFlexLexer} class, and you can also combine
3873 multiple C++ scanner classes together in the same program using the
3874 @samp{-P} option discussed above.
3876 Finally, note that the @code{%array} feature is not available to C++
3877 scanner classes; you must use @code{%pointer} (the default).
3879 Here is an example of a simple C++ scanner:
3881 @cindex C++ scanners, use of
3884 // An example of using the flex C++ scanner class.
3888 using namespace std;
3892 %option noyywrap c++
3900 name ({alpha}|{dig}|\$)({alpha}|{dig}|[_.\-/$])*
3901 num1 [-+]?{dig}+\.?([eE][-+]?{dig}+)?
3902 num2 [-+]?{dig}*\.{dig}+([eE][-+]?{dig}+)?
3903 number {num1}|{num2}
3907 {ws} /* skip blanks and tabs */
3912 while((c = yyinput()) != 0)
3919 if((c = yyinput()) == '/')
3927 {number} cout << "number " << YYText() << '\n';
3931 {name} cout << "name " << YYText() << '\n';
3933 {string} cout << "string " << YYText() << '\n';
3937 // This include is required if main() is an another source file.
3938 //#include <FlexLexer.h>
3940 int main( int /* argc */, char** /* argv */ )
3942 FlexLexer* lexer = new yyFlexLexer;
3943 while(lexer->yylex() != 0)
3950 @cindex C++, multiple different scanners
3951 If you want to create multiple (different) lexer classes, you use the
3952 @samp{-P} flag (or the @code{prefix=} option) to rename each
3953 @code{yyFlexLexer} to some other @samp{xxFlexLexer}. You then can
3954 include @file{<FlexLexer.h>} in your other sources once per lexer class,
3955 first renaming @code{yyFlexLexer} as follows:
3957 @cindex include files, with C++
3958 @cindex header files, with C++
3959 @cindex C++ scanners, including multiple scanners
3963 #define yyFlexLexer xxFlexLexer
3964 #include <FlexLexer.h>
3967 #define yyFlexLexer zzFlexLexer
3968 #include <FlexLexer.h>
3972 if, for example, you used @code{%option prefix="xx"} for one of your
3973 scanners and @code{%option prefix="zz"} for the other.
3975 @node Reentrant, Lex and Posix, Cxx, Top
3976 @chapter Reentrant C Scanners
3978 @cindex reentrant, explanation
3979 @code{flex} has the ability to generate a reentrant C scanner. This is
3980 accomplished by specifying @code{%option reentrant} (@samp{-R}) The generated
3981 scanner is both portable, and safe to use in one or more separate threads of
3982 control. The most common use for reentrant scanners is from within
3983 multi-threaded applications. Any thread may create and execute a reentrant
3984 @code{flex} scanner without the need for synchronization with other threads.
3988 * Reentrant Overview::
3989 * Reentrant Example::
3990 * Reentrant Detail::
3991 * Reentrant Functions::
3994 @node Reentrant Uses, Reentrant Overview, Reentrant, Reentrant
3995 @section Uses for Reentrant Scanners
3997 However, there are other uses for a reentrant scanner. For example, you
3998 could scan two or more files simultaneously to implement a @code{diff} at
3999 the token level (i.e., instead of at the character level):
4001 @cindex reentrant scanners, multiple interleaved scanners
4004 /* Example of maintaining more than one active scanner. */
4009 tok1 = yylex( scanner_1 );
4010 tok2 = yylex( scanner_2 );
4013 printf("Files are different.");
4015 } while ( tok1 && tok2 );
4019 Another use for a reentrant scanner is recursion.
4020 (Note that a recursive scanner can also be created using a non-reentrant scanner and
4021 buffer states. @xref{Multiple Input Buffers}.)
4023 The following crude scanner supports the @samp{eval} command by invoking
4024 another instance of itself.
4026 @cindex reentrant scanners, recursive invocation
4029 /* Example of recursive invocation. */
4036 YY_BUFFER_STATE buf;
4038 yylex_init( &scanner );
4039 yytext[yyleng-1] = ' ';
4041 buf = yy_scan_string( yytext + 5, scanner );
4044 yy_delete_buffer(buf,scanner);
4045 yylex_destroy( scanner );
4052 @node Reentrant Overview, Reentrant Example, Reentrant Uses, Reentrant
4053 @section An Overview of the Reentrant API
4055 @cindex reentrant, API explanation
4056 The API for reentrant scanners is different than for non-reentrant
4057 scanners. Here is a quick overview of the API:
4060 @code{%option reentrant} must be specified.
4063 All functions take one additional argument: @code{yyscanner}
4066 All global variables are replaced by their macro equivalents.
4067 (We tell you this because it may be important to you during debugging.)
4070 @code{yylex_init} and @code{yylex_destroy} must be called before and
4071 after @code{yylex}, respectively.
4074 Accessor methods (get/set functions) provide access to common
4075 @code{flex} variables.
4078 User-specific data can be stored in @code{yyextra}.
4081 @node Reentrant Example, Reentrant Detail, Reentrant Overview, Reentrant
4082 @section Reentrant Example
4084 First, an example of a reentrant scanner:
4085 @cindex reentrant, example of
4088 /* This scanner prints "//" comments. */
4090 %option reentrant stack noyywrap
4095 "//" yy_push_state( COMMENT, yyscanner);
4098 <COMMENT>\n yy_pop_state( yyscanner );
4099 <COMMENT>[^\n]+ fprintf( yyout, "%s\n", yytext);
4103 int main ( int argc, char * argv[] )
4107 yylex_init ( &scanner );
4109 yylex_destroy ( scanner );
4115 @node Reentrant Detail, Reentrant Functions, Reentrant Example, Reentrant
4116 @section The Reentrant API in Detail
4118 Here are the things you need to do or know to use the reentrant C API of
4122 * Specify Reentrant::
4123 * Extra Reentrant Argument::
4124 * Global Replacement::
4125 * Init and Destroy Functions::
4126 * Accessor Methods::
4131 @node Specify Reentrant, Extra Reentrant Argument, Reentrant Detail, Reentrant Detail
4132 @subsection Declaring a Scanner As Reentrant
4134 %option reentrant (--reentrant) must be specified.
4136 Notice that @code{%option reentrant} is specified in the above example
4137 (@pxref{Reentrant Example}. Had this option not been specified,
4138 @code{flex} would have happily generated a non-reentrant scanner without
4139 complaining. You may explicitly specify @code{%option noreentrant}, if
4140 you do @emph{not} want a reentrant scanner, although it is not
4141 necessary. The default is to generate a non-reentrant scanner.
4143 @node Extra Reentrant Argument, Global Replacement, Specify Reentrant, Reentrant Detail
4144 @subsection The Extra Argument
4146 @cindex reentrant, calling functions
4147 @vindex yyscanner (reentrant only)
4148 All functions take one additional argument: @code{yyscanner}.
4150 Notice that the calls to @code{yy_push_state} and @code{yy_pop_state}
4151 both have an argument, @code{yyscanner} , that is not present in a
4152 non-reentrant scanner. Here are the declarations of
4153 @code{yy_push_state} and @code{yy_pop_state} in the reentrant scanner:
4157 static void yy_push_state ( int new_state , yyscan_t yyscanner ) ;
4158 static void yy_pop_state ( yyscan_t yyscanner ) ;
4162 Notice that the argument @code{yyscanner} appears in the declaration of
4163 both functions. In fact, all @code{flex} functions in a reentrant
4164 scanner have this additional argument. It is always the last argument
4165 in the argument list, it is always of type @code{yyscan_t} (which is
4166 typedef'd to @code{void *}) and it is
4167 always named @code{yyscanner}. As you may have guessed,
4168 @code{yyscanner} is a pointer to an opaque data structure encapsulating
4169 the current state of the scanner. For a list of function declarations,
4170 see @ref{Reentrant Functions}. Note that preprocessor macros, such as
4171 @code{BEGIN}, @code{ECHO}, and @code{REJECT}, do not take this
4172 additional argument.
4174 @node Global Replacement, Init and Destroy Functions, Extra Reentrant Argument, Reentrant Detail
4175 @subsection Global Variables Replaced By Macros
4177 @cindex reentrant, accessing flex variables
4178 All global variables in traditional flex have been replaced by macro equivalents.
4180 Note that in the above example, @code{yyout} and @code{yytext} are
4181 not plain variables. These are macros that will expand to their equivalent lvalue.
4182 All of the familiar @code{flex} globals have been replaced by their macro
4183 equivalents. In particular, @code{yytext}, @code{yyleng}, @code{yylineno},
4184 @code{yyin}, @code{yyout}, @code{yyextra}, @code{yylval}, and @code{yylloc}
4185 are macros. You may safely use these macros in actions as if they were plain
4186 variables. We only tell you this so you don't expect to link to these variables
4187 externally. Currently, each macro expands to a member of an internal struct, e.g.,
4191 #define yytext (((struct yyguts_t*)yyscanner)->yytext_r)
4195 One important thing to remember about
4199 is not a global variable in a reentrant
4200 scanner, you can not access it directly from outside an action or from
4201 other functions. You must use an accessor method, e.g.,
4203 to accomplish this. (See below).
4205 @node Init and Destroy Functions, Accessor Methods, Global Replacement, Reentrant Detail
4206 @subsection Init and Destroy Functions
4208 @cindex memory, considerations for reentrant scanners
4209 @cindex reentrant, initialization
4211 @findex yylex_destroy
4213 @code{yylex_init} and @code{yylex_destroy} must be called before and
4214 after @code{yylex}, respectively.
4218 int yylex_init ( yyscan_t * ptr_yy_globals ) ;
4219 int yylex_init_extra ( YY_EXTRA_TYPE user_defined, yyscan_t * ptr_yy_globals ) ;
4220 int yylex ( yyscan_t yyscanner ) ;
4221 int yylex_destroy ( yyscan_t yyscanner ) ;
4225 The function @code{yylex_init} must be called before calling any other
4226 function. The argument to @code{yylex_init} is the address of an
4227 uninitialized pointer to be filled in by @code{yylex_init}, overwriting
4228 any previous contents. The function @code{yylex_init_extra} may be used
4229 instead, taking as its first argument a variable of type @code{YY_EXTRA_TYPE}.
4230 See the section on yyextra, below, for more details.
4232 The value stored in @code{ptr_yy_globals} should
4233 thereafter be passed to @code{yylex} and @code{yylex_destroy}. Flex
4234 does not save the argument passed to @code{yylex_init}, so it is safe to
4235 pass the address of a local pointer to @code{yylex_init} so long as it remains
4236 in scope for the duration of all calls to the scanner, up to and including
4237 the call to @code{yylex_destroy}.
4240 @code{yylex} should be familiar to you by now. The reentrant version
4241 takes one argument, which is the value returned (via an argument) by
4242 @code{yylex_init}. Otherwise, it behaves the same as the non-reentrant
4243 version of @code{yylex}.
4245 Both @code{yylex_init} and @code{yylex_init_extra} returns 0 (zero) on success,
4246 or non-zero on failure, in which case errno is set to one of the following values:
4250 Memory allocation error. @xref{memory-management}.
4256 The function @code{yylex_destroy} should be
4257 called to free resources used by the scanner. After @code{yylex_destroy}
4258 is called, the contents of @code{yyscanner} should not be used. Of
4259 course, there is no need to destroy a scanner if you plan to reuse it.
4260 A @code{flex} scanner (both reentrant and non-reentrant) may be
4261 restarted by calling @code{yyrestart}.
4263 Below is an example of a program that creates a scanner, uses it, then destroys
4273 yylex_init(&scanner);
4275 while ((tok=yylex(scanner)) > 0)
4276 printf("tok=%d yytext=%s\n", tok, yyget_text(scanner));
4278 yylex_destroy(scanner);
4284 @node Accessor Methods, Extra Data, Init and Destroy Functions, Reentrant Detail
4285 @subsection Accessing Variables with Reentrant Scanners
4287 @cindex reentrant, accessor functions
4288 Accessor methods (get/set functions) provide access to common
4289 @code{flex} variables.
4291 Many scanners that you build will be part of a larger project. Portions
4292 of your project will need access to @code{flex} values, such as
4293 @code{yytext}. In a non-reentrant scanner, these values are global, so
4294 there is no problem accessing them. However, in a reentrant scanner, there are no
4295 global @code{flex} values. You can not access them directly. Instead,
4296 you must access @code{flex} values using accessor methods (get/set
4297 functions). Each accessor method is named @code{yyget_NAME} or
4298 @code{yyset_NAME}, where @code{NAME} is the name of the @code{flex}
4299 variable you want. For example:
4301 @cindex accessor functions, use of
4304 /* Set the last character of yytext to NULL. */
4305 void chop ( yyscan_t scanner )
4307 int len = yyget_leng( scanner );
4308 yyget_text( scanner )[len - 1] = '\0';
4313 The above code may be called from within an action like this:
4318 .+\n { chop( yyscanner );}
4322 You may find that @code{%option header-file} is particularly useful for generating
4323 prototypes of all the accessor functions. @xref{option-header}.
4325 @node Extra Data, About yyscan_t, Accessor Methods, Reentrant Detail
4326 @subsection Extra Data
4328 @cindex reentrant, extra data
4330 User-specific data can be stored in @code{yyextra}.
4332 In a reentrant scanner, it is unwise to use global variables to
4333 communicate with or maintain state between different pieces of your program.
4334 However, you may need access to external data or invoke external functions
4335 from within the scanner actions.
4336 Likewise, you may need to pass information to your scanner
4337 (e.g., open file descriptors, or database connections).
4338 In a non-reentrant scanner, the only way to do this would be through the
4339 use of global variables.
4340 @code{Flex} allows you to store arbitrary, ``extra'' data in a scanner.
4341 This data is accessible through the accessor methods
4342 @code{yyget_extra} and @code{yyset_extra}
4343 from outside the scanner, and through the shortcut macro
4345 from within the scanner itself. They are defined as follows:
4347 @tindex YY_EXTRA_TYPE (reentrant only)
4352 #define YY_EXTRA_TYPE void*
4353 YY_EXTRA_TYPE yyget_extra ( yyscan_t scanner );
4354 void yyset_extra ( YY_EXTRA_TYPE arbitrary_data , yyscan_t scanner);
4358 In addition, an extra form of @code{yylex_init} is provided,
4359 @code{yylex_init_extra}. This function is provided so that the yyextra value can
4360 be accessed from within the very first yyalloc, used to allocate
4363 By default, @code{YY_EXTRA_TYPE} is defined as type @code{void *}. You
4364 may redefine this type using @code{%option extra-type="your_type"} in
4367 @cindex YY_EXTRA_TYPE, defining your own type
4370 /* An example of overriding YY_EXTRA_TYPE. */
4372 #include <sys/stat.h>
4376 %option extra-type="struct stat *"
4379 __filesize__ printf( "%ld", yyextra->st_size );
4380 __lastmod__ printf( "%ld", yyextra->st_mtime );
4382 void scan_file( char* filename )
4388 in = fopen( filename, "r" );
4389 stat( filename, &buf );
4391 yylex_init_extra( buf, &scanner );
4392 yyset_in( in, scanner );
4394 yylex_destroy( scanner );
4402 @node About yyscan_t, , Extra Data, Reentrant Detail
4403 @subsection About yyscan_t
4405 @tindex yyscan_t (reentrant only)
4406 @code{yyscan_t} is defined as:
4410 typedef void* yyscan_t;
4414 It is initialized by @code{yylex_init()} to point to
4415 an internal structure. You should never access this value
4416 directly. In particular, you should never attempt to free it
4417 (use @code{yylex_destroy()} instead.)
4419 @node Reentrant Functions, , Reentrant Detail, Reentrant
4420 @section Functions and Macros Available in Reentrant C Scanners
4422 The following Functions are available in a reentrant scanner:
4428 @findex yyget_lineno
4431 @findex yyset_lineno
4439 char *yyget_text ( yyscan_t scanner );
4440 int yyget_leng ( yyscan_t scanner );
4441 FILE *yyget_in ( yyscan_t scanner );
4442 FILE *yyget_out ( yyscan_t scanner );
4443 int yyget_lineno ( yyscan_t scanner );
4444 YY_EXTRA_TYPE yyget_extra ( yyscan_t scanner );
4445 int yyget_debug ( yyscan_t scanner );
4447 void yyset_debug ( int flag, yyscan_t scanner );
4448 void yyset_in ( FILE * in_str , yyscan_t scanner );
4449 void yyset_out ( FILE * out_str , yyscan_t scanner );
4450 void yyset_lineno ( int line_number , yyscan_t scanner );
4451 void yyset_extra ( YY_EXTRA_TYPE user_defined , yyscan_t scanner );
4455 There are no ``set'' functions for yytext and yyleng. This is intentional.
4457 The following Macro shortcuts are available in actions in a reentrant
4472 @cindex yylineno, in a reentrant scanner
4473 In a reentrant C scanner, support for yylineno is always present
4474 (i.e., you may access yylineno), but the value is never modified by
4475 @code{flex} unless @code{%option yylineno} is enabled. This is to allow
4476 the user to maintain the line count independently of @code{flex}.
4478 @anchor{bison-functions}
4479 The following functions and macros are made available when @code{%option
4480 bison-bridge} (@samp{--bison-bridge}) is specified:
4484 YYSTYPE * yyget_lval ( yyscan_t scanner );
4485 void yyset_lval ( YYSTYPE * yylvalp , yyscan_t scanner );
4490 The following functions and macros are made available
4491 when @code{%option bison-locations} (@samp{--bison-locations}) is specified:
4495 YYLTYPE *yyget_lloc ( yyscan_t scanner );
4496 void yyset_lloc ( YYLTYPE * yyllocp , yyscan_t scanner );
4501 Support for yylval assumes that @code{YYSTYPE} is a valid type. Support for
4502 yylloc assumes that @code{YYSLYPE} is a valid type. Typically, these types are
4503 generated by @code{bison}, and are included in section 1 of the @code{flex}
4506 @node Lex and Posix, Memory Management, Reentrant, Top
4507 @chapter Incompatibilities with Lex and Posix
4509 @cindex POSIX and lex
4510 @cindex lex (traditional) and POSIX
4512 @code{flex} is a rewrite of the AT&T Unix @emph{lex} tool (the two
4513 implementations do not share any code, though), with some extensions and
4514 incompatibilities, both of which are of concern to those who wish to
4515 write scanners acceptable to both implementations. @code{flex} is fully
4516 compliant with the POSIX @code{lex} specification, except that when
4517 using @code{%pointer} (the default), a call to @code{unput()} destroys
4518 the contents of @code{yytext}, which is counter to the POSIX
4519 specification. In this section we discuss all of the known areas of
4520 incompatibility between @code{flex}, AT&T @code{lex}, and the POSIX
4521 specification. @code{flex}'s @samp{-l} option turns on maximum
4522 compatibility with the original AT&T @code{lex} implementation, at the
4523 cost of a major loss in the generated scanner's performance. We note
4524 below which incompatibilities can be overcome using the @samp{-l}
4525 option. @code{flex} is fully compatible with @code{lex} with the
4526 following exceptions:
4530 The undocumented @code{lex} scanner internal variable @code{yylineno} is
4531 not supported unless @samp{-l} or @code{%option yylineno} is used.
4534 @code{yylineno} should be maintained on a per-buffer basis, rather than
4535 a per-scanner (single global variable) basis.
4538 @code{yylineno} is not part of the POSIX specification.
4541 The @code{input()} routine is not redefinable, though it may be called
4542 to read characters following whatever has been matched by a rule. If
4543 @code{input()} encounters an end-of-file the normal @code{yywrap()}
4544 processing is done. A ``real'' end-of-file is returned by
4545 @code{input()} as @code{EOF}.
4548 Input is instead controlled by defining the @code{YY_INPUT()} macro.
4551 The @code{flex} restriction that @code{input()} cannot be redefined is
4552 in accordance with the POSIX specification, which simply does not
4553 specify any way of controlling the scanner's input other than by making
4554 an initial assignment to @file{yyin}.
4557 The @code{unput()} routine is not redefinable. This restriction is in
4558 accordance with POSIX.
4561 @code{flex} scanners are not as reentrant as @code{lex} scanners. In
4562 particular, if you have an interactive scanner and an interrupt handler
4563 which long-jumps out of the scanner, and the scanner is subsequently
4564 called again, you may get the following message:
4566 @cindex error messages, end of buffer missed
4569 fatal flex scanner internal error--end of buffer missed
4573 To reenter the scanner, first use:
4575 @cindex restarting the scanner
4582 Note that this call will throw away any buffered input; usually this
4583 isn't a problem with an interactive scanner. @xref{Reentrant}, for
4584 @code{flex}'s reentrant API.
4587 Also note that @code{flex} C++ scanner classes
4589 reentrant, so if using C++ is an option for you, you should use
4590 them instead. @xref{Cxx}, and @ref{Reentrant} for details.
4593 @code{output()} is not supported. Output from the @b{ECHO} macro is
4594 done to the file-pointer @code{yyout} (default @file{stdout)}.
4597 @code{output()} is not part of the POSIX specification.
4600 @code{lex} does not support exclusive start conditions (%x), though they
4601 are in the POSIX specification.
4604 When definitions are expanded, @code{flex} encloses them in parentheses.
4605 With @code{lex}, the following:
4607 @cindex name definitions, not POSIX
4612 foo{NAME}? printf( "Found it\n" );
4617 will not match the string @samp{foo} because when the macro is expanded
4618 the rule is equivalent to @samp{foo[A-Z][A-Z0-9]*?} and the precedence
4619 is such that the @samp{?} is associated with @samp{[A-Z0-9]*}. With
4620 @code{flex}, the rule will be expanded to @samp{foo([A-Z][A-Z0-9]*)?}
4621 and so the string @samp{foo} will match.
4624 Note that if the definition begins with @samp{^} or ends with @samp{$}
4625 then it is @emph{not} expanded with parentheses, to allow these
4626 operators to appear in definitions without losing their special
4627 meanings. But the @samp{<s>}, @samp{/}, and @code{<<EOF>>} operators
4628 cannot be used in a @code{flex} definition.
4631 Using @samp{-l} results in the @code{lex} behavior of no parentheses
4632 around the definition.
4635 The POSIX specification is that the definition be enclosed in parentheses.
4638 Some implementations of @code{lex} allow a rule's action to begin on a
4639 separate line, if the rule's pattern has trailing whitespace:
4641 @cindex patterns and actions on different lines
4650 @code{flex} does not support this feature.
4653 The @code{lex} @code{%r} (generate a Ratfor scanner) option is not
4654 supported. It is not part of the POSIX specification.
4657 After a call to @code{unput()}, @emph{yytext} is undefined until the
4658 next token is matched, unless the scanner was built using @code{%array}.
4659 This is not the case with @code{lex} or the POSIX specification. The
4660 @samp{-l} option does away with this incompatibility.
4663 The precedence of the @samp{@{,@}} (numeric range) operator is
4664 different. The AT&T and POSIX specifications of @code{lex}
4665 interpret @samp{abc@{1,3@}} as match one, two,
4666 or three occurrences of @samp{abc}'', whereas @code{flex} interprets it
4667 as ``match @samp{ab} followed by one, two, or three occurrences of
4668 @samp{c}''. The @samp{-l} and @samp{--posix} options do away with this
4672 The precedence of the @samp{^} operator is different. @code{lex}
4673 interprets @samp{^foo|bar} as ``match either 'foo' at the beginning of a
4674 line, or 'bar' anywhere'', whereas @code{flex} interprets it as ``match
4675 either @samp{foo} or @samp{bar} if they come at the beginning of a
4676 line''. The latter is in agreement with the POSIX specification.
4679 The special table-size declarations such as @code{%a} supported by
4680 @code{lex} are not required by @code{flex} scanners.. @code{flex}
4683 The name @code{FLEX_SCANNER} is @code{#define}'d so scanners may be
4684 written for use with either @code{flex} or @code{lex}. Scanners also
4685 include @code{YY_FLEX_MAJOR_VERSION}, @code{YY_FLEX_MINOR_VERSION}
4686 and @code{YY_FLEX_SUBMINOR_VERSION}
4687 indicating which version of @code{flex} generated the scanner. For
4688 example, for the 2.5.22 release, these defines would be 2, 5 and 22
4689 respectively. If the version of @code{flex} being used is a beta
4690 version, then the symbol @code{FLEX_BETA} is defined.
4693 The symbols @samp{[[} and @samp{]]} in the code sections of the input
4694 may conflict with the m4 delimiters. @xref{M4 Dependency}.
4699 @cindex POSIX comp;compliance
4700 @cindex non-POSIX features of flex
4701 The following @code{flex} features are not included in @code{lex} or the
4702 POSIX specification:
4710 start condition scopes
4712 start condition stacks
4714 interactive/non-interactive scanners
4716 yy_scan_string() and friends
4720 yy_set_interactive()
4739 %@{@}'s around actions
4743 multiple actions on a line
4745 almost all of the @code{flex} command-line options
4748 The feature ``multiple actions on a line''
4749 refers to the fact that with @code{flex} you can put multiple actions on
4750 the same line, separated with semi-colons, while with @code{lex}, the
4755 foo handle_foo(); ++num_foos_seen;
4759 is (rather surprisingly) truncated to
4767 @code{flex} does not truncate the action. Actions that are not enclosed
4768 in braces are simply terminated at the end of the line.
4770 @node Memory Management, Serialized Tables, Lex and Posix, Top
4771 @chapter Memory Management
4773 @cindex memory management
4774 @anchor{memory-management}
4775 This chapter describes how flex handles dynamic memory, and how you can
4776 override the default behavior.
4779 * The Default Memory Management::
4780 * Overriding The Default Memory Management::
4781 * A Note About yytext And Memory::
4784 @node The Default Memory Management, Overriding The Default Memory Management, Memory Management, Memory Management
4785 @section The Default Memory Management
4787 Flex allocates dynamic memory during initialization, and once in a while from
4788 within a call to yylex(). Initialization takes place during the first call to
4789 yylex(). Thereafter, flex may reallocate more memory if it needs to enlarge a
4790 buffer. As of version 2.5.9 Flex will clean up all memory when you call @code{yylex_destroy}
4791 @xref{faq-memory-leak}.
4793 Flex allocates dynamic memory for four purposes, listed below @footnote{The
4794 quantities given here are approximate, and may vary due to host architecture,
4795 compiler configuration, or due to future enhancements to flex.}
4799 @item 16kB for the input buffer.
4800 Flex allocates memory for the character buffer used to perform pattern
4801 matching. Flex must read ahead from the input stream and store it in a large
4802 character buffer. This buffer is typically the largest chunk of dynamic memory
4803 flex consumes. This buffer will grow if necessary, doubling the size each time.
4804 Flex frees this memory when you call yylex_destroy(). The default size of this
4805 buffer (16384 bytes) is almost always too large. The ideal size for this
4806 buffer is the length of the longest token expected, in bytes, plus a little more. Flex will allocate a few
4807 extra bytes for housekeeping. Currently, to override the size of the input buffer
4808 you must @code{#define YY_BUF_SIZE} to whatever number of bytes you want. We don't plan
4809 to change this in the near future, but we reserve the right to do so if we ever add a more robust memory management
4812 @item 64kb for the REJECT state. This will only be allocated if you use REJECT.
4813 The size is large enough to hold the same number of states as characters in the input buffer. If you override the size of the
4814 input buffer (via @code{YY_BUF_SIZE}), then you automatically override the size of this buffer as well.
4816 @item 100 bytes for the start condition stack.
4817 Flex allocates memory for the start condition stack. This is the stack used
4818 for pushing start states, i.e., with yy_push_state(). It will grow if
4819 necessary. Since the states are simply integers, this stack doesn't consume
4820 much memory. This stack is not present if @code{%option stack} is not
4821 specified. You will rarely need to tune this buffer. The ideal size for this
4822 stack is the maximum depth expected. The memory for this stack is
4823 automatically destroyed when you call yylex_destroy(). @xref{option-stack}.
4825 @item 40 bytes for each YY_BUFFER_STATE.
4826 Flex allocates memory for each YY_BUFFER_STATE. The buffer state itself
4827 is about 40 bytes, plus an additional large character buffer (described above.)
4828 The initial buffer state is created during initialization, and with each call
4829 to yy_create_buffer(). You can't tune the size of this, but you can tune the
4830 character buffer as described above. Any buffer state that you explicitly
4831 create by calling yy_create_buffer() is @emph{NOT} destroyed automatically. You
4832 must call yy_delete_buffer() to free the memory. The exception to this rule is
4833 that flex will delete the current buffer automatically when you call
4834 yylex_destroy(). If you delete the current buffer, be sure to set it to NULL.
4835 That way, flex will not try to delete the buffer a second time (possibly
4836 crashing your program!) At the time of this writing, flex does not provide a
4837 growable stack for the buffer states. You have to manage that yourself.
4838 @xref{Multiple Input Buffers}.
4840 @item 84 bytes for the reentrant scanner guts
4841 Flex allocates about 84 bytes for the reentrant scanner structure when
4842 you call yylex_init(). It is destroyed when the user calls yylex_destroy().
4847 @node Overriding The Default Memory Management, A Note About yytext And Memory, The Default Memory Management, Memory Management
4848 @section Overriding The Default Memory Management
4850 @cindex yyalloc, overriding
4851 @cindex yyrealloc, overriding
4852 @cindex yyfree, overriding
4854 Flex calls the functions @code{yyalloc}, @code{yyrealloc}, and @code{yyfree}
4855 when it needs to allocate or free memory. By default, these functions are
4856 wrappers around the standard C functions, @code{malloc}, @code{realloc}, and
4857 @code{free}, respectively. You can override the default implementations by telling
4858 flex that you will provide your own implementations.
4860 To override the default implementations, you must do two things:
4864 @item Suppress the default implementations by specifying one or more of the
4869 @item @code{%option noyyalloc}
4870 @item @code{%option noyyrealloc}
4871 @item @code{%option noyyfree}.
4874 @item Provide your own implementation of the following functions: @footnote{It
4875 is not necessary to override all (or any) of the memory management routines.
4876 You may, for example, override @code{yyrealloc}, but not @code{yyfree} or
4881 // For a non-reentrant scanner
4882 void * yyalloc (size_t bytes);
4883 void * yyrealloc (void * ptr, size_t bytes);
4884 void yyfree (void * ptr);
4886 // For a reentrant scanner
4887 void * yyalloc (size_t bytes, void * yyscanner);
4888 void * yyrealloc (void * ptr, size_t bytes, void * yyscanner);
4889 void yyfree (void * ptr, void * yyscanner);
4895 In the following example, we will override all three memory routines. We assume
4896 that there is a custom allocator with garbage collection. In order to make this
4897 example interesting, we will use a reentrant scanner, passing a pointer to the
4898 custom allocator through @code{yyextra}.
4900 @cindex overriding the memory routines
4904 #include "some_allocator.h"
4907 /* Suppress the default implementations. */
4908 %option noyyalloc noyyrealloc noyyfree
4911 /* Initialize the allocator. */
4913 #define YY_EXTRA_TYPE struct allocator*
4914 #define YY_USER_INIT yyextra = allocator_create();
4921 /* Provide our own implementations. */
4922 void * yyalloc (size_t bytes, void* yyscanner) {
4923 return allocator_alloc (yyextra, bytes);
4926 void * yyrealloc (void * ptr, size_t bytes, void* yyscanner) {
4927 return allocator_realloc (yyextra, bytes);
4930 void yyfree (void * ptr, void * yyscanner) {
4931 /* Do nothing -- we leave it to the garbage collector. */
4938 @node A Note About yytext And Memory, , Overriding The Default Memory Management, Memory Management
4939 @section A Note About yytext And Memory
4941 @cindex yytext, memory considerations
4943 When flex finds a match, @code{yytext} points to the first character of the
4944 match in the input buffer. The string itself is part of the input buffer, and
4945 is @emph{NOT} allocated separately. The value of yytext will be overwritten the next
4946 time yylex() is called. In short, the value of yytext is only valid from within
4947 the matched rule's action.
4949 Often, you want the value of yytext to persist for later processing, i.e., by a
4950 parser with non-zero lookahead. In order to preserve yytext, you will have to
4951 copy it with strdup() or a similar function. But this introduces some headache
4952 because your parser is now responsible for freeing the copy of yytext. If you
4953 use a yacc or bison parser, (commonly used with flex), you will discover that
4954 the error recovery mechanisms can cause memory to be leaked.
4956 To prevent memory leaks from strdup'd yytext, you will have to track the memory
4957 somehow. Our experience has shown that a garbage collection mechanism or a
4958 pooled memory mechanism will save you a lot of grief when writing parsers.
4960 @node Serialized Tables, Diagnostics, Memory Management, Top
4961 @chapter Serialized Tables
4962 @cindex serialization
4963 @cindex memory, serialized tables
4965 @anchor{serialization}
4966 A @code{flex} scanner has the ability to save the DFA tables to a file, and
4967 load them at runtime when needed. The motivation for this feature is to reduce
4968 the runtime memory footprint. Traditionally, these tables have been compiled into
4969 the scanner as C arrays, and are sometimes quite large. Since the tables are
4970 compiled into the scanner, the memory used by the tables can never be freed.
4971 This is a waste of memory, especially if an application uses several scanners,
4972 but none of them at the same time.
4974 The serialization feature allows the tables to be loaded at runtime, before
4975 scanning begins. The tables may be discarded when scanning is finished.
4978 * Creating Serialized Tables::
4979 * Loading and Unloading Serialized Tables::
4980 * Tables File Format::
4983 @node Creating Serialized Tables, Loading and Unloading Serialized Tables, Serialized Tables, Serialized Tables
4984 @section Creating Serialized Tables
4985 @cindex tables, creating serialized
4986 @cindex serialization of tables
4988 You may create a scanner with serialized tables by specifying:
4992 %option tables-file=FILE
4998 These options instruct flex to save the DFA tables to the file @var{FILE}. The tables
4999 will @emph{not} be embedded in the generated scanner. The scanner will not
5000 function on its own. The scanner will be dependent upon the serialized tables. You must
5001 load the tables from this file at runtime before you can scan anything.
5003 If you do not specify a filename to @code{--tables-file}, the tables will be
5004 saved to @file{lex.yy.tables}, where @samp{yy} is the appropriate prefix.
5006 If your project uses several different scanners, you can concatenate the
5007 serialized tables into one file, and flex will find the correct set of tables,
5008 using the scanner prefix as part of the lookup key. An example follows:
5010 @cindex serialized tables, multiple scanners
5013 $ flex --tables-file --prefix=cpp cpp.l
5014 $ flex --tables-file --prefix=c c.l
5015 $ cat lex.cpp.tables lex.c.tables > all.tables
5019 The above example created two scanners, @samp{cpp}, and @samp{c}. Since we did
5020 not specify a filename, the tables were serialized to @file{lex.c.tables} and
5021 @file{lex.cpp.tables}, respectively. Then, we concatenated the two files
5022 together into @file{all.tables}, which we will distribute with our project. At
5023 runtime, we will open the file and tell flex to load the tables from it. Flex
5024 will find the correct tables automatically. (See next section).
5026 @node Loading and Unloading Serialized Tables, Tables File Format, Creating Serialized Tables, Serialized Tables
5027 @section Loading and Unloading Serialized Tables
5028 @cindex tables, loading and unloading
5029 @cindex loading tables at runtime
5030 @cindex tables, freeing
5031 @cindex freeing tables
5032 @cindex memory, serialized tables
5034 If you've built your scanner with @code{%option tables-file}, then you must
5035 load the scanner tables at runtime. This can be accomplished with the following
5038 @deftypefun int yytables_fload (FILE* @var{fp} [, yyscan_t @var{scanner}])
5039 Locates scanner tables in the stream pointed to by @var{fp} and loads them.
5040 Memory for the tables is allocated via @code{yyalloc}. You must call this
5041 function before the first call to @code{yylex}. The argument @var{scanner}
5042 only appears in the reentrant scanner.
5043 This function returns @samp{0} (zero) on success, or non-zero on error.
5046 The loaded tables are @strong{not} automatically destroyed (unloaded) when you
5047 call @code{yylex_destroy}. The reason is that you may create several scanners
5048 of the same type (in a reentrant scanner), each of which needs access to these
5049 tables. To avoid a nasty memory leak, you must call the following function:
5051 @deftypefun int yytables_destroy ([yyscan_t @var{scanner}])
5052 Unloads the scanner tables. The tables must be loaded again before you can scan
5053 any more data. The argument @var{scanner} only appears in the reentrant
5054 scanner. This function returns @samp{0} (zero) on success, or non-zero on
5058 @strong{The functions @code{yytables_fload} and @code{yytables_destroy} are not
5059 thread-safe.} You must ensure that these functions are called exactly once (for
5060 each scanner type) in a threaded program, before any thread calls @code{yylex}.
5061 After the tables are loaded, they are never written to, and no thread
5062 protection is required thereafter -- until you destroy them.
5064 @node Tables File Format, , Loading and Unloading Serialized Tables, Serialized Tables
5065 @section Tables File Format
5066 @cindex tables, file format
5067 @cindex file format, serialized tables
5069 This section defines the file format of serialized @code{flex} tables.
5071 The tables format allows for one or more sets of tables to be
5072 specified, where each set corresponds to a given scanner. Scanners are
5073 indexed by name, as described below. The file format is as follows:
5078 +-------------------------------+
5079 Header | uint32 th_magic; |
5080 | uint32 th_hsize; |
5081 | uint32 th_ssize; |
5082 | uint16 th_flags; |
5083 | char th_version[]; |
5085 | uint8 th_pad64[]; |
5086 +-------------------------------+
5087 Table 1 | uint16 td_id; |
5088 | uint16 td_flags; |
5089 | uint32 td_hilen; |
5090 | uint32 td_lolen; |
5092 | uint8 td_pad64[]; |
5093 +-------------------------------+
5100 +-------------------------------+
5109 The above diagram shows that a complete set of tables consists of a header
5110 followed by multiple individual tables. Furthermore, multiple complete sets may
5111 be present in the same file, each set with its own header and tables. The sets
5112 are contiguous in the file. The only way to know if another set follows is to
5113 check the next four bytes for the magic number (or check for EOF). The header
5114 and tables sections are padded to 64-bit boundaries. Below we describe each
5115 field in detail. This format does not specify how the scanner will expand the
5116 given data, i.e., data may be serialized as int8, but expanded to an int32
5117 array at runtime. This is to reduce the size of the serialized data where
5118 possible. Remember, @emph{all integer values are in network byte order}.
5121 Fields of a table header:
5125 Magic number, always 0xF13C57B1.
5128 Size of this entire header, in bytes, including all fields plus any padding.
5131 Size of this entire set, in bytes, including the header, all tables, plus
5135 Bit flags for this table set. Currently unused.
5138 Flex version in NULL-terminated string format. e.g., @samp{2.5.13a}. This is
5139 the version of flex that was used to create the serialized tables.
5142 Contains the name of this table set. The default is @samp{yytables},
5143 and is prefixed accordingly, e.g., @samp{footables}. Must be NULL-terminated.
5146 Zero or more NULL bytes, padding the entire header to the next 64-bit boundary
5147 as calculated from the beginning of the header.
5155 Specifies the table identifier. Possible values are:
5157 @item YYTD_ID_ACCEPT (0x01)
5159 @item YYTD_ID_BASE (0x02)
5161 @item YYTD_ID_CHK (0x03)
5163 @item YYTD_ID_DEF (0x04)
5165 @item YYTD_ID_EC (0x05)
5167 @item YYTD_ID_META (0x06)
5169 @item YYTD_ID_NUL_TRANS (0x07)
5171 @item YYTD_ID_NXT (0x08)
5172 @code{yy_nxt}. This array may be two dimensional. See the @code{td_hilen}
5174 @item YYTD_ID_RULE_CAN_MATCH_EOL (0x09)
5175 @code{yy_rule_can_match_eol}
5176 @item YYTD_ID_START_STATE_LIST (0x0A)
5177 @code{yy_start_state_list}. This array is handled specially because it is an
5178 array of pointers to structs. See the @code{td_flags} field below.
5179 @item YYTD_ID_TRANSITION (0x0B)
5180 @code{yy_transition}. This array is handled specially because it is an array of
5181 structs. See the @code{td_lolen} field below.
5182 @item YYTD_ID_ACCLIST (0x0C)
5187 Bit flags describing how to interpret the data in @code{td_data}.
5188 The data arrays are one-dimensional by default, but may be
5189 two dimensional as specified in the @code{td_hilen} field.
5192 @item YYTD_DATA8 (0x01)
5193 The data is serialized as an array of type int8.
5194 @item YYTD_DATA16 (0x02)
5195 The data is serialized as an array of type int16.
5196 @item YYTD_DATA32 (0x04)
5197 The data is serialized as an array of type int32.
5198 @item YYTD_PTRANS (0x08)
5199 The data is a list of indexes of entries in the expanded @code{yy_transition}
5200 array. Each index should be expanded to a pointer to the corresponding entry
5201 in the @code{yy_transition} array. We count on the fact that the
5202 @code{yy_transition} array has already been seen.
5203 @item YYTD_STRUCT (0x10)
5204 The data is a list of yy_trans_info structs, each of which consists of
5205 two integers. There is no padding between struct elements or between structs.
5206 The type of each member is determined by the @code{YYTD_DATA*} bits.
5210 If @code{td_hilen} is non-zero, then the data is a two-dimensional array.
5211 Otherwise, the data is a one-dimensional array. @code{td_hilen} contains the
5212 number of elements in the higher dimensional array, and @code{td_lolen} contains
5213 the number of elements in the lowest dimension.
5215 Conceptually, @code{td_data} is either @code{sometype td_data[td_lolen]}, or
5216 @code{sometype td_data[td_hilen][td_lolen]}, where @code{sometype} is specified
5217 by the @code{td_flags} field. It is possible for both @code{td_lolen} and
5218 @code{td_hilen} to be zero, in which case @code{td_data} is a zero length
5219 array, and no data is loaded, i.e., this table is simply skipped. Flex does not
5220 currently generate tables of zero length.
5223 Specifies the number of elements in the lowest dimension array. If this is
5224 a one-dimensional array, then it is simply the number of elements in this array.
5225 The element size is determined by the @code{td_flags} field.
5228 The table data. This array may be a one- or two-dimensional array, of type
5229 @code{int8}, @code{int16}, @code{int32}, @code{struct yy_trans_info}, or
5230 @code{struct yy_trans_info*}, depending upon the values in the
5231 @code{td_flags}, @code{td_hilen}, and @code{td_lolen} fields.
5234 Zero or more NULL bytes, padding the entire table to the next 64-bit boundary as
5235 calculated from the beginning of this table.
5238 @node Diagnostics, Limitations, Serialized Tables, Top
5239 @chapter Diagnostics
5241 @cindex error reporting, diagnostic messages
5242 @cindex warnings, diagnostic messages
5244 The following is a list of @code{flex} diagnostic messages:
5248 @samp{warning, rule cannot be matched} indicates that the given rule
5249 cannot be matched because it follows other rules that will always match
5250 the same text as it. For example, in the following @samp{foo} cannot be
5251 matched because it comes after an identifier ``catch-all'' rule:
5253 @cindex warning, rule cannot be matched
5256 [a-z]+ got_identifier();
5261 Using @code{REJECT} in a scanner suppresses this warning.
5264 @samp{warning, -s option given but default rule can be matched} means
5265 that it is possible (perhaps only in a particular start condition) that
5266 the default rule (match any single character) is the only one that will
5267 match a particular input. Since @samp{-s} was given, presumably this is
5271 @code{reject_used_but_not_detected undefined} or
5272 @code{yymore_used_but_not_detected undefined}. These errors can occur
5273 at compile time. They indicate that the scanner uses @code{REJECT} or
5274 @code{yymore()} but that @code{flex} failed to notice the fact, meaning
5275 that @code{flex} scanned the first two sections looking for occurrences
5276 of these actions and failed to find any, but somehow you snuck some in
5277 (via a #include file, for example). Use @code{%option reject} or
5278 @code{%option yymore} to indicate to @code{flex} that you really do use
5282 @samp{flex scanner jammed}. a scanner compiled with
5283 @samp{-s} has encountered an input string which wasn't matched by any of
5284 its rules. This error can also occur due to internal problems.
5287 @samp{token too large, exceeds YYLMAX}. your scanner uses @code{%array}
5288 and one of its rules matched a string longer than the @code{YYLMAX}
5289 constant (8K bytes by default). You can increase the value by
5290 #define'ing @code{YYLMAX} in the definitions section of your @code{flex}
5294 @samp{scanner requires -8 flag to use the character 'x'}. Your scanner
5295 specification includes recognizing the 8-bit character @samp{'x'} and
5296 you did not specify the -8 flag, and your scanner defaulted to 7-bit
5297 because you used the @samp{-Cf} or @samp{-CF} table compression options.
5298 See the discussion of the @samp{-7} flag, @ref{Scanner Options}, for
5302 @samp{flex scanner push-back overflow}. you used @code{unput()} to push
5303 back so much text that the scanner's buffer could not hold both the
5304 pushed-back text and the current token in @code{yytext}. Ideally the
5305 scanner should dynamically resize the buffer in this case, but at
5306 present it does not.
5309 @samp{input buffer overflow, can't enlarge buffer because scanner uses
5310 REJECT}. the scanner was working on matching an extremely large token
5311 and needed to expand the input buffer. This doesn't work with scanners
5312 that use @code{REJECT}.
5315 @samp{fatal flex scanner internal error--end of buffer missed}. This can
5316 occur in a scanner which is reentered after a long-jump has jumped out
5317 (or over) the scanner's activation frame. Before reentering the
5324 or, as noted above, switch to using the C++ scanner class.
5327 @samp{too many start conditions in <> construct!} you listed more start
5328 conditions in a <> construct than exist (so you must have listed at
5329 least one of them twice).
5332 @node Limitations, Bibliography, Diagnostics, Top
5333 @chapter Limitations
5335 @cindex limitations of flex
5337 Some trailing context patterns cannot be properly matched and generate
5338 warning messages (@samp{dangerous trailing context}). These are
5339 patterns where the ending of the first part of the rule matches the
5340 beginning of the second part, such as @samp{zx*/xy*}, where the 'x*'
5341 matches the 'x' at the beginning of the trailing context. (Note that
5342 the POSIX draft states that the text matched by such patterns is
5343 undefined.) For some trailing context rules, parts which are actually
5344 fixed-length are not recognized as such, leading to the abovementioned
5345 performance loss. In particular, parts using @samp{|} or @samp{@{n@}}
5346 (such as @samp{foo@{3@}}) are always considered variable-length.
5347 Combining trailing context with the special @samp{|} action can result
5348 in @emph{fixed} trailing context being turned into the more expensive
5349 @emph{variable} trailing context. For example, in the following:
5351 @cindex warning, dangerous trailing context
5360 Use of @code{unput()} invalidates yytext and yyleng, unless the
5361 @code{%array} directive or the @samp{-l} option has been used.
5362 Pattern-matching of @code{NUL}s is substantially slower than matching
5363 other characters. Dynamic resizing of the input buffer is slow, as it
5364 entails rescanning all the text matched so far by the current (generally
5365 huge) token. Due to both buffering of input and read-ahead, you cannot
5366 intermix calls to @file{<stdio.h>} routines, such as, @b{getchar()},
5367 with @code{flex} rules and expect it to work. Call @code{input()}
5368 instead. The total table entries listed by the @samp{-v} flag excludes
5369 the number of table entries needed to determine what rule has been
5370 matched. The number of entries is equal to the number of DFA states if
5371 the scanner does not use @code{REJECT}, and somewhat greater than the
5372 number of states if it does. @code{REJECT} cannot be used with the
5373 @samp{-f} or @samp{-F} options.
5375 The @code{flex} internal algorithms need documentation.
5377 @node Bibliography, FAQ, Limitations, Top
5378 @chapter Additional Reading
5380 You may wish to read more about the following programs:
5388 The following books may contain material of interest:
5390 John Levine, Tony Mason, and Doug Brown,
5392 O'Reilly and Associates. Be sure to get the 2nd edition.
5394 M. E. Lesk and E. Schmidt,
5395 @emph{LEX -- Lexical Analyzer Generator}
5397 Alfred Aho, Ravi Sethi and Jeffrey Ullman, @emph{Compilers: Principles,
5398 Techniques and Tools}, Addison-Wesley (1986). Describes the
5399 pattern-matching techniques used by @code{flex} (deterministic finite
5402 @node FAQ, Appendices, Bibliography, Top
5405 From time to time, the @code{flex} maintainer receives certain
5406 questions. Rather than repeat answers to well-understood problems, we
5410 * When was flex born?::
5411 * How do I expand backslash-escape sequences in C-style quoted strings?::
5412 * Why do flex scanners call fileno if it is not ANSI compatible?::
5413 * Does flex support recursive pattern definitions?::
5414 * How do I skip huge chunks of input (tens of megabytes) while using flex?::
5415 * Flex is not matching my patterns in the same order that I defined them.::
5416 * My actions are executing out of order or sometimes not at all.::
5417 * How can I have multiple input sources feed into the same scanner at the same time?::
5418 * Can I build nested parsers that work with the same input file?::
5419 * How can I match text only at the end of a file?::
5420 * How can I make REJECT cascade across start condition boundaries?::
5421 * Why cant I use fast or full tables with interactive mode?::
5422 * How much faster is -F or -f than -C?::
5423 * If I have a simple grammar cant I just parse it with flex?::
5424 * Why doesn't yyrestart() set the start state back to INITIAL?::
5425 * How can I match C-style comments?::
5426 * The period isn't working the way I expected.::
5427 * Can I get the flex manual in another format?::
5428 * Does there exist a "faster" NDFA->DFA algorithm?::
5429 * How does flex compile the DFA so quickly?::
5430 * How can I use more than 8192 rules?::
5431 * How do I abandon a file in the middle of a scan and switch to a new file?::
5432 * How do I execute code only during initialization (only before the first scan)?::
5433 * How do I execute code at termination?::
5434 * Where else can I find help?::
5435 * Can I include comments in the "rules" section of the file?::
5436 * I get an error about undefined yywrap().::
5437 * How can I change the matching pattern at run time?::
5438 * How can I expand macros in the input?::
5439 * How can I build a two-pass scanner?::
5440 * How do I match any string not matched in the preceding rules?::
5441 * I am trying to port code from AT&T lex that uses yysptr and yysbuf.::
5442 * Is there a way to make flex treat NULL like a regular character?::
5443 * Whenever flex can not match the input it says "flex scanner jammed".::
5444 * Why doesn't flex have non-greedy operators like perl does?::
5445 * Memory leak - 16386 bytes allocated by malloc.::
5446 * How do I track the byte offset for lseek()?::
5447 * How do I use my own I/O classes in a C++ scanner?::
5448 * How do I skip as many chars as possible?::
5450 * Are certain equivalent patterns faster than others?::
5451 * Is backing up a big deal?::
5452 * Can I fake multi-byte character support?::
5454 * Can you discuss some flex internals?::
5455 * unput() messes up yy_at_bol::
5456 * The | operator is not doing what I want::
5457 * Why can't flex understand this variable trailing context pattern?::
5458 * The ^ operator isn't working::
5459 * Trailing context is getting confused with trailing optional patterns::
5460 * Is flex GNU or not?::
5462 * I need to scan if-then-else blocks and while loops::
5466 * Is there a repository for flex scanners?::
5467 * How can I conditionally compile or preprocess my flex input file?::
5468 * Where can I find grammars for lex and yacc?::
5469 * I get an end-of-buffer message for each character scanned.::
5509 * What is the difference between YYLEX_PARAM and YY_DECL?::
5510 * Why do I get "conflicting types for yylex" error?::
5511 * How do I access the values set in a Flex action from within a Bison action?::
5514 @node When was flex born?
5515 @unnumberedsec When was flex born?
5517 Vern Paxson took over
5518 the @cite{Software Tools} lex project from Jef Poskanzer in 1982. At that point it
5519 was written in Ratfor. Around 1987 or so, Paxson translated it into C, and
5520 a legend was born :-).
5522 @node How do I expand backslash-escape sequences in C-style quoted strings?
5523 @unnumberedsec How do I expand backslash-escape sequences in C-style quoted strings?
5525 A key point when scanning quoted strings is that you cannot (easily) write
5526 a single rule that will precisely match the string if you allow things
5527 like embedded escape sequences and newlines. If you try to match strings
5528 with a single rule then you'll wind up having to rescan the string anyway
5529 to find any escape sequences.
5531 Instead you can use exclusive start conditions and a set of rules, one for
5532 matching non-escaped text, one for matching a single escape, one for
5533 matching an embedded newline, and one for recognizing the end of the
5534 string. Each of these rules is then faced with the question of where to
5535 put its intermediary results. The best solution is for the rules to
5536 append their local value of @code{yytext} to the end of a ``string literal''
5537 buffer. A rule like the escape-matcher will append to the buffer the
5538 meaning of the escape sequence rather than the literal text in @code{yytext}.
5539 In this way, @code{yytext} does not need to be modified at all.
5541 @node Why do flex scanners call fileno if it is not ANSI compatible?
5542 @unnumberedsec Why do flex scanners call fileno if it is not ANSI compatible?
5544 Flex scanners call @code{fileno()} in order to get the file descriptor
5545 corresponding to @code{yyin}. The file descriptor may be passed to
5546 @code{isatty()} or @code{read()}, depending upon which @code{%options} you specified.
5547 If your system does not have @code{fileno()} support, to get rid of the
5548 @code{read()} call, do not specify @code{%option read}. To get rid of the @code{isatty()}
5549 call, you must specify one of @code{%option always-interactive} or
5550 @code{%option never-interactive}.
5552 @node Does flex support recursive pattern definitions?
5553 @unnumberedsec Does flex support recursive pattern definitions?
5560 block "{"({block}|{statement})*"}"
5564 No. You cannot have recursive definitions. The pattern-matching power of
5565 regular expressions in general (and therefore flex scanners, too) is
5566 limited. In particular, regular expressions cannot ``balance'' parentheses
5567 to an arbitrary degree. For example, it's impossible to write a regular
5568 expression that matches all strings containing the same number of '@{'s
5569 as '@}'s. For more powerful pattern matching, you need a parser, such
5570 as @cite{GNU bison}.
5572 @node How do I skip huge chunks of input (tens of megabytes) while using flex?
5573 @unnumberedsec How do I skip huge chunks of input (tens of megabytes) while using flex?
5575 Use @code{fseek()} (or @code{lseek()}) to position yyin, then call @code{yyrestart()}.
5577 @node Flex is not matching my patterns in the same order that I defined them.
5578 @unnumberedsec Flex is not matching my patterns in the same order that I defined them.
5580 @code{flex} picks the
5581 rule that matches the most text (i.e., the longest possible input string).
5582 This is because @code{flex} uses an entirely different matching technique
5583 (``deterministic finite automata'') that actually does all of the matching
5584 simultaneously, in parallel. (Seems impossible, but it's actually a fairly
5585 simple technique once you understand the principles.)
5587 A side-effect of this parallel matching is that when the input matches more
5588 than one rule, @code{flex} scanners pick the rule that matched the @emph{most} text. This
5589 is explained further in the manual, in the section @xref{Matching}.
5591 If you want @code{flex} to choose a shorter match, then you can work around this
5592 behavior by expanding your short
5593 rule to match more text, then put back the extra:
5597 data_.* yyless( 5 ); BEGIN BLOCKIDSTATE;
5601 Another fix would be to make the second rule active only during the
5602 @code{<BLOCKIDSTATE>} start condition, and make that start condition exclusive
5603 by declaring it with @code{%x} instead of @code{%s}.
5605 A final fix is to change the input language so that the ambiguity for
5606 @samp{data_} is removed, by adding characters to it that don't match the
5607 identifier rule, or by removing characters (such as @samp{_}) from the
5608 identifier rule so it no longer matches @samp{data_}. (Of course, you might
5609 also not have the option of changing the input language.)
5611 @node My actions are executing out of order or sometimes not at all.
5612 @unnumberedsec My actions are executing out of order or sometimes not at all.
5614 Most likely, you have (in error) placed the opening @samp{@{} of the action
5615 block on a different line than the rule, e.g.,
5626 @code{flex} requires that the opening @samp{@{} of an action associated with a rule
5627 begin on the same line as does the rule. You need instead to write your rules
5632 ^(foo|bar) { // CORRECT!
5638 @node How can I have multiple input sources feed into the same scanner at the same time?
5639 @unnumberedsec How can I have multiple input sources feed into the same scanner at the same time?
5644 your scanner is free of backtracking (verified using @code{flex}'s @samp{-b} flag),
5646 AND you run your scanner interactively (@samp{-I} option; default unless using special table
5647 compression options),
5649 AND you feed it one character at a time by redefining @code{YY_INPUT} to do so,
5652 then every time it matches a token, it will have exhausted its input
5653 buffer (because the scanner is free of backtracking). This means you
5654 can safely use @code{select()} at the point and only call @code{yylex()} for another
5655 token if @code{select()} indicates there's data available.
5657 That is, move the @code{select()} out from the input function to a point where
5658 it determines whether @code{yylex()} gets called for the next token.
5660 With this approach, you will still have problems if your input can arrive
5661 piecemeal; @code{select()} could inform you that the beginning of a token is
5662 available, you call @code{yylex()} to get it, but it winds up blocking waiting
5663 for the later characters in the token.
5665 Here's another way: Move your input multiplexing inside of @code{YY_INPUT}. That
5666 is, whenever @code{YY_INPUT} is called, it @code{select()}'s to see where input is
5667 available. If input is available for the scanner, it reads and returns the
5668 next byte. If input is available from another source, it calls whatever
5669 function is responsible for reading from that source. (If no input is
5670 available, it blocks until some input is available.) I've used this technique in an
5671 interpreter I wrote that both reads keyboard input using a @code{flex} scanner and
5672 IPC traffic from sockets, and it works fine.
5674 @node Can I build nested parsers that work with the same input file?
5675 @unnumberedsec Can I build nested parsers that work with the same input file?
5677 This is not going to work without some additional effort. The reason is
5678 that @code{flex} block-buffers the input it reads from @code{yyin}. This means that the
5679 ``outermost'' @code{yylex()}, when called, will automatically slurp up the first 8K
5680 of input available on yyin, and subsequent calls to other @code{yylex()}'s won't
5681 see that input. You might be tempted to work around this problem by
5682 redefining @code{YY_INPUT} to only return a small amount of text, but it turns out
5683 that that approach is quite difficult. Instead, the best solution is to
5684 combine all of your scanners into one large scanner, using a different
5685 exclusive start condition for each.
5687 @node How can I match text only at the end of a file?
5688 @unnumberedsec How can I match text only at the end of a file?
5690 There is no way to write a rule which is ``match this text, but only if
5691 it comes at the end of the file''. You can fake it, though, if you happen
5692 to have a character lying around that you don't allow in your input.
5693 Then you redefine @code{YY_INPUT} to call your own routine which, if it sees
5694 an @samp{EOF}, returns the magic character first (and remembers to return a
5695 real @code{EOF} next time it's called). Then you could write:
5699 <COMMENT>(.|\n)*{EOF_CHAR} /* saw comment at EOF */
5703 @node How can I make REJECT cascade across start condition boundaries?
5704 @unnumberedsec How can I make REJECT cascade across start condition boundaries?
5706 You can do this as follows. Suppose you have a start condition @samp{A}, and
5707 after exhausting all of the possible matches in @samp{<A>}, you want to try
5708 matches in @samp{<INITIAL>}. Then you could use the following:
5714 <A>rule_that_is_long ...; REJECT;
5715 <A>rule ...; REJECT; /* shorter rule */
5719 /* Shortest and last rule in <A>, so
5720 * cascaded REJECTs will eventually
5721 * wind up matching this rule. We want
5722 * to now switch to the initial state
5723 * and try matching from there instead.
5725 yyless(0); /* put back matched text */
5731 @node Why cant I use fast or full tables with interactive mode?
5732 @unnumberedsec Why can't I use fast or full tables with interactive mode?
5734 One of the assumptions
5735 flex makes is that interactive applications are inherently slow (they're
5736 waiting on a human after all).
5737 It has to do with how the scanner detects that it must be finished scanning
5738 a token. For interactive scanners, after scanning each character the current
5739 state is looked up in a table (essentially) to see whether there's a chance
5740 of another input character possibly extending the length of the match. If
5741 not, the scanner halts. For non-interactive scanners, the end-of-token test
5742 is much simpler, basically a compare with 0, so no memory bus cycles. Since
5743 the test occurs in the innermost scanning loop, one would like to make it go
5744 as fast as possible.
5746 Still, it seems reasonable to allow the user to choose to trade off a bit
5747 of performance in this area to gain the corresponding flexibility. There
5748 might be another reason, though, why fast scanners don't support the
5751 @node How much faster is -F or -f than -C?
5752 @unnumberedsec How much faster is -F or -f than -C?
5754 Much faster (factor of 2-3).
5756 @node If I have a simple grammar cant I just parse it with flex?
5757 @unnumberedsec If I have a simple grammar can't I just parse it with flex?
5759 Is your grammar recursive? That's almost always a sign that you're
5760 better off using a parser/scanner rather than just trying to use a scanner
5763 @node Why doesn't yyrestart() set the start state back to INITIAL?
5764 @unnumberedsec Why doesn't yyrestart() set the start state back to INITIAL?
5766 There are two reasons. The first is that there might
5767 be programs that rely on the start state not changing across file changes.
5768 The second is that beginning with @code{flex} version 2.4, use of @code{yyrestart()} is no longer required,
5769 so fixing the problem there doesn't solve the more general problem.
5771 @node How can I match C-style comments?
5772 @unnumberedsec How can I match C-style comments?
5774 You might be tempted to try something like this:
5778 "/*".*"*/" // WRONG!
5786 "/*"(.|\n)"*/" // WRONG!
5790 The above rules will eat too much input, and blow up on things like:
5794 /* a comment */ do_my_thing( "oops */" );
5798 Here is one way which allows you to track line information:
5803 "/*" BEGIN(IN_COMMENT);
5806 "*/" BEGIN(INITIAL);
5807 [^*\n]+ // eat comment in chunks
5808 "*" // eat the lone star
5814 @node The period isn't working the way I expected.
5815 @unnumberedsec The '.' isn't working the way I expected.
5817 Here are some tips for using @samp{.}:
5821 A common mistake is to place the grouping parenthesis AFTER an operator, when
5822 you really meant to place the parenthesis BEFORE the operator, e.g., you
5823 probably want this @code{(foo|bar)+} and NOT this @code{(foo|bar+)}.
5825 The first pattern matches the words @samp{foo} or @samp{bar} any number of
5826 times, e.g., it matches the text @samp{barfoofoobarfoo}. The
5827 second pattern matches a single instance of @code{foo} or a single instance of
5828 @code{bar} followed by one or more @samp{r}s, e.g., it matches the text @code{barrrr} .
5830 A @samp{.} inside @samp{[]}'s just means a literal@samp{.} (period),
5831 and NOT ``any character except newline''.
5833 Remember that @samp{.} matches any character EXCEPT @samp{\n} (and @samp{EOF}).
5834 If you really want to match ANY character, including newlines, then use @code{(.|\n)}
5835 Beware that the regex @code{(.|\n)+} will match your entire input!
5837 Finally, if you want to match a literal @samp{.} (a period), then use @samp{[.]} or @samp{"."}
5840 @node Can I get the flex manual in another format?
5841 @unnumberedsec Can I get the flex manual in another format?
5843 The @code{flex} source distribution includes a texinfo manual. You are
5844 free to convert that texinfo into whatever format you desire. The
5845 @code{texinfo} package includes tools for conversion to a number of formats.
5847 @node Does there exist a "faster" NDFA->DFA algorithm?
5848 @unnumberedsec Does there exist a "faster" NDFA->DFA algorithm?
5850 There's no way around the potential exponential running time - it
5851 can take you exponential time just to enumerate all of the DFA states.
5852 In practice, though, the running time is closer to linear, or sometimes
5855 @node How does flex compile the DFA so quickly?
5856 @unnumberedsec How does flex compile the DFA so quickly?
5858 There are two big speed wins that @code{flex} uses:
5862 It analyzes the input rules to construct equivalence classes for those
5863 characters that always make the same transitions. It then rewrites the NFA
5864 using equivalence classes for transitions instead of characters. This cuts
5865 down the NFA->DFA computation time dramatically, to the point where, for
5866 uncompressed DFA tables, the DFA generation is often I/O bound in writing out
5869 It maintains hash values for previously computed DFA states, so testing
5870 whether a newly constructed DFA state is equivalent to a previously constructed
5871 state can be done very quickly, by first comparing hash values.
5874 @node How can I use more than 8192 rules?
5875 @unnumberedsec How can I use more than 8192 rules?
5877 @code{Flex} is compiled with an upper limit of 8192 rules per scanner.
5878 If you need more than 8192 rules in your scanner, you'll have to recompile @code{flex}
5879 with the following changes in @file{flexdef.h}:
5883 < #define YY_TRAILING_MASK 0x2000
5884 < #define YY_TRAILING_HEAD_MASK 0x4000
5886 > #define YY_TRAILING_MASK 0x20000000
5887 > #define YY_TRAILING_HEAD_MASK 0x40000000
5891 This should work okay as long as your C compiler uses 32 bit integers.
5892 But you might want to think about whether using such a huge number of rules
5893 is the best way to solve your problem.
5895 The following may also be relevant:
5897 With luck, you should be able to increase the definitions in flexdef.h for:
5901 #define JAMSTATE -32766 /* marks a reference to the state that always jams */
5902 #define MAXIMUM_MNS 31999
5903 #define BAD_SUBSCRIPT -32767
5907 recompile everything, and it'll all work. Flex only has these 16-bit-like
5908 values built into it because a long time ago it was developed on a machine
5909 with 16-bit ints. I've given this advice to others in the past but haven't
5910 heard back from them whether it worked okay or not...
5912 @node How do I abandon a file in the middle of a scan and switch to a new file?
5913 @unnumberedsec How do I abandon a file in the middle of a scan and switch to a new file?
5915 Just call @code{yyrestart(newfile)}. Be sure to reset the start state if you want a
5916 ``fresh start, since @code{yyrestart} does NOT reset the start state back to @code{INITIAL}.
5918 @node How do I execute code only during initialization (only before the first scan)?
5919 @unnumberedsec How do I execute code only during initialization (only before the first scan)?
5921 You can specify an initial action by defining the macro @code{YY_USER_INIT} (though
5922 note that @code{yyout} may not be available at the time this macro is executed). Or you
5923 can add to the beginning of your rules section:
5928 /* Must be indented! */
5929 static int did_init = 0;
5938 @node How do I execute code at termination?
5939 @unnumberedsec How do I execute code at termination?
5941 You can specify an action for the @code{<<EOF>>} rule.
5943 @node Where else can I find help?
5944 @unnumberedsec Where else can I find help?
5946 You can find the flex homepage on the web at
5947 @uref{http://flex.sourceforge.net/}. See that page for details about flex
5948 mailing lists as well.
5950 @node Can I include comments in the "rules" section of the file?
5951 @unnumberedsec Can I include comments in the "rules" section of the file?
5953 Yes, just about anywhere you want to. See the manual for the specific syntax.
5955 @node I get an error about undefined yywrap().
5956 @unnumberedsec I get an error about undefined yywrap().
5958 You must supply a @code{yywrap()} function of your own, or link to @file{libfl.a}
5959 (which provides one), or use
5967 in your source to say you don't want a @code{yywrap()} function.
5969 @node How can I change the matching pattern at run time?
5970 @unnumberedsec How can I change the matching pattern at run time?
5972 You can't, it's compiled into a static table when flex builds the scanner.
5974 @node How can I expand macros in the input?
5975 @unnumberedsec How can I expand macros in the input?
5977 The best way to approach this problem is at a higher level, e.g., in the parser.
5979 However, you can do this using multiple input buffers.
5985 /* Saw the macro "macro" followed by extra stuff. */
5986 main_buffer = YY_CURRENT_BUFFER;
5987 expansion_buffer = yy_scan_string(expand(yytext));
5988 yy_switch_to_buffer(expansion_buffer);
5992 if ( expansion_buffer )
5994 // We were doing an expansion, return to where
5996 yy_switch_to_buffer(main_buffer);
5997 yy_delete_buffer(expansion_buffer);
5998 expansion_buffer = 0;
6006 You probably will want a stack of expansion buffers to allow nested macros.
6007 From the above though hopefully the idea is clear.
6009 @node How can I build a two-pass scanner?
6010 @unnumberedsec How can I build a two-pass scanner?
6012 One way to do it is to filter the first pass to a temporary file,
6013 then process the temporary file on the second pass. You will probably see a
6014 performance hit, due to all the disk I/O.
6016 When you need to look ahead far forward like this, it almost always means
6017 that the right solution is to build a parse tree of the entire input, then
6018 walk it after the parse in order to generate the output. In a sense, this
6019 is a two-pass approach, once through the text and once through the parse
6020 tree, but the performance hit for the latter is usually an order of magnitude
6021 smaller, since everything is already classified, in binary format, and
6024 @node How do I match any string not matched in the preceding rules?
6025 @unnumberedsec How do I match any string not matched in the preceding rules?
6027 One way to assign precedence, is to place the more specific rules first. If
6028 two rules would match the same input (same sequence of characters) then the
6029 first rule listed in the @code{flex} input wins, e.g.,
6034 foo[a-zA-Z_]+ return FOO_ID;
6035 bar[a-zA-Z_]+ return BAR_ID;
6036 [a-zA-Z_]+ return GENERIC_ID;
6040 Note that the rule @code{[a-zA-Z_]+} must come *after* the others. It will match the
6041 same amount of text as the more specific rules, and in that case the
6042 @code{flex} scanner will pick the first rule listed in your scanner as the
6045 @node I am trying to port code from AT&T lex that uses yysptr and yysbuf.
6046 @unnumberedsec I am trying to port code from AT&T lex that uses yysptr and yysbuf.
6048 Those are internal variables pointing into the AT&T scanner's input buffer. I
6049 imagine they're being manipulated in user versions of the @code{input()} and @code{unput()}
6050 functions. If so, what you need to do is analyze those functions to figure out
6051 what they're doing, and then replace @code{input()} with an appropriate definition of
6052 @code{YY_INPUT}. You shouldn't need to (and must not) replace
6053 @code{flex}'s @code{unput()} function.
6055 @node Is there a way to make flex treat NULL like a regular character?
6056 @unnumberedsec Is there a way to make flex treat NULL like a regular character?
6058 Yes, @samp{\0} and @samp{\x00} should both do the trick. Perhaps you have an ancient
6059 version of @code{flex}. The latest release is version @value{VERSION}.
6061 @node Whenever flex can not match the input it says "flex scanner jammed".
6062 @unnumberedsec Whenever flex can not match the input it says "flex scanner jammed".
6064 You need to add a rule that matches the otherwise-unmatched text,
6071 [[a bunch of rules here]]
6073 . printf("bad input character '%s' at line %d\n", yytext, yylineno);
6077 See @code{%option default} for more information.
6079 @node Why doesn't flex have non-greedy operators like perl does?
6080 @unnumberedsec Why doesn't flex have non-greedy operators like perl does?
6082 A DFA can do a non-greedy match by stopping
6083 the first time it enters an accepting state, instead of consuming input until
6084 it determines that no further matching is possible (a ``jam'' state). This
6085 is actually easier to implement than longest leftmost match (which flex does).
6087 But it's also much less useful than longest leftmost match. In general,
6088 when you find yourself wishing for non-greedy matching, that's usually a
6089 sign that you're trying to make the scanner do some parsing. That's
6090 generally the wrong approach, since it lacks the power to do a decent job.
6091 Better is to either introduce a separate parser, or to split the scanner
6092 into multiple scanners using (exclusive) start conditions.
6095 a separate start state once you've seen the @samp{BEGIN}. In that state, you
6096 might then have a regex that will match @samp{END} (to kick you out of the
6097 state), and perhaps @samp{(.|\n)} to get a single character within the chunk ...
6099 This approach also has much better error-reporting properties.
6101 @node Memory leak - 16386 bytes allocated by malloc.
6102 @unnumberedsec Memory leak - 16386 bytes allocated by malloc.
6103 @anchor{faq-memory-leak}
6105 UPDATED 2002-07-10: As of @code{flex} version 2.5.9, this leak means that you did not
6106 call @code{yylex_destroy()}. If you are using an earlier version of @code{flex}, then read
6109 The leak is about 16426 bytes. That is, (8192 * 2 + 2) for the read-buffer, and
6110 about 40 for @code{struct yy_buffer_state} (depending upon alignment). The leak is in
6111 the non-reentrant C scanner only (NOT in the reentrant scanner, NOT in the C++
6112 scanner). Since @code{flex} doesn't know when you are done, the buffer is never freed.
6114 However, the leak won't multiply since the buffer is reused no matter how many
6115 times you call @code{yylex()}.
6117 If you want to reclaim the memory when you are completely done scanning, then
6122 /* For non-reentrant C scanner only. */
6123 yy_delete_buffer(YY_CURRENT_BUFFER);
6128 Note: @code{yy_init} is an "internal variable", and hasn't been tested in this
6129 situation. It is possible that some other globals may need resetting as well.
6131 @node How do I track the byte offset for lseek()?
6132 @unnumberedsec How do I track the byte offset for lseek()?
6136 > We thought that it would be possible to have this number through the
6137 > evaluation of the following expression:
6139 > seek_position = (no_buffers)*YY_READ_BUF_SIZE + yy_c_buf_p - YY_CURRENT_BUFFER->yy_ch_buf
6143 While this is the right idea, it has two problems. The first is that
6144 it's possible that @code{flex} will request less than @code{YY_READ_BUF_SIZE} during
6145 an invocation of @code{YY_INPUT} (or that your input source will return less
6146 even though @code{YY_READ_BUF_SIZE} bytes were requested). The second problem
6147 is that when refilling its internal buffer, @code{flex} keeps some characters
6148 from the previous buffer (because usually it's in the middle of a match,
6149 and needs those characters to construct @code{yytext} for the match once it's
6150 done). Because of this, @code{yy_c_buf_p - YY_CURRENT_BUFFER->yy_ch_buf} won't
6151 be exactly the number of characters already read from the current buffer.
6153 An alternative solution is to count the number of characters you've matched
6154 since starting to scan. This can be done by using @code{YY_USER_ACTION}. For
6159 #define YY_USER_ACTION num_chars += yyleng;
6163 (You need to be careful to update your bookkeeping if you use @code{yymore(}),
6164 @code{yyless()}, @code{unput()}, or @code{input()}.)
6166 @node How do I use my own I/O classes in a C++ scanner?
6167 @section How do I use my own I/O classes in a C++ scanner?
6169 When the flex C++ scanning class rewrite finally happens, then this sort of thing should become much easier.
6171 @cindex LexerOutput, overriding
6172 @cindex LexerInput, overriding
6173 @cindex overriding LexerOutput
6174 @cindex overriding LexerInput
6175 @cindex customizing I/O in C++ scanners
6176 @cindex C++ I/O, customizing
6177 You can do this by passing the various functions (such as @code{LexerInput()}
6178 and @code{LexerOutput()}) NULL @code{iostream*}'s, and then
6179 dealing with your own I/O classes surreptitiously (i.e., stashing them in
6180 special member variables). This works because the only assumption about
6181 the lexer regarding what's done with the iostream's is that they're
6182 ultimately passed to @code{LexerInput()} and @code{LexerOutput}, which then do whatever
6183 is necessary with them.
6185 @c faq edit stopped here
6186 @node How do I skip as many chars as possible?
6187 @unnumberedsec How do I skip as many chars as possible?
6189 How do I skip as many chars as possible -- without interfering with the other
6192 In the example below, we want to skip over characters until we see the phrase
6193 "endskip". The following will @emph{NOT} work correctly (do you see why not?)
6197 /* INCORRECT SCANNER */
6200 <INITIAL>startskip BEGIN(SKIP);
6202 <SKIP>"endskip" BEGIN(INITIAL);
6207 The problem is that the pattern .* will eat up the word "endskip."
6208 The simplest (but slow) fix is:
6212 <SKIP>"endskip" BEGIN(INITIAL);
6217 The fix involves making the second rule match more, without
6218 making it match "endskip" plus something else. So for example:
6222 <SKIP>"endskip" BEGIN(INITIAL);
6224 <SKIP>. ;/* so you eat up e's, too */
6228 @c TODO: Evaluate this faq.
6230 @unnumberedsec deleteme00
6236 Vern Paxson took over
6237 the Software Tools lex project from Jef Poskanzer in 1982. At that point it
6238 was written in Ratfor. Around 1987 or so, Paxson translated it into C, and
6239 a legend was born :-).
6243 @c TODO: Evaluate this faq.
6244 @node Are certain equivalent patterns faster than others?
6245 @unnumberedsec Are certain equivalent patterns faster than others?
6248 To: Adoram Rogel <adoram@orna.hybridge.com>
6249 Subject: Re: Flex 2.5.2 performance questions
6250 In-reply-to: Your message of Wed, 18 Sep 96 11:12:17 EDT.
6251 Date: Wed, 18 Sep 96 10:51:02 PDT
6252 From: Vern Paxson <vern>
6254 [Note, the most recent flex release is 2.5.4, which you can get from
6255 ftp.ee.lbl.gov. It has bug fixes over 2.5.2 and 2.5.3.]
6257 > 1. Using the pattern
6258 > ([Ff](oot)?)?[Nn](ote)?(\.)?
6260 > (((F|f)oot(N|n)ote)|((N|n)ote)|((N|n)\.)|((F|f)(N|n)(\.)))
6261 > (in a very complicated flex program) caused the program to slow from
6262 > 300K+/min to 100K/min (no other changes were done).
6264 These two are not equivalent. For example, the first can match "footnote."
6265 but the second can only match "footnote". This is almost certainly the
6266 cause in the discrepancy - the slower scanner run is matching more tokens,
6267 and/or having to do more backing up.
6269 > 2. Which of these two are better: [Ff]oot or (F|f)oot ?
6271 From a performance point of view, they're equivalent (modulo presumably
6272 minor effects such as memory cache hit rates; and the presence of trailing
6273 context, see below). From a space point of view, the first is slightly
6276 > 3. I have a pattern that look like this:
6277 > pats {p1}|{p2}|{p3}|...|{p50} (50 patterns ORd)
6279 > running yet another complicated program that includes the following rule:
6280 > <snext>{and}/{no4}{bb}{pats}
6282 > gets me to "too complicated - over 32,000 states"...
6284 I can't tell from this example whether the trailing context is variable-length
6285 or fixed-length (it could be the latter if {and} is fixed-length). If it's
6286 variable length, which flex -p will tell you, then this reflects a basic
6287 performance problem, and if you can eliminate it by restructuring your
6288 scanner, you will see significant improvement.
6290 > so I divided {pats} to {pats1}, {pats2},..., {pats5} each consists of about
6291 > 10 patterns and changed the rule to be 5 rules.
6292 > This did compile, but what is the rule of thumb here ?
6294 The rule is to avoid trailing context other than fixed-length, in which for
6295 a/b, either the 'a' pattern or the 'b' pattern have a fixed length. Use
6296 of the '|' operator automatically makes the pattern variable length, so in
6297 this case '[Ff]oot' is preferred to '(F|f)oot'.
6299 > 4. I changed a rule that looked like this:
6300 > <snext8>{and}{bb}/{ROMAN}[^A-Za-z] { BEGIN...
6302 > to the next 2 rules:
6303 > <snext8>{and}{bb}/{ROMAN}[A-Za-z] { ECHO;}
6304 > <snext8>{and}{bb}/{ROMAN} { BEGIN...
6306 > Again, I understand the using [^...] will cause a great performance loss
6308 Actually, it doesn't cause any sort of performance loss. It's a surprising
6309 fact about regular expressions that they always match in linear time
6310 regardless of how complex they are.
6312 > but are there any specific rules about it ?
6314 See the "Performance Considerations" section of the man page, and also
6315 the example in MISC/fastwc/.
6321 @c TODO: Evaluate this faq.
6322 @node Is backing up a big deal?
6323 @unnumberedsec Is backing up a big deal?
6326 To: Adoram Rogel <adoram@hybridge.com>
6327 Subject: Re: Flex 2.5.2 performance questions
6328 In-reply-to: Your message of Thu, 19 Sep 96 10:16:04 EDT.
6329 Date: Thu, 19 Sep 96 09:58:00 PDT
6330 From: Vern Paxson <vern>
6332 > a lot about the backing up problem.
6333 > I believe that there lies my biggest problem, and I'll try to improve
6336 Since you have variable trailing context, this is a bigger performance
6337 problem. Fixing it is usually easier than fixing backing up, which in a
6338 complicated scanner (yours seems to fit the bill) can be extremely
6339 difficult to do correctly.
6341 You also don't mention what flags you are using for your scanner.
6342 -f makes a large speed difference, and -Cfe buys you nearly as much
6343 speed but the resulting scanner is considerably smaller.
6345 > I have an | operator in {and} and in {pats} so both of them are variable
6348 -p should have reported this.
6350 > Is changing one of them to fixed-length is enough ?
6354 > Is it possible to change the 32,000 states limit ?
6356 Yes. I've appended instructions on how. Before you make this change,
6357 though, you should think about whether there are ways to fundamentally
6358 simplify your scanner - those are certainly preferable!
6362 To increase the 32K limit (on a machine with 32 bit integers), you increase
6363 the magnitude of the following in flexdef.h:
6365 #define JAMSTATE -32766 /* marks a reference to the state that always jams */
6366 #define MAXIMUM_MNS 31999
6367 #define BAD_SUBSCRIPT -32767
6368 #define MAX_SHORT 32700
6370 Adding a 0 or two after each should do the trick.
6374 @c TODO: Evaluate this faq.
6375 @node Can I fake multi-byte character support?
6376 @unnumberedsec Can I fake multi-byte character support?
6379 To: Heeman_Lee@hp.com
6380 Subject: Re: flex - multi-byte support?
6381 In-reply-to: Your message of Thu, 03 Oct 1996 17:24:04 PDT.
6382 Date: Fri, 04 Oct 1996 11:42:18 PDT
6383 From: Vern Paxson <vern>
6385 > I assume as long as my *.l file defines the
6386 > range of expected character code values (in octal format), flex will
6387 > scan the file and read multi-byte characters correctly. But I have no
6388 > confidence in this assumption.
6390 Your lack of confidence is justified - this won't work.
6392 Flex has in it a widespread assumption that the input is processed
6393 one byte at a time. Fixing this is on the to-do list, but is involved,
6394 so it won't happen any time soon. In the interim, the best I can suggest
6395 (unless you want to try fixing it yourself) is to write your rules in
6396 terms of pairs of bytes, using definitions in the first section:
6401 foo{X}bar found_foo_fe_c2_bar();
6403 etc. Definitely a pain - sorry about that.
6405 By the way, the email address you used for me is ancient, indicating you
6406 have a very old version of flex. You can get the most recent, 2.5.4, from
6413 @c TODO: Evaluate this faq.
6415 @unnumberedsec deleteme01
6418 To: moleary@primus.com
6419 Subject: Re: Flex / Unicode compatibility question
6420 In-reply-to: Your message of Tue, 22 Oct 1996 10:15:42 PDT.
6421 Date: Tue, 22 Oct 1996 11:06:13 PDT
6422 From: Vern Paxson <vern>
6424 Unfortunately flex at the moment has a widespread assumption within it
6425 that characters are processed 8 bits at a time. I don't see any easy
6426 fix for this (other than writing your rules in terms of double characters -
6427 a pain). I also don't know of a wider lex, though you might try surfing
6428 the Plan 9 stuff because I know it's a Unicode system, and also the PCCT
6429 toolkit (try searching say Alta Vista for "Purdue Compiler Construction
6432 Fixing flex to handle wider characters is on the long-term to-do list.
6433 But since flex is a strictly spare-time project these days, this probably
6434 won't happen for quite a while, unless someone else does it first.
6440 @c TODO: Evaluate this faq.
6441 @node Can you discuss some flex internals?
6442 @unnumberedsec Can you discuss some flex internals?
6445 To: Johan Linde <jl@theophys.kth.se>
6446 Subject: Re: translation of flex
6447 In-reply-to: Your message of Sun, 10 Nov 1996 09:16:36 PST.
6448 Date: Mon, 11 Nov 1996 10:33:50 PST
6449 From: Vern Paxson <vern>
6451 > I'm working for the Swedish team translating GNU program, and I'm currently
6452 > working with flex. I have a few questions about some of the messages which
6453 > I hope you can answer.
6455 All of the things you're wondering about, by the way, concerning flex
6456 internals - probably the only person who understands what they mean in
6457 English is me! So I wouldn't worry too much about getting them right.
6461 > msgid " %d protos created\n"
6463 > Does proto mean prototype?
6465 Yes - prototypes of state compression tables.
6468 > msgid " %d/%d (peak %d) template nxt-chk entries created\n"
6470 > Here I'm mainly puzzled by 'nxt-chk'. I guess it means 'next-check'. (?)
6471 > However, 'template next-check entries' doesn't make much sense to me. To be
6472 > able to find a good translation I need to know a little bit more about it.
6474 There is a scheme in the Aho/Sethi/Ullman compiler book for compressing
6475 scanner tables. It involves creating two pairs of tables. The first has
6476 "base" and "default" entries, the second has "next" and "check" entries.
6477 The "base" entry is indexed by the current state and yields an index into
6478 the next/check table. The "default" entry gives what to do if the state
6479 transition isn't found in next/check. The "next" entry gives the next
6480 state to enter, but only if the "check" entry verifies that this entry is
6481 correct for the current state. Flex creates templates of series of
6482 next/check entries and then encodes differences from these templates as a
6483 way to compress the tables.
6486 > msgid " %d/%d base-def entries created\n"
6488 > The same problem here for 'base-def'.
6496 @c TODO: Evaluate this faq.
6497 @node unput() messes up yy_at_bol
6498 @unnumberedsec unput() messes up yy_at_bol
6501 To: Xinying Li <xli@npac.syr.edu>
6503 In-reply-to: Your message of Wed, 13 Nov 1996 17:28:38 PST.
6504 Date: Wed, 13 Nov 1996 19:51:54 PST
6505 From: Vern Paxson <vern>
6507 > "unput()" them to input flow, question occurs. If I do this after I scan
6508 > a carriage, the variable "YY_CURRENT_BUFFER->yy_at_bol" is changed. That
6509 > means the carriage flag has gone.
6511 You can control this by calling yy_set_bol(). It's described in the manual.
6513 > And if in pre-reading it goes to the end of file, is anything done
6514 > to control the end of curren buffer and end of file?
6516 No, there's no way to put back an end-of-file.
6518 > By the way I am using flex 2.5.2 and using the "-l".
6520 The latest release is 2.5.4, by the way. It fixes some bugs in 2.5.2 and
6521 2.5.3. You can get it from ftp.ee.lbl.gov.
6527 @c TODO: Evaluate this faq.
6528 @node The | operator is not doing what I want
6529 @unnumberedsec The | operator is not doing what I want
6532 To: Alain.ISSARD@st.com
6533 Subject: Re: Start condition with FLEX
6534 In-reply-to: Your message of Mon, 18 Nov 1996 09:45:02 PST.
6535 Date: Mon, 18 Nov 1996 10:41:34 PST
6536 From: Vern Paxson <vern>
6538 > I am not able to use the start condition scope and to use the | (OR) with
6539 > rules having start conditions.
6541 The problem is that if you use '|' as a regular expression operator, for
6542 example "a|b" meaning "match either 'a' or 'b'", then it must *not* have
6543 any blanks around it. If you instead want the special '|' *action* (which
6544 from your scanner appears to be the case), which is a way of giving two
6545 different rules the same action:
6548 bar matched_foo_or_bar();
6550 then '|' *must* be separated from the first rule by whitespace and *must*
6551 be followed by a new line. You *cannot* write it as:
6553 foo | bar matched_foo_or_bar();
6555 even though you might think you could because yacc supports this syntax.
6556 The reason for this unfortunately incompatibility is historical, but it's
6557 unlikely to be changed.
6559 Your problems with start condition scope are simply due to syntax errors
6560 from your use of '|' later confusing flex.
6562 Let me know if you still have problems.
6568 @c TODO: Evaluate this faq.
6569 @node Why can't flex understand this variable trailing context pattern?
6570 @unnumberedsec Why can't flex understand this variable trailing context pattern?
6573 To: Gregory Margo <gmargo@newton.vip.best.com>
6574 Subject: Re: flex-2.5.3 bug report
6575 In-reply-to: Your message of Sat, 23 Nov 1996 16:50:09 PST.
6576 Date: Sat, 23 Nov 1996 17:07:32 PST
6577 From: Vern Paxson <vern>
6579 > Enclosed is a lex file that "real" lex will process, but I cannot get
6580 > flex to process it. Could you try it and maybe point me in the right direction?
6582 Your problem is that some of the definitions in the scanner use the '/'
6583 trailing context operator, and have it enclosed in ()'s. Flex does not
6584 allow this operator to be enclosed in ()'s because doing so allows undefined
6585 regular expressions such as "(a/b)+". So the solution is to remove the
6586 parentheses. Note that you must also be building the scanner with the -l
6587 option for AT&T lex compatibility. Without this option, flex automatically
6588 encloses the definitions in parentheses.
6594 @c TODO: Evaluate this faq.
6595 @node The ^ operator isn't working
6596 @unnumberedsec The ^ operator isn't working
6599 To: Thomas Hadig <hadig@toots.physik.rwth-aachen.de>
6600 Subject: Re: Flex Bug ?
6601 In-reply-to: Your message of Tue, 26 Nov 1996 14:35:01 PST.
6602 Date: Tue, 26 Nov 1996 11:15:05 PST
6603 From: Vern Paxson <vern>
6605 > In my lexer code, i have the line :
6608 > Thus all lines starting with an astrix (*) are comment lines.
6609 > This does not work !
6611 I can't get this problem to reproduce - it works fine for me. Note
6612 though that if what you have is slightly different:
6618 then it won't work, because flex pushes back macro definitions enclosed
6619 in ()'s, so the rule becomes
6623 and now that the '^' operator is not at the immediate beginning of the
6624 line, it's interpreted as just a regular character. You can avoid this
6625 behavior by using the "-l" lex-compatibility flag, or "%option lex-compat".
6631 @c TODO: Evaluate this faq.
6632 @node Trailing context is getting confused with trailing optional patterns
6633 @unnumberedsec Trailing context is getting confused with trailing optional patterns
6636 To: Adoram Rogel <adoram@hybridge.com>
6637 Subject: Re: Flex 2.5.4 BOF ???
6638 In-reply-to: Your message of Tue, 26 Nov 1996 16:10:41 PST.
6639 Date: Wed, 27 Nov 1996 10:56:25 PST
6640 From: Vern Paxson <vern>
6642 > Organization(s)?/[a-z]
6644 > This matched "Organizations" (looking in debug mode, the trailing s
6645 > was matched with trailing context instead of the optional (s) in the
6648 That should only happen with lex. Flex can properly match this pattern.
6649 (That might be what you're saying, I'm just not sure.)
6651 > Is there a way to avoid this dangerous trailing context problem ?
6653 Unfortunately, there's no easy way. On the other hand, I don't see why
6654 it should be a problem. Lex's matching is clearly wrong, and I'd hope
6655 that usually the intent remains the same as expressed with the pattern,
6656 so flex's matching will be correct.
6662 @c TODO: Evaluate this faq.
6663 @node Is flex GNU or not?
6664 @unnumberedsec Is flex GNU or not?
6667 To: Cameron MacKinnon <mackin@interlog.com>
6668 Subject: Re: Flex documentation bug
6669 In-reply-to: Your message of Mon, 02 Dec 1996 00:07:08 PST.
6670 Date: Sun, 01 Dec 1996 22:29:39 PST
6671 From: Vern Paxson <vern>
6673 > I'm not sure how or where to submit bug reports (documentation or
6674 > otherwise) for the GNU project stuff ...
6676 Well, strictly speaking flex isn't part of the GNU project. They just
6677 distribute it because no one's written a decent GPL'd lex replacement.
6678 So you should send bugs directly to me. Those sent to the GNU folks
6679 sometimes find there way to me, but some may drop between the cracks.
6681 > In GNU Info, under the section 'Start Conditions', and also in the man
6682 > page (mine's dated April '95) is a nice little snippet showing how to
6683 > parse C quoted strings into a buffer, defined to be MAX_STR_CONST in
6684 > size. Unfortunately, no overflow checking is ever done ...
6686 This is already mentioned in the manual:
6688 Finally, here's an example of how to match C-style quoted
6689 strings using exclusive start conditions, including expanded
6690 escape sequences (but not including checking for a string
6693 The reason for not doing the overflow checking is that it will needlessly
6694 clutter up an example whose main purpose is just to demonstrate how to
6697 The latest release is 2.5.4, by the way, available from ftp.ee.lbl.gov.
6703 @c TODO: Evaluate this faq.
6705 @unnumberedsec ERASEME53
6708 To: tsv@cs.UManitoba.CA
6709 Subject: Re: Flex (reg)..
6710 In-reply-to: Your message of Thu, 06 Mar 1997 23:50:16 PST.
6711 Date: Thu, 06 Mar 1997 15:54:19 PST
6712 From: Vern Paxson <vern>
6714 > [:alpha:] ([:alnum:] | \\_)*
6716 If your rule really has embedded blanks as shown above, then it won't
6717 work, as the first blank delimits the rule from the action. (It wouldn't
6718 even compile ...) You need instead:
6720 [:alpha:]([:alnum:]|\\_)*
6722 and that should work fine - there's no restriction on what can go inside
6723 of ()'s except for the trailing context operator, '/'.
6729 @c TODO: Evaluate this faq.
6730 @node I need to scan if-then-else blocks and while loops
6731 @unnumberedsec I need to scan if-then-else blocks and while loops
6734 To: "Mike Stolnicki" <mstolnic@ford.com>
6735 Subject: Re: FLEX help
6736 In-reply-to: Your message of Fri, 30 May 1997 13:33:27 PDT.
6737 Date: Fri, 30 May 1997 10:46:35 PDT
6738 From: Vern Paxson <vern>
6740 > We'd like to add "if-then-else", "while", and "for" statements to our
6742 > We've investigated many possible solutions. The one solution that seems
6743 > the most reasonable involves knowing the position of a TOKEN in yyin.
6745 I strongly advise you to instead build a parse tree (abstract syntax tree)
6746 and loop over that instead. You'll find this has major benefits in keeping
6747 your interpreter simple and extensible.
6749 That said, the functionality you mention for get_position and set_position
6750 have been on the to-do list for a while. As flex is a purely spare-time
6751 project for me, no guarantees when this will be added (in particular, it
6752 for sure won't be for many months to come).
6758 @c TODO: Evaluate this faq.
6760 @unnumberedsec ERASEME55
6763 To: Colin Paul Adams <colin@colina.demon.co.uk>
6764 Subject: Re: Flex C++ classes and Bison
6765 In-reply-to: Your message of 09 Aug 1997 17:11:41 PDT.
6766 Date: Fri, 15 Aug 1997 10:48:19 PDT
6767 From: Vern Paxson <vern>
6769 > #define YY_DECL int yylex (YYSTYPE *lvalp, struct parser_control
6772 > I have been trying to get this to work as a C++ scanner, but it does
6773 > not appear to be possible (warning that it matches no declarations in
6774 > yyFlexLexer, or something like that).
6776 > Is this supposed to be possible, or is it being worked on (I DID
6777 > notice the comment that scanner classes are still experimental, so I'm
6780 What you need to do is derive a subclass from yyFlexLexer that provides
6781 the above yylex() method, squirrels away lvalp and parm into member
6782 variables, and then invokes yyFlexLexer::yylex() to do the regular scanning.
6788 @c TODO: Evaluate this faq.
6790 @unnumberedsec ERASEME56
6793 To: Mikael.Latvala@lmf.ericsson.se
6794 Subject: Re: Possible mistake in Flex v2.5 document
6795 In-reply-to: Your message of Fri, 05 Sep 1997 16:07:24 PDT.
6796 Date: Fri, 05 Sep 1997 10:01:54 PDT
6797 From: Vern Paxson <vern>
6799 > In that example you show how to count comment lines when using
6800 > C style /* ... */ comments. My question is, shouldn't you take into
6801 > account a scenario where end of a comment marker occurs inside
6802 > character or string literals?
6804 The scanner certainly needs to also scan character and string literals.
6805 However it does that (there's an example in the man page for strings), the
6806 lexer will recognize the beginning of the literal before it runs across the
6807 embedded "/*". Consequently, it will finish scanning the literal before it
6808 even considers the possibility of matching "/*".
6812 '([^']*|{ESCAPE_SEQUENCE})'
6814 will match all the text between the ''s (inclusive). So the lexer
6815 considers this as a token beginning at the first ', and doesn't even
6816 attempt to match other tokens inside it.
6818 I thinnk this subtlety is not worth putting in the manual, as I suspect
6819 it would confuse more people than it would enlighten.
6825 @c TODO: Evaluate this faq.
6827 @unnumberedsec ERASEME57
6830 To: "Marty Leisner" <leisner@sdsp.mc.xerox.com>
6831 Subject: Re: flex limitations
6832 In-reply-to: Your message of Sat, 06 Sep 1997 11:27:21 PDT.
6833 Date: Mon, 08 Sep 1997 11:38:08 PDT
6834 From: Vern Paxson <vern>
6837 > [a-zA-Z]+ /* skip a line */
6838 > { printf("got %s\n", yytext); }
6841 What version of flex are you using? If I feed this to 2.5.4, it complains:
6843 "bug.l", line 5: EOF encountered inside an action
6844 "bug.l", line 5: unrecognized rule
6845 "bug.l", line 5: fatal parse error
6847 Not the world's greatest error message, but it manages to flag the problem.
6849 (With the introduction of start condition scopes, flex can't accommodate
6850 an action on a separate line, since it's ambiguous with an indented rule.)
6852 You can get 2.5.4 from ftp.ee.lbl.gov.
6858 @c TODO: Evaluate this faq.
6859 @node Is there a repository for flex scanners?
6860 @unnumberedsec Is there a repository for flex scanners?
6862 Not that we know of. You might try asking on comp.compilers.
6864 @c TODO: Evaluate this faq.
6865 @node How can I conditionally compile or preprocess my flex input file?
6866 @unnumberedsec How can I conditionally compile or preprocess my flex input file?
6869 Flex doesn't have a preprocessor like C does. You might try using m4, or the C
6870 preprocessor plus a sed script to clean up the result.
6873 @c TODO: Evaluate this faq.
6874 @node Where can I find grammars for lex and yacc?
6875 @unnumberedsec Where can I find grammars for lex and yacc?
6877 In the sources for flex and bison.
6879 @c TODO: Evaluate this faq.
6880 @node I get an end-of-buffer message for each character scanned.
6881 @unnumberedsec I get an end-of-buffer message for each character scanned.
6883 This will happen if your LexerInput() function returns only one character
6884 at a time, which can happen either if you're scanner is "interactive", or
6885 if the streams library on your platform always returns 1 for yyin->gcount().
6887 Solution: override LexerInput() with a version that returns whole buffers.
6889 @c TODO: Evaluate this faq.
6890 @node unnamed-faq-62
6891 @unnumberedsec unnamed-faq-62
6894 To: Georg.Rehm@CL-KI.Uni-Osnabrueck.DE
6895 Subject: Re: Flex maximums
6896 In-reply-to: Your message of Mon, 17 Nov 1997 17:16:06 PST.
6897 Date: Mon, 17 Nov 1997 17:16:15 PST
6898 From: Vern Paxson <vern>
6900 > I took a quick look into the flex-sources and altered some #defines in
6903 > #define INITIAL_MNS 64000
6904 > #define MNS_INCREMENT 1024000
6905 > #define MAXIMUM_MNS 64000
6907 The things to fix are to add a couple of zeroes to:
6909 #define JAMSTATE -32766 /* marks a reference to the state that always jams */
6910 #define MAXIMUM_MNS 31999
6911 #define BAD_SUBSCRIPT -32767
6912 #define MAX_SHORT 32700
6914 and, if you get complaints about too many rules, make the following change too:
6916 #define YY_TRAILING_MASK 0x200000
6917 #define YY_TRAILING_HEAD_MASK 0x400000
6923 @c TODO: Evaluate this faq.
6924 @node unnamed-faq-63
6925 @unnumberedsec unnamed-faq-63
6928 To: jimmey@lexis-nexis.com (Jimmey Todd)
6929 Subject: Re: FLEX question regarding istream vs ifstream
6930 In-reply-to: Your message of Mon, 08 Dec 1997 15:54:15 PST.
6931 Date: Mon, 15 Dec 1997 13:21:35 PST
6932 From: Vern Paxson <vern>
6934 > stdin_handle = YY_CURRENT_BUFFER;
6935 > ifstream fin( "aFile" );
6936 > yy_switch_to_buffer( yy_create_buffer( fin, YY_BUF_SIZE ) );
6938 > What I'm wanting to do, is pass the contents of a file thru one set
6939 > of rules and then pass stdin thru another set... It works great if, I
6940 > don't use the C++ classes. But since everything else that I'm doing is
6941 > in C++, I thought I'd be consistent.
6943 > The problem is that 'yy_create_buffer' is expecting an istream* as it's
6944 > first argument (as stated in the man page). However, fin is a ifstream
6945 > object. Any ideas on what I might be doing wrong? Any help would be
6946 > appreciated. Thanks!!
6948 You need to pass &fin, to turn it into an ifstream* instead of an ifstream.
6949 Then its type will be compatible with the expected istream*, because ifstream
6950 is derived from istream.
6956 @c TODO: Evaluate this faq.
6957 @node unnamed-faq-64
6958 @unnumberedsec unnamed-faq-64
6961 To: Enda Fadian <fadiane@piercom.ie>
6962 Subject: Re: Question related to Flex man page?
6963 In-reply-to: Your message of Tue, 16 Dec 1997 15:17:34 PST.
6964 Date: Tue, 16 Dec 1997 14:17:09 PST
6965 From: Vern Paxson <vern>
6967 > Can you explain to me what is ment by a long-jump in relation to flex?
6969 Using the longjmp() function while inside yylex() or a routine called by it.
6971 > what is the flex activation frame.
6973 Just yylex()'s stack frame.
6975 > As far as I can see yyrestart will bring me back to the sart of the input
6976 > file and using flex++ isnot really an option!
6978 No, yyrestart() doesn't imply a rewind, even though its name might sound
6979 like it does. It tells the scanner to flush its internal buffers and
6980 start reading from the given file at its present location.
6986 @c TODO: Evaluate this faq.
6987 @node unnamed-faq-65
6988 @unnumberedsec unnamed-faq-65
6991 To: hassan@larc.info.uqam.ca (Hassan Alaoui)
6992 Subject: Re: Need urgent Help
6993 In-reply-to: Your message of Sat, 20 Dec 1997 19:38:19 PST.
6994 Date: Sun, 21 Dec 1997 21:30:46 PST
6995 From: Vern Paxson <vern>
6997 > /usr/lib/yaccpar: In function `int yyparse()':
6998 > /usr/lib/yaccpar:184: warning: implicit declaration of function `int yylex(...)'
7000 > ld: Undefined symbol
7005 This is a known problem with Solaris C++ (and/or Solaris yacc). I believe
7006 the fix is to explicitly insert some 'extern "C"' statements for the
7007 corresponding routines/symbols.
7013 @c TODO: Evaluate this faq.
7014 @node unnamed-faq-66
7015 @unnumberedsec unnamed-faq-66
7018 To: mc0307@mclink.it
7019 Cc: gnu@prep.ai.mit.edu
7020 Subject: Re: [mc0307@mclink.it: Help request]
7021 In-reply-to: Your message of Fri, 12 Dec 1997 17:57:29 PST.
7022 Date: Sun, 21 Dec 1997 22:33:37 PST
7023 From: Vern Paxson <vern>
7025 > This is my definition for float and integer types:
7029 > I've tested my program on other lex version (on UNIX Sun Solaris an HP
7030 > UNIX) and it work well, so I think that my definitions are correct.
7031 > There are any differences between Lex and Flex?
7033 There are indeed differences, as discussed in the man page. The one
7034 you are probably running into is that when flex expands a name definition,
7035 it puts parentheses around the expansion, while lex does not. There's
7036 an example in the man page of how this can lead to different matching.
7037 Flex's behavior complies with the POSIX standard (or at least with the
7038 last POSIX draft I saw).
7044 @c TODO: Evaluate this faq.
7045 @node unnamed-faq-67
7046 @unnumberedsec unnamed-faq-67
7049 To: hassan@larc.info.uqam.ca (Hassan Alaoui)
7051 In-reply-to: Your message of Mon, 22 Dec 1997 16:06:35 PST.
7052 Date: Mon, 22 Dec 1997 14:35:05 PST
7053 From: Vern Paxson <vern>
7055 > Thank you very much for your help. I compile and link well with C++ while
7056 > declaring 'yylex ...' extern, But a little problem remains. I get a
7057 > segmentation default when executing ( I linked with lfl library) while it
7058 > works well when using LEX instead of flex. Do you have some ideas about the
7061 The one possible reason for this that comes to mind is if you've defined
7062 yytext as "extern char yytext[]" (which is what lex uses) instead of
7063 "extern char *yytext" (which is what flex uses). If it's not that, then
7064 I'm afraid I don't know what the problem might be.
7070 @c TODO: Evaluate this faq.
7071 @node unnamed-faq-68
7072 @unnumberedsec unnamed-faq-68
7075 To: "Bart Niswonger" <NISWONGR@almaden.ibm.com>
7076 Subject: Re: flex 2.5: c++ scanners & start conditions
7077 In-reply-to: Your message of Tue, 06 Jan 1998 10:34:21 PST.
7078 Date: Tue, 06 Jan 1998 19:19:30 PST
7079 From: Vern Paxson <vern>
7081 > The problem is that when I do this (using %option c++) start
7082 > conditions seem to not apply.
7084 The BEGIN macro modifies the yy_start variable. For C scanners, this
7085 is a static with scope visible through the whole file. For C++ scanners,
7086 it's a member variable, so it only has visible scope within a member
7087 function. Your lexbegin() routine is not a member function when you
7088 build a C++ scanner, so it's not modifying the correct yy_start. The
7089 diagnostic that indicates this is that you found you needed to add
7090 a declaration of yy_start in order to get your scanner to compile when
7091 using C++; instead, the correct fix is to make lexbegin() a member
7092 function (by deriving from yyFlexLexer).
7098 @c TODO: Evaluate this faq.
7099 @node unnamed-faq-69
7100 @unnumberedsec unnamed-faq-69
7103 To: "Boris Zinin" <boris@ippe.rssi.ru>
7104 Subject: Re: current position in flex buffer
7105 In-reply-to: Your message of Mon, 12 Jan 1998 18:58:23 PST.
7106 Date: Mon, 12 Jan 1998 12:03:15 PST
7107 From: Vern Paxson <vern>
7109 > The problem is how to determine the current position in flex active
7110 > buffer when a rule is matched....
7112 You will need to keep track of this explicitly, such as by redefining
7113 YY_USER_ACTION to count the number of characters matched.
7115 The latest flex release, by the way, is 2.5.4, available from ftp.ee.lbl.gov.
7121 @c TODO: Evaluate this faq.
7122 @node unnamed-faq-70
7123 @unnumberedsec unnamed-faq-70
7126 To: Bik.Dhaliwal@bis.org
7127 Subject: Re: Flex question
7128 In-reply-to: Your message of Mon, 26 Jan 1998 13:05:35 PST.
7129 Date: Tue, 27 Jan 1998 22:41:52 PST
7130 From: Vern Paxson <vern>
7132 > That requirement involves knowing
7133 > the character position at which a particular token was matched
7136 The way you have to do this is by explicitly keeping track of where
7137 you are in the file, by counting the number of characters scanned
7138 for each token (available in yyleng). It may prove convenient to
7139 do this by redefining YY_USER_ACTION, as described in the manual.
7145 @c TODO: Evaluate this faq.
7146 @node unnamed-faq-71
7147 @unnumberedsec unnamed-faq-71
7150 To: Vladimir Alexiev <vladimir@cs.ualberta.ca>
7151 Subject: Re: flex: how to control start condition from parser?
7152 In-reply-to: Your message of Mon, 26 Jan 1998 05:50:16 PST.
7153 Date: Tue, 27 Jan 1998 22:45:37 PST
7154 From: Vern Paxson <vern>
7156 > It seems useful for the parser to be able to tell the lexer about such
7157 > context dependencies, because then they don't have to be limited to
7158 > local or sequential context.
7160 One way to do this is to have the parser call a stub routine that's
7161 included in the scanner's .l file, and consequently that has access ot
7162 BEGIN. The only ugliness is that the parser can't pass in the state
7163 it wants, because those aren't visible - but if you don't have many
7164 such states, then using a different set of names doesn't seem like
7165 to much of a burden.
7167 While generating a .h file like you suggests is certainly cleaner,
7168 flex development has come to a virtual stand-still :-(, so a workaround
7169 like the above is much more pragmatic than waiting for a new feature.
7175 @c TODO: Evaluate this faq.
7176 @node unnamed-faq-72
7177 @unnumberedsec unnamed-faq-72
7180 To: Barbara Denny <denny@3com.com>
7181 Subject: Re: freebsd flex bug?
7182 In-reply-to: Your message of Fri, 30 Jan 1998 12:00:43 PST.
7183 Date: Fri, 30 Jan 1998 12:42:32 PST
7184 From: Vern Paxson <vern>
7186 > lex.yy.c:1996: parse error before `='
7188 This is the key, identifying this error. (It may help to pinpoint
7189 it by using flex -L, so it doesn't generate #line directives in its
7190 output.) I will bet you heavy money that you have a start condition
7191 name that is also a variable name, or something like that; flex spits
7192 out #define's for each start condition name, mapping them to a number,
7193 so you can wind up with:
7204 and the penultimate will turn into "int 1 = 3" after C preprocessing,
7205 since flex will put "#define foo 1" in the generated scanner.
7211 @c TODO: Evaluate this faq.
7212 @node unnamed-faq-73
7213 @unnumberedsec unnamed-faq-73
7216 To: Maurice Petrie <mpetrie@infoscigroup.com>
7217 Subject: Re: Lost flex .l file
7218 In-reply-to: Your message of Mon, 02 Feb 1998 14:10:01 PST.
7219 Date: Mon, 02 Feb 1998 11:15:12 PST
7220 From: Vern Paxson <vern>
7222 > I am curious as to
7223 > whether there is a simple way to backtrack from the generated source to
7224 > reproduce the lost list of tokens we are searching on.
7226 In theory, it's straight-forward to go from the DFA representation
7227 back to a regular-expression representation - the two are isomorphic.
7228 In practice, a huge headache, because you have to unpack all the tables
7229 back into a single DFA representation, and then write a program to munch
7230 on that and translate it into an RE.
7232 Sorry for the less-than-happy news ...
7238 @c TODO: Evaluate this faq.
7239 @node unnamed-faq-74
7240 @unnumberedsec unnamed-faq-74
7243 To: jimmey@lexis-nexis.com (Jimmey Todd)
7244 Subject: Re: Flex performance question
7245 In-reply-to: Your message of Thu, 19 Feb 1998 11:01:17 PST.
7246 Date: Thu, 19 Feb 1998 08:48:51 PST
7247 From: Vern Paxson <vern>
7249 > What I have found, is that the smaller the data chunk, the faster the
7250 > program executes. This is the opposite of what I expected. Should this be
7251 > happening this way?
7253 This is exactly what will happen if your input file has embedded NULs.
7256 A final note: flex is slow when matching NUL's, particularly
7257 when a token contains multiple NUL's. It's best to write
7258 rules which match short amounts of text if it's anticipated
7259 that the text will often include NUL's.
7261 So that's the first thing to look for.
7267 @c TODO: Evaluate this faq.
7268 @node unnamed-faq-75
7269 @unnumberedsec unnamed-faq-75
7272 To: jimmey@lexis-nexis.com (Jimmey Todd)
7273 Subject: Re: Flex performance question
7274 In-reply-to: Your message of Thu, 19 Feb 1998 11:01:17 PST.
7275 Date: Thu, 19 Feb 1998 15:42:25 PST
7276 From: Vern Paxson <vern>
7278 So there are several problems.
7280 First, to go fast, you want to match as much text as possible, which
7281 your scanners don't in the case that what they're scanning is *not*
7282 a <RN> tag. So you want a rule like:
7286 Second, C++ scanners are particularly slow if they're interactive,
7287 which they are by default. Using -B speeds it up by a factor of 3-4
7290 Third, C++ scanners that use the istream interface are slow, because
7291 of how poorly implemented istream's are. I built two versions of
7292 the following scanner:
7299 and the C version inhales a 2.5MB file on my workstation in 0.8 seconds.
7300 The C++ istream version, using -B, takes 3.8 seconds.
7306 @c TODO: Evaluate this faq.
7307 @node unnamed-faq-76
7308 @unnumberedsec unnamed-faq-76
7311 To: "Frescatore, David (CRD, TAD)" <frescatore@exc01crdge.crd.ge.com>
7312 Subject: Re: FLEX 2.5 & THE YEAR 2000
7313 In-reply-to: Your message of Wed, 03 Jun 1998 11:26:22 PDT.
7314 Date: Wed, 03 Jun 1998 10:22:26 PDT
7315 From: Vern Paxson <vern>
7317 > I am researching the Y2K problem with General Electric R&D
7318 > and need to know if there are any known issues concerning
7319 > the above mentioned software and Y2K regardless of version.
7321 There shouldn't be, all it ever does with the date is ask the system
7322 for it and then print it out.
7328 @c TODO: Evaluate this faq.
7329 @node unnamed-faq-77
7330 @unnumberedsec unnamed-faq-77
7333 To: "Hans Dermot Doran" <htd@ibhdoran.com>
7334 Subject: Re: flex problem
7335 In-reply-to: Your message of Wed, 15 Jul 1998 21:30:13 PDT.
7336 Date: Tue, 21 Jul 1998 14:23:34 PDT
7337 From: Vern Paxson <vern>
7339 > To overcome this, I gets() the stdin into a string and lex the string. The
7340 > string is lexed OK except that the end of string isn't lexed properly
7341 > (yy_scan_string()), that is the lexer dosn't recognise the end of string.
7343 Flex doesn't contain mechanisms for recognizing buffer endpoints. But if
7344 you use fgets instead (which you should anyway, to protect against buffer
7345 overflows), then the final \n will be preserved in the string, and you can
7346 scan that in order to find the end of the string.
7352 @c TODO: Evaluate this faq.
7353 @node unnamed-faq-78
7354 @unnumberedsec unnamed-faq-78
7357 To: soumen@almaden.ibm.com
7358 Subject: Re: Flex++ 2.5.3 instance member vs. static member
7359 In-reply-to: Your message of Mon, 27 Jul 1998 02:10:04 PDT.
7360 Date: Tue, 28 Jul 1998 01:10:34 PDT
7361 From: Vern Paxson <vern>
7371 > Now you'd expect mylineno to be a member of each instance of class
7372 > yyFlexLexer, but is this the case? A look at the lex.yy.cc file seems to
7373 > indicate otherwise; unless I am missing something the declaration of
7374 > mylineno seems to be outside any class scope.
7376 > How will this work if I want to run a multi-threaded application with each
7377 > thread creating a FlexLexer instance?
7379 Derive your own subclass and make mylineno a member variable of it.
7385 @c TODO: Evaluate this faq.
7386 @node unnamed-faq-79
7387 @unnumberedsec unnamed-faq-79
7390 To: Adoram Rogel <adoram@hybridge.com>
7391 Subject: Re: More than 32K states change hangs
7392 In-reply-to: Your message of Tue, 04 Aug 1998 16:55:39 PDT.
7393 Date: Tue, 04 Aug 1998 22:28:45 PDT
7394 From: Vern Paxson <vern>
7398 > I followed your advice, posted on Usenet bu you, and emailed to me
7399 > personally by you, on how to overcome the 32K states limit. I'm running
7400 > on Linux machines.
7401 > I took the full source of version 2.5.4 and did the following changes in
7403 > #define JAMSTATE -327660
7404 > #define MAXIMUM_MNS 319990
7405 > #define BAD_SUBSCRIPT -327670
7406 > #define MAX_SHORT 327000
7409 > All looked fine, including check and bigcheck, so I installed.
7411 Hmmm, you shouldn't increase MAX_SHORT, though looking through my email
7412 archives I see that I did indeed recommend doing so. Try setting it back
7413 to 32700; that should suffice that you no longer need -Ca. If it still
7414 hangs, then the interesting question is - where?
7416 > Compiling the same hanged program with a out-of-the-box (RedHat 4.2
7417 > distribution of Linux)
7418 > flex 2.5.4 binary works.
7420 Since Linux comes with source code, you should diff it against what
7421 you have to see what problems they missed.
7423 > Should I always compile with the -Ca option now ? even short and simple
7426 No, definitely not. It's meant to be for those situations where you
7427 absolutely must squeeze every last cycle out of your scanner.
7433 @c TODO: Evaluate this faq.
7434 @node unnamed-faq-80
7435 @unnumberedsec unnamed-faq-80
7438 To: "Schmackpfeffer, Craig" <Craig.Schmackpfeffer@usa.xerox.com>
7439 Subject: Re: flex output for static code portion
7440 In-reply-to: Your message of Tue, 11 Aug 1998 11:55:30 PDT.
7441 Date: Mon, 17 Aug 1998 23:57:42 PDT
7442 From: Vern Paxson <vern>
7444 > I would like to use flex under the hood to generate a binary file
7445 > containing the data structures that control the parse.
7447 This has been on the wish-list for a long time. In principle it's
7448 straight-forward - you redirect mkdata() et al's I/O to another file,
7449 and modify the skeleton to have a start-up function that slurps these
7450 into dynamic arrays. The concerns are (1) the scanner generation code
7451 is hairy and full of corner cases, so it's easy to get surprised when
7452 going down this path :-( ; and (2) being careful about buffering so
7453 that when the tables change you make sure the scanner starts in the
7454 correct state and reading at the right point in the input file.
7456 > I was wondering if you know of anyone who has used flex in this way.
7458 I don't - but it seems like a reasonable project to undertake (unlike
7459 numerous other flex tweaks :-).
7465 @c TODO: Evaluate this faq.
7466 @node unnamed-faq-81
7467 @unnumberedsec unnamed-faq-81
7470 Received: from 131.173.17.11 (131.173.17.11 [131.173.17.11])
7471 by ee.lbl.gov (8.9.1/8.9.1) with ESMTP id AAA03838
7472 for <vern@ee.lbl.gov>; Thu, 20 Aug 1998 00:47:57 -0700 (PDT)
7473 Received: from hal.cl-ki.uni-osnabrueck.de (hal.cl-ki.Uni-Osnabrueck.DE [131.173.141.2])
7474 by deimos.rz.uni-osnabrueck.de (8.8.7/8.8.8) with ESMTP id JAA34694
7475 for <vern@ee.lbl.gov>; Thu, 20 Aug 1998 09:47:55 +0200
7476 Received: (from georg@localhost) by hal.cl-ki.uni-osnabrueck.de (8.6.12/8.6.12) id JAA34834 for vern@ee.lbl.gov; Thu, 20 Aug 1998 09:47:54 +0200
7477 From: Georg Rehm <georg@hal.cl-ki.uni-osnabrueck.de>
7478 Message-Id: <199808200747.JAA34834@hal.cl-ki.uni-osnabrueck.de>
7479 Subject: "flex scanner push-back overflow"
7481 Date: Thu, 20 Aug 1998 09:47:54 +0200 (MEST)
7482 Reply-To: Georg.Rehm@CL-KI.Uni-Osnabrueck.DE
7483 X-NoJunk: Do NOT send commercial mail, spam or ads to this address!
7484 X-URL: http://www.cl-ki.uni-osnabrueck.de/~georg/
7485 X-Mailer: ELM [version 2.4ME+ PL28 (25)]
7487 Content-Type: text/plain; charset=US-ASCII
7488 Content-Transfer-Encoding: 7bit
7492 Yesterday, I encountered a strange problem: I use the macro processor m4
7493 to include some lengthy lists into a .l file. Following is a flex macro
7494 definition that causes some serious pain in my neck:
7496 AUTHOR ("A. Boucard / L. Boucard"|"A. Dastarac / M. Levent"|"A.Boucaud / L.Boucaud"|"Abderrahim Lamchichi"|"Achmat Dangor"|"Adeline Toullier"|"Adewale Maja-Pearce"|"Ahmed Ziri"|"Akram Ellyas"|"Alain Bihr"|"Alain Gresh"|"Alain Guillemoles"|"Alain Joxe"|"Alain Morice"|"Alain Renon"|"Alain Zecchini"|"Albert Memmi"|"Alberto Manguel"|"Alex De Waal"|"Alfonso Artico"| [...])
7498 The complete list contains about 10kB. When I try to "flex" this file
7499 (on a Solaris 2.6 machine, using a modified flex 2.5.4 (I only increased
7500 some of the predefined values in flexdefs.h) I get the error:
7502 myflex/flex -8 sentag.tmp.l
7503 flex scanner push-back overflow
7505 When I remove the slashes in the macro definition everything works fine.
7506 As I understand it, the double quotes escape the slash-character so it
7507 really means "/" and not "trailing context". Furthermore, I tried to
7508 escape the slashes with backslashes, but with no use, the same error message
7509 appeared when flexing the code.
7511 Do you have an idea what's going on here?
7513 Greetings from Germany,
7516 Georg Rehm georg@cl-ki.uni-osnabrueck.de
7517 Institute for Semantic Information Processing, University of Osnabrueck, FRG
7521 @c TODO: Evaluate this faq.
7522 @node unnamed-faq-82
7523 @unnumberedsec unnamed-faq-82
7526 To: Georg.Rehm@CL-KI.Uni-Osnabrueck.DE
7527 Subject: Re: "flex scanner push-back overflow"
7528 In-reply-to: Your message of Thu, 20 Aug 1998 09:47:54 PDT.
7529 Date: Thu, 20 Aug 1998 07:05:35 PDT
7530 From: Vern Paxson <vern>
7532 > myflex/flex -8 sentag.tmp.l
7533 > flex scanner push-back overflow
7535 Flex itself uses a flex scanner. That scanner is running out of buffer
7536 space when it tries to unput() the humongous macro you've defined. When
7537 you remove the '/'s, you make it small enough so that it fits in the buffer;
7538 removing spaces would do the same thing.
7540 The fix is to either rethink how come you're using such a big macro and
7541 perhaps there's another/better way to do it; or to rebuild flex's own
7542 scan.c with a larger value for
7544 #define YY_BUF_SIZE 16384
7550 @c TODO: Evaluate this faq.
7551 @node unnamed-faq-83
7552 @unnumberedsec unnamed-faq-83
7555 To: Jan Kort <jan@research.techforce.nl>
7557 In-reply-to: Your message of Fri, 04 Sep 1998 12:18:43 +0200.
7558 Date: Sat, 05 Sep 1998 00:59:49 PDT
7559 From: Vern Paxson <vern>
7563 > "TEST1\n" { fprintf(stderr, "TEST1\n"); yyless(5); }
7564 > ^\n { fprintf(stderr, "empty line\n"); }
7566 > \n { fprintf(stderr, "new line\n"); }
7569 > -- input ---------------------------------------
7571 > -- output --------------------------------------
7574 > ------------------------------------------------
7576 IMHO, it's not clear whether or not this is in fact a bug. It depends
7577 on whether you view yyless() as backing up in the input stream, or as
7578 pushing new characters onto the beginning of the input stream. Flex
7579 interprets it as the latter (for implementation convenience, I'll admit),
7580 and so considers the newline as in fact matching at the beginning of a
7581 line, as after all the last token scanned an entire line and so the
7582 scanner is now at the beginning of a new line.
7584 I agree that this is counter-intuitive for yyless(), given its
7585 functional description (it's less so for unput(), depending on whether
7586 you're unput()'ing new text or scanned text). But I don't plan to
7587 change it any time soon, as it's a pain to do so. Consequently,
7588 you do indeed need to use yy_set_bol() and YY_AT_BOL() to tweak
7589 your scanner into the behavior you desire.
7591 Sorry for the less-than-completely-satisfactory answer.
7597 @c TODO: Evaluate this faq.
7598 @node unnamed-faq-84
7599 @unnumberedsec unnamed-faq-84
7602 To: Patrick Krusenotto <krusenot@mac-info-link.de>
7603 Subject: Re: Problems with restarting flex-2.5.2-generated scanner
7604 In-reply-to: Your message of Thu, 24 Sep 1998 10:14:07 PDT.
7605 Date: Thu, 24 Sep 1998 23:28:43 PDT
7606 From: Vern Paxson <vern>
7608 > I am using flex-2.5.2 and bison 1.25 for Solaris and I am desperately
7609 > trying to make my scanner restart with a new file after my parser stops
7610 > with a parse error. When my compiler restarts, the parser always
7611 > receives the token after the token (in the old file!) that caused the
7614 I suspect the problem is that your parser has read ahead in order
7615 to attempt to resolve an ambiguity, and when it's restarted it picks
7616 up with that token rather than reading a fresh one. If you're using
7617 yacc, then the special "error" production can sometimes be used to
7618 consume tokens in an attempt to get the parser into a consistent state.
7624 @c TODO: Evaluate this faq.
7625 @node unnamed-faq-85
7626 @unnumberedsec unnamed-faq-85
7629 To: Henric Jungheim <junghelh@pe-nelson.com>
7630 Subject: Re: flex 2.5.4a
7631 In-reply-to: Your message of Tue, 27 Oct 1998 16:41:42 PST.
7632 Date: Tue, 27 Oct 1998 16:50:14 PST
7633 From: Vern Paxson <vern>
7635 > This brings up a feature request: How about a command line
7636 > option to specify the filename when reading from stdin? That way one
7637 > doesn't need to create a temporary file in order to get the "#line"
7638 > directives to make sense.
7640 Use -o combined with -t (per the man page description of -o).
7642 > P.S., Is there any simple way to use non-blocking IO to parse multiple
7647 One approach might be to return a magic character on EWOULDBLOCK and
7650 .*<magic-character> // put back .*, eat magic character
7652 This is off the top of my head, not sure it'll work.
7658 @c TODO: Evaluate this faq.
7659 @node unnamed-faq-86
7660 @unnumberedsec unnamed-faq-86
7663 To: "Repko, Billy D" <billy.d.repko@intel.com>
7664 Subject: Re: Compiling scanners
7665 In-reply-to: Your message of Wed, 13 Jan 1999 10:52:47 PST.
7666 Date: Thu, 14 Jan 1999 00:25:30 PST
7667 From: Vern Paxson <vern>
7669 > It appears that maybe it cannot find the lfl library.
7671 The Makefile in the distribution builds it, so you should have it.
7672 It's exceedingly trivial, just a main() that calls yylex() and
7673 a yyrap() that always returns 1.
7676 > \n ++num_lines; ++num_chars;
7679 You can't indent your rules like this - that's where the errors are coming
7680 from. Flex copies indented text to the output file, it's how you do things
7683 int num_lines_seen = 0;
7685 to declare local variables.
7691 @c TODO: Evaluate this faq.
7692 @node unnamed-faq-87
7693 @unnumberedsec unnamed-faq-87
7696 To: Erick Branderhorst <Erick.Branderhorst@asml.nl>
7697 Subject: Re: flex input buffer
7698 In-reply-to: Your message of Tue, 09 Feb 1999 13:53:46 PST.
7699 Date: Tue, 09 Feb 1999 21:03:37 PST
7700 From: Vern Paxson <vern>
7702 > In the flex.skl file the size of the default input buffers is set. Can you
7703 > explain why this size is set and why it is such a high number.
7705 It's large to optimize performance when scanning large files. You can
7706 safely make it a lot lower if needed.
7712 @c TODO: Evaluate this faq.
7713 @node unnamed-faq-88
7714 @unnumberedsec unnamed-faq-88
7717 To: "Guido Minnen" <guidomi@cogs.susx.ac.uk>
7718 Subject: Re: Flex error message
7719 In-reply-to: Your message of Wed, 24 Feb 1999 15:31:46 PST.
7720 Date: Thu, 25 Feb 1999 00:11:31 PST
7721 From: Vern Paxson <vern>
7723 > I'm extending a larger scanner written in Flex and I keep running into
7724 > problems. More specifically, I get the error message:
7725 > "flex: input rules are too complicated (>= 32000 NFA states)"
7727 Increase the definitions in flexdef.h for:
7729 #define JAMSTATE -32766 /* marks a reference to the state that always j
7731 #define MAXIMUM_MNS 31999
7732 #define BAD_SUBSCRIPT -32767
7734 recompile everything, and it should all work.
7740 @c TODO: Evaluate this faq.
7741 @node unnamed-faq-90
7742 @unnumberedsec unnamed-faq-90
7745 To: "Dmitriy Goldobin" <gold@ems.chel.su>
7746 Subject: Re: FLEX trouble
7747 In-reply-to: Your message of Mon, 31 May 1999 18:44:49 PDT.
7748 Date: Tue, 01 Jun 1999 00:15:07 PDT
7749 From: Vern Paxson <vern>
7751 > I have a trouble with FLEX. Why rule "/*".*"*/" work properly,=20
7752 > but rule "/*"(.|\n)*"*/" don't work ?
7754 The second of these will have to scan the entire input stream (because
7755 "(.|\n)*" matches an arbitrary amount of any text) in order to see if
7756 it ends with "*/", terminating the comment. That potentially will overflow
7759 > More complex rule "/*"([^*]|(\*/[^/]))*"*/ give an error
7760 > 'unrecognized rule'.
7762 You can't use the '/' operator inside parentheses. It's not clear
7763 what "(a/b)*" actually means.
7765 > I now use workaround with state <comment>, but single-rule is
7768 Single-rule is nice but will always have the problem of either setting
7769 restrictions on comments (like not allowing multi-line comments) and/or
7770 running the risk of consuming the entire input stream, as noted above.
7776 @c TODO: Evaluate this faq.
7777 @node unnamed-faq-91
7778 @unnumberedsec unnamed-faq-91
7781 Received: from mc-qout4.whowhere.com (mc-qout4.whowhere.com [209.185.123.18])
7782 by ee.lbl.gov (8.9.3/8.9.3) with SMTP id IAA05100
7783 for <vern@ee.lbl.gov>; Tue, 15 Jun 1999 08:56:06 -0700 (PDT)
7784 Received: from Unknown/Local ([?.?.?.?]) by my-deja.com; Tue Jun 15 08:55:43 1999
7786 Date: Tue, 15 Jun 1999 08:55:43 -0700
7787 From: "Aki Niimura" <neko@my-deja.com>
7788 Message-ID: <KNONDOHDOBGAEAAA@my-deja.com>
7793 X-Mailer: MailCity Service
7794 Subject: A question on flex C++ scanner
7795 X-Sender-Ip: 12.72.207.61
7796 Organization: My Deja Email (http://www.my-deja.com:80)
7797 Content-Type: text/plain; charset=us-ascii
7798 Content-Transfer-Encoding: 7bit
7802 I have been using flex for years.
7803 It works very well on many projects.
7804 Most case, I used it to generate a scanner on C language.
7805 However, one project I needed to generate a scanner
7806 on C++ lanuage. Thanks to your enhancement, flex did
7809 Currently, I'm working on enhancing my previous project.
7810 I need to deal with multiple input streams (recursive
7811 inclusion) in this scanner (C++).
7812 I did similar thing for another scanner (C) as you
7813 explained in your documentation.
7815 The generated scanner (C++) has necessary methods:
7816 - switch_to_buffer(struct yy_buffer_state *b)
7817 - yy_create_buffer(istream *is, int sz)
7818 - yy_delete_buffer(struct yy_buffer_state *b)
7820 However, I couldn't figure out how to access current
7821 buffer (yy_current_buffer).
7823 yy_current_buffer is a protected member of yyFlexLexer.
7824 I can't access it directly.
7825 Then, I thought yy_create_buffer() with is = 0 might
7826 return current stream buffer. But it seems not as far
7827 as I checked the source. (flex 2.5.4)
7829 I went through the Web in addition to Flex documentation.
7830 However, it hasn't been successful, so far.
7832 It is not my intention to bother you, but, can you
7833 comment about how to obtain the current stream buffer?
7835 Your response would be highly appreciated.
7840 --== Sent via Deja.com http://www.deja.com/ ==--
7841 Share what you know. Learn what you don't.
7845 @c TODO: Evaluate this faq.
7846 @node unnamed-faq-92
7847 @unnumberedsec unnamed-faq-92
7850 To: neko@my-deja.com
7851 Subject: Re: A question on flex C++ scanner
7852 In-reply-to: Your message of Tue, 15 Jun 1999 08:55:43 PDT.
7853 Date: Tue, 15 Jun 1999 09:04:24 PDT
7854 From: Vern Paxson <vern>
7856 > However, I couldn't figure out how to access current
7857 > buffer (yy_current_buffer).
7859 Derive your own subclass from yyFlexLexer.
7865 @c TODO: Evaluate this faq.
7866 @node unnamed-faq-93
7867 @unnumberedsec unnamed-faq-93
7870 To: "Stones, Darren" <Darren.Stones@nectech.co.uk>
7871 Subject: Re: You're the man to see?
7872 In-reply-to: Your message of Wed, 23 Jun 1999 11:10:29 PDT.
7873 Date: Wed, 23 Jun 1999 09:01:40 PDT
7874 From: Vern Paxson <vern>
7876 > I hope you can help me. I am using Flex and Bison to produce an interpreted
7877 > language. However all goes well until I try to implement an IF statement or
7878 > a WHILE. I cannot get this to work as the parser parses all the conditions
7879 > eg. the TRUE and FALSE conditons to check for a rule match. So I cannot
7882 You need to use the parser to build a parse tree (= abstract syntax trwee),
7883 and when that's all done you recursively evaluate the tree, binding variables
7884 to values at that time.
7890 @c TODO: Evaluate this faq.
7891 @node unnamed-faq-94
7892 @unnumberedsec unnamed-faq-94
7895 To: Petr Danecek <petr@ics.cas.cz>
7896 Subject: Re: flex - question
7897 In-reply-to: Your message of Mon, 28 Jun 1999 19:21:41 PDT.
7898 Date: Fri, 02 Jul 1999 16:52:13 PDT
7899 From: Vern Paxson <vern>
7901 > file, it takes an enormous amount of time. It is funny, because the
7902 > source code has only 12 rules!!! I think it looks like an exponencial
7905 Right, that's the problem - some patterns (those with a lot of
7906 ambiguity, where yours has because at any given time the scanner can
7907 be in the middle of all sorts of combinations of the different
7908 rules) blow up exponentially.
7910 For your rules, there is an easy fix. Change the ".*" that comes fater
7911 the directory name to "[^ ]*". With that in place, the rules are no
7912 longer nearly so ambiguous, because then once one of the directories
7913 has been matched, no other can be matched (since they all require a
7916 If that's not an acceptable solution, then you can enter a start state
7917 to pick up the .*\n after each directory is matched.
7919 Also note that for speed, you'll want to add a ".*" rule at the end,
7920 otherwise rules that don't match any of the patterns will be matched
7921 very slowly, a character at a time.
7927 @c TODO: Evaluate this faq.
7928 @node unnamed-faq-95
7929 @unnumberedsec unnamed-faq-95
7932 To: Tielman Koekemoer <tielman@spi.co.za>
7933 Subject: Re: Please help.
7934 In-reply-to: Your message of Thu, 08 Jul 1999 13:20:37 PDT.
7935 Date: Thu, 08 Jul 1999 08:20:39 PDT
7936 From: Vern Paxson <vern>
7938 > I was hoping you could help me with my problem.
7940 > I tried compiling (gnu)flex on a Solaris 2.4 machine
7941 > but when I ran make (after configure) I got an error.
7943 > --------------------------------------------------------------
7944 > gcc -c -I. -I. -g -O parse.c
7945 > ./flex -t -p ./scan.l >scan.c
7946 > sh: ./flex: not found
7948 > make: Fatal error: Command failed for target `scan.c'
7949 > -------------------------------------------------------------
7951 > What's strange to me is that I'm only
7952 > trying to install flex now. I then edited the Makefile to
7953 > and changed where it says "FLEX = flex" to "FLEX = lex"
7954 > ( lex: the native Solaris one ) but then it complains about
7955 > the "-p" option. Is there any way I can compile flex without
7956 > using flex or lex?
7958 > Thanks so much for your time.
7960 You managed to step on the bootstrap sequence, which first copies
7961 initscan.c to scan.c in order to build flex. Try fetching a fresh
7962 distribution from ftp.ee.lbl.gov. (Or you can first try removing
7963 ".bootstrap" and doing a make again.)
7969 @c TODO: Evaluate this faq.
7970 @node unnamed-faq-96
7971 @unnumberedsec unnamed-faq-96
7974 To: Tielman Koekemoer <tielman@spi.co.za>
7975 Subject: Re: Please help.
7976 In-reply-to: Your message of Fri, 09 Jul 1999 09:16:14 PDT.
7977 Date: Fri, 09 Jul 1999 00:27:20 PDT
7978 From: Vern Paxson <vern>
7980 > First I removed .bootstrap (and ran make) - no luck. I downloaded the
7981 > software but I still have the same problem. Is there anything else I
7986 cp initscan.c scan.c
7990 If this last tries to first build scan.c from scan.l using ./flex, then
7991 your "make" is broken, in which case compile scan.c to scan.o by hand.
7997 @c TODO: Evaluate this faq.
7998 @node unnamed-faq-97
7999 @unnumberedsec unnamed-faq-97
8002 To: Sumanth Kamenani <skamenan@crl.nmsu.edu>
8004 In-reply-to: Your message of Mon, 19 Jul 1999 23:08:41 PDT.
8005 Date: Tue, 20 Jul 1999 00:18:26 PDT
8006 From: Vern Paxson <vern>
8008 > I am getting a compilation error. The error is given as "unknown symbol- yylex".
8010 The parser relies on calling yylex(), but you're instead using the C++ scanning
8011 class, so you need to supply a yylex() "glue" function that calls an instance
8012 scanner of the scanner (e.g., "scanner->yylex()").
8018 @c TODO: Evaluate this faq.
8019 @node unnamed-faq-98
8020 @unnumberedsec unnamed-faq-98
8023 To: daniel@synchrods.synchrods.COM (Daniel Senderowicz)
8025 In-reply-to: Your message of Mon, 22 Nov 1999 11:19:04 PST.
8026 Date: Tue, 23 Nov 1999 15:54:30 PST
8027 From: Vern Paxson <vern>
8029 Well, your problem is the
8031 switch (yybgin-yysvec-1) { /* witchcraft */
8033 at the beginning of lex rules. "witchcraft" == "non-portable". It's
8034 assuming knowledge of the AT&T lex's internal variables.
8036 For flex, you can probably do the equivalent using a switch on YYSTATE.
8042 @c TODO: Evaluate this faq.
8043 @node unnamed-faq-99
8044 @unnumberedsec unnamed-faq-99
8047 To: archow@hss.hns.com
8048 Subject: Re: Regarding distribution of flex and yacc based grammars
8049 In-reply-to: Your message of Sun, 19 Dec 1999 17:50:24 +0530.
8050 Date: Wed, 22 Dec 1999 01:56:24 PST
8051 From: Vern Paxson <vern>
8053 > When we provide the customer with an object code distribution, is it
8054 > necessary for us to provide source
8055 > for the generated C files from flex and bison since they are generated by
8058 For flex, no. I don't know what the current state of this is for bison.
8060 > Also, is there any requrirement for us to neccessarily provide source for
8061 > the grammar files which are fed into flex and bison ?
8063 Again, for flex, no.
8065 See the file "COPYING" in the flex distribution for the legalese.
8071 @c TODO: Evaluate this faq.
8072 @node unnamed-faq-100
8073 @unnumberedsec unnamed-faq-100
8076 To: Martin Gallwey <gallweym@hyperion.moe.ul.ie>
8077 Subject: Re: Flex, and self referencing rules
8078 In-reply-to: Your message of Sun, 20 Feb 2000 01:01:21 PST.
8079 Date: Sat, 19 Feb 2000 18:33:16 PST
8080 From: Vern Paxson <vern>
8082 > However, I do not use unput anywhere. I do use self-referencing
8085 > UnaryExpr ({UnionExpr})|("-"{UnaryExpr})
8087 You can't do this - flex is *not* a parser like yacc (which does indeed
8088 allow recursion), it is a scanner that's confined to regular expressions.
8094 @c TODO: Evaluate this faq.
8095 @node unnamed-faq-101
8096 @unnumberedsec unnamed-faq-101
8099 To: slg3@lehigh.edu (SAMUEL L. GULDEN)
8100 Subject: Re: Flex problem
8101 In-reply-to: Your message of Thu, 02 Mar 2000 12:29:04 PST.
8102 Date: Thu, 02 Mar 2000 23:00:46 PST
8103 From: Vern Paxson <vern>
8105 If this is exactly your program:
8109 > whitespace [ \t\n]+
8112 > "[" { printf("open_brac\n");}
8113 > "]" { printf("close_brac\n");}
8114 > "+" { printf("addop\n");}
8115 > "*" { printf("multop\n");}
8116 > {digits} { printf("NUMBER = %s\n", yytext);}
8119 then the problem is that the last rule needs to be "{whitespace}" !
8125 @node What is the difference between YYLEX_PARAM and YY_DECL?
8126 @unnumberedsec What is the difference between YYLEX_PARAM and YY_DECL?
8128 YYLEX_PARAM is not a flex symbol. It is for Bison. It tells Bison to pass extra
8129 params when it calls yylex() from the parser.
8131 YY_DECL is the Flex declaration of yylex. The default is similar to this:
8135 #define int yy_lex ()
8140 @node Why do I get "conflicting types for yylex" error?
8141 @unnumberedsec Why do I get "conflicting types for yylex" error?
8143 This is a compiler error regarding a generated Bison parser, not a Flex scanner.
8144 It means you need a prototype of yylex() in the top of the Bison file.
8145 Be sure the prototype matches YY_DECL.
8147 @node How do I access the values set in a Flex action from within a Bison action?
8148 @unnumberedsec How do I access the values set in a Flex action from within a Bison action?
8150 With $1, $2, $3, etc. These are called "Semantic Values" in the Bison manual.
8151 See @ref{Top, , , bison, the GNU Bison Manual}.
8153 @node Appendices, Indices, FAQ, Top
8154 @appendix Appendices
8157 * Makefiles and Flex::
8163 @node Makefiles and Flex, Bison Bridge, Appendices, Appendices
8164 @appendixsec Makefiles and Flex
8166 @cindex Makefile, syntax
8168 In this appendix, we provide tips for writing Makefiles to build your scanners.
8170 In a traditional build environment, we say that the @file{.c} files are the
8171 sources, and the @file{.o} files are the intermediate files. When using
8172 @code{flex}, however, the @file{.l} files are the sources, and the generated
8173 @file{.c} files (along with the @file{.o} files) are the intermediate files.
8174 This requires you to carefully plan your Makefile.
8176 Modern @command{make} programs understand that @file{foo.l} is intended to
8177 generate @file{lex.yy.c} or @file{foo.c}, and will behave
8178 accordingly@footnote{GNU @command{make} and GNU @command{automake} are two such
8179 programs that provide implicit rules for flex-generated scanners.}@footnote{GNU @command{automake}
8180 may generate code to execute flex in lex-compatible mode, or to stdout. If this is not what you want,
8181 then you should provide an explicit rule in your Makefile.am}. The
8182 following Makefile does not explicitly instruct @command{make} how to build
8183 @file{foo.c} from @file{foo.l}. Instead, it relies on the implicit rules of the
8184 @command{make} program to build the intermediate file, @file{scan.c}:
8186 @cindex Makefile, example of implicit rules
8189 # Basic Makefile -- relies on implicit rules
8190 # Creates "myprogram" from "scan.l" and "myprogram.c"
8193 myprogram: scan.o myprogram.o
8200 For simple cases, the above may be sufficient. For other cases,
8201 you may have to explicitly instruct @command{make} how to build your scanner.
8202 The following is an example of a Makefile containing explicit rules:
8204 @cindex Makefile, explicit example
8207 # Basic Makefile -- provides explicit rules
8208 # Creates "myprogram" from "scan.l" and "myprogram.c"
8211 myprogram: scan.o myprogram.o
8212 $(CC) -o $@ $(LDFLAGS) $^
8214 myprogram.o: myprogram.c
8215 $(CC) $(CPPFLAGS) $(CFLAGS) -o $@ -c $^
8218 $(CC) $(CPPFLAGS) $(CFLAGS) -o $@ -c $^
8221 $(LEX) $(LFLAGS) -o $@ $^
8229 Notice in the above example that @file{scan.c} is in the @code{clean} target.
8230 This is because we consider the file @file{scan.c} to be an intermediate file.
8232 Finally, we provide a realistic example of a @code{flex} scanner used with a
8233 @code{bison} parser@footnote{This example also applies to yacc parsers.}.
8234 There is a tricky problem we have to deal with. Since a @code{flex} scanner
8235 will typically include a header file (e.g., @file{y.tab.h}) generated by the
8236 parser, we need to be sure that the header file is generated BEFORE the scanner
8237 is compiled. We handle this case in the following example:
8241 # Makefile example -- scanner and parser.
8242 # Creates "myprogram" from "scan.l", "parse.y", and "myprogram.c"
8247 objects = scan.o parse.o myprogram.o
8249 myprogram: $(objects)
8250 scan.o: scan.l parse.c
8252 myprogram.o: myprogram.c
8257 In the above example, notice the line,
8261 scan.o: scan.l parse.c
8265 , which lists the file @file{parse.c} (the generated parser) as a dependency of
8266 @file{scan.o}. We want to ensure that the parser is created before the scanner
8267 is compiled, and the above line seems to do the trick. Feel free to experiment
8268 with your specific implementation of @command{make}.
8271 For more details on writing Makefiles, see @ref{Top, , , make, The
8274 @node Bison Bridge, M4 Dependency, Makefiles and Flex, Appendices
8275 @section C Scanners with Bison Parsers
8277 @cindex bison, bridging with flex
8283 This section describes the @code{flex} features useful when integrating
8284 @code{flex} with @code{GNU bison}@footnote{The features described here are
8285 purely optional, and are by no means the only way to use flex with bison.
8286 We merely provide some glue to ease development of your parser-scanner pair.}.
8287 Skip this section if you are not using
8288 @code{bison} with your scanner. Here we discuss only the @code{flex}
8289 half of the @code{flex} and @code{bison} pair. We do not discuss
8290 @code{bison} in any detail. For more information about generating
8291 @code{bison} parsers, see @ref{Top, , , bison, the GNU Bison Manual}.
8293 A compatible @code{bison} scanner is generated by declaring @samp{%option
8294 bison-bridge} or by supplying @samp{--bison-bridge} when invoking @code{flex}
8295 from the command line. This instructs @code{flex} that the macro
8296 @code{yylval} may be used. The data type for
8297 @code{yylval}, @code{YYSTYPE},
8298 is typically defined in a header file, included in section 1 of the
8299 @code{flex} input file. For a list of functions and macros
8300 available, @xref{bison-functions}.
8302 The declaration of yylex becomes,
8304 @findex yylex (reentrant version)
8307 int yylex ( YYSTYPE * lvalp, yyscan_t scanner );
8311 If @code{%option bison-locations} is specified, then the declaration
8314 @findex yylex (reentrant version)
8317 int yylex ( YYSTYPE * lvalp, YYLTYPE * llocp, yyscan_t scanner );
8321 Note that the macros @code{yylval} and @code{yylloc} evaluate to pointers.
8322 Support for @code{yylloc} is optional in @code{bison}, so it is optional in
8323 @code{flex} as well. The following is an example of a @code{flex} scanner that
8324 is compatible with @code{bison}.
8326 @cindex bison, scanner to be called from bison
8329 /* Scanner for "C" assignment statements... sort of. */
8331 #include "y.tab.h" /* Generated by bison. */
8334 %option bison-bridge bison-locations
8337 [[:digit:]]+ { yylval->num = atoi(yytext); return NUMBER;}
8338 [[:alnum:]]+ { yylval->str = strdup(yytext); return STRING;}
8339 "="|";" { return yytext[0];}
8345 As you can see, there really is no magic here. We just use
8346 @code{yylval} as we would any other variable. The data type of
8347 @code{yylval} is generated by @code{bison}, and included in the file
8348 @file{y.tab.h}. Here is the corresponding @code{bison} parser:
8350 @cindex bison, parser
8353 /* Parser to convert "C" assignments to lisp. */
8355 /* Pass the argument to yyparse through to yylex. */
8356 #define YYPARSE_PARAM scanner
8357 #define YYLEX_PARAM scanner
8369 STRING '=' NUMBER ';' {
8370 printf( "(setf %s %d)", $1, $3 );
8376 @node M4 Dependency, Common Patterns, Bison Bridge, Appendices
8377 @section M4 Dependency
8379 The macro processor @code{m4}@footnote{The use of m4 is subject to change in
8380 future revisions of flex. It is not part of the public API of flex. Do not depend on it.}
8381 must be installed wherever flex is installed.
8382 @code{flex} invokes @samp{m4}, found by searching the directories in the
8383 @code{PATH} environment variable. Any code you place in section 1 or in the
8384 actions will be sent through m4. Please follow these rules to protect your
8385 code from unwanted @code{m4} processing.
8389 @item Do not use symbols that begin with, @samp{m4_}, such as, @samp{m4_define},
8390 or @samp{m4_include}, since those are reserved for @code{m4} macro names. If for
8391 some reason you need m4_ as a prefix, use a preprocessor #define to get your
8392 symbol past m4 unmangled.
8394 @item Do not use the strings @samp{[[} or @samp{]]} anywhere in your code. The
8395 former is not valid in C, except within comments and strings, but the latter is valid in
8396 code such as @code{x[y[z]]}. The solution is simple. To get the literal string
8397 @code{"]]"}, use @code{"]""]"}. To get the array notation @code{x[y[z]]},
8398 use @code{x[y[z] ]}. Flex will attempt to detect these sequences in user code, and
8399 escape them. However, it's best to avoid this complexity where possible, by
8400 removing such sequences from your code.
8404 @code{m4} is only required at the time you run @code{flex}. The generated
8405 scanner is ordinary C or C++, and does @emph{not} require @code{m4}.
8407 @node Common Patterns, ,M4 Dependency, Appendices
8408 @section Common Patterns
8409 @cindex patterns, common
8411 This appendix provides examples of common regular expressions you might use
8417 * Quoted Constructs::
8422 @node Numbers, Identifiers, ,Common Patterns
8427 @item C99 decimal constant
8428 @code{([[:digit:]]@{-@}[0])[[:digit:]]*}
8430 @item C99 hexadecimal constant
8431 @code{0[xX][[:xdigit:]]+}
8433 @item C99 octal constant
8436 @item C99 floating point constant
8438 {dseq} ([[:digit:]]+)
8439 {dseq_opt} ([[:digit:]]*)
8440 {frac} (({dseq_opt}"."{dseq})|{dseq}".")
8441 {exp} ([eE][+-]?{dseq})
8444 {fsuff_opt} ({fsuff}?)
8446 {hdseq} ([[:xdigit:]]+)
8447 {hdseq_opt} ([[:xdigit:]]*)
8448 {hfrac} (({hdseq_opt}"."{hdseq})|({hdseq}"."))
8449 {bexp} ([pP][+-]?{dseq})
8450 {dfc} (({frac}{exp_opt}{fsuff_opt})|({dseq}{exp}{fsuff_opt}))
8451 {hfc} (({hpref}{hfrac}{bexp}{fsuff_opt})|({hpref}{hdseq}{bexp}{fsuff_opt}))
8453 {c99_floating_point_constant} ({dfc}|{hfc})
8456 See C99 section 6.4.4.2 for the gory details.
8460 @node Identifiers, Quoted Constructs, Numbers, Common Patterns
8461 @subsection Identifiers
8465 @item C99 Identifier
8467 ucn ((\\u([[:xdigit:]]{4}))|(\\U([[:xdigit:]]{8})))
8468 nondigit [_[:alpha:]]
8469 c99_id ([_[:alpha:]]|{ucn})([_[:alnum:]]|{ucn})*
8472 Technically, the above pattern does not encompass all possible C99 identifiers, since C99 allows for
8473 "implementation-defined" characters. In practice, C compilers follow the above pattern, with the
8474 addition of the @samp{$} character.
8476 @item UTF-8 Encoded Unicode Code Point
8478 [\x09\x0A\x0D\x20-\x7E]|[\xC2-\xDF][\x80-\xBF]|\xE0[\xA0-\xBF][\x80-\xBF]|[\xE1-\xEC\xEE\xEF]([\x80-\xBF]{2})|\xED[\x80-\x9F][\x80-\xBF]|\xF0[\x90-\xBF]([\x80-\xBF]{2})|[\xF1-\xF3]([\x80-\xBF]{3})|\xF4[\x80-\x8F]([\x80-\xBF]{2})
8483 @node Quoted Constructs, Addresses, Identifiers, Common Patterns
8484 @subsection Quoted Constructs
8487 @item C99 String Literal
8488 @code{L?\"([^\"\\\n]|(\\['\"?\\abfnrtv])|(\\([0123456]@{1,3@}))|(\\x[[:xdigit:]]+)|(\\u([[:xdigit:]]@{4@}))|(\\U([[:xdigit:]]@{8@})))*\"}
8491 @code{("/*"([^*]|"*"[^/])*"*/")|("/"(\\\n)*"/"[^\n]*)}
8493 Note that in C99, a @samp{//}-style comment may be split across lines, and, contrary to popular belief,
8494 does not include the trailing @samp{\n} character.
8496 A better way to scan @samp{/* */} comments is by line, rather than matching
8497 possibly huge comments all at once. This will allow you to scan comments of
8498 unlimited length, as long as line breaks appear at sane intervals. This is also
8499 more efficient when used with automatic line number processing. @xref{option-yylineno}.
8503 "/*" BEGIN(COMMENT);
8515 @node Addresses, ,Quoted Constructs, Common Patterns
8516 @subsection Addresses
8522 dec-octet [0-9]|[1-9][0-9]|1[0-9][0-9]|2[0-4][0-9]|25[0-5]
8523 IPv4address {dec-octet}\.{dec-octet}\.{dec-octet}\.{dec-octet}
8528 h16 [0-9A-Fa-f]{1,4}
8529 ls32 {h16}:{h16}|{IPv4address}
8530 IPv6address ({h16}:){6}{ls32}|
8531 ::({h16}:){5}{ls32}|
8532 ({h16})?::({h16}:){4}{ls32}|
8533 (({h16}:){0,1}{h16})?::({h16}:){3}{ls32}|
8534 (({h16}:){0,2}{h16})?::({h16}:){2}{ls32}|
8535 (({h16}:){0,3}{h16})?::{h16}:{ls32}|
8536 (({h16}:){0,4}{h16})?::{ls32}|
8537 (({h16}:){0,5}{h16})?::{h16}|
8538 (({h16}:){0,6}{h16})?::
8541 See @uref{http://www.ietf.org/rfc/rfc2373.txt, RFC 2373} for details.
8542 Note that you have to fold the definition of @code{IPv6address} into one
8543 line and that it also matches the ``unspecified address'' ``::''.
8546 @code{(([^:/?#]+):)?("//"([^/?#]*))?([^?#]*)(\?([^#]*))?(#(.*))?}
8548 This pattern is nearly useless, since it allows just about any character
8549 to appear in a URI, including spaces and control characters. See
8550 @uref{http://www.ietf.org/rfc/rfc2396.txt, RFC 2396} for details.
8555 @node Indices, , Appendices, Top
8560 * Index of Functions and Macros::
8561 * Index of Variables::
8562 * Index of Data Types::
8564 * Index of Scanner Options::
8567 @node Concept Index, Index of Functions and Macros, Indices, Indices
8568 @unnumberedsec Concept Index
8572 @node Index of Functions and Macros, Index of Variables, Concept Index, Indices
8573 @unnumberedsec Index of Functions and Macros
8575 This is an index of functions and preprocessor macros that look like functions.
8576 For macros that expand to variables or constants, see @ref{Index of Variables}.
8580 @node Index of Variables, Index of Data Types, Index of Functions and Macros, Indices
8581 @unnumberedsec Index of Variables
8583 This is an index of variables, constants, and preprocessor macros
8584 that expand to variables or constants.
8588 @node Index of Data Types, Index of Hooks, Index of Variables, Indices
8589 @unnumberedsec Index of Data Types
8592 @node Index of Hooks, Index of Scanner Options, Index of Data Types, Indices
8593 @unnumberedsec Index of Hooks
8595 This is an index of "hooks" that the user may define. These hooks typically correspond
8596 to specific locations in the generated scanner, and may be used to insert arbitrary code.
8600 @node Index of Scanner Options, , Index of Hooks, Indices
8601 @unnumberedsec Index of Scanner Options
8605 @c A vim script to name the faq entries. delete this when faqs are no longer
8606 @c named "unnamed-faq-XXX".
8608 @c fu! Faq2 () range abort
8609 @c let @r=input("Rename to: ")
8610 @c exe "%s/" . @w . "/" . @r . "/g"
8613 @c nnoremap <F5> 1G/@node\s\+unnamed-faq-\d\+<cr>mfww"wy5ezt:call Faq2()<cr>