Version 2.3 ----------------------------- 02/20/07: beazley Fixed a bug with character literals if the literal '.' appeared as the last symbol of a grammar rule. Reported by Ales Smrcka. 02/19/07: beazley Warning messages are now redirected to stderr instead of being printed to standard output. 02/19/07: beazley Added a warning message to lex.py if it detects a literal backslash character inside the t_ignore declaration. This is to help problems that might occur if someone accidentally defines t_ignore as a Python raw string. For example: t_ignore = r' \t' The idea for this is from an email I received from David Cimimi who reported bizarre behavior in lexing as a result of defining t_ignore as a raw string by accident. 02/18/07: beazley Performance improvements. Made some changes to the internal table organization and LR parser to improve parsing performance. 02/18/07: beazley Automatic tracking of line number and position information must now be enabled by a special flag to parse(). For example: yacc.parse(data,tracking=True) In many applications, it's just not that important to have the parser automatically track all line numbers. By making this an optional feature, it allows the parser to run significantly faster (more than a 20% speed increase in many cases). Note: positional information is always available for raw tokens---this change only applies to positional information associated with nonterminal grammar symbols. *** POTENTIAL INCOMPATIBILITY *** 02/18/07: beazley Yacc no longer supports extended slices of grammar productions. However, it does support regular slices. For example: def p_foo(p): '''foo: a b c d e''' p[0] = p[1:3] This change is a performance improvement to the parser--it streamlines normal access to the grammar values since slices are now handled in a __getslice__() method as opposed to __getitem__(). 02/12/07: beazley Fixed a bug in the handling of token names when combined with start conditions. Bug reported by Todd O'Bryan. Version 2.2 ------------------------------ 11/01/06: beazley Added lexpos() and lexspan() methods to grammar symbols. These mirror the same functionality of lineno() and linespan(). For example: def p_expr(p): 'expr : expr PLUS expr' p.lexpos(1) # Lexing position of left-hand-expression p.lexpos(1) # Lexing position of PLUS start,end = p.lexspan(3) # Lexing range of right hand expression 11/01/06: beazley Minor change to error handling. The recommended way to skip characters in the input is to use t.lexer.skip() as shown here: def t_error(t): print "Illegal character '%s'" % t.value[0] t.lexer.skip(1) The old approach of just using t.skip(1) will still work, but won't be documented. 10/31/06: beazley Discarded tokens can now be specified as simple strings instead of functions. To do this, simply include the text "ignore_" in the token declaration. For example: t_ignore_cppcomment = r'//.*' Previously, this had to be done with a function. For example: def t_ignore_cppcomment(t): r'//.*' pass If start conditions/states are being used, state names should appear before the "ignore_" text. 10/19/06: beazley The Lex module now provides support for flex-style start conditions as described at http://www.gnu.org/software/flex/manual/html_chapter/flex_11.html. Please refer to this document to understand this change note. Refer to the PLY documentation for PLY-specific explanation of how this works. To use start conditions, you first need to declare a set of states in your lexer file: states = ( ('foo','exclusive'), ('bar','inclusive') ) This serves the same role as the %s and %x specifiers in flex. One a state has been declared, tokens for that state can be declared by defining rules of the form t_state_TOK. For example: t_PLUS = '\+' # Rule defined in INITIAL state t_foo_NUM = '\d+' # Rule defined in foo state t_bar_NUM = '\d+' # Rule defined in bar state t_foo_bar_NUM = '\d+' # Rule defined in both foo and bar t_ANY_NUM = '\d+' # Rule defined in all states In addition to defining tokens for each state, the t_ignore and t_error specifications can be customized for specific states. For example: t_foo_ignore = " " # Ignored characters for foo state def t_bar_error(t): # Handle errors in bar state With token rules, the following methods can be used to change states def t_TOKNAME(t): t.lexer.begin('foo') # Begin state 'foo' t.lexer.push_state('foo') # Begin state 'foo', push old state # onto a stack t.lexer.pop_state() # Restore previous state t.lexer.current_state() # Returns name of current state These methods mirror the BEGIN(), yy_push_state(), yy_pop_state(), and yy_top_state() functions in flex. The use of start states can be used as one way to write sub-lexers. For example, the lexer or parser might instruct the lexer to start generating a different set of tokens depending on the context. example/yply/ylex.py shows the use of start states to grab C/C++ code fragments out of traditional yacc specification files. *** NEW FEATURE *** Suggested by Daniel Larraz with whom I also discussed various aspects of the design. 10/19/06: beazley Minor change to the way in which yacc.py was reporting shift/reduce conflicts. Although the underlying LALR(1) algorithm was correct, PLY was under-reporting the number of conflicts compared to yacc/bison when precedence rules were in effect. This change should make PLY report the same number of conflicts as yacc. 10/19/06: beazley Modified yacc so that grammar rules could also include the '-' character. For example: def p_expr_list(p): 'expression-list : expression-list expression' Suggested by Oldrich Jedlicka. 10/18/06: beazley Attribute lexer.lexmatch added so that token rules can access the re match object that was generated. For example: def t_FOO(t): r'some regex' m = t.lexer.lexmatch # Do something with m This may be useful if you want to access named groups specified within the regex for a specific token. Suggested by Oldrich Jedlicka. 10/16/06: beazley Changed the error message that results if an illegal character is encountered and no default error function is defined in lex. The exception is now more informative about the actual cause of the error. Version 2.1 ------------------------------ 10/02/06: beazley The last Lexer object built by lex() can be found in lex.lexer. The last Parser object built by yacc() can be found in yacc.parser. 10/02/06: beazley New example added: examples/yply This example uses PLY to convert Unix-yacc specification files to PLY programs with the same grammar. This may be useful if you want to convert a grammar from bison/yacc to use with PLY. 10/02/06: beazley Added support for a start symbol to be specified in the yacc input file itself. Just do this: start = 'name' where 'name' matches some grammar rule. For example: def p_name(p): 'name : A B C' ... This mirrors the functionality of the yacc %start specifier. 09/30/06: beazley Some new examples added.: examples/GardenSnake : A simple indentation based language similar to Python. Shows how you might handle whitespace. Contributed by Andrew Dalke. examples/BASIC : An implementation of 1964 Dartmouth BASIC. Contributed by Dave against his better judgement. 09/28/06: beazley Minor patch to allow named groups to be used in lex regular expression rules. For example: t_QSTRING = r'''(?P['"]).*?(?P=quote)''' Patch submitted by Adam Ring. 09/28/06: beazley LALR(1) is now the default parsing method. To use SLR, use yacc.yacc(method="SLR"). Note: there is no performance impact on parsing when using LALR(1) instead of SLR. However, constructing the parsing tables will take a little longer. 09/26/06: beazley Change to line number tracking. To modify line numbers, modify the line number of the lexer itself. For example: def t_NEWLINE(t): r'\n' t.lexer.lineno += 1 This modification is both cleanup and a performance optimization. In past versions, lex was monitoring every token for changes in the line number. This extra processing is unnecessary for a vast majority of tokens. Thus, this new approach cleans it up a bit. *** POTENTIAL INCOMPATIBILITY *** You will need to change code in your lexer that updates the line number. For example, "t.lineno += 1" becomes "t.lexer.lineno += 1" 09/26/06: beazley Added the lexing position to tokens as an attribute lexpos. This is the raw index into the input text at which a token appears. This information can be used to compute column numbers and other details (e.g., scan backwards from lexpos to the first newline to get a column position). 09/25/06: beazley Changed the name of the __copy__() method on the Lexer class to clone(). This is used to clone a Lexer object (e.g., if you're running different lexers at the same time). 09/21/06: beazley Limitations related to the use of the re module have been eliminated. Several users reported problems with regular expressions exceeding more than 100 named groups. To solve this, lex.py is now capable of automatically splitting its master regular regular expression into smaller expressions as needed. This should, in theory, make it possible to specify an arbitrarily large number of tokens. 09/21/06: beazley Improved error checking in lex.py. Rules that match the empty string are now rejected (otherwise they cause the lexer to enter an infinite loop). An extra check for rules containing '#' has also been added. Since lex compiles regular expressions in verbose mode, '#' is interpreted as a regex comment, it is critical to use '\#' instead. 09/18/06: beazley Added a @TOKEN decorator function to lex.py that can be used to define token rules where the documentation string might be computed in some way. digit = r'([0-9])' nondigit = r'([_A-Za-z])' identifier = r'(' + nondigit + r'(' + digit + r'|' + nondigit + r')*)' from ply.lex import TOKEN @TOKEN(identifier) def t_ID(t): # Do whatever The @TOKEN decorator merely sets the documentation string of the associated token function as needed for lex to work. Note: An alternative solution is the following: def t_ID(t): # Do whatever t_ID.__doc__ = identifier Note: Decorators require the use of Python 2.4 or later. If compatibility with old versions is needed, use the latter solution. The need for this feature was suggested by Cem Karan. 09/14/06: beazley Support for single-character literal tokens has been added to yacc. These literals must be enclosed in quotes. For example: def p_expr(p): "expr : expr '+' expr" ... def p_expr(p): 'expr : expr "-" expr' ... In addition to this, it is necessary to tell the lexer module about literal characters. This is done by defining the variable 'literals' as a list of characters. This should be defined in the module that invokes the lex.lex() function. For example: literals = ['+','-','*','/','(',')','='] or simply literals = '+=*/()=' It is important to note that literals can only be a single character. When the lexer fails to match a token using its normal regular expression rules, it will check the current character against the literal list. If found, it will be returned with a token type set to match the literal character. Otherwise, an illegal character will be signalled. 09/14/06: beazley Modified PLY to install itself as a proper Python package called 'ply'. This will make it a little more friendly to other modules. This changes the usage of PLY only slightly. Just do this to import the modules import ply.lex as lex import ply.yacc as yacc Alternatively, you can do this: from ply import * Which imports both the lex and yacc modules. Change suggested by Lee June. 09/13/06: beazley Changed the handling of negative indices when used in production rules. A negative production index now accesses already parsed symbols on the parsing stack. For example, def p_foo(p): "foo: A B C D" print p[1] # Value of 'A' symbol print p[2] # Value of 'B' symbol print p[-1] # Value of whatever symbol appears before A # on the parsing stack. p[0] = some_val # Sets the value of the 'foo' grammer symbol This behavior makes it easier to work with embedded actions within the parsing rules. For example, in C-yacc, it is possible to write code like this: bar: A { printf("seen an A = %d\n", $1); } B { do_stuff; } In this example, the printf() code executes immediately after A has been parsed. Within the embedded action code, $1 refers to the A symbol on the stack. To perform this equivalent action in PLY, you need to write a pair of rules like this: def p_bar(p): "bar : A seen_A B" do_stuff def p_seen_A(p): "seen_A :" print "seen an A =", p[-1] The second rule "seen_A" is merely a empty production which should be reduced as soon as A is parsed in the "bar" rule above. The use of the negative index p[-1] is used to access whatever symbol appeared before the seen_A symbol. This feature also makes it possible to support inherited attributes. For example: def p_decl(p): "decl : scope name" def p_scope(p): """scope : GLOBAL | LOCAL""" p[0] = p[1] def p_name(p): "name : ID" if p[-1] == "GLOBAL": # ... else if p[-1] == "LOCAL": #... In this case, the name rule is inheriting an attribute from the scope declaration that precedes it. *** POTENTIAL INCOMPATIBILITY *** If you are currently using negative indices within existing grammar rules, your code will break. This should be extremely rare if non-existent in most cases. The argument to various grammar rules is not usually not processed in the same way as a list of items. Version 2.0 ------------------------------ 09/07/06: beazley Major cleanup and refactoring of the LR table generation code. Both SLR and LALR(1) table generation is now performed by the same code base with only minor extensions for extra LALR(1) processing. 09/07/06: beazley Completely reimplemented the entire LALR(1) parsing engine to use the DeRemer and Pennello algorithm for calculating lookahead sets. This significantly improves the performance of generating LALR(1) tables and has the added feature of actually working correctly! If you experienced weird behavior with LALR(1) in prior releases, this should hopefully resolve all of those problems. Many thanks to Andrew Waters and Markus Schoepflin for submitting bug reports and helping me test out the revised LALR(1) support. Version 1.8 ------------------------------ 08/02/06: beazley Fixed a problem related to the handling of default actions in LALR(1) parsing. If you experienced subtle and/or bizarre behavior when trying to use the LALR(1) engine, this may correct those problems. Patch contributed by Russ Cox. Note: This patch has been superceded by revisions for LALR(1) parsing in Ply-2.0. 08/02/06: beazley Added support for slicing of productions in yacc. Patch contributed by Patrick Mezard. Version 1.7 ------------------------------ 03/02/06: beazley Fixed infinite recursion problem ReduceToTerminals() function that would sometimes come up in LALR(1) table generation. Reported by Markus Schoepflin. 03/01/06: beazley Added "reflags" argument to lex(). For example: lex.lex(reflags=re.UNICODE) This can be used to specify optional flags to the re.compile() function used inside the lexer. This may be necessary for special situations such as processing Unicode (e.g., if you want escapes like \w and \b to consult the Unicode character property database). The need for this suggested by Andreas Jung. 03/01/06: beazley Fixed a bug with an uninitialized variable on repeated instantiations of parser objects when the write_tables=0 argument was used. Reported by Michael Brown. 03/01/06: beazley Modified lex.py to accept Unicode strings both as the regular expressions for tokens and as input. Hopefully this is the only change needed for Unicode support. Patch contributed by Johan Dahl. 03/01/06: beazley Modified the class-based interface to work with new-style or old-style classes. Patch contributed by Michael Brown (although I tweaked it slightly so it would work with older versions of Python). Version 1.6 ------------------------------ 05/27/05: beazley Incorporated patch contributed by Christopher Stawarz to fix an extremely devious bug in LALR(1) parser generation. This patch should fix problems numerous people reported with LALR parsing. 05/27/05: beazley Fixed problem with lex.py copy constructor. Reported by Dave Aitel, Aaron Lav, and Thad Austin. 05/27/05: beazley Added outputdir option to yacc() to control output directory. Contributed by Christopher Stawarz. 05/27/05: beazley Added rununit.py test script to run tests using the Python unittest module. Contributed by Miki Tebeka. Version 1.5 ------------------------------ 05/26/04: beazley Major enhancement. LALR(1) parsing support is now working. This feature was implemented by Elias Ioup (ezioup@alumni.uchicago.edu) and optimized by David Beazley. To use LALR(1) parsing do the following: yacc.yacc(method="LALR") Computing LALR(1) parsing tables takes about twice as long as the default SLR method. However, LALR(1) allows you to handle more complex grammars. For example, the ANSI C grammar (in example/ansic) has 13 shift-reduce conflicts with SLR, but only has 1 shift-reduce conflict with LALR(1). 05/20/04: beazley Added a __len__ method to parser production lists. Can be used in parser rules like this: def p_somerule(p): """a : B C D | E F" if (len(p) == 3): # Must have been first rule elif (len(p) == 2): # Must be second rule Suggested by Joshua Gerth and others. Version 1.4 ------------------------------ 04/23/04: beazley Incorporated a variety of patches contributed by Eric Raymond. These include: 0. Cleans up some comments so they don't wrap on an 80-column display. 1. Directs compiler errors to stderr where they belong. 2. Implements and documents automatic line counting when \n is ignored. 3. Changes the way progress messages are dumped when debugging is on. The new format is both less verbose and conveys more information than the old, including shift and reduce actions. 04/23/04: beazley Added a Python setup.py file to simply installation. Contributed by Adam Kerrison. 04/23/04: beazley Added patches contributed by Adam Kerrison. - Some output is now only shown when debugging is enabled. This means that PLY will be completely silent when not in debugging mode. - An optional parameter "write_tables" can be passed to yacc() to control whether or not parsing tables are written. By default, it is true, but it can be turned off if you don't want the yacc table file. Note: disabling this will cause yacc() to regenerate the parsing table each time. 04/23/04: beazley Added patches contributed by David McNab. This patch addes two features: - The parser can be supplied as a class instead of a module. For an example of this, see the example/classcalc directory. - Debugging output can be directed to a filename of the user's choice. Use yacc(debugfile="somefile.out") Version 1.3 ------------------------------ 12/10/02: jmdyck Various minor adjustments to the code that Dave checked in today. Updated test/yacc_{inf,unused}.exp to reflect today's changes. 12/10/02: beazley Incorporated a variety of minor bug fixes to empty production handling and infinite recursion checking. Contributed by Michael Dyck. 12/10/02: beazley Removed bogus recover() method call in yacc.restart() Version 1.2 ------------------------------ 11/27/02: beazley Lexer and parser objects are now available as an attribute of tokens and slices respectively. For example: def t_NUMBER(t): r'\d+' print t.lexer def p_expr_plus(t): 'expr: expr PLUS expr' print t.lexer print t.parser This can be used for state management (if needed). 10/31/02: beazley Modified yacc.py to work with Python optimize mode. To make this work, you need to use yacc.yacc(optimize=1) Furthermore, you need to first run Python in normal mode to generate the necessary parsetab.py files. After that, you can use python -O or python -OO. Note: optimized mode turns off a lot of error checking. Only use when you are sure that your grammar is working. Make sure parsetab.py is up to date! 10/30/02: beazley Added cloning of Lexer objects. For example: import copy l = lex.lex() lc = copy.copy(l) l.input("Some text") lc.input("Some other text") ... This might be useful if the same "lexer" is meant to be used in different contexts---or if multiple lexers are running concurrently. 10/30/02: beazley Fixed subtle bug with first set computation and empty productions. Patch submitted by Michael Dyck. 10/30/02: beazley Fixed error messages to use "filename:line: message" instead of "filename:line. message". This makes error reporting more friendly to emacs. Patch submitted by François Pinard. 10/30/02: beazley Improvements to parser.out file. Terminals and nonterminals are sorted instead of being printed in random order. Patch submitted by François Pinard. 10/30/02: beazley Improvements to parser.out file output. Rules are now printed in a way that's easier to understand. Contributed by Russ Cox. 10/30/02: beazley Added 'nonassoc' associativity support. This can be used to disable the chaining of operators like a < b < c. To use, simply specify 'nonassoc' in the precedence table precedence = ( ('nonassoc', 'LESSTHAN', 'GREATERTHAN'), # Nonassociative operators ('left', 'PLUS', 'MINUS'), ('left', 'TIMES', 'DIVIDE'), ('right', 'UMINUS'), # Unary minus operator ) Patch contributed by Russ Cox. 10/30/02: beazley Modified the lexer to provide optional support for Python -O and -OO modes. To make this work, Python *first* needs to be run in unoptimized mode. This reads the lexing information and creates a file "lextab.py". Then, run lex like this: # module foo.py ... ... lex.lex(optimize=1) Once the lextab file has been created, subsequent calls to lex.lex() will read data from the lextab file instead of using introspection. In optimized mode (-O, -OO) everything should work normally despite the loss of doc strings. To change the name of the file 'lextab.py' use the following: lex.lex(lextab="footab") (this creates a file footab.py) Version 1.1 October 25, 2001 ------------------------------ 10/25/01: beazley Modified the table generator to produce much more compact data. This should greatly reduce the size of the parsetab.py[c] file. Caveat: the tables still need to be constructed so a little more work is done in parsetab on import. 10/25/01: beazley There may be a possible bug in the cycle detector that reports errors about infinite recursion. I'm having a little trouble tracking it down, but if you get this problem, you can disable the cycle detector as follows: yacc.yacc(check_recursion = 0) 10/25/01: beazley Fixed a bug in lex.py that sometimes caused illegal characters to be reported incorrectly. Reported by Sverre Jørgensen. 7/8/01 : beazley Added a reference to the underlying lexer object when tokens are handled by functions. The lexer is available as the 'lexer' attribute. This was added to provide better lexing support for languages such as Fortran where certain types of tokens can't be conveniently expressed as regular expressions (and where the tokenizing function may want to perform a little backtracking). Suggested by Pearu Peterson. 6/20/01 : beazley Modified yacc() function so that an optional starting symbol can be specified. For example: yacc.yacc(start="statement") Normally yacc always treats the first production rule as the starting symbol. However, if you are debugging your grammar it may be useful to specify an alternative starting symbol. Idea suggested by Rich Salz. Version 1.0 June 18, 2001 -------------------------- Initial public offering