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 The Scannerless Boolean Parser (SBP) is a scannerless parser for <a
51 href=http://www.cs.queensu.ca/home/okhotin/boolean/>boolean
52 grammars</a> (a superset of context-free grammars). It is written in
53 Java and emits Java source code.
55 <h1>What is interesting about it?</h1>
57 SBP deliberately sacrifices performance in favor of ease of extensibility.
60 Since it is an implementation of the (modified) <a
61 href=http://www.program-transformation.org/Sdf/GeneralizedLR>Lang-Tomita
62 GLR algorithm</a>, SBP supports all context-free languages.
66 href=http://en.wikipedia.org/wiki/Lexerless_parsing>scannerless</a>
67 (does not require a lexer). This allows it to easily handle languages
68 which have non-regular lexical structure or lack a clear lexer-parser
69 distinction, such as TeX, XML, RFC1738 (URLs), ASN.1, SMTP headers,
73 In addition to the juxtaposition and union operators provided in
74 context-free languages, SBP supports grammars which use the
75 intersection operator (<a
76 href=http://www.cs.queensu.ca/home/okhotin/conjunctive/>conjunctive
77 grammars</a>) and the complement operator (<a
78 href=http://www.cs.queensu.ca/home/okhotin/boolean/>boolean
81 <h1>What features does it have?</h1>
83 Features fully implemented are in <font color=green>green</font>;
84 those partially implemented are in <font color=orange>orange</font>;
85 those unimplemented (but planned) are in <font color=red>red</font>.
87 <ul> <li> <b>An implementation of the Lang-Tomita GLR parsing algorithm</b>
89 <li> Including <font color=green>Johnstone & Scott's RNGLR algorithm</font> for epsilon-productions</a>
91 <li> <a href=http://citeseer.ist.psu.edu/vandenbrand02disambiguation.html><font color=green>Visser's</font> extensions</a>
92 for <font color=green>scannerless parsing</font>
93 <ul> <li> <font color=green>Follow</font>, <font color=green>Avoid, Prefer</font>, <font color=green>Reject</font> constraints
94 <li> <font color=green>Character ranges</font>
95 <li> Automatic insertion of <font color=green>whitespace/comments</font>
98 <li> <font color=green>Any topological space</font> can be
99 used as an alphabet (need not be discrete)
100 <ul> <li> <font color=green>Unicode</font>
101 <li> <font color=orange>Trees</font>
104 <li> <font color=green>Associativity constraints</font> on <font color=green><i>n</i>-ary operators</font>
108 <li> <b>Ability to parse a wide variety of grammars in
109 </b> O(n<sup>3</sup>) time:
112 <li> <font color=green>all context-free grammars</font>
114 <li> <font color=green>epsilon productions</font>, <font
115 color=green>included in the parse forest</font>
117 <li> <font color=green>circularities</font>, <font
118 color=red>included in the parse forest</font>.
120 <li> Regular expression operators (
121 <tt><font color=green>*</font></tt>,
122 <tt><font color=green>?</font></tt>,
123 <tt><font color=green>+</font></tt>
126 <li> <font color=green>conjunctive grammars</font>
127 (<font color=green>intersection</font> operator)
129 <li> <font color=orange>boolean grammars</font> (<font
130 color=green>intersection</font>, <font
131 color=green>intersect-with-complement</font>, and
132 <font color=orange>generalized-complement</font>)
136 <li> <b>Facilitates experimenting with grammars</b>
139 <li> <font color=green>Interpreted mode</font>, in which the
140 parse table is interpreted directly, eliminating the
141 need for a compiler and making it easier for grammars
142 to operate on grammars.
144 <li> <font color=green>Simple
145 <a href=api/edu/berkeley/sbp/package-summary.html>API</a></font>
146 makes it easy to generate, analyze, and modify grammars
150 <li> Components of a grammar (nonterminals,
151 productions, etc) <font
152 color=green>represented as objects</font>
153 <li> composite elements implement <font color=green><tt>Iterable<T></tt></font>
156 <li> <font color=red>Compiled mode</font>, in which Java
157 source code is emitted; compiling this code yields a
158 parser. The resulting parser is <i>much</i> faster.
164 <h1>What is it deliberately missing?</h1>
166 <ul> <li> Semantic actions; the only option is to return a parse forest.
168 <li> This keeps the grammar specification language-neutral.
169 <li> A grammar can, however, indicate that certain parts of the parse tree should be dropped.
173 <h1>What features would be nice to have?</h1>
176 <li> <strike>Drop Farshi's algorithm and use <a
177 href=http://doi.ieeecomputersociety.org/10.1109/HICSS.2002.994495>GRMLR</a></strike>.
178 <font color=green>Done!</font>
180 <li> An implementation of the <a
181 href=http://www.cs.berkeley.edu/~smcpeak/elkhound/sources/elkhound/algorithm.html>McPeak-Necula
182 optimization</a> for bounded-depth determinism.
184 <li> Lazy parse trees, to decrease the space requirements from
185 o(n) to o(1) [but still O(n)].
187 <li> Consider implementing <a
188 href=http://www.cs.uvic.ca/~nigelh/Publications/cc99-paper.pdf>
189 Aycock-Horspool</a> unrolling. Improves performance with
190 only highly localized increase in algorithmic complexity.
191 Subsumes many other optimizations.
195 <h1>What are the long term goals?</h1>
197 As we come to a more mature understanding of the pragmatic aspects of
198 boolean grammars, a long-term goal is to migrate support for these
199 features to existing high-performance GLR implementations (<a
200 href=http://www.cs.berkeley.edu/~smcpeak/elkhound/>Elkhound</a>, <a
201 href=http://www.delorie.com/gnu/docs/bison/bison_90.html>bison-glr</a>).
203 <h1>Where can I read more about it?</h1>
205 <ul> <li> The <a href=../README>README</a> file is the best place to start
206 <li> After that, be sure to read <a href=jargon.txt>jargon.txt</a>
208 href=api/edu/berkeley/sbp/package-summary.html>javadoc</a>
209 is the best description of the API
210 <li> There's a <a href=../tests/meta.g>tentative metagrammar</a>,
212 <li> You can also get <a href=osq.lunch.talk.pdf>slides</a>
213 from my talk at the OSQ Lunch on 02-Nov-2005, though some of
214 the stuff (specifically what SBP can and cannot do) is
216 <li> A <a href=preprint.pdf>preprint</a> of one of my conference
220 <h1>Where can I get it?</h1>
222 The color coding above accurately reflects the state of the
223 implementation (<font color=green>11-Dec-2005</font>). However, in its current state it is a
224 bit messy, and may require a bit of fiddling to get it to do what you
225 want. This situation should improve in the next few weeks as I am
226 done adding features (for now) and am currently focusing on
227 reliability, cleanliness, and performance.
230 SBP is available under the BSD license.
233 You can download a snapshot (<font color=green>11-Dec-2005</font>) <a
234 href=../../sbp/edu.berkeley.sbp.tar.gz>here</a>. The parser-generator
235 requires Java 1.5 or later; the Java code it emits <font
236 color=orange>should run on any Java 1.1+ JVM</font>. After unpacking
237 the archive, simply type <tt>make</tt> to compile SBP and run the
240 </td></tr></table></center>