This is Info file flex.info, produced by Makeinfo-1.55 from the input file flex.texinfo. START-INFO-DIR-ENTRY * flex: (flex). Fast lexical analyzer generator. END-INFO-DIR-ENTRY  File: flex.info, Node: Performance, Next: Incompatibilities, Prev: Invoking, Up: Top Performance Considerations ************************** The main design goal of `flex' is that it generate high performance scanners. It has been optimized for dealing well with large sets of rules. Aside from the effects of table compression on scanner speed outlined above, there are a number of options/actions which degrade performance. These are, from most expensive to least: `REJECT' pattern sets that require backtracking arbitrary trailing context `^' beginning-of-line operator `yymore' with the first three all being quite expensive and the last two being quite cheap. `REJECT' should be avoided at all costs when performance is important. It is a particularly expensive option. Getting rid of backtracking is messy and often may be an enormous amount of work for a complicated scanner. In principle, one begins by using the `-b' flag to generate a `lex.backtrack' file. For example, on the input %% foo return TOK_KEYWORD; foobar return TOK_KEYWORD; the file looks like: State #6 is non-accepting - associated rule line numbers: 2 3 out-transitions: [ o ] jam-transitions: EOF [ \001-n p-\177 ] State #8 is non-accepting - associated rule line numbers: 3 out-transitions: [ a ] jam-transitions: EOF [ \001-` b-\177 ] State #9 is non-accepting - associated rule line numbers: 3 out-transitions: [ r ] jam-transitions: EOF [ \001-q s-\177 ] Compressed tables always backtrack. The first few lines tell us that there's a scanner state in which it can make a transition on an `o' but not on any other character, and that in that state the currently scanned text does not match any rule. The state occurs when trying to match the rules found at lines 2 and 3 in the input file. If the scanner is in that state and then reads something other than an `o', it will have to backtrack to find a rule which is matched. With a bit of headscratching one can see that this must be the state it's in when it has seen `fo'. When this has happened, if anything other than another `o' is seen, the scanner will have to back up to simply match the `f' (by the default rule). The comment regarding State #8 indicates there's a problem when `foob' has been scanned. Indeed, on any character other than a `b', the scanner will have to back up to accept `foo'. Similarly, the comment for State #9 concerns when `fooba' has been scanned. The final comment reminds us that there's no point going to all the trouble of removing backtracking from the rules unless we're using `-f' or `-F', since there's no performance gain doing so with compressed scanners. The way to remove the backtracking is to add "error" rules: %% foo return TOK_KEYWORD; foobar return TOK_KEYWORD; fooba | foob | fo { /* false alarm, not really a keyword */ return TOK_ID; } Eliminating backtracking among a list of keywords can also be done using a "catch-all" rule: %% foo return TOK_KEYWORD; foobar return TOK_KEYWORD; [a-z]+ return TOK_ID; This is usually the best solution when appropriate. Backtracking messages tend to cascade. With a complicated set of rules it's not uncommon to get hundreds of messages. If one can decipher them, though, it often only takes a dozen or so rules to eliminate the backtracking (though it's easy to make a mistake and have an error rule accidentally match a valid token. A possible future `flex' feature will be to automatically add rules to eliminate backtracking). Variable trailing context (where both the leading and trailing parts do not have a fixed length) entails almost the same performance loss as `REJECT' (i.e., substantial). So when possible a rule like: %% mouse|rat/(cat|dog) run(); is better written: %% mouse/cat|dog run(); rat/cat|dog run(); or as %% mouse|rat/cat run(); mouse|rat/dog run(); Note that here the special `|' action does not provide any savings, and can even make things worse (*note Deficiencies and Bugs: Bugs.). Another area where the user can increase a scanner's performance (and one that's easier to implement) arises from the fact that the longer the tokens matched, the faster the scanner will run. This is because with long tokens the processing of most input characters takes place in the (short) inner scanning loop, and does not often have to go through the additional work of setting up the scanning environment (e.g., `yytext') for the action. Recall the scanner for C comments: %x comment %% int line_num = 1; "/*" BEGIN(comment); [^*\n]* "*"+[^*/\n]* \n ++line_num; "*"+"/" BEGIN(INITIAL); This could be sped up by writing it as: %x comment %% int line_num = 1; "/*" BEGIN(comment); [^*\n]* [^*\n]*\n ++line_num; "*"+[^*/\n]* "*"+[^*/\n]*\n ++line_num; "*"+"/" BEGIN(INITIAL); Now instead of each newline requiring the processing of another action, recognizing the newlines is "distributed" over the other rules to keep the matched text as long as possible. Note that adding rules does not slow down the scanner! The speed of the scanner is independent of the number of rules or (modulo the considerations given at the beginning of this section) how complicated the rules are with regard to operators such as `*' and `|'. A final example in speeding up a scanner: suppose you want to scan through a file containing identifiers and keywords, one per line and with no other extraneous characters, and recognize all the keywords. A natural first approach is: %% asm | auto | break | ... etc ... volatile | while /* it's a keyword */ .|\n /* it's not a keyword */ To eliminate the back-tracking, introduce a catch-all rule: %% asm | auto | break | ... etc ... volatile | while /* it's a keyword */ [a-z]+ | .|\n /* it's not a keyword */ Now, if it's guaranteed that there's exactly one word per line, then we can reduce the total number of matches by a half by merging in the recognition of newlines with that of the other tokens: %% asm\n | auto\n | break\n | ... etc ... volatile\n | while\n /* it's a keyword */ [a-z]+\n | .|\n /* it's not a keyword */ One has to be careful here, as we have now reintroduced backtracking into the scanner. In particular, while we know that there will never be any characters in the input stream other than letters or newlines, `flex' can't figure this out, and it will plan for possibly needing backtracking when it has scanned a token like `auto' and then the next character is something other than a newline or a letter. Previously it would then just match the `auto' rule and be done, but now it has no `auto' rule, only a `auto\n' rule. To eliminate the possibility of backtracking, we could either duplicate all rules but without final newlines, or, since we never expect to encounter such an input and therefore don't how it's classified, we can introduce one more catch-all rule, this one which doesn't include a newline: %% asm\n | auto\n | break\n | ... etc ... volatile\n | while\n /* it's a keyword */ [a-z]+\n | [a-z]+ | .|\n /* it's not a keyword */ Compiled with `-Cf', this is about as fast as one can get a `flex' scanner to go for this particular problem. A final note: `flex' is slow when matching `NUL''s, particularly when a token contains multiple `NUL''s. It's best to write rules which match short amounts of text if it's anticipated that the text will often include `NUL''s.  File: flex.info, Node: Incompatibilities, Next: Diagnostics, Prev: Performance, Up: Top Incompatibilities with `lex' and POSIX ************************************** `flex' is a rewrite of the Unix tool `lex' (the two implementations do not share any code, though), with some extensions and incompatibilities, both of which are of concern to those who wish to write scanners acceptable to either implementation. At present, the POSIX `lex' draft is very close to the original `lex' implementation, so some of these incompatibilities are also in conflict with the POSIX draft. But the intent is that except as noted below, `flex' as it presently stands will ultimately be POSIX conformant (i.e., that those areas of conflict with the POSIX draft will be resolved in `flex''s favor). Please bear in mind that all the comments which follow are with regard to the POSIX draft standard of Summer 1989, and not the final document (or subsequent drafts); they are included so `flex' users can be aware of the standardization issues and those areas where `flex' may in the near future undergo changes incompatible with its current definition. `flex' is fully compatible with `lex' with the following exceptions: * The undocumented `lex' scanner internal variable `yylineno' is not supported. It is difficult to support this option efficiently, since it requires examining every character scanned and reexamining the characters when the scanner backs up. Things get more complicated when the end of buffer or file is reached or a `NUL' is scanned (since the scan must then be restarted with the proper line number count), or the user uses the `yyless', `unput', or `REJECT' actions, or the multiple input buffer functions. The fix is to add rules which, upon seeing a newline, increment `yylineno'. This is usually an easy process, though it can be a drag if some of the patterns can match multiple newlines along with other characters. `yylineno' is not part of the POSIX draft. * The `input' routine is not redefinable, though it may be called to read characters following whatever has been matched by a rule. If `input' encounters an end-of-file the normal `yywrap' processing is done. A "real" end-of-file is returned by `input' as `EOF'. Input is instead controlled by redefining the `YY_INPUT' macro. The `flex' restriction that `input' cannot be redefined is in accordance with the POSIX draft, but `YY_INPUT' has not yet been accepted into the draft (and probably won't; it looks like the draft will simply not specify any way of controlling the scanner's input other than by making an initial assignment to `yyin'). * `flex' scanners do not use `stdio' for input. Because of this, when writing an interactive scanner one must explicitly call `fflush' on the stream associated with the terminal after writing out a prompt. With `lex' such writes are automatically flushed since `lex' scanners use `getchar' for their input. Also, when writing interactive scanners with `flex', the `-I' flag must be used. * `flex' scanners are not as reentrant as `lex' scanners. In particular, if you have an interactive scanner and an interrupt handler which long-jumps out of the scanner, and the scanner is subsequently called again, you may get the following message: fatal flex scanner internal error--end of buffer missed To reenter the scanner, first use yyrestart( yyin ); * `output' is not supported. Output from the `ECHO' macro is done to the file-pointer `yyout' (default `stdout'). The POSIX draft mentions that an `output' routine exists but currently gives no details as to what it does. * `lex' does not support exclusive start conditions (`%x'), though they are in the current POSIX draft. * When definitions are expanded, `flex' encloses them in parentheses. With `lex', the following: NAME [A-Z][A-Z0-9]* %% foo{NAME}? printf( "Found it\n" ); %% will not match the string `foo' because, when the macro is expanded, the rule is equivalent to `foo[A-Z][A-Z0-9]*?' and the precedence is such that the `?' is associated with `[A-Z0-9]*'. With `flex', the rule will be expanded to `foo([A-Z][A-Z0-9]*)?' and so the string `foo' will match. Note that because of this, the `^', `$', `', `/', and `<>' operators cannot be used in a `flex' definition. The POSIX draft interpretation is the same as in `flex'. * To specify a character class which matches anything but a left bracket (`]'), in `lex' one can use `[^]]' but with `flex' one must use `[^\]]'. The latter works with `lex', too. * The lex `%r' (generate a Ratfor scanner) option is not supported. It is not part of the POSIX draft. * If you are providing your own `yywrap' routine, you must include a `#undef yywrap' in the definitions section (section 1). Note that the `#undef' will have to be enclosed in `%{}'. The POSIX draft specifies that `yywrap' is a function, and this is very unlikely to change; so `flex' users are warned that `yywrap' is likely to be changed to a function in the near future. * After a call to `unput', `yytext' and `yyleng' are undefined until the next token is matched. This is not the case with `lex' or the present POSIX draft. * The precedence of the `{}' (numeric range) operator is different. `lex' interprets `abc{1,3}' as "match one, two, or three occurrences of `abc'," whereas `flex' interprets it as "match `ab' followed by one, two, or three occurrences of `c'." The latter is in agreement with the current POSIX draft. * The precedence of the `^' operator is different. `lex' interprets `^foo|bar' as "match either `foo' at the beginning of a line, or `bar' anywhere", whereas `flex' interprets it as "match either `foo' or `bar' if they come at the beginning of a line". The latter is in agreement with the current POSIX draft. * To refer to `yytext' outside of the scanner source file, the correct definition with `flex' is `extern char *yytext' rather than `extern char yytext[]'. This is contrary to the current POSIX draft but a point on which `flex' will not be changing, as the array representation entails a serious performance penalty. It is hoped that the POSIX draft will be amended to support the `flex' variety of declaration (as this is a fairly painless change to require of `lex' users). * `yyin' is initialized by `lex' to be `stdin'; `flex', on the other hand, initializes `yyin' to `NULL' and then assigns it to stdin the first time the scanner is called, providing `yyin' has not already been assigned to a non-`NULL' value. The difference is subtle, but the net effect is that with `flex' scanners, `yyin' does not have a valid value until the scanner has been called. * The special table-size declarations such as `%a' supported by `lex' are not required by `flex' scanners; `flex' ignores them. * The name `FLEX_SCANNER' is `#define''d so scanners may be written for use with either `flex' or `lex'. The following `flex' features are not included in lex or the POSIX draft standard: `yyterminate()' `<>' `YY_DECL' `#line' directives `%{}' around actions `yyrestart()' comments beginning with `#' (deprecated) multiple actions on a line This last feature refers to the fact that with `flex' you can put multiple actions on the same line, separated with semicolons, while with `lex', the following foo handle_foo(); ++num_foos_seen; is (rather surprisingly) truncated to foo handle_foo(); `flex' does not truncate the action. Actions that are not enclosed in braces are simply terminated at the end of the line.  File: flex.info, Node: Diagnostics, Next: Bugs, Prev: Incompatibilities, Up: Top Diagnostic Messages ******************* `reject_used_but_not_detected undefined' `yymore_used_but_not_detected undefined' These errors can occur at compile time. They indicate that the scanner uses `REJECT' or `yymore' but that `flex' failed to notice the fact, meaning that `flex' scanned the first two sections looking for occurrences of these actions and failed to find any, but somehow you snuck some in (via a `#include' file, for example). Make an explicit reference to the action in your `flex' input file. (Note that previously `flex' supported a `%used'/`%unused' mechanism for dealing with this problem; this feature is still supported but now deprecated, and will go away soon unless the author hears from people who can argue compellingly that they need it.) `flex scanner jammed' A scanner compiled with `-s' has encountered an input string which wasn't matched by any of its rules. `flex input buffer overflowed' A scanner rule matched a string long enough to overflow the scanner's internal input buffer (16K bytes by default--controlled by `YY_BUF_SIZE' in `flex.skel'. Note that to redefine this macro, you must first `#undefine' it). `scanner requires -8 flag' Your scanner specification includes recognizing 8-bit characters and you did not specify the `-8' flag (and your site has not installed `flex' with `-8' as the default). `fatal flex scanner internal error--end of buffer missed' This can occur in an scanner which is reentered after a long-jump has jumped out (or over) the scanner's activation frame. Before reentering the scanner, use: yyrestart( yyin ); `too many %t classes!' You managed to put every single character into its own `%t' class. `flex' requires that at least one of the classes share characters.  File: flex.info, Node: Bugs, Next: Acknowledgements, Prev: Diagnostics, Up: Top Deficiencies and Bugs ********************* Some trailing context patterns cannot be properly matched and generate warning messages (`Dangerous trailing context'). These are patterns where the ending of the first part of the rule matches the beginning of the second part, such as `zx*/xy*', where the `x*' matches the `x' at the beginning of the trailing context. (Note that the POSIX draft states that the text matched by such patterns is undefined.) For some trailing context rules, parts which are actually fixed-length are not recognized as such, leading to the abovementioned performance loss. In particular, parts using `|' or {n} (such as `foo{3}') are always considered variable-length. Combining trailing context with the special `|' action can result in fixed trailing context being turned into the more expensive variable trailing context. For example, this happens in the following example: %% abc | xyz/def Use of `unput' invalidates `yytext' and `yyleng'. Use of `unput' to push back more text than was matched can result in the pushed-back text matching a beginning-of-line (`^') rule even though it didn't come at the beginning of the line (though this is rare!). Pattern-matching of `NUL''s is substantially slower than matching other characters. `flex' does not generate correct `#line' directives for code internal to the scanner; thus, bugs in `flex.skel' yield bogus line numbers. Due to both buffering of input and read-ahead, you cannot intermix calls to `stdio.h' routines, such as, for example, `getchar', with `flex' rules and expect it to work. Call `input' instead. The total table entries listed by the `-v' flag excludes the number of table entries needed to determine what rule has been matched. The number of entries is equal to the number of DFA states if the scanner does not use `REJECT', and somewhat greater than the number of states if it does. `REJECT' cannot be used with the `-f' or `-F' options. Some of the macros, such as `yywrap', may in the future become functions which live in the `-lfl' library. This will doubtless break a lot of code, but may be required for POSIX compliance. The `flex' internal algorithms need documentation.  File: flex.info, Node: Acknowledgements, Prev: Bugs, Up: Top Contributors to `flex' ********************** The author of `flex' is Vern Paxson, with the help of many ideas and much inspiration from Van Jacobson. Original version by Jef Poskanzer. The fast table representation is a partial implementation of a design done by Van Jacobson. The implementation was done by Kevin Gong and Vern Paxson. Thanks to the many `flex' beta-testers, feedbackers, and contributors, especially Casey Leedom, `benson@odi.com', Keith Bostic, Frederic Brehm, Nick Christopher, Jason Coughlin, Scott David Daniels, Leo Eskin, Chris Faylor, Eric Goldman, Eric Hughes, Jeffrey R. Jones, Kevin B. Kenny, Ronald Lamprecht, Greg Lee, Craig Leres, Mohamed el Lozy, Jim Meyering, Marc Nozell, Esmond Pitt, Jef Poskanzer, Jim Roskind, Dave Tallman, Frank Whaley, Ken Yap, and those whose names have slipped my marginal mail-archiving skills but whose contributions are appreciated all the same. Thanks to Keith Bostic, John Gilmore, Craig Leres, Bob Mulcahy, Rich Salz, and Richard Stallman for help with various distribution headaches. Thanks to Esmond Pitt and Earle Horton for 8-bit character support; to Benson Margulies and Fred Burke for C++ support; to Ove Ewerlid for the basics of support for `NUL''s; and to Eric Hughes for the basics of support for multiple buffers. Work is being done on extending `flex' to generate scanners in which the state machine is directly represented in C code rather than tables. These scanners may well be substantially faster than those generated using `-f' or `-F'. If you are working in this area and are interested in comparing notes and seeing whether redundant work can be avoided, contact Ove Ewerlid (`ewerlid@mizar.DoCS.UU.SE'). This work was primarily done when I was at the Real Time Systems Group at the Lawrence Berkeley Laboratory in Berkeley, CA. Many thanks to all there for the support I received. Send comments to: Vern Paxson Computer Systems Engineering Bldg. 46A, Room 1123 Lawrence Berkeley Laboratory Berkeley, CA 94720 vern@ee.lbl.gov .