++ed by:
DBOEHMER

1 PAUSE user

Terrence Brannon

NAME

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

OVERVIEW-TYPE QUESTIONS

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

Answer by Yves Orton:

I have a data structure which is

a hash of entries where 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.

Also see the "Left-recursion" section under "PARSER BEHAVIOR"

DEBUGGING

PARSER BEHAVIOR

Insuring a top-level rule match

I have a question regarding subrules within grammars used with Parse::RecDescent - The following code best illustrates the best test case that I can identify which highlights my problem.

Note that with this test code, each of the lines within the __DATA__ section are tested against the grammar and the expected result, either pass or fail, is indicated by the 1 or 0 at the start of the line respectively.

However, despite these expected results, the grammar does not reject the lines arg1, and arg1,arg2,, which ideally should be rejected due to the incomplete match of the subrule (comma element) - In these cases, a trace shows that the subrule terminal comma is matched and consumed, despite the entire subrule, consisting of comma and element, not being matched. I am sure this is a relatively straight-forward oversight within the grammar on my part, but I am at a loss as to how to correct this.

 #!/usr/bin/perl
 
 use Parse::RecDescent;
 use Test::More 'no_plan';
 
 my $grammar = q
 {
 argument:       element ((comma element)(s))(?)
 
 element:        /[\w.-_]+/
 
 comma:          ','
 };
 
 my $sql = Parse::RecDescent->new( $grammar );
 while ( chomp( my $test = <DATA> ) )
 {
     my ( $result, $statement ) = split /,/, $test, 2;
     ok( defined $sql->argument( $statement ) == $result, $test );
 }
 
 __DATA__
 0,
 1,arg1
 0,arg1,
 1,arg1,arg2
 0,arg1,arg2,
 1,arg1,arg2,arg3
 

Answer by Randal L. Schwartz

It's matching a portion of the string, which is legal unless you also anchor the end of the pattern. I typically use /\z/ at the end of my top-level pattern.

Also, your comma-separated string can be parsed simply with

 argument: element(s /,/)

as shown in the P::RD examples on the manpage.

Backtracking (PRD doesn't do it)

Take the following regex:

  /(aa|a)a/

It will match this text:

  aaa

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" 

or

 "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

  $Parse::RecDescent::skip 

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.

Left-recursion

  • On elimination of left-recursion Randal Schwartz states:

    I had a fun time eliminating a mutual left-recursion problem for my "Data-Undumper" program, which used P::RD as an essential component. See my discussion of such at

     http://www.stonehenge.com/merlyn/LinuxMag/col29.html
  • Also, regarding elimination see this Perlmonks node:

     http://www.perlmonks.org/index.pl?lastnode_id=6364&node_id=153155
  • Regarding detection of left-recursion, Conway states:

    RecDescent does a complete graph traversal looking for n-ary left-recursion loops and fails to compile the grammar if it finds any left-recursive loop involving any number of rules. It has done this since its earliest versions.

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>

    or

      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.

Making warning line numbers correspond to your grammar

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

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).

IGNORABLE TOKENS

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.

  • Answer by Conway

    Sure. Treat them as whitespace!

    Do something like this:

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

NEWLINE PROCESSING

Parsing Windows Init Files

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

 [SECTION]
 parameter 1=value 1
 parameter2=value 2

 [NEXTSECTION]
 other parameter=other value
  • Answer by Damian

           q{
                    config:
                              section(s)
    
                    section:
                              <skip: ''>  section_label parameter(s)
                            | <error>
    
                    section_label:
                              /\[[A-Z]+\]\s*/
    
                    parameter:
                              key "=" value
    
                    key:
                              /[A-Za-z0-9 ]+/
    
                    value:
                              /.*\n/
            }
  • Another Answer

    Also take a look at the example in section 11.2 in "Data Munging with Perl" by Dave Cross.

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 =
     q{
         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);

     $parse->line($data);

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

  • Answer by Damian

    By default, P::RD skips all whitespace (*including* newlines) before tokens.

    Try this instead:

             line:    <skip: qr/[ \t]*/> word(s) newline 
                                    { print "Found a line\n"; }
             word:    /\w+/
             newline: /\n/

COLUMN-ORIENTED PROCESSING

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)

  • Answer by Damian Conway

    You really do want to use a regex here (to get the lookahead/backtracking that RecDescent doesn't do).

       line: m/^(\s*                        # leading whitespace
                  (.*?)                     # the title
                  (?:\s+(?=\d+\.\d+\s*$))   # the space preceeding the numbers
                )
                (\d+)                       # the major version number
                \.
                (\d+)                       # the minor version number
             /x
             <reject: length $1 != 47>
             { @{$return}{title major minor)} = ($2,$3,$4) }

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".

  • A simple answer that is wrong:

    Hmm, I guess I could make this simpler and do this:

    line: word <reject: $thiscolumn != 22> word word: /\S+/

    right?

    Wrong. And the reason why is that The

      <reject:...> 

    won't skip any whitespace after the first word.

    You instead would want:

            line: word '' <reject: $thiscolumn != 22> word
  • Restating it in the positive can be a GOTCHA:

    I'd state that in the positive instead:

        line: word '' { $thiscolumn == 22 } word 

    This seems nice and more to the point, but unfortunately a failing conditional yields a false value but not necessarily an undef value. So in this case, you might get back a 0 from evaluating this conditional, but unfortunately, that does not lead to failure.

    On the other hand, <reject> is exactly the same as the action { undef } and is guaranteed to make a production fail immediately.

    So if you would like to state the test in the positive, then do this:

       line: word '' { $thiscolumn == 22 || undef } word 

MODULAR / GENERATIVE / CREATIVE / HAIRY PARSING

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.

  • And the answer is...

     http://perlmonks.org/index.pl?node_id=176399

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.

  • Answer by Yves Orton

    Heres the required patch

     # Line 1692 of Parse::RecDescent reads
    
     my $nextnamespace = "namespace000001";
    
     # Add This Method
     sub set_namespace {
       local $_=$_[1];
       croak "$_ is not a legal namespace." if /[^\w]/;
       $_.="00" unless /\d+$/; # Ensure our namespace has a number at the end
       $nextnamespace=$_;
     }
    
     # And then just call
     Parse::RecDescent->set_namespace("MyNameSpace");

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.

 http://www.stonehenge.com/merlyn/LinuxMag/col04.html

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?

  • yes, here's an example

     #!/usr/local/bin/perl -w
     use strict;
     use Parse::RecDescent;
     
     $::RD_ERRORS = 1;
     $::RD_WARN = 1;
     $::RD_HINT = 1;
     
     our $text_to_parse = "";
     
     my $grammar1 = q{
     [...]
     }
     
     our $inner_parser = new Parse::RecDescent($grammar1);
     
     my $grammar2 = q{
     [...]
     
     rule: TEXT
            {
              $text_to_parse = $item{TEXT};
               if (defined $text_to_parse) { print "executing inner parse...\n"; }
               my $p_text = $inner_parser->startrule($text_to_parse);
            }
     
     [...]
     
     }
     

Incremental generation of data structure representing parse

I have a data structure which is

a hash of entries where 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 Marcel Grunaer

    Try this grammar, which you have to feed the input as one big string. It uses a global variable, $::res into which the results are assembled. At the end the variable is also returned for convenience.

    It basically parses a phrase and a list of meanings. Instead of reconstructing what it just parsed at each step, it checks the remaining text at various stages (using an idea taken from Parse::RecDescent::Consumer) to see what the 'phrase' or 'meaning' subrules just matched. The 'meanings' subrule then (implicitly) returns a reference to an array of 'meaning' strings. That arrayref is stored at the proper slot in the result hash.

    (Hope that explanation makes sense. I'm sure Damian can come up with a grammar that's way more elegant and efficient...)

     { sub consumer {
              my $text = shift;
              my $closure = sub { substr $text, 0, length($text) - 
     length($_[0]) }
     } }
     
     start : entry(s) { $::res }
     
     entry :
                comment
              | 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

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?

  • Answer by Randal Schwartz

    Use a set.

     use Parse::RecDescent;
     use Data::Dumper; $|++;
     my $parser = Parse::RecDescent->new(q{
     
     line: word(s) /\z/ {
     my @words = @{$item[1]};
     my %count;
     (grep ++$count{$_} > 1, @words) ? undef : \@words;
     }
     
     word: "one" | "two" | "three"
     
     }) or die;
     
     for ("one two", "one one", "two three one", "three one two one") {
       print "$_ =>\n";
       print Dumper($parser->line($_));
     }
     
     # which generates:
     
     one two =>
       $VAR1 = [
               'one',
               'two'
              ];
     one one =>
       $VAR1 = undef;
     two three one =>
       $VAR1 = [
               'two',
               'three',
               'one'
              ];
     three one two one =>
       $VAR1 = undef;
  • Embellishment by Damian Conway

    Furthermore, if uniqueness was something you needed to enforce more widely in your grammar, you could factor it out into a parametric rule:

            use Parse::RecDescent;
            use Data::Dumper; $|++;
            my $parser = Parse::RecDescent->new(q{
    
            line: unique['word'] /\z/ { $return = $item[1] }
    
            unique: <matchrule: $arg[0]>(s)
                {
                  my %seen;
                  foreach (@{$item[1]}) { undef $item[1] and last if
    $seen{$_}++ }
                  $return = $item[1];
                }
    
            word: "one" | "two" | "three"
    
            }) or die;
    
            for ("one two", "one one", "two three one", "three one two
    one") {
              print "$_ =>\n";
              print Dumper($parser->line($_));
            }

CLEANING UP YOUR GRAMMARS

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:

 CSVLine:
      NonFinalToken(s?) QuotedText
 NonFinalToken: 
      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!

OPTIMIZING YOUR GRAMMARS

Eliminate backtracking when possible

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

 expression      :       unary_expr PLUS_OP expression
                 |       unary_expr

versus

 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?

  • Reduce the "depth" of the grammar. Use fewer levels of nested subrules.

  • Where possible, use regex terminals instead of subrules. For example, instead of:

                    string: '"' char(s?) '"'
    
                    char:   /[^"\\]/
                        |   '\\"'
                        |   '\\\\'

    write:

                    string: /"([^"\\]|\\["\\])*"/
  • Where possible, use string terminals instead of regexes.

  • Use repetitions or <leftop>/<rightop> instead of recursion.

    For example, instead of:

                    list:  '(' elems ')'
                    elems: elem ',' elems
                         | elem

    write:

                    list: '(' <leftop: elem ',' elem> ')'

    or:

                    list: '('  elem(s /,/)  ')'
  • Factor out common prefixes in a set of alternate productions.

    For example, instead of:

                    id: name rank serial_num
                      | name rank hair_colour
                      | name rank shoe_size

    write:

                    id: name rank (serial_num | haircolour | shoe_size)
  • Pre-parse the input somehow to break it into smaller sections. Parse each section separately.

  • Precompile they grammar. This won't speed up the parsing, but it will speed up the parser construction for big grammars (which Really Big Files often require).

  • Consider whether you would be better porting your grammar to Parse::Yapp instead.

CAPTURING MATCHES

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:

 QuotedText: 
       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.

 QuotedText: 
       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?

  • Answer by Damian Conway

    To turn off whitespace skipping so I can handle it manually, I always use:

            <skip:''>

    See:

            demo_decomment.pl
            demo_embedding.pl
            demo_textgen.pl

    for examples.

My grammar is not returning any data!

What's wrong?!

  • Answer by Brent Dax:

    This is a clue; either something is wrong with your actions or the grammar isn't parsing the data correctly. Try adding | <error>

    clauses to the end of each top-level rule. This will tell you if there's a parsing error, and possibly what the error is. If this doesn't show anything, look hard at the actions. You may want to explicitly set the $return variable in the actions.

THINGS NOT TO DO

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:

        christmas:
                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.

ERROR HANDLING

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 ??

  • Like this:

            use Parse::RecDescent;
            open (Parse::RecDescent::ERROR, ">errfile")
                    or die "Can't redirect errors to file 'errfile'";
    
            # your program here

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.

  • Answer by Damian Conway

    You can get access to the error data by referring to the attribute $thisparser-{errors}> within a rule.

    $thisparser-{errors}> is a reference to an array of arrays. Each of the inner arrays has two elements: the error message, and the line number.

    So, for example, if your top-level rule is:

     start: subrule1 subrule2

    then you could intercept any errors and print them to a log file, like so:

     start: subrule1 subrule2
          | { foreach (@{$thisparser->{errors}}) {
                  print LOGFILE "Line $_->[1]:$_->[0]\n";
              }
              $thisparser->{errors} = undef;
            }

    Note that the line:

     $thisparser->{errors} = undef;

    is doing double duty. By resetting $thisparser-{errors}>, it's preventing those annoying error messages from being automagically printed. And by having the value undef as the last value of the action, it's causing the action to fail, which means the second production fails, which means the top rule fails, which means errors still cause the parse to fail and return undef.

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(?)
 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?

  • You can just do:

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

    and when you try to parse "attack bastards", you will get:

        ERROR (line 1): Invalid Command: Was expecting Number but found
            "bastards" instead

    You might want to use <error?>, which will only print the error when it saw '#atlantis' or 'attack' (because then you are committed).

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?

  • Answer by Damian

    There's no "official" way (yet). A future release will allow you to create pluggable OO error handlers. At the moment the best you can do is:

            startrule: try_this
                     | try_that
                     | try_the_other
                     | { ::do_something_with( $thisparser->{errors} ) }

    $thisparser->{errors} will contain a reference to an array-of-arrays, where each inner array contains the error message and the line number it occurred at.

OTHER Parse::RecDescent QUESTIONS

Matching line continuation characters

I need to parse a grammar that includes line continuation characters. For example:

 // COMMAND ARG1-VALUE,ARG2-VALUE, +
    ARG3-VALUE,ARG4-VALUE, +
    EVEN-MORE-ARGS
 // ANOTHERCOMMAND
 * 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.

  • Answer by Damian Conway

     use Parse::RecDescent;
     
     my @lines = << 'EOINST';
     // COMMAND ARG1-VALUE,ARG2-VALUE, +
        ARG3-VALUE,ARG4-VALUE, +
        EVEN-MORE-ARGS
     // ANOTHERCOMMAND
     * and a comment
     * or two
     EOINST
     
     my $parse = Parse::RecDescent->new(join '', <DATA>) or die "Bad Grammar!";
     
     use Data::Dumper 'Dumper';
     print Dumper [
     $parse->Instructions("@lines") or die "NOT parsable!!\n"
     ];
     
     __DATA__
     
     Instructions: command(s)
     
     command: multiline_command
            | singleline_command
            | comment
     
     singleline_command: 
            '//'  /.*/
                    { {command => $item[-1]} }
     
     multiline_command:  
            '//' /(.*?[+][ \t]*\n)+.*/
                    { $item[-1] =~ s/[+][ \t]*\n//g; {command => $item[-1]} }
     
     comment:
            '*'  /.*/
                    { {comment => $item[-1]} }
     

How can I match parenthetical expressions to arbitrary depth?

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

  • Answer by FAQ author

    Maybe Text::Balanced is enough for your needs. See it on search.CPAN.org under author id DCONWAY.

  • Answer by lhoward of perlmonks.org:

    Parse::RecDescent implements a full-featured recursive-descent parser. A real parser (as opposed to parsing a string with a regular expression alone) is much more powerful and can be more apropriate for parsing highly structured/nested data like you have. It has been a while since I've written a grammer so it may look a bit rough.

     use Parse::RecDescent;
     my $teststr="blah1,blah2(blah3,blah4(blah5,blah6(blah7))),blah8";
     my $grammar = q {
             content:        /[^\)\(\,]+/
             function:       content '(' list ')'
             value:          content
             item:           function | value
             list:           item ',' list | item
             startrule:      list
     };
     my $parser = new Parse::RecDescent ($grammar) or die "Bad grammar!\n";
     
     defined $parser->startrule($teststr) or print "Bad text!\n";

    To which merlyn (Randal Schwartz) of perlmonks.org says:

    Simplifying the grammar, we get:

     use Parse::RecDescent;  
     my $teststr="blah1,blah2(blah3,blah4(blah5,blah6(blah7))),blah8";  
     my $grammar = q {
      list: <leftop: item ',' item> 
      item: word '(' list ')' <commit>
          | word 
      word: /\w+/  
     };  
     my $parser = new Parse::RecDescent ($grammar) or die "Bad grammar!\n";
     
     defined $parser->list($teststr) or print "Bad text!\n";
    
     

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 documentation.

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+/

 EndGrammar

My grammar matches a filename, ie:

 %reb.html

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

 /\s*/

Yet it did not gobble up the white space between the '[' terminal and the <skip:''>'%' file production

  • Answer by Randal Schwartz

    That's the expected behavior. The outer prefix is in effect until changed, but you changed it early in the rule, so the previous "whitespace skip" is effectively gone by the time you hunt for '%'.

    To get what you want, you want:

     '%' <skip:''> file

    in your rule. back

Matching blank lines

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

  • Answer by Damian Conway

    Unless you use the /m suffix, the trailing $ means "end of string", not "end of line". You want:

       blank_line:  /^\s+?$/m

    or

       blank_line:  /^\s+?\n/

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>
 EOGRAMMAR
 
 Why does '@' have to be escaped? And whatever reason
 that may be, why doesnt it apply to '\n'?
 
  • Answer by Damian Conway

    Because you're using an interpolating here document. You almost certainly want this instead:

     my $grammar = <<'EOGRAMMAR';           # The quotes are critical!
        
     
      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>
     EOGRAMMAR
     

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 $/;
              }

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 $&;

would yield "ban" instead of "banan"?

  •     $a =~ /b.*?n/;

Programming Topics Germane to Parse::RecDescent Use

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:

------ 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{1.2.3.5.8};

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:

How does this work this way?

  • Damian says:

    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.

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 difference.

   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 - 
$rule_offset);

                 # 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?

  • The answer is on CPAN

    Parse::RecDescent::Consumer, on CPAN, prints out the text consumed between stages of a parse... even if that part may fail later. The implementation is straightforward, it creates closures containing $text and evaluates them later to get the text consumed.

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.

A tutorial on Shallow versus Deep Copying

Written by "Philip 'Yes, that's my address' Newton" <nospam.newton@gmx.li>

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 <nospam.newton@gmx.li> 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 demo_calc.pl 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 demo_calc.pl 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 demo_calc.pl
            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 demo_calc.pl and you'll see different output. The line from demo_calc.pl 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 or 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.

    However...

       Even Damian can make a mistake

    is by far the likeliest explanation.

RESOURCES

  • MySql to Oracle schema conversion utility

    Written in Parse::RecDescent by Tim Bunce:

     http://groups.google.com/groups?q=recdescent&start=40&hl=en&scoring=d&rnum=41&selm=9km7b1%246vc%241%40FreeBSD.csie.NCTU.edu.tw
  • Data::MultiValuedHash on search.cpan.org

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

  • Parse::Recdescent tutorial at www.perl.com

    http://www.perl.com/pub/a/2001/06/13/recdecent.html

  • "Safe undumping"

    In this Linux Magazine article 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

  • py

    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 http://perl.plover.com/py/

  • Parse::YAPP

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

  • 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 http://perlmonks.org/index.pl?node_id=208285 for an example of where it would be better suited for the task than Parse::RecDescent.

  • "Mastering Regular Expressions" by Jeffrey Freidl

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

  • "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.

  • http://www.PerlMonks.org

    A useful site to get fast help on Perl.

AUTHOR

The author of Parse::RecDescent::FAQ is Terrence Brannon <tbone@cpan.org>.

The author of Parse::RecDescent is Damian Conway.

The (unwitting) contributors to this FAQ

  • Me

  • Damian Conway

  • Marcel Grunaer

  • Brent Dax

  • Randal L. Schwartz (merlyn), Perl hacker

  • lhoward of Perlmonks

  • Matthew Wickline

  • Gwynn Judd

7 POD Errors

The following errors were encountered while parsing the POD:

Around line 984:

=over should be: '=over' or '=over positive_number'

Around line 1463:

=over should be: '=over' or '=over positive_number'

Around line 1557:

=back doesn't take any parameters, but you said =back =head2 More on blank lines

Around line 1567:

=over should be: '=over' or '=over positive_number'

Around line 1632:

=over should be: '=over' or '=over positive_number'

Around line 2032:

=over should be: '=over' or '=over positive_number'

You can't have =items (as at line 2092) unless the first thing after the =over is an =item

Around line 2145:

Expected '=item *'