jepler/yapps
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
yapps/doc/yapps2.html
Mike Kazantsev 7f948bae0e Cumulative debian patches 2.1.1-17.2 (circa debian-specific stuff) 
By Matthias Urlichs and debian project maintainers/contributors
2013-01-19 09:21:27 +06:00

1206 lines
57 KiB
HTML
Raw Blame History

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN"
"http://www.w3.org/TR/REC-html40/loose.dtd">
<HTML>
<HEAD>
<META http-equiv="Content-Type" content="text/html; charset=UTF-8">
<META name="GENERATOR" content="hevea 1.08">
<STYLE type="text/css">
.toc{list-style:none;}
.title{margin:auto;text-align:center}
.center{text-align:center;margin-left:auto;margin-right:auto;}
.flushleft{text-align:left;margin-left:0ex;margin-right:auto;}
.flushright{text-align:right;margin-left:auto;margin-right:0ex;}
DIV TABLE{margin-left:inherit;margin-right:inherit;}
PRE{text-align:left;margin-left:0ex;margin-right:auto;}
BLOCKQUOTE{margin-left:4ex;margin-right:4ex;text-align:left;}
.part{margin:auto;text-align:center}
</STYLE>
</HEAD>
<BODY >
<!--HEVEA command line is: /usr/bin/hevea yapps2.tex -->
<!--HTMLHEAD-->
<!--ENDHTML-->
<!--PREFIX <ARG ></ARG>-->
<!--CUT DEF section 1 -->
<BR>
<BR>
<DIV CLASS="center">
<TABLE CELLSPACING=2 CELLPADDING=0>
<TR><TD ALIGN=center NOWRAP><FONT SIZE=5>The <EM>Yapps</EM> Parser Generator System</FONT></TD>
</TR>
<TR><TD ALIGN=center NOWRAP><CODE>http://theory.stanford.edu/~amitp/Yapps/</CODE></TD>
</TR>
<TR><TD ALIGN=center NOWRAP>Version 2</TD>
</TR>
<TR><TD ALIGN=center NOWRAP>&nbsp;</TD>
</TR>
<TR><TD ALIGN=center NOWRAP>Amit J. Patel</TD>
</TR>
<TR><TD ALIGN=center NOWRAP>http://www-cs-students.stanford.edu/&nbsp;amitp/
http://www-cs-students.stanford.edu/&nbsp;amitp/</TD>
</TR></TABLE> <HR SIZE=2>
</DIV><BR>
<BR>
<!--TOC section Introduction-->
<H2 CLASS="section"><A NAME="htoc1">1</A>&nbsp;&nbsp;Introduction</H2><!--SEC END -->
<EM>Yapps</EM> (<U>Y</U>et <U>A</U>nother <U>P</U>ython
<U>P</U>arser <U>S</U>ystem) is an easy to use parser
generator that is written in Python and generates Python code. There
are several parser generator systems already available for Python,
including <TT>PyLR, kjParsing, PyBison,</TT> and <TT>mcf.pars,</TT>
but I had different goals for my parser. Yapps is simple, is easy to
use, and produces human-readable parsers. It is not the fastest or
most powerful parser. Yapps is designed to be used when regular
expressions are not enough and other parser systems are too much:
situations where you may write your own recursive descent parser.<BR>
<BR>
Some unusual features of Yapps that may be of interest are:
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">Yapps produces recursive descent parsers that are readable by
humans, as opposed to table-driven parsers that are difficult to
read. A Yapps parser for a simple calculator looks similar to the
one that Mark Lutz wrote by hand for <EM>Programming Python.</EM><BR>
<BR>
<LI CLASS="li-enumerate">Yapps also allows for rules that accept parameters and pass
arguments to be used while parsing subexpressions. Grammars that
allow for arguments to be passed to subrules and for values to be
passed back are often called <EM>attribute grammars.</EM> In many
cases parameterized rules can be used to perform actions at &#8220;parse
time&#8221; that are usually delayed until later. For example,
information about variable declarations can be passed into the
rules that parse a procedure body, so that undefined variables can
be detected at parse time. The types of defined variables can be
used in parsing as well&mdash;for example, if the type of <TT>X</TT> is
known, we can determine whether <TT>X(1)</TT> is an array reference or
a function call.<BR>
<BR>
<LI CLASS="li-enumerate">Yapps grammars are fairly easy to write, although there are
some inconveniences having to do with ELL(1) parsing that have to be
worked around. For example, rules have to be left factored and
rules may not be left recursive. However, neither limitation seems
to be a problem in practice. <BR>
<BR>
Yapps grammars look similar to the notation used in the Python
reference manual, with operators like <CODE>*</CODE>, <CODE>+</CODE>, <CODE>|</CODE>,
<CODE>[]</CODE>, and <CODE>()</CODE> for patterns, names (<TT>tim</TT>) for rules,
regular expressions (<CODE>"[a-z]+"</CODE>) for tokens, and <CODE>#</CODE> for
comments.<BR>
<BR>
<LI CLASS="li-enumerate">The Yapps parser generator is written as a single Python module
with no C extensions. Yapps produces parsers that are written
entirely in Python, and require only the Yapps run-time module (5k)
for support.<BR>
<BR>
<LI CLASS="li-enumerate">Yapps's scanner is context-sensitive, picking tokens based on
the types of the tokens accepted by the parser. This can be
helpful when implementing certain kinds of parsers, such as for a
preprocessor.</OL>
There are several disadvantages of using Yapps over another parser system:
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">Yapps parsers are <TT>ELL(1)</TT> (Extended LL(1)), which is
less powerful than <TT>LALR</TT> (used by <TT>PyLR</TT>) or
<TT>SLR</TT> (used by <TT>kjParsing</TT>), so Yapps would not be a
good choice for parsing complex languages. For example, allowing
both <TT>x := 5;</TT> and <TT>x;</TT> as statements is difficult
because we must distinguish based on only one token of lookahead.
Seeing only <TT>x</TT>, we cannot decide whether we have an
assignment statement or an expression statement. (Note however
that this kind of grammar can be matched with backtracking; see
section <A HREF="#sec:future">F</A>.)<BR>
<BR>
<LI CLASS="li-enumerate">The scanner that Yapps provides can only read from strings, not
files, so an entire file has to be read in before scanning can
begin. It is possible to build a custom scanner, though, so in
cases where stream input is needed (from the console, a network, or
a large file are examples), the Yapps parser can be given a custom
scanner that reads from a stream instead of a string.<BR>
<BR>
<LI CLASS="li-enumerate">Yapps is not designed with efficiency in mind.</OL>
Yapps provides an easy to use parser generator that produces parsers
similar to what you might write by hand. It is not meant to be a
solution for all parsing problems, but instead an aid for those times
you would write a parser by hand rather than using one of the more
powerful parsing packages available.<BR>
<BR>
Yapps 2.0 is easier to use than Yapps 1.0. New features include a
less restrictive input syntax, which allows mixing of sequences,
choices, terminals, and nonterminals; optional matching; the ability
to insert single-line statements into the generated parser; and
looping constructs <CODE>*</CODE> and <CODE>+</CODE> similar to the repetitive
matching constructs in regular expressions. Unfortunately, the
addition of these constructs has made Yapps 2.0 incompatible with
Yapps 1.0, so grammars will have to be rewritten. See section
<A HREF="#sec:Upgrading">??</A> for tips on changing Yapps 1.0 grammars for use
with Yapps 2.0.<BR>
<BR>
<!--TOC section Examples-->
<H2 CLASS="section"><A NAME="htoc2">2</A>&nbsp;&nbsp;Examples</H2><!--SEC END -->
In this section are several examples that show the use of Yapps.
First, an introduction shows how to construct grammars and write them
in Yapps form. This example can be skipped by someone familiar with
grammars and parsing. Next is a Lisp expression grammar that produces
a parse tree as output. This example demonstrates the use of tokens
and rules, as well as returning values from rules. The third example
is a expression evaluation grammar that evaluates during parsing
(instead of producing a parse tree).<BR>
<BR>
<!--TOC subsection Introduction to Grammars-->
<H3 CLASS="subsection"><A NAME="htoc3">2.1</A>&nbsp;&nbsp;Introduction to Grammars</H3><!--SEC END -->
A <EM>grammar</EM> for a natural language specifies how words can be put
together to form large structures, such as phrases and sentences. A
grammar for a computer language is similar in that it specifies how
small components (called <EM>tokens</EM>) can be put together to form
larger structures. In this section we will write a grammar for a tiny
subset of English.<BR>
<BR>
Simple English sentences can be described as being a noun phrase
followed by a verb followed by a noun phrase. For example, in the
sentence, &#8220;Jack sank the blue ship,&#8221; the word &#8220;Jack&#8221; is the first
noun phrase, &#8220;sank&#8221; is the verb, and &#8220;the blue ship&#8221; is the second
noun phrase. In addition we should say what a noun phrase is; for
this example we shall say that a noun phrase is an optional article
(a, an, the) followed by any number of adjectives followed by a noun.
The tokens in our language are the articles, nouns, verbs, and
adjectives. The <EM>rules</EM> in our language will tell us how to
combine the tokens together to form lists of adjectives, noun phrases,
and sentences:
<UL CLASS="itemize"><LI CLASS="li-itemize">
<TT>sentence: noun_phrase verb noun_phrase</TT>
<LI CLASS="li-itemize"><TT>noun_phrase: [article] adjective* noun</TT>
</UL>
Notice that some things that we said easily in English, such as
&#8220;optional article,&#8221; are expressed using special syntax, such as
brackets. When we said, &#8220;any number of adjectives,&#8221; we wrote
<TT>adjective*</TT>, where the <TT>*</TT> means &#8220;zero or more of the
preceding pattern&#8221;.<BR>
<BR>
The grammar given above is close to a Yapps grammar. We also have to
specify what the tokens are, and what to do when a pattern is matched.
For this example, we will do nothing when patterns are matched; the
next example will explain how to perform match actions.
<PRE CLASS="verbatim">
parser TinyEnglish:
ignore: "\\W+"
token noun: "(Jack|spam|ship)"
token verb: "(sank|threw)"
token article: "(a|an|the)"
token adjective: "(blue|red|green)"
rule sentence: noun_phrase verb noun_phrase
rule noun_phrase: [article] adjective* noun
</PRE>
The tokens are specified as Python <EM>regular expressions</EM>. Since
Yapps produces Python code, you can write any regular expression that
would be accepted by Python. (<EM>Note:</EM> These are Python 1.5
regular expressions from the <TT>re</TT> module, not Python 1.4
regular expressions from the <TT>regex</TT> module.) In addition to
tokens that you want to see (which are given names), you can also
specify tokens to ignore, marked by the <TT>ignore</TT> keyword. In
this parser we want to ignore whitespace.<BR>
<BR>
The TinyEnglish grammar shows how you define tokens and rules, but it
does not specify what should happen once we've matched the rules. In
the next example, we will take a grammar and produce a <EM>parse
tree</EM> from it.<BR>
<BR>
<!--TOC subsection Lisp Expressions-->
<H3 CLASS="subsection"><A NAME="htoc4">2.2</A>&nbsp;&nbsp;Lisp Expressions</H3><!--SEC END -->
Lisp syntax, although hated by many, has a redeeming quality: it is
simple to parse. In this section we will construct a Yapps grammar to
parse Lisp expressions and produce a parse tree as output.<BR>
<BR>
<!--TOC subsubsection Defining the Grammar-->
<H4 CLASS="subsubsection">Defining the Grammar</H4><!--SEC END -->
The syntax of Lisp is simple. It has expressions, which are
identifiers, strings, numbers, and lists. A list is a left
parenthesis followed by some number of expressions (separated by
spaces) followed by a right parenthesis. For example, <CODE>5</CODE>,
<CODE>"ni"</CODE>, and <CODE>(print "1+2 = " (+ 1 2))</CODE> are Lisp expressions.
Written as a grammar,
<PRE CLASS="verbatim">
expr: ID | STR | NUM | list
list: ( expr* )
</PRE>
In addition to having a grammar, we need to specify what to do every
time something is matched. For the tokens, which are strings, we just
want to get the &#8220;value&#8221; of the token, attach its type (identifier,
string, or number) in some way, and return it. For the lists, we want
to construct and return a Python list.<BR>
<BR>
Once some pattern is matched, we enclose a return statement enclosed
in <CODE>{{...}}</CODE>. The braces allow us to insert any one-line
statement into the parser. Within this statement, we can refer to the
values returned by matching each part of the rule. After matching a
token such as <TT>ID</TT>, &#8220;ID&#8221; will be bound to the text of the
matched token. Let's take a look at the rule:
<PRE CLASS="verbatim">
rule expr: ID {{ return ('id', ID) }}
...
</PRE>
In a rule, tokens return the text that was matched. For identifiers,
we just return the identifier, along with a &#8220;tag&#8221; telling us that
this is an identifier and not a string or some other value. Sometimes
we may need to convert this text to a different form. For example, if
a string is matched, we want to remove quotes and handle special forms
like <CODE>\n</CODE>. If a number is matched, we want to convert it into a
number. Let's look at the return values for the other tokens:
<PRE CLASS="verbatim">
...
| STR {{ return ('str', eval(STR)) }}
| NUM {{ return ('num', atoi(NUM)) }}
...
</PRE>
If we get a string, we want to remove the quotes and process any
special backslash codes, so we run <TT>eval</TT> on the quoted string.
If we get a number, we convert it to an integer with <TT>atoi</TT> and
then return the number along with its type tag.<BR>
<BR>
For matching a list, we need to do something slightly more
complicated. If we match a Lisp list of expressions, we want to
create a Python list with those values.
<PRE CLASS="verbatim">
rule list: "\\(" # Match the opening parenthesis
{{ result = [] }} # Create a Python list
(
expr # When we match an expression,
{{ result.append(expr) }} # add it to the list
)* # * means repeat this if needed
"\\)" # Match the closing parenthesis
{{ return result }} # Return the Python list
</PRE>
In this rule we first match the opening parenthesis, then go into a
loop. In this loop we match expressions and add them to the list.
When there are no more expressions to match, we match the closing
parenthesis and return the resulting. Note that <CODE>#</CODE> is used for
comments, just as in Python.<BR>
<BR>
The complete grammar is specified as follows:
<PRE CLASS="verbatim">
parser Lisp:
ignore: '\\s+'
token NUM: '[0-9]+'
token ID: '[-+*/!@%^&amp;=.a-zA-Z0-9_]+'
token STR: '"([^\\"]+|\\\\.)*"'
rule expr: ID {{ return ('id', ID) }}
| STR {{ return ('str', eval(STR)) }}
| NUM {{ return ('num', atoi(NUM)) }}
| list {{ return list }}
rule list: "\\(" {{ result = [] }}
( expr {{ result.append(expr) }}
)*
"\\)" {{ return result }}
</PRE>
One thing you may have noticed is that <CODE>"\\("</CODE> and <CODE>"\\)"</CODE>
appear in the <TT>list</TT> rule. These are <EM>inline tokens</EM>:
they appear in the rules without being given a name with the
<TT>token</TT> keyword. Inline tokens are more convenient to use, but
since they do not have a name, the text that is matched cannot be used
in the return value. They are best used for short simple patterns
(usually punctuation or keywords).<BR>
<BR>
Another thing to notice is that the number and identifier tokens
overlap. For example, &#8220;487&#8221; matches both NUM and ID. In Yapps, the
scanner only tries to match tokens that are acceptable to the parser.
This rule doesn't help here, since both NUM and ID can appear in the
same place in the grammar. There are two rules used to pick tokens if
more than one matches. One is that the <EM>longest</EM> match is
preferred. For example, &#8220;487x&#8221; will match as an ID (487x) rather
than as a NUM (487) followed by an ID (x). The second rule is that if
the two matches are the same length, the <EM>first</EM> one listed in
the grammar is preferred. For example, &#8220;487&#8221; will match as an NUM
rather than an ID because NUM is listed first in the grammar. Inline
tokens have preference over any tokens you have listed.<BR>
<BR>
Now that our grammar is defined, we can run Yapps to produce a parser,
and then run the parser to produce a parse tree.<BR>
<BR>
<!--TOC subsubsection Running Yapps-->
<H4 CLASS="subsubsection">Running Yapps</H4><!--SEC END -->
In the Yapps module is a function <TT>generate</TT> that takes an
input filename and writes a parser to another file. We can use this
function to generate the Lisp parser, which is assumed to be in
<TT>lisp.g</TT>.
<PRE CLASS="verbatim">
% python
Python 1.5.1 (#1, Sep 3 1998, 22:51:17) [GCC 2.7.2.3] on linux-i386
Copyright 1991-1995 Stichting Mathematisch Centrum, Amsterdam
&gt;&gt;&gt; import yapps
&gt;&gt;&gt; yapps.generate('lisp.g')
</PRE>
At this point, Yapps has written a file <TT>lisp.py</TT> that contains
the parser. In that file are two classes (one scanner and one parser)
and a function (called <TT>parse</TT>) that puts things together for
you.<BR>
<BR>
Alternatively, we can run Yapps from the command line to generate the
parser file:
<PRE CLASS="verbatim">
% python yapps.py lisp.g
</PRE>
After running Yapps either from within Python or from the command
line, we can use the Lisp parser by calling the <TT>parse</TT>
function. The first parameter should be the rule we want to match,
and the second parameter should be the string to parse.
<PRE CLASS="verbatim">
&gt;&gt;&gt; import lisp
&gt;&gt;&gt; lisp.parse('expr', '(+ 3 4)')
[('id', '+'), ('num', 3), ('num', 4)]
&gt;&gt;&gt; lisp.parse('expr', '(print "3 = " (+ 1 2))')
[('id', 'print'), ('str', '3 = '), [('id', '+'), ('num', 1), ('num', 2)]]
</PRE>
The <TT>parse</TT> function is not the only way to use the parser;
section <A HREF="#sec:Parser-Objects">5.1</A> describes how to access parser objects
directly.<BR>
<BR>
We've now gone through the steps in creating a grammar, writing a
grammar file for Yapps, producing a parser, and using the parser. In
the next example we'll see how rules can take parameters and also how
to do computations instead of just returning a parse tree.<BR>
<BR>
<!--TOC subsection Calculator-->
<H3 CLASS="subsection"><A NAME="htoc5">2.3</A>&nbsp;&nbsp;Calculator</H3><!--SEC END -->
A common example parser given in many textbooks is that for simple
expressions, with numbers, addition, subtraction, multiplication,
division, and parenthesization of subexpressions. We'll write this
example in Yapps, evaluating the expression as we parse.<BR>
<BR>
Unlike <TT>yacc</TT>, Yapps does not have any way to specify
precedence rules, so we have to do it ourselves. We say that an
expression is the sum of terms, and that a term is the product of
factors, and that a factor is a number or a parenthesized expression:
<PRE CLASS="verbatim">
expr: factor ( ("+"|"-") factor )*
factor: term ( ("*"|"/") term )*
term: NUM | "(" expr ")"
</PRE>
In order to evaluate the expression as we go, we should keep along an
accumulator while evaluating the lists of terms or factors. Just as
we kept a &#8220;result&#8221; variable to build a parse tree for Lisp
expressions, we will use a variable to evaluate numerical
expressions. The full grammar is given below:
<PRE CLASS="verbatim">
parser Calculator:
token END: "$" # $ means end of string
token NUM: "[0-9]+"
rule goal: expr END {{ return expr }}
# An expression is the sum and difference of factors
rule expr: factor {{ v = factor }}
( "[+]" factor {{ v = v+factor }}
| "-" factor {{ v = v-factor }}
)* {{ return v }}
# A factor is the product and division of terms
rule factor: term {{ v = term }}
( "[*]" term {{ v = v*term }}
| "/" term {{ v = v/term }}
)* {{ return v }}
# A term is either a number or an expression surrounded by parentheses
rule term: NUM {{ return atoi(NUM) }}
| "\\(" expr "\\)" {{ return expr }}
</PRE>
The top-level rule is <EM>goal</EM>, which says that we are looking for
an expression followed by the end of the string. The <TT>END</TT>
token is needed because without it, it isn't clear when to stop
parsing. For example, the string &#8220;1+3&#8221; could be parsed either as
the expression &#8220;1&#8221; followed by the string &#8220;+3&#8221; or it could be
parsed as the expression &#8220;1+3&#8221;. By requiring expressions to end
with <TT>END</TT>, the parser is forced to take &#8220;1+3&#8221;.<BR>
<BR>
In the two rules with repetition, the accumulator is named <TT>v</TT>.
After reading in one expression, we initialize the accumulator. Each
time through the loop, we modify the accumulator by adding,
subtracting, multiplying by, or dividing the previous accumulator by
the expression that has been parsed. At the end of the rule, we
return the accumulator.<BR>
<BR>
The calculator example shows how to process lists of elements using
loops, as well as how to handle precedence of operators.<BR>
<BR>
<EM>Note:</EM> It's often important to put the <TT>END</TT> token in, so
put it in unless you are sure that your grammar has some other
non-ambiguous token marking the end of the program.<BR>
<BR>
<!--TOC subsection Calculator with Memory-->
<H3 CLASS="subsection"><A NAME="htoc6">2.4</A>&nbsp;&nbsp;Calculator with Memory</H3><!--SEC END -->
In the previous example we learned how to write a calculator that
evaluates simple numerical expressions. In this section we will
extend the example to support both local and global variables.<BR>
<BR>
To support global variables, we will add assignment statements to the
&#8220;goal&#8221; rule.
<PRE CLASS="verbatim">
rule goal: expr END {{ return expr }}
| 'set' ID expr END {{ global_vars[ID] = expr }}
{{ return expr }}
</PRE>
To use these variables, we need a new kind of terminal:
<PRE CLASS="verbatim">
rule term: ... | ID {{ return global_vars[ID] }}
</PRE>
So far, these changes are straightforward. We simply have a global
dictionary <TT>global_vars</TT> that stores the variables and values,
we modify it when there is an assignment statement, and we look up
variables in it when we see a variable name.<BR>
<BR>
To support local variables, we will add variable declarations to the
set of allowed expressions.
<PRE CLASS="verbatim">
rule term: ... | 'let' VAR '=' expr 'in' expr ...
</PRE>
This is where it becomes tricky. Local variables should be stored in
a local dictionary, not in the global one. One trick would be to save
a copy of the global dictionary, modify it, and then restore it
later. In this example we will instead use <EM>attributes</EM> to
create local information and pass it to subrules.<BR>
<BR>
A rule can optionally take parameters. When we invoke the rule, we
must pass in arguments. For local variables, let's use a single
parameter, <TT>local_vars</TT>:
<PRE CLASS="verbatim">
rule expr&lt;&lt;local_vars&gt;&gt;: ...
rule factor&lt;&lt;local_vars&gt;&gt;: ...
rule term&lt;&lt;local_vars&gt;&gt;: ...
</PRE>
Each time we want to match <TT>expr</TT>, <TT>factor</TT>, or
<TT>term</TT>, we will pass the local variables in the current rule to
the subrule. One interesting case is when we pass as an argument
something <EM>other</EM> than <TT>local_vars</TT>:
<PRE CLASS="verbatim">
rule term&lt;&lt;local_vars&gt;&gt;: ...
| 'let' VAR '=' expr&lt;&lt;local_vars&gt;&gt;
{{ local_vars = [(VAR, expr)] + local_vars }}
'in' expr&lt;&lt;local_vars&gt;&gt;
{{ return expr }}
</PRE>
Note that the assignment to the local variables list does not modify
the original list. This is important to keep local variables from
being seen outside the &#8220;let&#8221;.<BR>
<BR>
The other interesting case is when we find a variable:
<PRE CLASS="verbatim">
global_vars = {}
def lookup(map, name):
for x,v in map: if x==name: return v
return global_vars[name]
%%
...
rule term&lt;&lt;local_vars&gt;: ...
| VAR {{ return lookup(local_vars, VAR) }}
</PRE>
The lookup function will search through the local variable list, and
if it cannot find the name there, it will look it up in the global
variable dictionary.<BR>
<BR>
A complete grammar for this example, including a read-eval-print loop
for interacting with the calculator, can be found in the examples
subdirectory included with Yapps.<BR>
<BR>
In this section we saw how to insert code before the parser. We also
saw how to use attributes to transmit local information from one rule
to its subrules.<BR>
<BR>
<!--TOC section Grammars-->
<H2 CLASS="section"><A NAME="htoc7">3</A>&nbsp;&nbsp;Grammars</H2><!--SEC END -->
Each Yapps grammar has a name, a list of tokens, and a set of
production rules. A grammar named <TT>X</TT> will be used to produce
a parser named <TT>X</TT> and a scanner anmed <TT>XScanner</TT>. As
in Python, names are case sensitive, start with a letter, and contain
letters, numbers, and underscores (_).<BR>
<BR>
There are three kinds of tokens in Yapps: named, inline, and ignored.
As their name implies, named tokens are given a name, using the token
construct: <TT>token <EM>name</EM> : <EM>regexp</EM></TT>. In a rule, the
token can be matched by using the name. Inline tokens are regular
expressions that are used in rules without being declared. Ignored
tokens are declared using the ignore construct: <TT>ignore:
<EM>regexp</EM></TT>. These tokens are ignored by the scanner, and are
not seen by the parser. Often whitespace is an ignored token. The
regular expressions used to define tokens should use the syntax
defined in the <TT>re</TT> module, so some symbols may have to be
backslashed.<BR>
<BR>
Production rules in Yapps have a name and a pattern to match. If the
rule is parameterized, the name should be followed by a list of
parameter names in <CODE>&lt;&lt;...&gt;&gt;</CODE>. A pattern can be a simple pattern
or a compound pattern. Simple patterns are the name of a named token,
a regular expression in quotes (inline token), the name of a
production rule (followed by arguments in <CODE>&lt;&lt;...&gt;&gt;</CODE>, if the rule
has parameters), and single line Python statements (<CODE>{{...}}</CODE>).
Compound patterns are sequences (<CODE>A B C ...</CODE>), choices (
<CODE>A | B | C | ...</CODE>), options (<CODE>[...]</CODE>), zero-or-more repetitions
(<CODE>...*</CODE>), and one-or-more repetitions (<CODE>...+</CODE>). Like
regular expressions, repetition operators have a higher precedence
than sequences, and sequences have a higher precedence than choices.<BR>
<BR>
Whenever <CODE>{{...}}</CODE> is used, a legal one-line Python statement
should be put inside the braces. The token <CODE>}}</CODE> should not
appear within the <CODE>{{...}}</CODE> section, even within a string, since
Yapps does not attempt to parse the Python statement. A workaround
for strings is to put two strings together (<CODE>"}" "}"</CODE>), or to use
backslashes (<CODE>"}\}"</CODE>). At the end of a rule you should use a
<CODE>{{ return X }}</CODE> statement to return a value. However, you
should <EM>not</EM> use any control statements (<TT>return</TT>,
<TT>continue</TT>, <TT>break</TT>) in the middle of a rule. Yapps
needs to make assumptions about the control flow to generate a parser,
and any changes to the control flow will confuse Yapps.<BR>
<BR>
The <CODE>&lt;&lt;...&gt;&gt;</CODE> form can occur in two places: to define parameters
to a rule and to give arguments when matching a rule. Parameters use
the syntax used for Python functions, so they can include default
arguments and the special forms (<CODE>*args</CODE> and <CODE>**kwargs</CODE>).
Arguments use the syntax for Python function call arguments, so they
can include normal arguments and keyword arguments. The token
<CODE>&gt;&gt;</CODE> should not appear within the <CODE>&lt;&lt;...&gt;&gt;</CODE> section.<BR>
<BR>
In both the statements and rule arguments, you can use names defined
by the parser to refer to matched patterns. You can refer to the text
matched by a named token by using the token name. You can use the
value returned by a production rule by using the name of that rule.
If a name <TT>X</TT> is matched more than once (such as in loops), you
will have to save the earlier value(s) in a temporary variable, and
then use that temporary variable in the return value. The next
section has an example of a name that occurs more than once.<BR>
<BR>
<!--TOC subsection Left Factoring-->
<H3 CLASS="subsection"><A NAME="htoc8">3.1</A>&nbsp;&nbsp;Left Factoring</H3><!--SEC END -->
<A NAME="sec:Left-Factoring"></A>
Yapps produces ELL(1) parsers, which determine which clause to match
based on the first token available. Sometimes the leftmost tokens of
several clauses may be the same. The classic example is the
<EM>if/then/else</EM> construct in Pascal:
<PRE CLASS="verbatim">
rule stmt: "if" expr "then" stmt {{ then_part = stmt }}
"else" stmt {{ return ('If',expr,then_part,stmt) }}
| "if" expr "then" stmt {{ return ('If',expr,stmt,[]) }}
</PRE>
(Note that we have to save the first <TT>stmt</TT> into a variable
because there is another <TT>stmt</TT> that will be matched.) The
left portions of the two clauses are the same, which presents a
problem for the parser. The solution is <EM>left-factoring</EM>: the
common parts are put together, and <EM>then</EM> a choice is made about
the remaining part:
<PRE CLASS="verbatim">
rule stmt: "if" expr
"then" stmt {{ then_part = stmt }}
{{ else_part = [] }}
[ "else" stmt {{ else_part = stmt }} ]
{{ return ('If', expr, then_part, else_part) }}
</PRE>
Unfortunately, the classic <EM>if/then/else</EM> situation is
<EM>still</EM> ambiguous when you left-factor. Yapps can deal with this
situation, but will report a warning; see section
<A HREF="#sec:Ambiguous-Grammars">3.3</A> for details.<BR>
<BR>
In general, replace rules of the form:
<PRE CLASS="verbatim">
rule A: a b1 {{ return E1 }}
| a b2 {{ return E2 }}
| c3 {{ return E3 }}
| c4 {{ return E4 }}
</PRE>
with rules of the form:
<PRE CLASS="verbatim">
rule A: a ( b1 {{ return E1 }}
| b2 {{ return E2 }}
)
| c3 {{ return E3 }}
| c4 {{ return E4 }}
</PRE>
<!--TOC subsection Left Recursion-->
<H3 CLASS="subsection"><A NAME="htoc9">3.2</A>&nbsp;&nbsp;Left Recursion</H3><!--SEC END -->
A common construct in grammars is for matching a list of patterns,
sometimes separated with delimiters such as commas or semicolons. In
LR-based parser systems, we can parse a list with something like this:
<PRE CLASS="verbatim">
rule sum: NUM {{ return NUM }}
| sum "+" NUM {{ return (sum, NUM) }}
</PRE>
Parsing <TT>1+2+3+4</TT> would produce the output
<TT>(((1,2),3),4)</TT>, which is what we want from a left-associative
addition operator. Unfortunately, this grammar is <EM>left
recursive,</EM> because the <TT>sum</TT> rule contains a clause that
begins with <TT>sum</TT>. (The recursion occurs at the left side of
the clause.)<BR>
<BR>
We must restructure this grammar to be <EM>right recursive</EM> instead:
<PRE CLASS="verbatim">
rule sum: NUM {{ return NUM }}
| NUM "+" sum {{ return (NUM, sum) }}
</PRE>
Unfortunately, using this grammar, <TT>1+2+3+4</TT> would be parsed as
<TT>(1,(2,(3,4)))</TT>, which no longer follows left associativity.
The rule also needs to be left-factored. Instead, we write the
pattern as a loop instead:
<PRE CLASS="verbatim">
rule sum: NUM {{ v = NUM }}
( "[+]" NUM {{ v = (v,NUM) }} )*
{{ return v }}
</PRE>
In general, replace rules of the form:
<PRE CLASS="verbatim">
rule A: A a1 -&gt; &lt;&lt; E1 &gt;&gt;
| A a2 -&gt; &lt;&lt; E2 &gt;&gt;
| b3 -&gt; &lt;&lt; E3 &gt;&gt;
| b4 -&gt; &lt;&lt; E4 &gt;&gt;
</PRE>
with rules of the form:
<PRE CLASS="verbatim">
rule A: ( b3 {{ A = E3 }}
| b4 {{ A = E4 }} )
( a1 {{ A = E1 }}
| a2 {{ A = E2 }} )*
{{ return A }}
</PRE>
We have taken a rule that proved problematic for with recursion and
turned it into a rule that works well with looping constructs.<BR>
<BR>
<!--TOC subsection Ambiguous Grammars-->
<H3 CLASS="subsection"><A NAME="htoc10">3.3</A>&nbsp;&nbsp;Ambiguous Grammars</H3><!--SEC END -->
<A NAME="sec:Ambiguous-Grammars"></A>
In section <A HREF="#sec:Left-Factoring">3.1</A> we saw the classic if/then/else
ambiguity, which occurs because the &#8220;else ...&#8221; portion of an &#8220;if
...then ...else ...&#8221; construct is optional. Programs with
nested if/then/else constructs can be ambiguous when one of the else
clauses is missing:
<PRE CLASS="verbatim">
if 1 then if 1 then
if 5 then if 5 then
x := 1; x := 1;
else else
y := 9; y := 9;
</PRE>
The indentation shows that the program can be parsed in two different
ways. (Of course, if we all would adopt Python's indentation-based
structuring, this would never happen!) Usually we want the parsing on
the left: the &#8220;else&#8221; should be associated with the closest &#8220;if&#8221;
statement. In section <A HREF="#sec:Left-Factoring">3.1</A> we &#8220;solved&#8221; the
problem by using the following grammar:
<PRE CLASS="verbatim">
rule stmt: "if" expr
"then" stmt {{ then_part = stmt }}
{{ else_part = [] }}
[ "else" stmt {{ else_part = stmt }} ]
{{ return ('If', expr, then_part, else_part) }}
</PRE>
Here, we have an optional match of &#8220;else&#8221; followed by a statement.
The ambiguity is that if an &#8220;else&#8221; is present, it is not clear
whether you want it parsed immediately or if you want it to be parsed
by the outer &#8220;if&#8221;.<BR>
<BR>
Yapps will deal with the situation by matching when the else pattern
when it can. The parser will work in this case because it prefers the
<EM>first</EM> matching clause, which tells Yapps to parse the &#8220;else&#8221;.
That is exactly what we want!<BR>
<BR>
For ambiguity cases with choices, Yapps will choose the <EM>first</EM>
matching choice. However, remember that Yapps only looks at the first
token to determine its decision, so <TT>(a b | a c)</TT> will result in
Yapps choosing <TT>a b</TT> even when the input is <TT>a c</TT>. It only
looks at the first token, <TT>a</TT>, to make its decision.<BR>
<BR>
<!--TOC section Customization-->
<H2 CLASS="section"><A NAME="htoc11">4</A>&nbsp;&nbsp;Customization</H2><!--SEC END -->
Both the parsers and the scanners can be customized. The parser is
usually extended by subclassing, and the scanner can either be
subclassed or completely replaced.<BR>
<BR>
<!--TOC subsection Customizing Parsers-->
<H3 CLASS="subsection"><A NAME="htoc12">4.1</A>&nbsp;&nbsp;Customizing Parsers</H3><!--SEC END -->
If additional fields and methods are needed in order for a parser to
work, Python subclassing can be used. (This is unlike parser classes
written in static languages, in which these fields and methods must be
defined in the generated parser class.) We simply subclass the
generated parser, and add any fields or methods required. Expressions
in the grammar can call methods of the subclass to perform any actions
that cannot be expressed as a simple expression. For example,
consider this simple grammar:
<PRE CLASS="verbatim">
parser X:
rule goal: "something" {{ self.printmsg() }}
</PRE>
The <TT>printmsg</TT> function need not be implemented in the parser
class <TT>X</TT>; it can be implemented in a subclass:
<PRE CLASS="verbatim">
import Xparser
class MyX(Xparser.X):
def printmsg(self):
print "Hello!"
</PRE>
<!--TOC subsection Customizing Scanners-->
<H3 CLASS="subsection"><A NAME="htoc13">4.2</A>&nbsp;&nbsp;Customizing Scanners</H3><!--SEC END -->
The generated parser class is not dependent on the generated scanner
class. A scanner object is passed to the parser object's constructor
in the <TT>parse</TT> function. To use a different scanner, write
your own function to construct parser objects, with an instance of a
different scanner. Scanner objects must have a <TT>token</TT> method
that accepts an integer <TT>N</TT> as well as a list of allowed token
types, and returns the Nth token, as a tuple. The default scanner
raises <TT>NoMoreTokens</TT> if no tokens are available, and
<TT>SyntaxError</TT> if no token could be matched. However, the
parser does not rely on these exceptions; only the <TT>parse</TT>
convenience function (which calls <TT>wrap_error_reporter</TT>) and
the <TT>print_error</TT> error display function use those exceptions.<BR>
<BR>
The tuples representing tokens have four elements. The first two are
the beginning and ending indices of the matched text in the input
string. The third element is the type tag, matching either the name
of a named token or the quoted regexp of an inline or ignored token.
The fourth element of the token tuple is the matched text. If the
input string is <TT>s</TT>, and the token tuple is
<TT>(b,e,type,val)</TT>, then <TT>val</TT> should be equal to
<TT>s[b:e]</TT>.<BR>
<BR>
The generated parsers do not the beginning or ending index. They use
only the token type and value. However, the default error reporter
uses the beginning and ending index to show the user where the error
is.<BR>
<BR>
<!--TOC section Parser Mechanics-->
<H2 CLASS="section"><A NAME="htoc14">5</A>&nbsp;&nbsp;Parser Mechanics</H2><!--SEC END -->
The base parser class (Parser) defines two methods, <TT>_scan</TT>
and <TT>_peek</TT>, and two fields, <TT>_pos</TT> and
<TT>_scanner</TT>. The generated parser inherits from the base
parser, and contains one method for each rule in the grammar. To
avoid name clashes, do not use names that begin with an underscore
(<TT>_</TT>).<BR>
<BR>
<!--TOC subsection Parser Objects-->
<H3 CLASS="subsection"><A NAME="htoc15">5.1</A>&nbsp;&nbsp;Parser Objects</H3><!--SEC END -->
<A NAME="sec:Parser-Objects"></A>
Yapps produces as output two exception classes, a scanner class, a
parser class, and a function <TT>parse</TT> that puts everything
together. The <TT>parse</TT> function does not have to be used;
instead, one can create a parser and scanner object and use them
together for parsing.
<PRE CLASS="verbatim">
def parse(rule, text):
P = X(XScanner(text))
return wrap_error_reporter(P, rule)
</PRE>
The <TT>parse</TT> function takes a name of a rule and an input string
as input. It creates a scanner and parser object, then calls
<TT>wrap_error_reporter</TT> to execute the method in the parser
object named <TT>rule</TT>. The wrapper function will call the
appropriate parser rule and report any parsing errors to standard
output.<BR>
<BR>
There are several situations in which the <TT>parse</TT> function
would not be useful. If a different parser or scanner is being used,
or exceptions are to be handled differently, a new <TT>parse</TT>
function would be required. The supplied <TT>parse</TT> function can
be used as a template for writing a function for your own needs. An
example of a custom parse function is the <TT>generate</TT> function
in <TT>Yapps.py</TT>.<BR>
<BR>
<!--TOC subsection Context Sensitive Scanner-->
<H3 CLASS="subsection"><A NAME="htoc16">5.2</A>&nbsp;&nbsp;Context Sensitive Scanner</H3><!--SEC END -->
Unlike most scanners, the scanner produced by Yapps can take into
account the context in which tokens are needed, and try to match only
good tokens. For example, in the grammar:
<PRE CLASS="verbatim">
parser IniFile:
token ID: "[a-zA-Z_0-9]+"
token VAL: ".*"
rule pair: ID "[ \t]*=[ \t]*" VAL "\n"
</PRE>
we would like to scan lines of text and pick out a name/value pair.
In a conventional scanner, the input string <TT>shell=progman.exe</TT>
would be turned into a single token of type <TT>VAL</TT>. The Yapps
scanner, however, knows that at the beginning of the line, an
<TT>ID</TT> is expected, so it will return <TT>"shell"</TT> as a token
of type <TT>ID</TT>. Later, it will return <TT>"progman.exe"</TT> as
a token of type <TT>VAL</TT>.<BR>
<BR>
Context sensitivity decreases the separation between scanner and
parser, but it is useful in parsers like <TT>IniFile</TT>, where the
tokens themselves are not unambiguous, but <EM>are</EM> unambiguous
given a particular stage in the parsing process.<BR>
<BR>
Unfortunately, context sensitivity can make it more difficult to
detect errors in the input. For example, in parsing a Pascal-like
language with &#8220;begin&#8221; and &#8220;end&#8221; as keywords, a context sensitive
scanner would only match &#8220;end&#8221; as the END token if the parser is in
a place that will accept the END token. If not, then the scanner
would match &#8220;end&#8221; as an identifier. To disable the context
sensitive scanner in Yapps, add the
<TT>context-insensitive-scanner</TT> option to the grammar:
<PRE CLASS="verbatim">
Parser X:
option: "context-insensitive-scanner"
</PRE>
Context-insensitive scanning makes the parser look cleaner as well.<BR>
<BR>
<!--TOC subsection Internal Variables-->
<H3 CLASS="subsection"><A NAME="htoc17">5.3</A>&nbsp;&nbsp;Internal Variables</H3><!--SEC END -->
There are two internal fields that may be of use. The parser object
has two fields, <TT>_pos</TT>, which is the index of the current
token being matched, and <TT>_scanner</TT>, which is the scanner
object. The token itself can be retrieved by accessing the scanner
object and calling the <TT>token</TT> method with the token index. However, if you call <TT>token</TT> before the token has been requested by the parser, it may mess up a context-sensitive scanner.<SUP><A NAME="text1" HREF="#note1">1</A></SUP> A
potentially useful combination of these fields is to extract the
portion of the input matched by the current rule. To do this, just save the scanner state (<TT>_scanner.pos</TT>) before the text is matched and then again after the text is matched:
<PRE CLASS="verbatim">
rule R:
{{ start = self._scanner.pos }}
a b c
{{ end = self._scanner.pos }}
{{ print 'Text is', self._scanner.input[start:end] }}
</PRE>
<!--TOC subsection Pre- and Post-Parser Code-->
<H3 CLASS="subsection"><A NAME="htoc18">5.4</A>&nbsp;&nbsp;Pre- and Post-Parser Code</H3><!--SEC END -->
Sometimes the parser code needs to rely on helper variables,
functions, and classes. A Yapps grammar can optionally be surrounded
by double percent signs, to separate the grammar from Python code.
<PRE CLASS="verbatim">
... Python code ...
%%
... Yapps grammar ...
%%
... Python code ...
</PRE>
The second <CODE>%%</CODE> can be omitted if there is no Python code at the
end, and the first <CODE>%%</CODE> can be omitted if there is no extra
Python code at all. (To have code only at the end, both separators
are required.)<BR>
<BR>
If the second <CODE>%%</CODE> is omitted, Yapps will insert testing code
that allows you to use the generated parser to parse a file.<BR>
<BR>
The extended calculator example in the Yapps examples subdirectory
includes both pre-parser and post-parser code.<BR>
<BR>
<!--TOC subsection Representation of Grammars-->
<H3 CLASS="subsection"><A NAME="htoc19">5.5</A>&nbsp;&nbsp;Representation of Grammars</H3><!--SEC END -->
For each kind of pattern there is a class derived from Pattern. Yapps
has classes for Terminal, NonTerminal, Sequence, Choice, Option, Plus,
Star, and Eval. Each of these classes has the following interface:
<UL CLASS="itemize"><LI CLASS="li-itemize">
setup(<EM>gen</EM>) Set accepts-&#1108;, and call
<EM>gen.changed()</EM> if it changed. This function can change the
flag from false to true but <EM>not</EM> from true to false.
<LI CLASS="li-itemize">update(<EM>(</EM>gen)) Set <SPAN STYLE="font-variant:small-caps">first</SPAN>and <SPAN STYLE="font-variant:small-caps">follow</SPAN>, and call
<EM>gen.changed()</EM> if either changed. This function can add to
the sets but <EM>not</EM> remove from them.
<LI CLASS="li-itemize">output(<EM>gen</EM>, <EM>indent</EM>) Generate code for matching
this rule, using <EM>indent</EM> as the current indentation level.
Writes are performed using <EM>gen.write</EM>.
<LI CLASS="li-itemize">used(<EM>vars</EM>) Given a list of variables <EM>vars</EM>,
return two lists: one containing the variables that are used, and
one containing the variables that are assigned. This function is
used for optimizing the resulting code.
</UL>
Both <EM>setup</EM> and <EM>update</EM> monotonically increase the
variables they modify. Since the variables can only increase a finite
number of times, we can repeatedly call the function until the
variable stabilized. The <EM>used</EM> function is not currently
implemented.<BR>
<BR>
With each pattern in the grammar Yapps associates three pieces of
information: the <SPAN STYLE="font-variant:small-caps">first</SPAN>set, the <SPAN STYLE="font-variant:small-caps">follow</SPAN>set, and the
accepts-&#1108; flag.<BR>
<BR>
The <SPAN STYLE="font-variant:small-caps">first</SPAN>set contains the tokens that can appear as we start
matching the pattern. The <SPAN STYLE="font-variant:small-caps">follow</SPAN>set contains the tokens that can
appear immediately after we match the pattern. The accepts-&#1108;
flag is true if the pattern can match no tokens. In this case, <SPAN STYLE="font-variant:small-caps">first</SPAN>will contain all the elements in <SPAN STYLE="font-variant:small-caps">follow</SPAN>. The <SPAN STYLE="font-variant:small-caps">follow</SPAN>set is not
needed when accepts-&#1108; is false, and may not be accurate in
those cases.<BR>
<BR>
Yapps does not compute these sets precisely. Its approximation can
miss certain cases, such as this one:
<PRE CLASS="verbatim">
rule C: ( A* | B )
rule B: C [A]
</PRE>
Yapps will calculate <TT>C</TT>'s <SPAN STYLE="font-variant:small-caps">follow</SPAN>set to include <TT>A</TT>.
However, <TT>C</TT> will always match all the <TT>A</TT>'s, so <TT>A</TT> will
never follow it. Yapps 2.0 does not properly handle this construct,
but if it seems important, I may add support for it in a future
version.<BR>
<BR>
Yapps also cannot handle constructs that depend on the calling
sequence. For example:
<PRE CLASS="verbatim">
rule R: U | 'b'
rule S: | 'c'
rule T: S 'b'
rule U: S 'a'
</PRE>
The <SPAN STYLE="font-variant:small-caps">follow</SPAN>set for <TT>S</TT> includes <TT>a</TT> and <TT>b</TT>. Since <TT>S</TT> can be empty, the <SPAN STYLE="font-variant:small-caps">first</SPAN>set for <TT>S</TT> should include <TT>a</TT>,
<TT>b</TT>, and <TT>c</TT>. However, when parsing <TT>R</TT>, if the lookahead
is <TT>b</TT> we should <EM>not</EM> parse <TT>U</TT>. That's because in <TT>U</TT>, <TT>S</TT> is followed by <TT>a</TT> and not <TT>b</TT>. Therefore in
<TT>R</TT>, we should choose rule <TT>U</TT> only if there is an <TT>a</TT> or
<TT>c</TT>, but not if there is a <TT>b</TT>. Yapps and many other LL(1)
systems do not distinguish <TT>S b</TT> and <TT>S a</TT>, making <TT>S</TT>'s <SPAN STYLE="font-variant:small-caps">follow</SPAN>set <TT>a, b</TT>, and making <TT>R</TT> always try to match
<TT>U</TT>. In this case we can solve the problem by changing <TT>R</TT> to
<CODE>'b' | U</CODE> but it may not always be possible to solve all such
problems in this way.<BR>
<BR>
<!--TOC section Grammar for Parsers-->
<H2 CLASS="section"><A NAME="htoc20">A</A>&nbsp;&nbsp;Grammar for Parsers</H2><!--SEC END -->
This is the grammar for parsers, without any Python code mixed in.
The complete grammar can be found in <TT>parsedesc.g</TT> in the Yapps
distribution.
<PRE CLASS="verbatim">
parser ParserDescription:
ignore: "\\s+"
ignore: "#.*?\r?\n"
token END: "$" # $ means end of string
token ATTR: "&lt;&lt;.+?&gt;&gt;"
token STMT: "{{.+?}}"
token ID: '[a-zA-Z_][a-zA-Z_0-9]*'
token STR: '[rR]?\'([^\\n\'\\\\]|\\\\.)*\'|[rR]?"([^\\n"\\\\]|\\\\.)*"'
rule Parser: "parser" ID ":"
Options
Tokens
Rules
END
rule Options: ( "option" ":" STR )*
rule Tokens: ( "token" ID ":" STR | "ignore" ":" STR )*
rule Rules: ( "rule" ID OptParam ":" ClauseA )*
rule ClauseA: ClauseB ( '[|]' ClauseB )*
rule ClauseB: ClauseC*
rule ClauseC: ClauseD [ '[+]' | '[*]' ]
rule ClauseD: STR | ID [ATTR] | STMT
| '\\(' ClauseA '\\) | '\\[' ClauseA '\\]'
</PRE>
<!--TOC section Upgrading-->
<H2 CLASS="section"><A NAME="htoc21">B</A>&nbsp;&nbsp;Upgrading</H2><!--SEC END -->
Yapps 2.0 is not backwards compatible with Yapps 1.0. In this section
are some tips for upgrading:
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">
Yapps 1.0 was distributed as a single file. Yapps 2.0 is
instead distributed as two Python files: a <EM>parser generator</EM>
(26k) and a <EM>parser runtime</EM> (5k). You need both files to
create parsers, but you need only the runtime (<TT>yappsrt.py</TT>)
to use the parsers.<BR>
<BR>
<LI CLASS="li-enumerate">Yapps 1.0 supported Python 1.4 regular expressions from the
<TT>regex</TT> module. Yapps 2.0 uses Python 1.5 regular
expressions from the <TT>re</TT> module. <EM>The new syntax for
regular expressions is not compatible with the old syntax.</EM>
Andrew Kuchling has a guide to converting
regular
expressionshttp://www.python.org/doc/howto/regex-to-re/ on his
web page.<BR>
<BR>
<LI CLASS="li-enumerate">Yapps 1.0 wants a pattern and then a return value in <CODE>-&gt;</CODE>
<CODE>&lt;&lt;...&gt;&gt;</CODE>. Yapps 2.0 allows patterns and Python statements to
be mixed. To convert a rule like this:
<PRE CLASS="verbatim">
rule R: A B C -&gt; &lt;&lt; E1 &gt;&gt;
| X Y Z -&gt; &lt;&lt; E2 &gt;&gt;
</PRE>to Yapps 2.0 form, replace the return value specifiers with return
statements:
<PRE CLASS="verbatim">
rule R: A B C {{ return E1 }}
| X Y Z {{ return E2 }}
</PRE><LI CLASS="li-enumerate">Yapps 2.0 does not perform tail recursion elimination. This
means any recursive rules you write will be turned into recursive
methods in the parser. The parser will work, but may be slower.
It can be made faster by rewriting recursive rules, using instead
the looping operators <CODE>*</CODE> and <CODE>+</CODE> provided in Yapps 2.0.</OL>
<!--TOC section Troubleshooting-->
<H2 CLASS="section"><A NAME="htoc22">C</A>&nbsp;&nbsp;Troubleshooting</H2><!--SEC END -->
<UL CLASS="itemize"><LI CLASS="li-itemize">
A common error is to write a grammar that doesn't have an END
token. End tokens are needed when it is not clear when to stop
parsing. For example, when parsing the expression <TT>3+5</TT>, it is
not clear after reading <TT>3</TT> whether to treat it as a complete
expression or whether the parser should continue reading.
Therefore the grammar for numeric expressions should include an end
token. Another example is the grammar for Lisp expressions. In
Lisp, it is always clear when you should stop parsing, so you do
<EM>not</EM> need an end token. In fact, it may be more useful not
to have an end token, so that you can read in several Lisp expressions.
<LI CLASS="li-itemize">If there is a chance of ambiguity, make sure to put the choices
in the order you want them checked. Usually the most specific
choice should be first. Empty sequences should usually be last.
<LI CLASS="li-itemize">The context sensitive scanner is not appropriate for all
grammars. You might try using the insensitive scanner with the
<TT>context-insensitive-scanner</TT> option in the grammar.
<LI CLASS="li-itemize">If performance turns out to be a problem, try writing a custom
scanner. The Yapps scanner is rather slow (but flexible and easy
to understand).
</UL>
<!--TOC section History-->
<H2 CLASS="section"><A NAME="htoc23">D</A>&nbsp;&nbsp;History</H2><!--SEC END -->
Yapps 1 had several limitations that bothered me while writing
parsers:
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">
It was not possible to insert statements into the generated
parser. A common workaround was to write an auxilliary function
that executed those statements, and to call that function as part
of the return value calculation. For example, several of my
parsers had an &#8220;append(x,y)&#8221; function that existed solely to call
&#8220;x.append(y)&#8221;.
<LI CLASS="li-enumerate">The way in which grammars were specified was rather
restrictive: a rule was a choice of clauses. Each clause was a
sequence of tokens and rule names, followed by a return value.
<LI CLASS="li-enumerate">Optional matching had to be put into a separate rule because
choices were only made at the beginning of a rule.
<LI CLASS="li-enumerate">Repetition had to be specified in terms of recursion. Not only
was this awkward (sometimes requiring additional rules), I had to
add a tail recursion optimization to Yapps to transform the
recursion back into a loop.
</OL>
Yapps 2 addresses each of these limitations.
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">
Statements can occur anywhere within a rule. (However, only
one-line statements are allowed; multiline blocks marked by
indentation are not.)
<LI CLASS="li-enumerate">Grammars can be specified using any mix of sequences, choices,
tokens, and rule names. To allow for complex structures,
parentheses can be used for grouping.
<LI CLASS="li-enumerate">Given choices and parenthesization, optional matching can be
expressed as a choice between some pattern and nothing. In
addition, Yapps 2 has the convenience syntax <CODE>[A B ...]</CODE> for
matching <CODE>A B ...</CODE> optionally.
<LI CLASS="li-enumerate">Repetition operators <CODE>*</CODE> for zero or more and <CODE>+</CODE> for
one or more make it easy to specify repeating patterns.
</OL>
It is my hope that Yapps 2 will be flexible enough to meet my needs
for another year, yet simple enough that I do not hesitate to use it.<BR>
<BR>
<!--TOC section Debian Extensions-->
<H2 CLASS="section"><A NAME="htoc24">E</A>&nbsp;&nbsp;Debian Extensions</H2><!--SEC END -->
<A NAME="sec:debian"></A>
The Debian version adds the following enhancements to the original
Yapps code. They were written by Matthias Urlichs.
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">
Yapps can stack input sources ("include files"). A usage example
is supplied with the calc.g sample program.
<LI CLASS="li-enumerate">Yapps now understands augmented ignore-able patterns.
This means that Yapps can parse multi-line C comments; this wasn't
possible before.
<LI CLASS="li-enumerate">Better error reporting.
<LI CLASS="li-enumerate">Yapps now reads its input incrementally.
</OL>
The generated parser has been renamed to <TT>yapps/runtime.py</TT>.
In Debian, this file is provided by the <TT>yapps2-runtime</TT> package.
You need to depend on it if you Debianize Python programs which use
yapps.<BR>
<BR>
<!--TOC section Future Extensions-->
<H2 CLASS="section"><A NAME="htoc25">F</A>&nbsp;&nbsp;Future Extensions</H2><!--SEC END -->
<A NAME="sec:future"></A>
I am still investigating the possibility of LL(2) and higher
lookahead. However, it looks like the resulting parsers will be
somewhat ugly. <BR>
<BR>
It would be nice to control choices with user-defined predicates.<BR>
<BR>
The most likely future extension is backtracking. A grammar pattern
like <CODE>(VAR ':=' expr)? {{ return Assign(VAR,expr) }} : expr {{ return expr }}</CODE>
would turn into code that attempted to match <CODE>VAR ':=' expr</CODE>. If
it succeeded, it would run <CODE>{{ return ... }}</CODE>. If it failed, it
would match <CODE>expr {{ return expr }}</CODE>. Backtracking may make it
less necessary to write LL(2) grammars.<BR>
<BR>
<!--TOC section References-->
<H2 CLASS="section"><A NAME="htoc26">G</A>&nbsp;&nbsp;References</H2><!--SEC END -->
<OL CLASS="enumerate" type=1><LI CLASS="li-enumerate">
The Python-Parser
SIGhttp://www.python.org/sigs/parser-sig/ is the first place
to look for a list of parser systems for Python.<BR>
<BR>
<LI CLASS="li-enumerate">ANTLR/PCCTS, by Terrence Parr, is available at
The ANTLR Home Pagehttp://www.antlr.org/.<BR>
<BR>
<LI CLASS="li-enumerate">PyLR, by Scott Cotton, is at his Starship
pagehttp://starship.skyport.net/crew/scott/PyLR.html.<BR>
<BR>
<LI CLASS="li-enumerate">John Aycock's Compiling Little Languages
Frameworkhttp://www.foretec.com/python/workshops/1998-11/proceedings/papers/aycock-little/aycock-little.html.<BR>
<BR>
<LI CLASS="li-enumerate">PyBison, by Scott Hassan, can be found at
his Python Projects
pagehttp://coho.stanford.edu/~hassan/Python/.<BR>
<BR>
<LI CLASS="li-enumerate">mcf.pars, by Mike C. Fletcher, is available at
his web
pagehttp://members.rogers.com/mcfletch/programming/simpleparse/simpleparse.html.<BR>
<BR>
<LI CLASS="li-enumerate">kwParsing, by Aaron Watters, is available at
his Starship
pagehttp://starship.skyport.net/crew/aaron_watters/kwParsing/.
</OL>
<!--BEGIN NOTES document-->
<HR WIDTH="50%" SIZE=1><DL CLASS="list"><DT CLASS="dt-list"><A NAME="note1" HREF="#text1"><FONT SIZE=5>1</FONT></A><DD CLASS="dd-list">When using a context-sensitive scanner, the parser tells the scanner what the valid token types are at each point. If you call <TT>token</TT> before the parser can tell the scanner the valid token types, the scanner will attempt to match without considering the context.
</DL>
<!--END NOTES-->
<!--HTMLFOOT-->
<!--ENDHTML-->
<!--FOOTER-->
<HR SIZE=2><BLOCKQUOTE CLASS="quote"><EM>This document was translated from L<sup>A</sup>T<sub>E</sub>X by
</EM><A HREF="http://pauillac.inria.fr/~maranget/hevea/index.html"><EM>H<FONT SIZE=2><sup>E</sup></FONT>V<FONT SIZE=2><sup>E</sup></FONT>A</EM></A><EM>.</EM></BLOCKQUOTE></BODY>
</HTML>
Reference in a new issue View git blame Copy permalink