874 lines
28 KiB
HTML
874 lines
28 KiB
HTML
<html>
|
|
<head>
|
|
<title>PLY Internals</title>
|
|
</head>
|
|
<body bgcolor="#ffffff">
|
|
|
|
<h1>PLY Internals</h1>
|
|
|
|
<b>
|
|
David M. Beazley <br>
|
|
dave@dabeaz.com<br>
|
|
</b>
|
|
|
|
<p>
|
|
<b>PLY Version: 3.0</b>
|
|
<p>
|
|
|
|
<!-- INDEX -->
|
|
<div class="sectiontoc">
|
|
<ul>
|
|
<li><a href="#internal_nn1">Introduction</a>
|
|
<li><a href="#internal_nn2">Grammar Class</a>
|
|
<li><a href="#internal_nn3">Productions</a>
|
|
<li><a href="#internal_nn4">LRItems</a>
|
|
<li><a href="#internal_nn5">LRTable</a>
|
|
<li><a href="#internal_nn6">LRGeneratedTable</a>
|
|
<li><a href="#internal_nn7">LRParser</a>
|
|
<li><a href="#internal_nn8">ParserReflect</a>
|
|
<li><a href="#internal_nn9">High-level operation</a>
|
|
</ul>
|
|
</div>
|
|
<!-- INDEX -->
|
|
|
|
|
|
<H2><a name="internal_nn1"></a>1. Introduction</H2>
|
|
|
|
|
|
This document describes classes and functions that make up the internal
|
|
operation of PLY. Using this programming interface, it is possible to
|
|
manually build an parser using a different interface specification
|
|
than what PLY normally uses. For example, you could build a gramar
|
|
from information parsed in a completely different input format. Some of
|
|
these objects may be useful for building more advanced parsing engines
|
|
such as GLR.
|
|
|
|
<p>
|
|
It should be stressed that using PLY at this level is not for the
|
|
faint of heart. Generally, it's assumed that you know a bit of
|
|
the underlying compiler theory and how an LR parser is put together.
|
|
|
|
<H2><a name="internal_nn2"></a>2. Grammar Class</H2>
|
|
|
|
|
|
The file <tt>ply.yacc</tt> defines a class <tt>Grammar</tt> that
|
|
is used to hold and manipulate information about a grammar
|
|
specification. It encapsulates the same basic information
|
|
about a grammar that is put into a YACC file including
|
|
the list of tokens, precedence rules, and grammar rules.
|
|
Various operations are provided to perform different validations
|
|
on the grammar. In addition, there are operations to compute
|
|
the first and follow sets that are needed by the various table
|
|
generation algorithms.
|
|
|
|
<p>
|
|
<tt><b>Grammar(terminals)</b></tt>
|
|
|
|
<blockquote>
|
|
Creates a new grammar object. <tt>terminals</tt> is a list of strings
|
|
specifying the terminals for the grammar. An instance <tt>g</tt> of
|
|
<tt>Grammar</tt> has the following methods:
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.set_precedence(term,assoc,level)</tt></b>
|
|
<blockquote>
|
|
Sets the precedence level and associativity for a given terminal <tt>term</tt>.
|
|
<tt>assoc</tt> is one of <tt>'right'</tt>,
|
|
<tt>'left'</tt>, or <tt>'nonassoc'</tt> and <tt>level</tt> is a positive integer. The higher
|
|
the value of <tt>level</tt>, the higher the precedence. Here is an example of typical
|
|
precedence settings:
|
|
|
|
<pre>
|
|
g.set_precedence('PLUS', 'left',1)
|
|
g.set_precedence('MINUS', 'left',1)
|
|
g.set_precedence('TIMES', 'left',2)
|
|
g.set_precedence('DIVIDE','left',2)
|
|
g.set_precedence('UMINUS','left',3)
|
|
</pre>
|
|
|
|
This method must be called prior to adding any productions to the
|
|
grammar with <tt>g.add_production()</tt>. The precedence of individual grammar
|
|
rules is determined by the precedence of the right-most terminal.
|
|
|
|
</blockquote>
|
|
<p>
|
|
<b><tt>g.add_production(name,syms,func=None,file='',line=0)</tt></b>
|
|
<blockquote>
|
|
Adds a new grammar rule. <tt>name</tt> is the name of the rule,
|
|
<tt>syms</tt> is a list of symbols making up the right hand
|
|
side of the rule, <tt>func</tt> is the function to call when
|
|
reducing the rule. <tt>file</tt> and <tt>line</tt> specify
|
|
the filename and line number of the rule and are used for
|
|
generating error messages.
|
|
|
|
<p>
|
|
The list of symbols in <tt>syms</tt> may include character
|
|
literals and <tt>%prec</tt> specifiers. Here are some
|
|
examples:
|
|
|
|
<pre>
|
|
g.add_production('expr',['expr','PLUS','term'],func,file,line)
|
|
g.add_production('expr',['expr','"+"','term'],func,file,line)
|
|
g.add_production('expr',['MINUS','expr','%prec','UMINUS'],func,file,line)
|
|
</pre>
|
|
|
|
<p>
|
|
If any kind of error is detected, a <tt>GrammarError</tt> exception
|
|
is raised with a message indicating the reason for the failure.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.set_start(start=None)</tt></b>
|
|
<blockquote>
|
|
Sets the starting rule for the grammar. <tt>start</tt> is a string
|
|
specifying the name of the start rule. If <tt>start</tt> is omitted,
|
|
the first grammar rule added with <tt>add_production()</tt> is taken to be
|
|
the starting rule. This method must always be called after all
|
|
productions have been added.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.find_unreachable()</tt></b>
|
|
<blockquote>
|
|
Diagnostic function. Returns a list of all unreachable non-terminals
|
|
defined in the grammar. This is used to identify inactive parts of
|
|
the grammar specification.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.infinite_cycle()</tt></b>
|
|
<blockquote>
|
|
Diagnostic function. Returns a list of all non-terminals in the
|
|
grammar that result in an infinite cycle. This condition occurs if
|
|
there is no way for a grammar rule to expand to a string containing
|
|
only terminal symbols.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.undefined_symbols()</tt></b>
|
|
<blockquote>
|
|
Diagnostic function. Returns a list of tuples <tt>(name, prod)</tt>
|
|
corresponding to undefined symbols in the grammar. <tt>name</tt> is the
|
|
name of the undefined symbol and <tt>prod</tt> is an instance of
|
|
<tt>Production</tt> which has information about the production rule
|
|
where the undefined symbol was used.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.unused_terminals()</tt></b>
|
|
<blockquote>
|
|
Diagnostic function. Returns a list of terminals that were defined,
|
|
but never used in the grammar.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.unused_rules()</tt></b>
|
|
<blockquote>
|
|
Diagnostic function. Returns a list of <tt>Production</tt> instances
|
|
corresponding to production rules that were defined in the grammar,
|
|
but never used anywhere. This is slightly different
|
|
than <tt>find_unreachable()</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.unused_precedence()</tt></b>
|
|
<blockquote>
|
|
Diagnostic function. Returns a list of tuples <tt>(term, assoc)</tt>
|
|
corresponding to precedence rules that were set, but never used the
|
|
grammar. <tt>term</tt> is the terminal name and <tt>assoc</tt> is the
|
|
precedence associativity (e.g., <tt>'left'</tt>, <tt>'right'</tt>,
|
|
or <tt>'nonassoc'</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.compute_first()</tt></b>
|
|
<blockquote>
|
|
Compute all of the first sets for all symbols in the grammar. Returns a dictionary
|
|
mapping symbol names to a list of all first symbols.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.compute_follow()</tt></b>
|
|
<blockquote>
|
|
Compute all of the follow sets for all non-terminals in the grammar.
|
|
The follow set is the set of all possible symbols that might follow a
|
|
given non-terminal. Returns a dictionary mapping non-terminal names
|
|
to a list of symbols.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.build_lritems()</tt></b>
|
|
<blockquote>
|
|
Calculates all of the LR items for all productions in the grammar. This
|
|
step is required before using the grammar for any kind of table generation.
|
|
See the section on LR items below.
|
|
</blockquote>
|
|
|
|
<p>
|
|
The following attributes are set by the above methods and may be useful
|
|
in code that works with the grammar. All of these attributes should be
|
|
assumed to be read-only. Changing their values directly will likely
|
|
break the grammar.
|
|
|
|
<p>
|
|
<b><tt>g.Productions</tt></b>
|
|
<blockquote>
|
|
A list of all productions added. The first entry is reserved for
|
|
a production representing the starting rule. The objects in this list
|
|
are instances of the <tt>Production</tt> class, described shortly.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.Prodnames</tt></b>
|
|
<blockquote>
|
|
A dictionary mapping the names of nonterminals to a list of all
|
|
productions of that nonterminal.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.Terminals</tt></b>
|
|
<blockquote>
|
|
A dictionary mapping the names of terminals to a list of the
|
|
production numbers where they are used.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.Nonterminals</tt></b>
|
|
<blockquote>
|
|
A dictionary mapping the names of nonterminals to a list of the
|
|
production numbers where they are used.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.First</tt></b>
|
|
<blockquote>
|
|
A dictionary representing the first sets for all grammar symbols. This is
|
|
computed and returned by the <tt>compute_first()</tt> method.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.Follow</tt></b>
|
|
<blockquote>
|
|
A dictionary representing the follow sets for all grammar rules. This is
|
|
computed and returned by the <tt>compute_follow()</tt> method.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>g.Start</tt></b>
|
|
<blockquote>
|
|
Starting symbol for the grammar. Set by the <tt>set_start()</tt> method.
|
|
</blockquote>
|
|
|
|
For the purposes of debugging, a <tt>Grammar</tt> object supports the <tt>__len__()</tt> and
|
|
<tt>__getitem__()</tt> special methods. Accessing <tt>g[n]</tt> returns the nth production
|
|
from the grammar.
|
|
|
|
|
|
<H2><a name="internal_nn3"></a>3. Productions</H2>
|
|
|
|
|
|
<tt>Grammar</tt> objects store grammar rules as instances of a <tt>Production</tt> class. This
|
|
class has no public constructor--you should only create productions by calling <tt>Grammar.add_production()</tt>.
|
|
The following attributes are available on a <tt>Production</tt> instance <tt>p</tt>.
|
|
|
|
<p>
|
|
<b><tt>p.name</tt></b>
|
|
<blockquote>
|
|
The name of the production. For a grammar rule such as <tt>A : B C D</tt>, this is <tt>'A'</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.prod</tt></b>
|
|
<blockquote>
|
|
A tuple of symbols making up the right-hand side of the production. For a grammar rule such as <tt>A : B C D</tt>, this is <tt>('B','C','D')</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.number</tt></b>
|
|
<blockquote>
|
|
Production number. An integer containing the index of the production in the grammar's <tt>Productions</tt> list.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.func</tt></b>
|
|
<blockquote>
|
|
The name of the reduction function associated with the production.
|
|
This is the function that will execute when reducing the entire
|
|
grammar rule during parsing.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.callable</tt></b>
|
|
<blockquote>
|
|
The callable object associated with the name in <tt>p.func</tt>. This is <tt>None</tt>
|
|
unless the production has been bound using <tt>bind()</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.file</tt></b>
|
|
<blockquote>
|
|
Filename associated with the production. Typically this is the file where the production was defined. Used for error messages.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.lineno</tt></b>
|
|
<blockquote>
|
|
Line number associated with the production. Typically this is the line number in <tt>p.file</tt> where the production was defined. Used for error messages.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.prec</tt></b>
|
|
<blockquote>
|
|
Precedence and associativity associated with the production. This is a tuple <tt>(assoc,level)</tt> where
|
|
<tt>assoc</tt> is one of <tt>'left'</tt>,<tt>'right'</tt>, or <tt>'nonassoc'</tt> and <tt>level</tt> is
|
|
an integer. This value is determined by the precedence of the right-most terminal symbol in the production
|
|
or by use of the <tt>%prec</tt> specifier when adding the production.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.usyms</tt></b>
|
|
<blockquote>
|
|
A list of all unique symbols found in the production.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.lr_items</tt></b>
|
|
<blockquote>
|
|
A list of all LR items for this production. This attribute only has a meaningful value if the
|
|
<tt>Grammar.build_lritems()</tt> method has been called. The items in this list are
|
|
instances of <tt>LRItem</tt> described below.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.lr_next</tt></b>
|
|
<blockquote>
|
|
The head of a linked-list representation of the LR items in <tt>p.lr_items</tt>.
|
|
This attribute only has a meaningful value if the <tt>Grammar.build_lritems()</tt>
|
|
method has been called. Each <tt>LRItem</tt> instance has a <tt>lr_next</tt> attribute
|
|
to move to the next item. The list is terminated by <tt>None</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.bind(dict)</tt></b>
|
|
<blockquote>
|
|
Binds the production function name in <tt>p.func</tt> to a callable object in
|
|
<tt>dict</tt>. This operation is typically carried out in the last step
|
|
prior to running the parsing engine and is needed since parsing tables are typically
|
|
read from files which only include the function names, not the functions themselves.
|
|
</blockquote>
|
|
|
|
<P>
|
|
<tt>Production</tt> objects support
|
|
the <tt>__len__()</tt>, <tt>__getitem__()</tt>, and <tt>__str__()</tt>
|
|
special methods.
|
|
<tt>len(p)</tt> returns the number of symbols in <tt>p.prod</tt>
|
|
and <tt>p[n]</tt> is the same as <tt>p.prod[n]</tt>.
|
|
|
|
<H2><a name="internal_nn4"></a>4. LRItems</H2>
|
|
|
|
|
|
The construction of parsing tables in an LR-based parser generator is primarily
|
|
done over a set of "LR Items". An LR item represents a stage of parsing one
|
|
of the grammar rules. To compute the LR items, it is first necessary to
|
|
call <tt>Grammar.build_lritems()</tt>. Once this step, all of the productions
|
|
in the grammar will have their LR items attached to them.
|
|
|
|
<p>
|
|
Here is an interactive example that shows what LR items look like if you
|
|
interactively experiment. In this example, <tt>g</tt> is a <tt>Grammar</tt>
|
|
object.
|
|
|
|
<blockquote>
|
|
<pre>
|
|
>>> <b>g.build_lritems()</b>
|
|
>>> <b>p = g[1]</b>
|
|
>>> <b>p</b>
|
|
Production(statement -> ID = expr)
|
|
>>>
|
|
</pre>
|
|
</blockquote>
|
|
|
|
In the above code, <tt>p</tt> represents the first grammar rule. In
|
|
this case, a rule <tt>'statement -> ID = expr'</tt>.
|
|
|
|
<p>
|
|
Now, let's look at the LR items for <tt>p</tt>.
|
|
|
|
<blockquote>
|
|
<pre>
|
|
>>> <b>p.lr_items</b>
|
|
[LRItem(statement -> . ID = expr),
|
|
LRItem(statement -> ID . = expr),
|
|
LRItem(statement -> ID = . expr),
|
|
LRItem(statement -> ID = expr .)]
|
|
>>>
|
|
</pre>
|
|
</blockquote>
|
|
|
|
In each LR item, the dot (.) represents a specific stage of parsing. In each LR item, the dot
|
|
is advanced by one symbol. It is only when the dot reaches the very end that a production
|
|
is successfully parsed.
|
|
|
|
<p>
|
|
An instance <tt>lr</tt> of <tt>LRItem</tt> has the following
|
|
attributes that hold information related to that specific stage of
|
|
parsing.
|
|
|
|
<p>
|
|
<b><tt>lr.name</tt></b>
|
|
<blockquote>
|
|
The name of the grammar rule. For example, <tt>'statement'</tt> in the above example.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.prod</tt></b>
|
|
<blockquote>
|
|
A tuple of symbols representing the right-hand side of the production, including the
|
|
special <tt>'.'</tt> character. For example, <tt>('ID','.','=','expr')</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.number</tt></b>
|
|
<blockquote>
|
|
An integer representing the production number in the grammar.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.usyms</tt></b>
|
|
<blockquote>
|
|
A set of unique symbols in the production. Inherited from the original <tt>Production</tt> instance.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_index</tt></b>
|
|
<blockquote>
|
|
An integer representing the position of the dot (.). You should never use <tt>lr.prod.index()</tt>
|
|
to search for it--the result will be wrong if the grammar happens to also use (.) as a character
|
|
literal.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_after</tt></b>
|
|
<blockquote>
|
|
A list of all productions that can legally appear immediately to the right of the
|
|
dot (.). This list contains <tt>Production</tt> instances. This attribute
|
|
represents all of the possible branches a parse can take from the current position.
|
|
For example, suppose that <tt>lr</tt> represents a stage immediately before
|
|
an expression like this:
|
|
|
|
<pre>
|
|
>>> <b>lr</b>
|
|
LRItem(statement -> ID = . expr)
|
|
>>>
|
|
</pre>
|
|
|
|
Then, the value of <tt>lr.lr_after</tt> might look like this, showing all productions that
|
|
can legally appear next:
|
|
|
|
<pre>
|
|
>>> <b>lr.lr_after</b>
|
|
[Production(expr -> expr PLUS expr),
|
|
Production(expr -> expr MINUS expr),
|
|
Production(expr -> expr TIMES expr),
|
|
Production(expr -> expr DIVIDE expr),
|
|
Production(expr -> MINUS expr),
|
|
Production(expr -> LPAREN expr RPAREN),
|
|
Production(expr -> NUMBER),
|
|
Production(expr -> ID)]
|
|
>>>
|
|
</pre>
|
|
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_before</tt></b>
|
|
<blockquote>
|
|
The grammar symbol that appears immediately before the dot (.) or <tt>None</tt> if
|
|
at the beginning of the parse.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_next</tt></b>
|
|
<blockquote>
|
|
A link to the next LR item, representing the next stage of the parse. <tt>None</tt> if <tt>lr</tt>
|
|
is the last LR item.
|
|
</blockquote>
|
|
|
|
<tt>LRItem</tt> instances also support the <tt>__len__()</tt> and <tt>__getitem__()</tt> special methods.
|
|
<tt>len(lr)</tt> returns the number of items in <tt>lr.prod</tt> including the dot (.). <tt>lr[n]</tt>
|
|
returns <tt>lr.prod[n]</tt>.
|
|
|
|
<p>
|
|
It goes without saying that all of the attributes associated with LR
|
|
items should be assumed to be read-only. Modifications will very
|
|
likely create a small black-hole that will consume you and your code.
|
|
|
|
<H2><a name="internal_nn5"></a>5. LRTable</H2>
|
|
|
|
|
|
The <tt>LRTable</tt> class is used to represent LR parsing table data. This
|
|
minimally includes the production list, action table, and goto table.
|
|
|
|
<p>
|
|
<b><tt>LRTable()</tt></b>
|
|
<blockquote>
|
|
Create an empty LRTable object. This object contains only the information needed to
|
|
run an LR parser.
|
|
</blockquote>
|
|
|
|
An instance <tt>lrtab</tt> of <tt>LRTable</tt> has the following methods:
|
|
|
|
<p>
|
|
<b><tt>lrtab.read_table(module)</tt></b>
|
|
<blockquote>
|
|
Populates the LR table with information from the module specified in <tt>module</tt>.
|
|
<tt>module</tt> is either a module object already loaded with <tt>import</tt> or
|
|
the name of a Python module. If it's a string containing a module name, it is
|
|
loaded and parsing data is extracted. Returns the signature value that was used
|
|
when initially writing the tables. Raises a <tt>VersionError</tt> exception if
|
|
the module was created using an incompatible version of PLY.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lrtab.bind_callables(dict)</tt></b>
|
|
<blockquote>
|
|
This binds all of the function names used in productions to callable objects
|
|
found in the dictionary <tt>dict</tt>. During table generation and when reading
|
|
LR tables from files, PLY only uses the names of action functions such as <tt>'p_expr'</tt>,
|
|
<tt>'p_statement'</tt>, etc. In order to actually run the parser, these names
|
|
have to be bound to callable objects. This method is always called prior to
|
|
running a parser.
|
|
</blockquote>
|
|
|
|
After <tt>lrtab</tt> has been populated, the following attributes are defined.
|
|
|
|
<p>
|
|
<b><tt>lrtab.lr_method</tt></b>
|
|
<blockquote>
|
|
The LR parsing method used (e.g., <tt>'LALR'</tt>)
|
|
</blockquote>
|
|
|
|
|
|
<p>
|
|
<b><tt>lrtab.lr_productions</tt></b>
|
|
<blockquote>
|
|
The production list. If the parsing tables have been newly
|
|
constructed, this will be a list of <tt>Production</tt> instances. If
|
|
the parsing tables have been read from a file, it's a list
|
|
of <tt>MiniProduction</tt> instances. This, together
|
|
with <tt>lr_action</tt> and <tt>lr_goto</tt> contain all of the
|
|
information needed by the LR parsing engine.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lrtab.lr_action</tt></b>
|
|
<blockquote>
|
|
The LR action dictionary that implements the underlying state machine.
|
|
The keys of this dictionary are the LR states.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lrtab.lr_goto</tt></b>
|
|
<blockquote>
|
|
The LR goto table that contains information about grammar rule reductions.
|
|
</blockquote>
|
|
|
|
|
|
<H2><a name="internal_nn6"></a>6. LRGeneratedTable</H2>
|
|
|
|
|
|
The <tt>LRGeneratedTable</tt> class represents constructed LR parsing tables on a
|
|
grammar. It is a subclass of <tt>LRTable</tt>.
|
|
|
|
<p>
|
|
<b><tt>LRGeneratedTable(grammar, method='LALR',log=None)</tt></b>
|
|
<blockquote>
|
|
Create the LR parsing tables on a grammar. <tt>grammar</tt> is an instance of <tt>Grammar</tt>,
|
|
<tt>method</tt> is a string with the parsing method (<tt>'SLR'</tt> or <tt>'LALR'</tt>), and
|
|
<tt>log</tt> is a logger object used to write debugging information. The debugging information
|
|
written to <tt>log</tt> is the same as what appears in the <tt>parser.out</tt> file created
|
|
by yacc. By supplying a custom logger with a different message format, it is possible to get
|
|
more information (e.g., the line number in <tt>yacc.py</tt> used for issuing each line of
|
|
output in the log). The result is an instance of <tt>LRGeneratedTable</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
An instance <tt>lr</tt> of <tt>LRGeneratedTable</tt> has the following attributes.
|
|
|
|
<p>
|
|
<b><tt>lr.grammar</tt></b>
|
|
<blockquote>
|
|
A link to the Grammar object used to construct the parsing tables.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_method</tt></b>
|
|
<blockquote>
|
|
The LR parsing method used (e.g., <tt>'LALR'</tt>)
|
|
</blockquote>
|
|
|
|
|
|
<p>
|
|
<b><tt>lr.lr_productions</tt></b>
|
|
<blockquote>
|
|
A reference to <tt>grammar.Productions</tt>. This, together with <tt>lr_action</tt> and <tt>lr_goto</tt>
|
|
contain all of the information needed by the LR parsing engine.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_action</tt></b>
|
|
<blockquote>
|
|
The LR action dictionary that implements the underlying state machine. The keys of this dictionary are
|
|
the LR states.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.lr_goto</tt></b>
|
|
<blockquote>
|
|
The LR goto table that contains information about grammar rule reductions.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.sr_conflicts</tt></b>
|
|
<blockquote>
|
|
A list of tuples <tt>(state,token,resolution)</tt> identifying all shift/reduce conflicts. <tt>state</tt> is the LR state
|
|
number where the conflict occurred, <tt>token</tt> is the token causing the conflict, and <tt>resolution</tt> is
|
|
a string describing the resolution taken. <tt>resolution</tt> is either <tt>'shift'</tt> or <tt>'reduce'</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>lr.rr_conflicts</tt></b>
|
|
<blockquote>
|
|
A list of tuples <tt>(state,rule,rejected)</tt> identifying all reduce/reduce conflicts. <tt>state</tt> is the
|
|
LR state number where the conflict occurred, <tt>rule</tt> is the production rule that was selected
|
|
and <tt>rejected</tt> is the production rule that was rejected. Both <tt>rule</tt> and </tt>rejected</tt> are
|
|
instances of <tt>Production</tt>. They can be inspected to provide the user with more information.
|
|
</blockquote>
|
|
|
|
<p>
|
|
There are two public methods of <tt>LRGeneratedTable</tt>.
|
|
|
|
<p>
|
|
<b><tt>lr.write_table(modulename,outputdir="",signature="")</tt></b>
|
|
<blockquote>
|
|
Writes the LR parsing table information to a Python module. <tt>modulename</tt> is a string
|
|
specifying the name of a module such as <tt>"parsetab"</tt>. <tt>outputdir</tt> is the name of a
|
|
directory where the module should be created. <tt>signature</tt> is a string representing a
|
|
grammar signature that's written into the output file. This can be used to detect when
|
|
the data stored in a module file is out-of-sync with the the grammar specification (and that
|
|
the tables need to be regenerated). If <tt>modulename</tt> is a string <tt>"parsetab"</tt>,
|
|
this function creates a file called <tt>parsetab.py</tt>. If the module name represents a
|
|
package such as <tt>"foo.bar.parsetab"</tt>, then only the last component, <tt>"parsetab"</tt> is
|
|
used.
|
|
</blockquote>
|
|
|
|
|
|
<H2><a name="internal_nn7"></a>7. LRParser</H2>
|
|
|
|
|
|
The <tt>LRParser</tt> class implements the low-level LR parsing engine.
|
|
|
|
|
|
<p>
|
|
<b><tt>LRParser(lrtab, error_func)</tt></b>
|
|
<blockquote>
|
|
Create an LRParser. <tt>lrtab</tt> is an instance of <tt>LRTable</tt>
|
|
containing the LR production and state tables. <tt>error_func</tt> is the
|
|
error function to invoke in the event of a parsing error.
|
|
</blockquote>
|
|
|
|
An instance <tt>p</tt> of <tt>LRParser</tt> has the following methods:
|
|
|
|
<p>
|
|
<b><tt>p.parse(input=None,lexer=None,debug=0,tracking=0,tokenfunc=None)</tt></b>
|
|
<blockquote>
|
|
Run the parser. <tt>input</tt> is a string, which if supplied is fed into the
|
|
lexer using its <tt>input()</tt> method. <tt>lexer</tt> is an instance of the
|
|
<tt>Lexer</tt> class to use for tokenizing. If not supplied, the last lexer
|
|
created with the <tt>lex</tt> module is used. <tt>debug</tt> is a boolean flag
|
|
that enables debugging. <tt>tracking</tt> is a boolean flag that tells the
|
|
parser to perform additional line number tracking. <tt>tokenfunc</tt> is a callable
|
|
function that returns the next token. If supplied, the parser will use it to get
|
|
all tokens.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.restart()</tt></b>
|
|
<blockquote>
|
|
Resets the parser state for a parse already in progress.
|
|
</blockquote>
|
|
|
|
<H2><a name="internal_nn8"></a>8. ParserReflect</H2>
|
|
|
|
|
|
<p>
|
|
The <tt>ParserReflect</tt> class is used to collect parser specification data
|
|
from a Python module or object. This class is what collects all of the
|
|
<tt>p_rule()</tt> functions in a PLY file, performs basic error checking,
|
|
and collects all of the needed information to build a grammar. Most of the
|
|
high-level PLY interface as used by the <tt>yacc()</tt> function is actually
|
|
implemented by this class.
|
|
|
|
<p>
|
|
<b><tt>ParserReflect(pdict, log=None)</tt></b>
|
|
<blockquote>
|
|
Creates a <tt>ParserReflect</tt> instance. <tt>pdict</tt> is a dictionary
|
|
containing parser specification data. This dictionary typically corresponds
|
|
to the module or class dictionary of code that implements a PLY parser.
|
|
<tt>log</tt> is a logger instance that will be used to report error
|
|
messages.
|
|
</blockquote>
|
|
|
|
An instance <tt>p</tt> of <tt>ParserReflect</tt> has the following methods:
|
|
|
|
<p>
|
|
<b><tt>p.get_all()</tt></b>
|
|
<blockquote>
|
|
Collect and store all required parsing information.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.validate_all()</tt></b>
|
|
<blockquote>
|
|
Validate all of the collected parsing information. This is a seprate step
|
|
from <tt>p.get_all()</tt> as a performance optimization. In order to
|
|
increase parser start-up time, a parser can elect to only validate the
|
|
parsing data when regenerating the parsing tables. The validation
|
|
step tries to collect as much information as possible rather than
|
|
raising an exception at the first sign of trouble. The attribute
|
|
<tt>p.error</tt> is set if there are any validation errors. The
|
|
value of this attribute is also returned.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.signature()</tt></b>
|
|
<blockquote>
|
|
Compute a signature representing the contents of the collected parsing
|
|
data. The signature value should change if anything in the parser
|
|
specification has changed in a way that would justify parser table
|
|
regeneration. This method can be called after <tt>p.get_all()</tt>,
|
|
but before <tt>p.validate_all()</tt>.
|
|
</blockquote>
|
|
|
|
The following attributes are set in the process of collecting data:
|
|
|
|
<p>
|
|
<b><tt>p.start</tt></b>
|
|
<blockquote>
|
|
The grammar start symbol, if any. Taken from <tt>pdict['start']</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.error_func</tt></b>
|
|
<blockquote>
|
|
The error handling function or <tt>None</tt>. Taken from <tt>pdict['p_error']</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.tokens</tt></b>
|
|
<blockquote>
|
|
The token list. Taken from <tt>pdict['tokens']</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.prec</tt></b>
|
|
<blockquote>
|
|
The precedence specifier. Taken from <tt>pdict['precedence']</tt>.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.preclist</tt></b>
|
|
<blockquote>
|
|
A parsed version of the precedence specified. A list of tuples of the form
|
|
<tt>(token,assoc,level)</tt> where <tt>token</tt> is the terminal symbol,
|
|
<tt>assoc</tt> is the associativity (e.g., <tt>'left'</tt>) and <tt>level</tt>
|
|
is a numeric precedence level.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.grammar</tt></b>
|
|
<blockquote>
|
|
A list of tuples <tt>(name, rules)</tt> representing the grammar rules. <tt>name</tt> is the
|
|
name of a Python function or method in <tt>pdict</tt> that starts with <tt>"p_"</tt>.
|
|
<tt>rules</tt> is a list of tuples <tt>(filename,line,prodname,syms)</tt> representing
|
|
the grammar rules found in the documentation string of that function. <tt>filename</tt> and <tt>line</tt> contain location
|
|
information that can be used for debugging. <tt>prodname</tt> is the name of the
|
|
production. <tt>syms</tt> is the right-hand side of the production. If you have a
|
|
function like this
|
|
|
|
<pre>
|
|
def p_expr(p):
|
|
'''expr : expr PLUS expr
|
|
| expr MINUS expr
|
|
| expr TIMES expr
|
|
| expr DIVIDE expr'''
|
|
</pre>
|
|
|
|
then the corresponding entry in <tt>p.grammar</tt> might look like this:
|
|
|
|
<pre>
|
|
('p_expr', [ ('calc.py',10,'expr', ['expr','PLUS','expr']),
|
|
('calc.py',11,'expr', ['expr','MINUS','expr']),
|
|
('calc.py',12,'expr', ['expr','TIMES','expr']),
|
|
('calc.py',13,'expr', ['expr','DIVIDE','expr'])
|
|
])
|
|
</pre>
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.pfuncs</tt></b>
|
|
<blockquote>
|
|
A sorted list of tuples <tt>(line, file, name, doc)</tt> representing all of
|
|
the <tt>p_</tt> functions found. <tt>line</tt> and <tt>file</tt> give location
|
|
information. <tt>name</tt> is the name of the function. <tt>doc</tt> is the
|
|
documentation string. This list is sorted in ascending order by line number.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.files</tt></b>
|
|
<blockquote>
|
|
A dictionary holding all of the source filenames that were encountered
|
|
while collecting parser information. Only the keys of this dictionary have
|
|
any meaning.
|
|
</blockquote>
|
|
|
|
<p>
|
|
<b><tt>p.error</tt></b>
|
|
<blockquote>
|
|
An attribute that indicates whether or not any critical errors
|
|
occurred in validation. If this is set, it means that that some kind
|
|
of problem was detected and that no further processing should be
|
|
performed.
|
|
</blockquote>
|
|
|
|
|
|
<H2><a name="internal_nn9"></a>9. High-level operation</H2>
|
|
|
|
|
|
Using all of the above classes requires some attention to detail. The <tt>yacc()</tt>
|
|
function carries out a very specific sequence of operations to create a grammar.
|
|
This same sequence should be emulated if you build an alternative PLY interface.
|
|
|
|
<ol>
|
|
<li>A <tt>ParserReflect</tt> object is created and raw grammar specification data is
|
|
collected.
|
|
<li>A <tt>Grammar</tt> object is created and populated with information
|
|
from the specification data.
|
|
<li>A <tt>LRGenerator</tt> object is created to run the LALR algorithm over
|
|
the <tt>Grammar</tt> object.
|
|
<li>Productions in the LRGenerator and bound to callables using the <tt>bind_callables()</tt>
|
|
method.
|
|
<li>A <tt>LRParser</tt> object is created from from the information in the
|
|
<tt>LRGenerator</tt> object.
|
|
</ol>
|
|
|
|
</body>
|
|
</html>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|