2 <head><title>The Scannerless Boolean Parser (SBP)</title>
8 font-family: helvetica, verdana, arial, sans-serif;
13 border-top-width: 2pt;
14 border-top-style: solid;
18 font-family: helvetica, verdana, arial, sans-serif;
25 font-family: helvetica, verdana, arial, sans-serif;
31 font-family: helvetica, verdana, arial, sans-serif;
36 LI { margin-top: 5px; }
41 <center><table><tr><td width=600>
44 <font style='font-size:24pt; font-family:helvetica, verdana, arial, sans-serif'>
45 <b>SBP: the Scannerless Boolean Parser</b></font>
50 <table width=400><tr><td>
51 <font color=red><b>Update:</b></font> [05-July-2006]<br><br>
53 The reflective grammar-to-java bindings are complete, so SBP is now
54 vastly easier to use. You can find example code <a
55 href=../src/edu/berkeley/sbp/misc/Demo.java>here</a>
56 and the companion grammar <a
57 href=../tests/demo.g>here</a>.
58 I will be writing up a tutorial tomorrow (the 7th). -- Adam
65 The Scannerless Boolean Parser (SBP) is a scannerless parser for <a
66 href=http://www.cs.queensu.ca/home/okhotin/boolean/>boolean
67 grammars</a> (a superset of context-free grammars). It is written in
68 Java and emits Java source code.
70 <h1>What is interesting about it?</h1>
72 SBP deliberately sacrifices performance in favor of ease of extensibility.
75 Since it is an implementation of the (modified) <a
76 href=http://www.program-transformation.org/Sdf/GeneralizedLR>Lang-Tomita
77 GLR algorithm</a>, SBP supports all context-free languages.
81 href=http://en.wikipedia.org/wiki/Lexerless_parsing>scannerless</a>
82 (does not require a lexer). This allows it to easily handle languages
83 which have non-regular lexical structure or lack a clear lexer-parser
84 distinction, such as TeX, XML, RFC1738 (URLs), ASN.1, SMTP headers,
88 In addition to the juxtaposition and union operators provided in
89 context-free languages, SBP supports grammars which use the
90 intersection operator (<a
91 href=http://www.cs.queensu.ca/home/okhotin/conjunctive/>conjunctive
92 grammars</a>) and the complement operator (<a
93 href=http://www.cs.queensu.ca/home/okhotin/boolean/>boolean
96 <h1>What features does it have?</h1>
98 Features fully implemented are in <font color=green>green</font>;
99 those partially implemented are in <font color=orange>orange</font>;
100 those unimplemented (but planned) are in <font color=red>red</font>.
102 <ul> <li> <b>An implementation of the Lang-Tomita GLR parsing algorithm</b>
104 <li> Including <font color=green>Johnstone & Scott's RNGLR algorithm</font> for epsilon-productions</a>
106 <li> <a href=http://citeseer.ist.psu.edu/vandenbrand02disambiguation.html><font color=green>Visser's</font> extensions</a>
107 for <font color=green>scannerless parsing</font>
108 <ul> <li> <font color=green>Follow</font>, <font color=green>Avoid, Prefer</font>, <font color=green>Reject</font> constraints
109 <li> <font color=green>Character ranges</font>
110 <li> Automatic insertion of <font color=green>whitespace/comments</font>
113 <li> <font color=green>Any topological space</font> can be
114 used as an alphabet (need not be discrete)
115 <ul> <li> <font color=green>Unicode</font>
116 <li> <font color=orange>Trees</font>
119 <li> <font color=green>Associativity constraints</font> on <font color=green><i>n</i>-ary operators</font>
123 <li> <b>Ability to parse a wide variety of grammars in
124 </b> O(n<sup>3</sup>) time:
127 <li> <font color=green>all context-free grammars</font>
129 <li> <font color=green>epsilon productions</font>, <font
130 color=green>included in the parse forest</font>
132 <li> <font color=green>circularities</font>, <font
133 color=red>included in the parse forest</font>.
135 <li> Regular expression operators (
136 <tt><font color=green>*</font></tt>,
137 <tt><font color=green>?</font></tt>,
138 <tt><font color=green>+</font></tt>
141 <li> <font color=green>conjunctive grammars</font>
142 (<font color=green>intersection</font> operator)
144 <li> <font color=orange>boolean grammars</font> (<font
145 color=green>intersection</font>, <font
146 color=green>intersect-with-complement</font>, and
147 <font color=orange>generalized-complement</font>)
151 <li> <b>Facilitates experimenting with grammars</b>
154 <li> <font color=green>Interpreted mode</font>, in which the
155 parse table is interpreted directly, eliminating the
156 need for a compiler and making it easier for grammars
157 to operate on grammars.
159 <li> <font color=green>Simple
160 <a href=api/edu/berkeley/sbp/package-summary.html>API</a></font>
161 makes it easy to generate, analyze, and modify grammars
165 <li> Components of a grammar (nonterminals,
166 productions, etc) <font
167 color=green>represented as objects</font>
168 <li> composite elements implement <font color=green><tt>Iterable<T></tt></font>
171 <li> <font color=red>Compiled mode</font>, in which Java
172 source code is emitted; compiling this code yields a
173 parser. The resulting parser is <i>much</i> faster.
179 <h1>What is it deliberately missing?</h1>
181 <ul> <li> Semantic actions; the only option is to return a parse forest.
183 <li> This keeps the grammar specification language-neutral.
184 <li> A grammar can, however, indicate that certain parts of the parse tree should be dropped.
188 <h1>What features would be nice to have?</h1>
191 <li> <strike>Drop Farshi's algorithm and use <a
192 href=http://doi.ieeecomputersociety.org/10.1109/HICSS.2002.994495>GRMLR</a></strike>.
193 <font color=green>Done!</font>
195 <li> An implementation of the <a
196 href=http://www.cs.berkeley.edu/~smcpeak/elkhound/sources/elkhound/algorithm.html>McPeak-Necula
197 optimization</a> for bounded-depth determinism.
199 <li> Lazy parse trees, to decrease the space requirements from
200 o(n) to o(1) [but still O(n)].
202 <li> Consider implementing <a
203 href=http://www.cs.uvic.ca/~nigelh/Publications/cc99-paper.pdf>
204 Aycock-Horspool</a> unrolling. Improves performance with
205 only highly localized increase in algorithmic complexity.
206 Subsumes many other optimizations.
210 <h1>What are the long term goals?</h1>
212 As we come to a more mature understanding of the pragmatic aspects of
213 boolean grammars, a long-term goal is to migrate support for these
214 features to existing high-performance GLR implementations (<a
215 href=http://www.cs.berkeley.edu/~smcpeak/elkhound/>Elkhound</a>, <a
216 href=http://www.delorie.com/gnu/docs/bison/bison_90.html>bison-glr</a>).
218 <h1>Where can I read more about it?</h1>
220 <ul> <li> The <a href=../README>README</a> file is the best place to start
221 <li> After that, be sure to read <a href=jargon.txt>jargon.txt</a>
223 href=api/edu/berkeley/sbp/package-summary.html>javadoc</a>
224 is the best description of the API
225 <li> There's a <a href=../tests/meta.g>tentative metagrammar</a>,
227 <li> You can also get <a href=osq.lunch.talk.pdf>slides</a>
228 from my talk at the OSQ Lunch on 02-Nov-2005, though some of
229 the stuff (specifically what SBP can and cannot do) is
231 <li> A <a href=preprint.pdf>preprint</a> of one of my conference
235 <h1>Where can I get it?</h1>
237 The color coding above accurately reflects the state of the
238 implementation (<font color=green>11-Dec-2005</font>). However, in its current state it is a
239 bit messy, and may require a bit of fiddling to get it to do what you
240 want. This situation should improve in the next few weeks as I am
241 done adding features (for now) and am currently focusing on
242 reliability, cleanliness, and performance.
245 SBP is available under the BSD license.
248 You can download a snapshot (<font color=green>11-Dec-2005</font>) <a
249 href=../../sbp/edu.berkeley.sbp.tar.gz>here</a>. The parser-generator
250 requires Java 1.5 or later; the Java code it emits <font
251 color=orange>should run on any Java 1.1+ JVM</font>. After unpacking
252 the archive, simply type <tt>make</tt> to compile SBP and run the
255 </td></tr></table></center>