View on
MetaCPAN is shutting down
For details read Perl NOC. After June 25th this page will redirect to
Terrence Brannon > Parse-RecDescent-FAQ-7.5 > Parse::RecDescent::FAQ::Original



Annotate this POD


New  10
Open  0
View/Report Bugs
Module Version: 6.0.j   Source  


Parse::RecDescent::FAQ::Original - the original, official, authorized FAQ for Parse::RecDescent.


Is Parse::RecDescent LL(1)? LL(N)? LR(1)? LR(N)?

The best answer to this question is this thread:

Bu also see the "Left-recursion" section under "PARSER BEHAVIOR"

    I have a data structure which is
    a hash of entries
    an entry is a list/array of sets
    I have also a grammar that can parse the syntax of the text files that
    contain the data I want to fill this structure with. Until here
    everything is ok.
    Problem: I cannot figure out how to actually FILL the parsed data into
    the structure. I can only decide if a string is grammatically correct
    or not.
Answer by Yves Orton:
    { sub consumer {
             my $text = shift;
             my $closure = sub { substr $text, 0, length($text) - 
    length($_[0]) }
    } }
    start : entry(s) { $::res }
    entry :
             | def
             | <error>
    def : <rulevar: local $p_cons>
    def : <rulevar: local $p_text>
    # The // skips initial whitespace so it won't end up in $p_text
    def :
         // { $p_cons = consumer($text) } phrase { $p_text = 
    $p_cons->($text) }
         '=' meanings ';'
         { $::res->{$p_text} = $item{meanings} }
    comment : /#.*(?=\n)/m
    phrase  : ident(s)
    ident   : /[\w&\.'-]+/
    meanings : meaning(s /:/)
    meaning : <rulevar: local $m_cons>
    meaning : // { $m_cons = consumer($text) } element(s /,?/) 
    { $m_cons->($text) }
    element : alternation(s /\|/)
    alternation : expr(s /[+>]/)
    expr : /!?/ term
    term : ident '(' meaning ')' | ident


all I did was add an action!

It may seem like a trivial thing to add a print action to the end of a rule, but that can cause problems.

From we read:

 My code quit working when I added
   {print "Character: $item[1]\n"}
 to the end of the C<character> rule in the grammer below... any idea why?
 use strict;
 use warnings;
 use Parse::RecDescent;
 my $grammar = q {
   start    :  character character character(s)
               {print "Found: ", $item[1], $item[2], join "", @{$item[3]}, "\n"; }
   character:  /\w/ {print "Character: $item[1]\n"}
 my $parser = Parse::RecDescent->new($grammar);
 defined $parser->start("ABCD") or die "didn't match";
Answer by ikegami
    My code quit working when I added
      {print "Character: $item[1]\n"}
    to the end of the C<character> rule in the grammer below... any idea why?
    use strict;
    use warnings;
    use Parse::RecDescent;
    my $grammar = q {
      start    :  character character character(s)
                  {print "Found: ", $item[1], $item[2], join "", @{$item[3]}, "\n"; }
      character:  /\w/ {print "Character: $item[1]\n"}
    my $parser = Parse::RecDescent->new($grammar);
    defined $parser->start("ABCD") or die "didn't match";


Flags you can set

Paraphrased from Yves Orton:

To understand more about why your parser is not behaving as it should take advantage of the following variables:

   $::RD_ERRORS       # unless undefined, report fatal errors
   $::RD_WARN         # unless undefined, also report non-fatal problems
   $::RD_HINT         # if defined, also suggestion remedies
   $::RD_TRACE        # if defined, also trace parsers' behaviour

Making warning line numbers correspond to your grammar

How do I match the line numbers with the actual contents of my script?

Solution 1

(from Holger Helmuth)

A possible solution is to put anything more complicated than a trivial statement into a subroutine outside of the parser. To call it you have to qualify it with the package name.

Without this (maybe trivial?) method I probably would have given up on P:RD on my last project. Disadvantage is that the code is distributed into two locations and there should be a speed penality too, but it was definitely worth it.

Solution 2

At present, you can't (but that's on the ToDo list). Setting $::RD_TRACE can be useful though:

Once you've run with $RD_TRACE, do this:

        perl -w RD_TRACE

Then go and examine the actual line numbers given for the error in the file RD_TRACE.

That will show you the actual generated code that's the problem.

That code will, in turn, give you a hint where the problem is in the grammar (e.g. find out which subroutine it's in, which will tell you the name of the offending rule).


Floating or Substring Matches

Oftentimes, you may want to find if a grammar pattern exists anywhere in the input string, not necessarily right from the start.

In this post:

titled "Re: floating / backtracking - style match", Ted Zlatnov provides an elegant solution to getting the job done.


Take the following regex:


It will match this text:


Now take the following PRD grammar:

 use strict;
 use warnings;
 use Parse::RecDescent;
 my $grammar = 'startrule: ( "aa" | "a" ) "a"';
 my $parser  = Parse::RecDescent->new($grammar);
 my $text    = 'aa';
 print defined($parser->startrule($text)) ? "Good!\n" : "Bad!\n";

It *will* print "Bad", meaning that PRD did not match it.

The reason is that the first branch of the alternation matched, then the next subrule failed.

You may ask: is there any way to persuade a top-down parser like P::RD to accept the above text as valid? I know that, in this simple example, I could easily rewrite the grammar (either by saying

 ("a" |"aa" ) "a" 


 "aa" a" | "a" "a"

but what I mean is: Is there any additional feature I have missed which would allow the grammar as is to parse the text successfully?

To put the question another way, can I get P::RD to behave more like a regex engine? After all, even an NFA engine would backtrack to try all possible alternatives before failing :-) (Perhaps parsers just do not backtrack past individual subrules under any circumstances.)

And the answer is...

RecDescent parsers do not work that way. They don't backtrack on failure; they just fail. Of course, there's nothing to prevent a recursive descent parser from incorporating backtracking too, but RecDescent doesn't.

So, if you need backtracking in part of your grammar, you need to use plain old regexes there.

Skipping over things

The variable


is what you need. It takes strings and qr-quoted regular expressions. See the Parse::RecDescent docs for details.

Also, see the Terminal Separators parts of the docs.

Here is an example of removing C comments in program text (courtesy Damian Conway) by treating them as whitespace!

        program: <skip: qr{\s* (/[*] .*? [*]/ \s*)*}x> statement(s)

        statement: # etc...

However, thanks to Chas Owens, we have an update that works for him. Instead of just x as a regular expression modifier he had to add s as well:

 program: <skip: qr{\s* (/[*] .*? [*]/ \s*)*}xs> statement(s)

Quoting within a skip expression:

Here is my script:

 ------ Start Script ------
 use strict;
 use warnings;
 $::RD_TRACE = 1;
 use Parse::RecDescent;
 my $grammar = q{
    input:  number(s) { $return = $item{ number } } | <error>
    number: <skip: '\.*'> /\d+/ 

 my $parser = new Parse::RecDescent($grammar);
 my $test_string = qq{};

 print join( "\n", @{ $parser -> input( $test_string ) } );
 ------ End Script ------

This script works great. However, if I change the value of the skip directive so that it uses double quotes instead of single quotes:

 <skip: "\.*">

the grammar fails to parse the input. However, if I put square brackets around the escaped dot:

 <skip: "[\.]*">

the grammar starts working again...

and here's why:

This small test program may help you figure out what's going wrong:

        print "\.*", "\n";
        print '\.*', "\n";

Backslash works differently inside single and double quotes. Try:

      <skip: "\\.*">

The reason the third variant:

      <skip: "[\.]*">

works is because it becomes the pattern:


which is a literal dot.


Commit in subrule which is optional in rule

Ques: When a subrule in a rule that has a "zero or more" repetition specifier (ie. ? or s?) has a <commit> directive in its production followed by the conditional <error?> <reject> production, if that subrule's production becomes committed, does that error cause the rule containing the subrule to fail also? It should right, if we have committed? It does not seem to work.

Here is what I mean:

 myrule: 'stuff' mysubrule(?)

 mysubrule: ID <commit> '[' ']'
       | <error?> <reject>

If the 1st production of mysubrule has committed, then myrule should fail. It doesn't seem to. If this is not a bug, how do I get this behavior?

Answer by Damian

The optional nature of the reference to mysubrule(?) means that, when the subrule fails (whether committed or not) the failure doesn't matter, since myrule can match if it finds zero mysubrules, which it just did.

The usual way to get the rule-of-the-subrule to fail upon subrule failure is by "anchoring" the end of the match. That might be:

  myrule: 'stuff' mysubrule(?) ...!ID

  mysubrule: ID <commit> '[' ']'
           | <error?> <reject>


  myrule: 'stuff' mysubrule(?) /\s*\Z/

  mysubrule: ID <commit> '[' ']'
           | <error?> <reject>

or whatever addition confirms that there really wasn't anything else after 'stuff'.

Now that you think you know the answer...

That answer is partially wrong, as was pointed out by Marcel Grunaer. In this phrase:

  myrule: 'stuff' mysubrule(?) ...!ID

it is necessary to return a { 1 }, as the rule fails otherwise, presumably because of the negative lookahead:

  myrule: 'stuff' mysubrule(?) ...!ID { 1 }

Marcel went on to point out an optimization:

another option would be the use of a rulevar:

  myrule : <rulevar: local $failed>
  myrule : 'stuff' mysubrule(?) <reject:$failed>

  mysubrule: ID <commit> '[' ']'
    | <error?> { $failed++ }

this way you don't have to specify a potentially complex negative lookahead, and the method works over several levels of subrules as well.


Removing C comments

Since there is no separate lexer in recdescent. And it is top down. Is there anyway to deal w/ removing C comments that could be anywhere.


Parsing Windows Init (.ini) Files

I'm trying ot use Parse::RecDescent to parse a configuration file which looks like this:

 parameter 1=value 1
 parameter2=value 2

 other parameter=other value

As end of line

I'm trying to parse a text line by line using Parse::RecDescent. Each line is terminated by a "\n".

Although the task setting up a grammar for this case is straightforward the following program doesn't produce any results.

     use Parse::RecDescent;

     $grammar =
         line:       word(s) newline { print "Found a line\n"; }
         word:       /\w+/
         newline : /\n/

     $parse = new Parse::RecDescent ($grammar);

     $data =
     qq(This is line one\nAnd this is line two\n);


RecDescent doesn't recognize the newlines. Does anybody know what I'm getting wrong?

Example In a line-delimited grammar

 I want to use Parse::RecDescent to parse a certain assembly
 language. In assembly, data is "line oriented", that is newline is the
 statement separator. Simplified demonstration: 
 command1 arg1 \n
 command2 arg2 \n
 Defines two statements. More than one statement on a line is illegal,
 a statement broken to two lines is also illegal. 
 Now, P::RD has an issue with newlines. It swallows them by
 default. There is a workaround using the "skip" directive, but it
 became tedious and my grammar code is full of <skip: qr/ \t*/>
 Is there an easier way ?
 What I want, for starters, is a trivial "command arg" pairs for statements:
 line -> command arg "\n"


Whitespace, text, column N, period, number (some reference to lookahead)

Ok, now the line I'm trying to deal with is:

"some amount of whitespace, then some text, then starting at column 48 a number, followed by a period, followed by another number". I want to capture the text (the "title"), and the two numbers (major and minor versions)

Another example

I'm parsing some lines where the "column formatting" is fixed, i.e. a particular line might be formally described as "a single word followed by some amount of whitespace followed by another word whose first character begins at column 22".

Parsing Quotes within .... Quotes! ^

A recent node on Perlmonks had a question on how to parse:

    A statement "with a quoted ';'" ;

The best answer is the perl_quotelike section in the offical docs for Parse::RecDescent. An earlier node on Perlmonks: covers the use of perl_quotelike in detail.


Be sure to see the example see the "RESOURCES" section at the end for even more nosebleed parsing examples.

Subrule matches N times, where N is parsed on the fly

Per this link: we have a case where the future-matching is done as a function of the current matching.

Rule-building by interpolating an array of choices

Let's say you had to parse phrases like "6 red balls". You see this as "$quantity $color $item". Now, the colors which are acceptable for parsing vary from program to program and so the rule which matches the colors needs to be variable.

We see a fruitful discussion and solution of this problem here:

Macro processing

In a RecDescent grammar, I'd like to add support in my language for "macros". These macros would be very much like #define macros in C and would support arguments that are instantiated into the expansion. In C, this looks something like:

 #define   add(a,b)        (a+b)

In my language, I'd like something like:

 myMacro(x,y):        any_expression

where "any_expression" is any old (deep) expression defined elsewhere in the grammar, and (x,y) would be replaced/instantiated in that expression. This may seem straighforward, but, well, it ain't to me. (I am, after all, mush4brains.) In my P::RD grammar, the "macro" rule without arguments would look like:

 macro:         identifier ':' expr

where, again, expr is defined fully in the grammar. Adding arguments to this... I need to instantiate my (x,y) into the "expr" string before the subrules interpret it. Now... I thought of using standard Perl substitution in a pre-scan, but this seems less-than-ideal, since it requires some knowledge of the grammar in the Perl REs. I thought of using two separate grammars, the first of which is a simplified version of the full second grammar, but this also seems a bit redundant.

Cloning parsers but giving each parser it own package

It seems that the namespace for pre-compiled parsers (ie. compiled into a perl module) have hard-coded namespaces (IE. namespace00001). I was trying to clone one of these parsers by calling its 'new' method, but each parser is sharing the same namespace and thus any global variables I have in that namespace within a startcode block of my grammar before the first rule get overwritten by the other corresponding parser.

Parsing sentences to generate sentences

In this column, Randal shows how to read text to generate more text. He parses sentences to make Parse::RecDescent parse trees which he then re-walks with random weightings to create new sentences.

Calling a parser within a grammar

I have a script that uses Parse::RecDescent, in which I want to define 2 parsers. The grammar for the second parser has to call the first parser.

Can I do this?

Incremental generation of data structure representing parse

    I have a data structure which is
    a hash of entries
    an entry is a list/array of sets
    I have also a grammar that can parse the syntax of the text files that
    contain the data I want to fill this structure with. Until here
    everything is ok.
    Problem: I cannot figure out how to actually FILL the parsed data into
    the structure. I can only decide if a string is grammatically correct
    or not.

XOR as opposed IOR alternation matching

I'm using alternations in some productions but in contrast to the definition of the |-operator, I'm looking for a behaviour which is XOR (^) not OR (|). So far I used the <reject> directive to simulate such a behaviour.

Is there any easy solution to this?


In honor of the original (and greatest) Perl book on cleaning up your Perl code, this section is written in the style of Joseph Hall's "Effective Perl Programming"

Use repetition modifiers with a separator pattern to match CSV-like data

The intuitive way to match CSV data is this:

      NonFinalToken(s?) QuotedText
      QuotedText Comma
      { $return = $item[1] }

or, in other (merlyn's) words, "many comma terminated items followed by one standalone item".

Instead, take the approach shown by merlyn:

 CSVLine: QuotedText(s Comma) { use Data::Dumper; Dumper($item[1]) }

Then just define QuotedText, Comma, and you're done!


Dan Sugalski's Practice and Theory of Optimizing PRD Grammars

This link is excellent. So excellent in fact, that I took the time to mirror it to my site to avoid it ever being lost:

This link provides an excellent discussion of hits and misses in optimizing a grammar whose initial parse time started out at around minutes. After whittling it down to 5 minutes, Dan is still not happy and continues with even more aggressive optimizations! Well worth the read.

Eliminate backtracking when possible

Let's take a look at two computationally equivalent grammars:

 expression      :       unary_expr PLUS_OP expression
                 |       unary_expr


 expression      :       unary_expr plus_expression
 plus_expression :       PLUS_OP expression
                 |       # nothing

The second one is more efficient because it does not have to do backtracking.

The first one is more readable and more maintainable though. It is more readable because it doesnt have an empty rule. It is more maintainable because as you add more expression types (minus_expression, mult_expression...) you don't have to add an empty rule to each of them. The top level description scales without change.

But, if speed is what you want then the second one is the way to go.

Precompiling Grammars for Speed of Execution

Take a look at Parse::RecDescent's precompilation option under the section titled "Precompiling parsers".

Parse::RecDescent is slow on Really Big Files. How can I speed it up?


Hey! I'm getting back ARRAY(0x355300) instead of what I set $return to!

Here's a prime example of when this mistake is made:

       DoubleQuote TextChar(s?) DoubleQuote
       { my $chars = scalar(@item) - 1;  
         $return = join ('', @item[2..$chars]) }

This rule is incorrectly written. The author thinks that @item will have one TextChar from position 2 until all TextChars are matched. However, the true structure of @item is:

position one: the string matched by rule DoubleQuote
position two: array reference representing parse tree for TextChar(s?)
position three: the string matched by rule DoubleQuote

Note that position two is an array reference. So the rule must be rewritten in this way.

       DoubleQuote TextChar(s?) DoubleQuote
       { $return = join ( '', @{$item[2]} ) }

Getting text from subrule matches

I can't seem to get the text from my subrule matches...

Your problem is in this rule:

    tuple : (number dot)(2)

is the same as:

    tuple        : anon_subrule(2)

    anon_subrule : number dot

Like all subrules, this anonymous subrule returns only its last item (namely, the dot). If you want just the number back, write this:

    tuple : (number dot {$item[1]})(2)

If you want both number and dot back (in a nested array), write this:

    tuple : (number dot {\@item})(2)

Capturing whitespace between tokens

I need to capture the whitespace between tokens using Parse::RecDescent. I've tried modifying the $skip expression to // or /\b/ (so I can tokenize whitespace), but that doesn't seem to have the desired effect.

Just having a variable where all skipped whitespace is stored would be sufficient.

Does anybody know how to trick Parse::RecDescent into doing this?

My grammar is not returning any data!

What's wrong?!


Don't think that rule: statement and rule: statement(1) are the same

Even though in pure Perl, the repetition modifier returns the same data structure without or without an argument of one:

 use Data::Dumper;

 my @a = (x);
 my @b = (x) x 1;

 my $x = 'x';
 my $y = 'x' x 1;

In this first case below, rule_one returns a scalar upon matching, while rule_two returns an arrayref with 1 element upon matching:

 rule_one: statement
 rule_two: statement(1)

However counter-intuitive this may at first sound, Damian provides us with some insight:

I don't think x is the right analogy for RecDescent repetition operators. The behaviour of * and + in regexes is a closer model.

I guess it depends on your poitn of view. I would have said that the absolute consistency with which *every* repetition (regardless of its number) returns an array ref is better than the alternative:

                gold_rings(5)                   # returns array ref
                calling_birds(4)                # returns array ref
                French_hens(3)                  # returns array ref
                turtle_doves(2)                 # returns array ref
                partridge_in_a_pear_tree(1)     # returns scalar value

Especially if you have to change the count at some later point, which would mess up any code relying on the type of value returned.

Do not follow <resync> with <reject> to skip errors

resync is used to allow a rule which would normally fail to "pass" so that parsing can continue. If you add the reject, then it unconditionally fails.

Do not assume that %item contains an array ref of all text matched for a particular subrule

For example:

        range: '(' number '..' number )'
                        { $return = $item{number} }

will return only the value corresponding to the last match of the number subrule.

To get each value for the number subrule, you have a couple of choices, both documented in the Parse::RecDescent manpage under @item and %item.

use <rulevar: local $x> not <rulevar: $x>

If you say:

        somerule: <rulevar: $x>

you get a lexical $x within the rule (only). If you say:

        somerule: <rulevar: local $x>

you get a localized $x within the rule (and any subrules it calls).

Don't use $::RD_AUTOACTION to print while you are parsing

You can't print out your result while you are parsing it, because you can't "unprint" a backtrack.

Instead, have the final top-level rule do all the diagnostic printing, or alternatively use P::RD's tracing functionality to observe parsing in action.


Propagating a failure up after a <commit> on a subrule

See Commit in subrule which is optional in rule

In a non-shell (e.g. CGI) environment

I need to parse a file with a parser that I cooked up from Parse::Recdescent. My problem, it must work in a cgi environment and the script must be able to handle errors. I tried eval, piping, forking, Tie::STDERR, but the errors msgs from Parse::Recdescent seem unstoppable.

I can catch them only by redirecting the script's STDERR from the shell. But howdo I catch them from within ??

Accessing error data

What I want to do is export a <error: foo bar> condition to the calling program. Currently _error writes the error message to (what essentially is STDOUT) which means a parsing error prints a nice message, but it is up to the reader to DO anything about it.

I have a Tk viewer that will parse the file. When the parse fails, I would like to capture the line number and error message returned and position the viewer (a text widget) to that line, highlighted. And no STDOUT message.

Something kind of like the eval function where the return value is the result but $@ is set as a "side effect".

Am I missing something about the built in capability? Is the solution as simple as overloading the Parse::RecDescent::_error subroutine with my own copy which might look like this:

%Parse::RecDescent::ERROR=(); ... sub Parse::RecDescent::_error($;$) { $ERRORS++; return 0 if ! _verbosity("ERRORS"); %Parse::RecDescent::ERROR=('line'=>$_[1],'msg'=>$_[0]); return 1; }

It seems like it should work and I tried it and it did. BUT this is an extremely complex bit of code and I'm concerned about unforseen consequences.

Of course, I could make ERROR a my variable and provide a method to access it so it is not a "global", and this would be OO code of a higher purity (or something), but that is not really the point.

Simple Error Handling

I'm trying to write a parser for orders for Atlantis (PBEM game). Syntax is pretty simple: one line per command, each command starts with name, followed by list of parameters. Basically it's something like this (grammar for parsing one line):

 Statement:Comment | Command Comment(?)
 Command:'#atlantis' <commit> FactionID String
    Command:'attack' <commit> Number(s)

However I have problems to make it work as I want:

1) In case of failed parsing (syntax error, not allowey keyword, ...) I want to store error messages in variable (not to be just printed), so I can process them later.

I don't think Parse::RecDescent has a hook for that (Damian, something for the todo list?), but you can always install a $SIG {__WARN__} handler and process the generated warnings.

2) In case if user types "attack bastards" I want to give him error message that "list of numbers expected" instead of just saying the "cannot parse this line". The only thing that I came up with now was defining every command like this: Command:Attack Attack:'attack' AttackParams AttackParams:Number(s) | <error> ... Any better solutions?

Collecting all error messages for processing

Is there a way to to collect the parser-generated errors and use them later on in my script?

OTHER Parse::RecDescent QUESTIONS ^

How to get the old article to work?

Jethro of Perlmonks answers:

The problem is that return() finishes parsing.

Parse::RecDescent works with a variable $return to give back results from subrules to parent rules. If you change the three occurences of return inside the grammar (but not inside the sub 'expression') to $return= the parser works.

Matching line continuation characters

    I need to parse a grammar that includes line continuation
    characters.  For example:
        ARG3-VALUE,ARG4-VALUE, +
     * and a comment
     * or two
    How do I formulate a rule (or rules) to treat the first command
    as if all 5 arguments were specified on a single line?  I need to
    skip over the /\s*+\n\s*/ sequence.  It seems like skip or resync
    should do this for me, but if so, I haven't discovered the
    correct technique, yet.

How can I match parenthetical expressions to arbitrary depth?

Example: a, (b ,c, (e,f , [h, i], j) )

Switching out of first-match-wins mode

I have a set of alternatives on which I want to avoid the default first-match-wins behavior of Parse::RecDescent. How do I do it?

    Use a scored grammar. For example, this scoring directive
     opcode: /$match_text1/  <score: { length join '' @item}>
     opcode: /$match_text2/  <score: { length join '' @item}>
     opcode: /$match_text3/  <score: { length join '' @item}>
    would return the opcode with the longest length, as opposed to which
    one matched first.
    Just look for the section "Scored productions" in the .pod

I'm having problems with the inter-token separator:

     my $parse = Parse::RecDescent->new(<<'EndGrammar');
     rebol   : block  { dump_item('block', \@item)  }
             | scalar { dump_item('scalar', \@item) }
     block       : '[' block_stuff(s?) ']'
     block_stuff : scalar
     scalar      : <skip:''> '%' file
     file        : /w+/
    My grammar matches a filename, ie: 
    just fine. However, it does not match a filename within a block, ie: 
     [ %reb.html ]
    and I know exactly why after tracing the grammar. 
    It is trying the 
     <skip:''> '%' file
    production with the input text 
     " %reb.html"
    note the space in the input text. 
    The reason this distresses me is that I have not changed the universal token 
    separator from 
    Yet it did not gobble up the white space between the '[' terminal and the <skip:''>'%' file production

Matching blank lines

How do I match an arbitrary number of blank lines in Parse::RecDescent?

I have a rule which MUST be failing, but it isn't. Why?

   blank_line:    { $text =~ /silly-regex/ }
          parses with no error.

The pattern match still fails, but returns the empty string (""). Since that's not undef, the rule matches (even though it doesn't do what you want).

How can I get at the text remaining to be parsed?

See the documentation for the $text variable.

You don't escape Perl symbols in your grammars. Why did I have to?

 my $grammar = <<EOGRAMMAR;
 export_line:   stock_symbol    COMMA   # 1
                stock_name      COMMA2  # 2
                stock_code      COMMA3  # 3
                trade_side      COMMA4  # 4
                trade_volume    COMMA5  # 5
                floating_point  COMMA6  # 6
                tc                      # 7
 { print "got \@item\n"; }
     | <error>
 Why does '@' have to be escaped? And whatever reason
 that may be, why doesnt it apply to '\n'?

Other modules appear to not work when used with P::RD

Such-and-such a module works fine when I don't use Parse::RecDescent

 Did you alter the value of undef with your parser code? 

The problem has nothing to do with Parse::RecDescent.

Rather, it was caused by your having set $/ to undef, which seems to have caused Mail::POP3 to over-read from its socket (that might be considered a bug in the Mail::POP3 module).

As a rule-of-thumb, *never* alter $/ without local-izing it. In other words, change things like this:

         $/ = undef;

to this:

          local $/;


Getting large files into memory

An issue related to PRD is how to get huge files into memory for parsing. Check out this link:

for some approaches to this problems.

Double vs Single-quoted strings

I'm playing around with the <skip:> directive and I've noticed something interesting that I can't explain to myself.

Here is my script:

     use strict;
     use warnings;
     $::RD_TRACE = 1;
     use Parse::RecDescent;
     my $grammar = q{
        input:  number(s) { $return = $item{ number } } | <error>
        number: <skip: '\.*'> /\d+/ 
     my $parser = new Parse::RecDescent($grammar);
     my $test_string = qq{};
     print join( "\n", @{ $parser -> input( $test_string ) } );

This script works great. However, if I change the value of the skip directive so that it uses double quotes instead of single quotes:

    <skip: "\.*">

the grammar fails to parse the input. However, if I put square brackets around the escaped dot:

    <skip: "[\.]*">

the grammar starts working again:

How does this work this way?

Tracking text parsed between phases of the parse

    I wanted to know, after matching a rule, what text the rule matched.
    So I used two variables to remember what the remaining text and
    offset were before and after the rule and just determined the
       report : <rulevar: local $rule_text>
       report : <rulevar: local $rule_offset>
       report :
                     $rule_text   = $text;
                     $rule_offset = $thisoffset;
             ...some subrules...
                     my $str = substr($rule_text, 0, $thisoffset - 
                     # remove all sorts of whitespace
                     $str =~ s/^\s*//s;
                     $str =~ s/\s*$//s;
                     $str =~ s/\s+/ /gs;
                     # Now $str contains the text matched by this rule
    This is the kind of thing I thought would have been possible a lot
    easier. Did I miss something?
    If not, is there a way to make this available in every parser,
    e.g. by providing a new directive or something like that?

Unconditionally listifying scalars

Quite often when using Parse::RecDescent, I want to treat the return value of a production the same regardless of whether P::RD returns a string or a list of string.

Shallow versus Deep Copying

An article by Randal Schwartz


A tutorial written by Philip Newton <>

    Start off with an array of (references to) arrays: 
        @array = ( [1,2,3], ['a', 'u', 'B', 'Q', 'M'], ['%'] );
    Now a shallow copy looks like this: 
        @shallow = ( $array[0], $array[1], $array[2] );
    This copies the references over from @array to @shallow. Now @shallow
    is ( [1,2,3], ['a', 'u', 'B', 'Q', 'M'], ['%'] ) -- the same as
    @array. But there's only one 2 and one 'Q', since there are two
    references pointing to the same place.  
    Here's what it looks like in the debugger: 
      DB<5> x \@array
     0  ARRAY(0x10e5560)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'B'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
      DB<6> x \@shallow
     0  ARRAY(0xcaef60)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'B'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
    You can see that @array lives somewhere around 0x10e5560, whereas
    @shallow lives around 0xcaef60, but the three references point to
    arrays in the same place. If I now change $array[1][2] to 'C', watch
    what happens:  
      DB<7> $array[1][2] = 'C'
      DB<8> x \@array
     0  ARRAY(0x10e5560)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
      DB<9> x \@shallow
     0  ARRAY(0xcaef60)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
    $shallow[1][2] is now also 'C'! This is because it just followed the
    pointer to the array at 0x10e5638 and found the modified data there.  
    Now see what happens when I do a copy that's one level deeper -- not
    just copying the references but the data behind the references:  
     @deep = ( [ @{$array[0]} ], [ @{$array[1]} ], [ @{$array[2]} ] );
    This uses the knowledge that @array[0..2] are all references to
    arrays, and it only goes one level deeper. A more general algorithm
    (such as Storable's dclone, mentioned in `perldoc -q copy`) would do a
    walk and copy differently depending on the type of reference it
    encounters at each stage.  
    Now watch: 
      DB<12> x \@array
     0  ARRAY(0x10e5560)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
      DB<13> x \@deep
     0  ARRAY(0x10ef89c)
       0  ARRAY(0x10eb298)
          0  1
          1  2
          2  3
       1  ARRAY(0x10eb2c4)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10ef07c)
          0  '%'
    The references point to different places. 
    Now if you change @array, @deep doesn't change: 
      DB<14> push @{$array[2]}, '$'
      DB<15> x \@array
     0  ARRAY(0x10e5560)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
          1  '$'
      DB<16> x \@shallow
     0  ARRAY(0xcaef60)
       0  ARRAY(0x10e5464)
          0  1
          1  2
          2  3
       1  ARRAY(0x10e5638)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10e568c)
          0  '%'
          1  '$'
      DB<17> x \@deep
     0  ARRAY(0x10ef89c)
       0  ARRAY(0x10eb298)
          0  1
          1  2
          2  3
       1  ARRAY(0x10eb2c4)
          0  'a'
          1  'u'
          2  'C'
          3  'Q'
          4  'M'
       2  ARRAY(0x10ef07c)
          0  '%'
    @deep didn't change, since it's got its own value of the anonymous
    array containing '%', but @shallow did.  
    Hope this helps a bit. 
    Cheers, Philip -- Philip Newton <> 
    If you're not part of the solution, you're part of the precipitate

Apparent, but not really deep copying: my (@list) = @{[@{$_[0]}]};

I was meandering through in the Parse::RecDescent demo directory and came across this

 sub evalop
        my (@list) = @{[@{$_[0]}]};
        my $val = shift(@list)->();

I took the line that confused me step-by-step and don't get the purpose of this. Working from inner to outer:

   @{$_[0]}     # easy --- deference an array reference
 [  @{$_[0]} ]    # ok --- turn it back into an array ref.. why?
 @{ [ @{$_[0]} ] } # umm -- uh.... well, the @sign implies
                             # we have an array, but how is it 
                             # different from the first array we
                             # dereferenced?
    The line from is in
    fact not doing any deep copying.
        #!/usr/bin/perl -w
        my @original = (
            [0],  [1,2,3],  [4,5,6],  [7,8,9]
        my @copy = &some_kind_of_copy( \@original );
        sub some_kind_of _copy {
            # here's that line from
            my (@list) = @{[@{$_[0]}]};
            return @list;
     $original[0][0]         = 'zero';
     @{ $original[1] }[0..2] = qw(one   two   three);
     @{ $original[2] }[0..2] = qw(four  five  six);
     @{ $original[3] }[0..2] = qw(seven eight nine);
        # now use the debugger to look at the addresses,
        # or use Data::Dumper to look at @copy, or just
        # compare one of the items...
     if (  $copy[1][2] eq 'three'  ) {
        print "Shallow Copy\n";
     } elsif (  $copy[1][2] == 3  ) {
        print "Deep Copy\n";
     } else {
            print "This should never happen!!!\n"
    If you wanted that line to do deep copying of a list of anon arrays,
    then the line should read
        my @list = map  { [@$_] }  @{$_[0]};
                   # turn $_[0] into a list (of arrayrefs)
                   # turn each (arrayref) element of that list
                   # into an anonymous array containing
                   # a list found by derefrencing the arrarref
    Try plugging that line into above script instead of the line from the and you'll see different output. The line from
    is in fact doing extra useless work. My guess is that the extra
        @{[    ]}
    around there is one of two things:
        1) a momentary lapse of attention
           resulting in a copy/paste error, or duplicate typing
        2) an artifact of earlier code wherein something extra was
           going on in there and has since been deleted.
    Even Damian can make a mistake, but it's not a mistake that affects
    output... it just makes for a tiny bit of wasted work (or maybe Perl is
    smart enough to optimze away the wasted work, I dunno).

* Damian Conway says:

I have no recollection of why I did this (see children, that's why you should *always* comment your code!).

I *suspect* it's vestigal -- from a time when contents of the argument array reference were somehow modified in situ, but it was important that the original argument's contents not be changed.

The ungainly @{[@{$_[0]}]} syntax is a way of (shallow) copying the array referenced in $_[0] without declaring a new variable. So another possible explanation is that evalop may originally have been a one-liner, in which case I might have used this "inlined copy" to keep the subroutine's body to a single expression.


   Even Damian can make a mistake

is by far the likeliest explanation.

Regular Expressions

Shortest match instead of longest match

What is the Perl idiom for getting the leftmost shortest match? For instance, so:

 $a = "banana";
 $a =~ /b.*n/;  # but different
 print $&;

yields banan. How can I get it to yield ban?

Tutorial on positive and negative lookahead and lookbehind regexps:

Japhy's Book on Regular Expressions

Ok, so he hasn't named it yet. It's available online for free!

"Mastering Regular Expressions" by Jeffrey Freidl

You still need to know when to use /.*/ or /.+/ or /[^x]*/

Perlmonks Nodes ^

Basically the number of posts on Perlmonks about Parse::RecDescent are growing faster than I have time to study them closely and categorize them. But it is important that you have rapid access to them instead of slogging through Super Search or a google site search, so here you go:


Web Links

"Some Parse::RecDescent Tutorials"

Craig's Parse::Recdescent Area

Contains an article (in English) published during the Zweiter Perl Workshop as well as some slides for a presentation:

Hugh Myer's tips on Parse::RecDescent

17 tips you cannot do without... recently updated!

Parse::Recdescent tutorial at

Yes, I see the mis-spelling, but the link works. And in the Perl community, we are all about pragmatism!

Published articles and books

Parse Data::Dumper output

In the Linux Magazine article "Safe undumping" by Randal Schwartz uses Parse::RecDescent to parse Data::Dumper output. Not fast, but quite complete.

"Parsing Interesting Things"

SysAdminMag.COM, December 2001, Randal Schwartz

"Data Munging with Perl" by Dave Cross

Chapter 11 is focused on parsing with special emphasis on practical use of Parse::RecDescent.

"Object-Oriented Perl" by Damian Conway

This book will aid you in complexity management for large grammars.

CPAN Modules of Wonderful Utility!


Data::Match is an excellent tool for parse tree foraging. A typical autoaction is $::RD_AUTOACTION = q { [@item] } ;

which leads to deeply nested parse trees.

I first inquired about various tools for spelunking in such trees:

And am very happy so far with Data::Match. Here is a sample parse tree from my parse:

 $VAR1 = [

Now, my goal is to get at the inner-most array in $VAR1->[1]:


And with Data::Match, I can do so in a very definitional fashion:

 my $match = match
     # The parse tree
     # The Data::Match pattern match template
     FIND ( 
           COLLECT (
                     EXPR(q{! ref}),
                     EXPR(q{! ref})


The pattern is basically saying: "match an array ref consisting of 2 elements where each element is not an reference of any sort". Since the strings 'city' and 'taipei' both fulfill that criterion that is what matches.


Transparent manipulation of single or multiple-valued Perl hash values.


Parse::RecDescent::Consumer makes it easy to keep track of how much text has been consumed through successful matches.

Regexp::List (part of Regexp::Optimizer)

If you have a lot of alternations, it will be more efficient to test a single pre-compiled, optimized regular expression than a bunch of strings. And Regexp::List is designed to help you do just that.


Regexp:Assemble|Regexp::Assemble is a similar tool to Regexp::List, except that instead of just providing plain strings, you can give it a series of regular expressions, for it to assemble into one large regular expression.

Text::Balanced and Regexp::Common::balanced

Use this instead of writing hairy regular expressions to match certain common "balanced" forms of text, such as tags and parenthesized text.

See for an example of where it would be better suited for the task than Parse::RecDescent.


Sometimes in grammars, you are not sure if you will end up with a scalar match or an array of results. But you will often want to process them both the same. Scalar::Listify facilitates this process.

Practical Parser Examples

DBFeeder / Data::Generate - database test data generation tool is a a powerful data generation tool. Internally, it uses Data::Generate to parse "inverted regular expressions" using Parse::RecDescent.

Tangential new directions

In a recent perlmonks thread: the author of DBFeeder (Davide Conti) discusses SQL parsing with Parse::RecDescent (among other tools). This technology is not currently a part of DBFeeder and Data::Generate, but an avenue he is exploring.

Article comparing Gene Parsing with PRD and other approaches

"The Making of Entrez Gene parsers in Perl using Parse::RecDescent, Parse::Yapp, Perl-byacc and Regex" by Dr. Mingyi Liu

is available from:

Module dependency expression parser

This code by Marc Prewitt was a response to Mark Dominus' Perl 'Expert' Quiz of the Week #24 (Module dependency evaluation). This quiz of the week was a practical challenge to find code that could end up useable within Module::Build. The requirement was to allow for the module prerequisite to be specified using boolean logic, e.g.:

        requires => q[
                 (DBD::Pg && DateTime::Format::Pg)
                 (DBD::mysql && DateTime::Format::mysql)

If we need to, we can also include version specifications:

             requires => q[
                 ( DBD::Pg > 1.1 && DateTime::Format::Pg )
                 ( DBD::mysql <= 1.2 && DateTime::Format::mysql )

The neat thing about the solution is that after coding it, Marc had PRD generate a parser that ran independantly from PRD itself.

This solution, according to the judge, "... is probably the best solution to use if full expression parsing is desired for Module::Build."

Marc's outline of his solution and link to full code is available here:

XSH - The XML Editing Shell


uses Parse::RecDescent

Parsing Symbolic Expressions / Boolean Logic

Says Stefan Mueller:

Parse::RecDescent can easily parse such expressions. You can find a working example grammar for more involved *arithmetic* expressions in the Math::Symbolic::Parser module that is part of the Math::Symbolic distribution. It should be fairly easy to modify it to work for boolean logic.

MySql to Oracle schema conversion utility

Written in Parse::RecDescent by Tim Bunce:

Other Parsing Engines


py, which I assume is short for parse yacc, is a program by Mark-Jason Dominus which parses GNU Bison output to produce Perl parsers.

It is obtainable from


A bottom-up parser which will be familiar to those who have used Lex and Yacc. Parse::RecDescent is a top-down parser.






Domain-specific parsers

On CPAN there are many parsers for particular tasks such as CSV, HTML, XML, etc.

a VIM Syntax File for Parse::RecDescent provides a VIM syntax file for prd.



The author of Parse::RecDescent is Damian Conway.

The author of Parse::RecDescent::FAQ is Terrence Brannon <>.

The (unwitting) contributors to this FAQ

Sources for FAQ Material

I try to regularly scan all sources of Perl question-answer for FAQ material. If you can think of another source that is not listed below, I would appreciate knowing about it.

Source Repository for this FAQ ^

syntax highlighting: