PLY Internals

David M. Beazley
dave@dabeaz.com

PLY Version: 3.0

1. Introduction

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.

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.

2. Grammar Class

The file ply.yacc defines a class Grammar 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.

Grammar(terminals)

Creates a new grammar object. terminals is a list of strings specifying the terminals for the grammar. An instance g of Grammar has the following methods:

g.set_precedence(term,assoc,level)

Sets the precedence level and associativity for a given terminal term. assoc is one of 'right', 'left', or 'nonassoc' and level is a positive integer. The higher the value of level, the higher the precedence. Here is an example of typical precedence settings:
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)
This method must be called prior to adding any productions to the grammar with g.add_production(). The precedence of individual grammar rules is determined by the precedence of the right-most terminal.

g.add_production(name,syms,func=None,file='',line=0)

Adds a new grammar rule. name is the name of the rule, syms is a list of symbols making up the right hand side of the rule, func is the function to call when reducing the rule. file and line specify the filename and line number of the rule and are used for generating error messages.

The list of symbols in syms may include character literals and %prec specifiers. Here are some examples:

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)

If any kind of error is detected, a GrammarError exception is raised with a message indicating the reason for the failure.

g.set_start(start=None)

Sets the starting rule for the grammar. start is a string specifying the name of the start rule. If start is omitted, the first grammar rule added with add_production() is taken to be the starting rule. This method must always be called after all productions have been added.

g.find_unreachable()

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.

g.infinite_cycle()

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.

g.undefined_symbols()

Diagnostic function. Returns a list of tuples (name, prod) corresponding to undefined symbols in the grammar. name is the name of the undefined symbol and prod is an instance of Production which has information about the production rule where the undefined symbol was used.

g.unused_terminals()

Diagnostic function. Returns a list of terminals that were defined, but never used in the grammar.

g.unused_rules()

Diagnostic function. Returns a list of Production instances corresponding to production rules that were defined in the grammar, but never used anywhere. This is slightly different than find_unreachable().

g.unused_precedence()

Diagnostic function. Returns a list of tuples (term, assoc) corresponding to precedence rules that were set, but never used the grammar. term is the terminal name and assoc is the precedence associativity (e.g., 'left', 'right', or 'nonassoc'.

g.compute_first()

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.

g.compute_follow()

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.

g.build_lritems()

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.

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.

g.Productions

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 Production class, described shortly.

g.Prodnames

A dictionary mapping the names of nonterminals to a list of all productions of that nonterminal.

g.Terminals

A dictionary mapping the names of terminals to a list of the production numbers where they are used.

g.Nonterminals

A dictionary mapping the names of nonterminals to a list of the production numbers where they are used.

g.First

A dictionary representing the first sets for all grammar symbols. This is computed and returned by the compute_first() method.

g.Follow

A dictionary representing the follow sets for all grammar rules. This is computed and returned by the compute_follow() method.

g.Start

Starting symbol for the grammar. Set by the set_start() method.
For the purposes of debugging, a Grammar object supports the __len__() and __getitem__() special methods. Accessing g[n] returns the nth production from the grammar.

3. Productions

Grammar objects store grammar rules as instances of a Production class. This class has no public constructor--you should only create productions by calling Grammar.add_production(). The following attributes are available on a Production instance p.

p.name

The name of the production. For a grammar rule such as A : B C D, this is 'A'.

p.prod

A tuple of symbols making up the right-hand side of the production. For a grammar rule such as A : B C D, this is ('B','C','D').

p.number

Production number. An integer containing the index of the production in the grammar's Productions list.

p.func

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.

p.callable

The callable object associated with the name in p.func. This is None unless the production has been bound using bind().

p.file

Filename associated with the production. Typically this is the file where the production was defined. Used for error messages.

p.lineno

Line number associated with the production. Typically this is the line number in p.file where the production was defined. Used for error messages.

p.prec

Precedence and associativity associated with the production. This is a tuple (assoc,level) where assoc is one of 'left','right', or 'nonassoc' and level is an integer. This value is determined by the precedence of the right-most terminal symbol in the production or by use of the %prec specifier when adding the production.

p.usyms

A list of all unique symbols found in the production.

p.lr_items

A list of all LR items for this production. This attribute only has a meaningful value if the Grammar.build_lritems() method has been called. The items in this list are instances of LRItem described below.

p.lr_next

The head of a linked-list representation of the LR items in p.lr_items. This attribute only has a meaningful value if the Grammar.build_lritems() method has been called. Each LRItem instance has a lr_next attribute to move to the next item. The list is terminated by None.

p.bind(dict)

Binds the production function name in p.func to a callable object in dict. 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.

Production objects support the __len__(), __getitem__(), and __str__() special methods. len(p) returns the number of symbols in p.prod and p[n] is the same as p.prod[n].

4. LRItems

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 Grammar.build_lritems(). Once this step, all of the productions in the grammar will have their LR items attached to them.

Here is an interactive example that shows what LR items look like if you interactively experiment. In this example, g is a Grammar object.

>>> g.build_lritems()
>>> p = g[1]
>>> p
Production(statement -> ID = expr)
>>>
In the above code, p represents the first grammar rule. In this case, a rule 'statement -> ID = expr'.

Now, let's look at the LR items for p.

>>> p.lr_items
[LRItem(statement -> . ID = expr), 
 LRItem(statement -> ID . = expr), 
 LRItem(statement -> ID = . expr), 
 LRItem(statement -> ID = expr .)]
>>>
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.

An instance lr of LRItem has the following attributes that hold information related to that specific stage of parsing.

lr.name

The name of the grammar rule. For example, 'statement' in the above example.

lr.prod

A tuple of symbols representing the right-hand side of the production, including the special '.' character. For example, ('ID','.','=','expr').

lr.number

An integer representing the production number in the grammar.

lr.usyms

A set of unique symbols in the production. Inherited from the original Production instance.

lr.lr_index

An integer representing the position of the dot (.). You should never use lr.prod.index() to search for it--the result will be wrong if the grammar happens to also use (.) as a character literal.

lr.lr_after

A list of all productions that can legally appear immediately to the right of the dot (.). This list contains Production instances. This attribute represents all of the possible branches a parse can take from the current position. For example, suppose that lr represents a stage immediately before an expression like this:
>>> lr
LRItem(statement -> ID = . expr)
>>>
Then, the value of lr.lr_after might look like this, showing all productions that can legally appear next:
>>> lr.lr_after
[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)]
>>>

lr.lr_before

The grammar symbol that appears immediately before the dot (.) or None if at the beginning of the parse.

lr.lr_next

A link to the next LR item, representing the next stage of the parse. None if lr is the last LR item.
LRItem instances also support the __len__() and __getitem__() special methods. len(lr) returns the number of items in lr.prod including the dot (.). lr[n] returns lr.prod[n].

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.

5. LRTable

The LRTable class is used to represent LR parsing table data. This minimally includes the production list, action table, and goto table.

LRTable()

Create an empty LRTable object. This object contains only the information needed to run an LR parser.
An instance lrtab of LRTable has the following methods:

lrtab.read_table(module)

Populates the LR table with information from the module specified in module. module is either a module object already loaded with import 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 VersionError exception if the module was created using an incompatible version of PLY.

lrtab.bind_callables(dict)

This binds all of the function names used in productions to callable objects found in the dictionary dict. During table generation and when reading LR tables from files, PLY only uses the names of action functions such as 'p_expr', 'p_statement', 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.
After lrtab has been populated, the following attributes are defined.

lrtab.lr_method

The LR parsing method used (e.g., 'LALR')

lrtab.lr_productions

The production list. If the parsing tables have been newly constructed, this will be a list of Production instances. If the parsing tables have been read from a file, it's a list of MiniProduction instances. This, together with lr_action and lr_goto contain all of the information needed by the LR parsing engine.

lrtab.lr_action

The LR action dictionary that implements the underlying state machine. The keys of this dictionary are the LR states.

lrtab.lr_goto

The LR goto table that contains information about grammar rule reductions.

6. LRGeneratedTable

The LRGeneratedTable class represents constructed LR parsing tables on a grammar. It is a subclass of LRTable.

LRGeneratedTable(grammar, method='LALR',log=None)

Create the LR parsing tables on a grammar. grammar is an instance of Grammar, method is a string with the parsing method ('SLR' or 'LALR'), and log is a logger object used to write debugging information. The debugging information written to log is the same as what appears in the parser.out 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 yacc.py used for issuing each line of output in the log). The result is an instance of LRGeneratedTable.

An instance lr of LRGeneratedTable has the following attributes.

lr.grammar

A link to the Grammar object used to construct the parsing tables.

lr.lr_method

The LR parsing method used (e.g., 'LALR')

lr.lr_productions

A reference to grammar.Productions. This, together with lr_action and lr_goto contain all of the information needed by the LR parsing engine.

lr.lr_action

The LR action dictionary that implements the underlying state machine. The keys of this dictionary are the LR states.

lr.lr_goto

The LR goto table that contains information about grammar rule reductions.

lr.sr_conflicts

A list of tuples (state,token,resolution) identifying all shift/reduce conflicts. state is the LR state number where the conflict occurred, token is the token causing the conflict, and resolution is a string describing the resolution taken. resolution is either 'shift' or 'reduce'.

lr.rr_conflicts

A list of tuples (state,rule,rejected) identifying all reduce/reduce conflicts. state is the LR state number where the conflict occurred, rule is the production rule that was selected and rejected is the production rule that was rejected. Both rule and rejected are instances of Production. They can be inspected to provide the user with more information.

There are two public methods of LRGeneratedTable.

lr.write_table(modulename,outputdir="",signature="")

Writes the LR parsing table information to a Python module. modulename is a string specifying the name of a module such as "parsetab". outputdir is the name of a directory where the module should be created. signature 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 modulename is a string "parsetab", this function creates a file called parsetab.py. If the module name represents a package such as "foo.bar.parsetab", then only the last component, "parsetab" is used.

7. LRParser

The LRParser class implements the low-level LR parsing engine.

LRParser(lrtab, error_func)

Create an LRParser. lrtab is an instance of LRTable containing the LR production and state tables. error_func is the error function to invoke in the event of a parsing error.
An instance p of LRParser has the following methods:

p.parse(input=None,lexer=None,debug=0,tracking=0,tokenfunc=None)

Run the parser. input is a string, which if supplied is fed into the lexer using its input() method. lexer is an instance of the Lexer class to use for tokenizing. If not supplied, the last lexer created with the lex module is used. debug is a boolean flag that enables debugging. tracking is a boolean flag that tells the parser to perform additional line number tracking. tokenfunc is a callable function that returns the next token. If supplied, the parser will use it to get all tokens.

p.restart()

Resets the parser state for a parse already in progress.

8. ParserReflect

The ParserReflect class is used to collect parser specification data from a Python module or object. This class is what collects all of the p_rule() 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 yacc() function is actually implemented by this class.

ParserReflect(pdict, log=None)

Creates a ParserReflect instance. pdict is a dictionary containing parser specification data. This dictionary typically corresponds to the module or class dictionary of code that implements a PLY parser. log is a logger instance that will be used to report error messages.
An instance p of ParserReflect has the following methods:

p.get_all()

Collect and store all required parsing information.

p.validate_all()

Validate all of the collected parsing information. This is a seprate step from p.get_all() 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 p.error is set if there are any validation errors. The value of this attribute is also returned.

p.signature()

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 p.get_all(), but before p.validate_all().
The following attributes are set in the process of collecting data:

p.start

The grammar start symbol, if any. Taken from pdict['start'].

p.error_func

The error handling function or None. Taken from pdict['p_error'].

p.tokens

The token list. Taken from pdict['tokens'].

p.prec

The precedence specifier. Taken from pdict['precedence'].

p.preclist

A parsed version of the precedence specified. A list of tuples of the form (token,assoc,level) where token is the terminal symbol, assoc is the associativity (e.g., 'left') and level is a numeric precedence level.

p.grammar

A list of tuples (name, rules) representing the grammar rules. name is the name of a Python function or method in pdict that starts with "p_". rules is a list of tuples (filename,line,prodname,syms) representing the grammar rules found in the documentation string of that function. filename and line contain location information that can be used for debugging. prodname is the name of the production. syms is the right-hand side of the production. If you have a function like this
def p_expr(p):
    '''expr : expr PLUS expr
            | expr MINUS expr
            | expr TIMES expr
            | expr DIVIDE expr'''
then the corresponding entry in p.grammar might look like this:
('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'])
           ])

p.pfuncs

A sorted list of tuples (line, file, name, doc) representing all of the p_ functions found. line and file give location information. name is the name of the function. doc is the documentation string. This list is sorted in ascending order by line number.

p.files

A dictionary holding all of the source filenames that were encountered while collecting parser information. Only the keys of this dictionary have any meaning.

p.error

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.

9. High-level operation

Using all of the above classes requires some attention to detail. The yacc() 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.
  1. A ParserReflect object is created and raw grammar specification data is collected.
  2. A Grammar object is created and populated with information from the specification data.
  3. A LRGenerator object is created to run the LALR algorithm over the Grammar object.
  4. Productions in the LRGenerator and bound to callables using the bind_callables() method.
  5. A LRParser object is created from from the information in the LRGenerator object.