NAME Parser::MGC - build simple recursive-descent parsers SYNOPSIS package My::Grammar::Parser; use base qw( Parser::MGC ); sub parse { my $self = shift; $self->sequence_of( sub { $self->any_of( sub { $self->token_int }, sub { $self->token_string }, sub { \$self->token_ident }, sub { $self->scope_of( "(", \&parse, ")" ) } ); } ); } my $parser = My::Grammar::Parser->new; my $tree = $parser->from_file( $ARGV[0] ); ... DESCRIPTION This base class provides a low-level framework for building recursive-descent parsers that consume a given input string from left to right, returning a parse structure. It takes its name from the m//gc regexps used to implement the token parsing behaviour. It provides a number of token-parsing methods, which each extract a grammatical token from the string. It also provides wrapping methods that can be used to build up a possibly-recursive grammar structure, by applying a structure around other parts of parsing code. Backtracking Each method, both token and structural, atomically either consumes a prefix of the string and returns its result, or fails and consumes nothing. This makes it simple to implement grammars that require backtracking. Several structure-forming methods have some form of "optional" behaviour; they can optionally consume some amount of input or take some particular choice, but if the code invoked inside that subsequently fails, the structure can backtrack and take some different behaviour. This is usually what is required when testing whether the structure of the input string matches some part of the grammar that is optional, or has multiple choices. However, once the choice of grammar has been made, it is often useful to be able to fix on that one choice, thus making subsequent failures propagate up rather than taking that alternative behaviour. Control of this backtracking is given by the commit method; and careful use of this method is one of the key advantages that Parser::MGC has over more simple parsing using single regexps alone. Stall Detection Most of the methods in this class have bounded execution time, but some methods ("list_of" and "sequence_of") repeatedly recuse into other code to build up a list of results until some ending condition is reached. A possible class of bug is that whatever they recurse into might successfully match an empty string, and thus make no progress. These methods will automatically detect this situation if they repeatedly encounter the same string position more than a certain number of times (given by the stallcount argument). If this count is reached, the entire parse attempt will be aborted by the "die" method. CONSTRUCTOR new $parser = Parser::MGC->new( %args ); Returns a new instance of a Parser::MGC object. This must be called on a subclass that provides method of the name provided as toplevel, by default called parse. Takes the following named arguments toplevel => STRING Name of the toplevel method to use to start the parse from. If not supplied, will try to use a method called parse. patterns => HASH Keys in this hash should map to quoted regexp (qr//) references, to override the default patterns used to match tokens. See PATTERNS below accept_0o_oct => BOOL If true, the token_int method will also accept integers with a 0o prefix as octal. stallcount => INT Since version 0.21. The number of times that the stall-detector would have to see the same position before it aborts the parse attempt. If not supplied, a default of 10 will apply. PATTERNS The following pattern names are recognised. They may be passed to the constructor in the patterns hash, or provided as a class method under the name pattern_name. * ws Pattern used to skip whitespace between tokens. Defaults to /[\s\n\t]+/ * comment Pattern used to skip comments between tokens. Undefined by default. * int Pattern used to parse an integer by token_int. Defaults to /-?(?:0x[[:xdigit:]]+|[[:digit:]]+)/. If accept_0o_oct is given, then this will be expanded to match /0o[0-7]+/ as well. * float Pattern used to parse a floating-point number by token_float. Defaults to /-?(?:\d*\.\d+|\d+\.)(?:e-?\d+)?|-?\d+e-?\d+/i. * ident Pattern used to parse an identifier by token_ident. Defaults to /[[:alpha:]_]\w*/ * string_delim Pattern used to delimit a string by token_string. Defaults to /["']/. SUBCLASSING METHODS The following optional methods may be defined by subclasses, to customise their parsing. on_parse_start $parser->on_parse_start; Since version 0.21. If defined, is invoked by the from_* method that begins a new parse operation, just before invoking the toplevel structure method. on_parse_end $result = $parser->on_parse_end( $result ); Since version 0.21. If defined, is invoked by the from_* method once it has finished the toplevel structure method. This is passed the tentative result from the structure method, and whatever it returns becomes the result of the from_* method itself. METHODS from_string $result = $parser->from_string( $str ); Parse the given literal string and return the result from the toplevel method. from_file $result = $parser->from_file( $file, %opts ); Parse the given file, which may be a pathname in a string, or an opened IO handle, and return the result from the toplevel method. The following options are recognised: binmode => STRING If set, applies the given binmode to the filehandle before reading. Typically this can be used to set the encoding of the file. $parser->from_file( $file, binmode => ":encoding(UTF-8)" ); filename $filename = $parser->filename; Since version 0.20. Returns the name of the file currently being parsed, if invoked from within "from_file". from_reader $result = $parser->from_reader( \&reader ); Since version 0.05. Parse the input which is read by the reader function. This function will be called in scalar context to generate portions of string to parse, being passed the $parser object. The function should return undef when it has no more string to return. $reader->( $parser ); Note that because it is not generally possible to detect exactly when more input may be required due to failed regexp parsing, the reader function is only invoked during searching for skippable whitespace. This makes it suitable for reading lines of a file in the common case where lines are considered as skippable whitespace, or for reading lines of input interactively from a user. It cannot be used in all cases (for example, reading fixed-size buffers from a file) because two successive invocations may split a single token across the buffer boundaries, and cause parse failures. pos $pos = $parser->pos; Since version 0.09. Returns the current parse position, as a character offset from the beginning of the file or string. take $str = $parser->take( $len ); Since version 0.16. Returns the next $len characters directly from the input, prior to any whitespace or comment skipping. This does not take account of any end-of-scope marker that may be pending. It is intended for use by parsers of partially-binary protocols, or other situations in which it would be incorrect for the end-of-scope marker to take effect at this time. where ( $lineno, $col, $text ) = $parser->where; Returns the current parse position, as a line and column number, and the entire current line of text. The first line is numbered 1, and the first column is numbered 0. fail fail_from $parser->fail( $message ); $parser->fail_from( $pos, $message ); fail_from since version 0.09. Aborts the current parse attempt with the given message string. The failure message will include the line and column position, and the line of input that failed at the current parse position (fail), or a position earlier obtained using the pos method (fail_from). This failure will propagate up to the inner-most structure parsing method that has not been committed; or will cause the entire parser to fail if there are no further options to take. die die_from $parser->die( $message ); $parser->die_from( $pos, $message ); Since version 0.20. Throws an exception that propagates as normal for die, entirely out of the entire parser and to the caller of the toplevel from_* method that invoked it, bypassing all of the back-tracking logic. This is much like using core's die directly, except that the message string will include the line and column position, and the line of input that the parser was working on, as it does in the "fail" method. This method is intended for reporting fatal errors where the parsed input was correctly recognised at a grammar level, but is requesting something that cannot be fulfilled semantically. warn warn_from $parser->warn( $message ); $parser->warn_from( $pos, $message ); Since version 0.22. Issues a warning as normal for the warn core Perl function, appending information to the message string giving the current line and column position and the line of input the parser was working on, as it does in the "fail" method. at_eos $eos = $parser->at_eos; Returns true if the input string is at the end of the string. scope_level $level = $parser->scope_level; Since version 0.05. Returns the number of nested scope_of calls that have been made. include_string $result = $parser->include_string( $str, %opts ); Since version 0.21. Parses a given string into the existing parser object. The current parser state is moved aside from the duration of this method, and is replaced by the given string. Then the toplevel parser method (or a different as specified) is invoked over it. Its result is returned by this method. This would typically be used to handle some sort of "include" or "macro expansion" ability, by injecting new content in as if the current parse location had encountered it. Other than the internal parser state, other object fields are not altered, so whatever effects the invoked parsing methods will have on it can continue to inspect and alter it as required. The following options are recognised: filename => STRING If set, provides a filename (or other descriptive text) to pretend for the source of this string. It need not be a real file on the filesystem; it could for example explain the source of the string in some other way. It is the value reported by the "filename" method and printed in failure messages. toplevel => STRING | CODE If set, provides the toplevel parser method to use within this inclusion, overriding the object's defined default. STRUCTURE-FORMING METHODS The following methods may be used to build a grammatical structure out of the defined basic token-parsing methods. Each takes at least one code reference, which will be passed the actual $parser object as its first argument. Anywhere that a code reference is expected also permits a plain string giving the name of a method to invoke. This is sufficient in many simple cases, such as $self->any_of( 'token_int', 'token_string', ... ); maybe $ret = $parser->maybe( $code ); Attempts to execute the given $code in scalar context, and returns what it returned, accepting that it might fail. $code may either be a CODE reference or a method name given as a string. $ret = $parser->maybe( $code, @args ); Since version 0.22 this method passes any additional arguments into the invoked code. This is especially useful if the $code is a method name. If the code fails (either by calling fail itself, or by propagating a failure from another method it invoked) before it has invoked commit, then none of the input string will be consumed; the current parsing position will be restored. undef will be returned in this case. If it calls commit then any subsequent failure will be propagated to the caller, rather than returning undef. This may be considered to be similar to the ? regexp qualifier. sub parse_declaration { my $self = shift; [ $self->parse_type, $self->token_ident, $self->maybe( sub { $self->expect( "=" ); $self->parse_expression } ), ]; } scope_of $ret = $parser->scope_of( $start, $code, $stop ); Expects to find the $start pattern, then attempts to execute the given $code, then expects to find the $stop pattern. Returns whatever the code returned. $code may either be a CODE reference of a method name given as a string. While the code is being executed, the $stop pattern will be used by the token parsing methods as an end-of-scope marker; causing them to raise a failure if called at the end of a scope. sub parse_block { my $self = shift; $self->scope_of( "{", 'parse_statements', "}" ); } If the $start pattern is undefined, it is presumed the caller has already checked for this. This is useful when the stop pattern needs to be calculated based on the start pattern. sub parse_bracketed { my $self = shift; my $delim = $self->expect( qr/[\(\[\<\{]/ ); $delim =~ tr/([<{/)]>}/; $self->scope_of( undef, 'parse_body', $delim ); } This method does not have any optional parts to it; any failures are immediately propagated to the caller. committed_scope_of $ret = $parser->committed_scope_of( $start, $code, $stop ); Since version 0.16. A variant of "scope_of" that calls "commit" after a successful match of the start pattern. This is usually what you want if using scope_of from within an any_of choice, if no other alternative following this one could possibly match if the start pattern has. list_of $ret = $parser->list_of( $sep, $code ); Expects to find a list of instances of something parsed by $code, separated by the $sep pattern. Returns an ARRAY ref containing a list of the return values from the $code. A single trailing delimiter is allowed, and does not affect the return value. $code may either be a CODE reference or a method name given as a string. It is called in list context, and whatever values it returns are appended to the eventual result - similar to perl's map. This method does not consider it an error if the returned list is empty; that is, that the scope ended before any item instances were parsed from it. sub parse_numbers { my $self = shift; $self->list_of( ",", 'token_int' ); } If the code fails (either by invoking fail itself, or by propagating a failure from another method it invoked) before it has invoked commit on a particular item, then the item is aborted and the parsing position will be restored to the beginning of that failed item. The list of results from previous successful attempts will be returned. If it calls commit within an item then any subsequent failure for that item will cause the entire list_of to fail, propagating that to the caller. sequence_of $ret = $parser->sequence_of( $code ); A shortcut for calling list_of with an empty string as separator; expects to find at least one instance of something parsed by $code, separated only by skipped whitespace. This may be considered to be similar to the + or * regexp qualifiers. sub parse_statements { my $self = shift; $self->sequence_of( 'parse_statement' ); } The interaction of failures in the code and the commit method is identical to that of list_of. any_of $ret = $parser->any_of( @codes ); Since version 0.06. Expects that one of the given code instances can parse something from the input, returning what it returned. Each code instance may indicate a failure to parse by calling the fail method or otherwise propagating a failure. Each code instance may either be a CODE reference or a method name given as a string. This may be considered to be similar to the | regexp operator for forming alternations of possible parse trees. sub parse_statement { my $self = shift; $self->any_of( sub { $self->parse_declaration; $self->expect(";") }, sub { $self->parse_expression; $self->expect(";") }, sub { $self->parse_block }, ); } If the code for a given choice fails (either by invoking fail itself, or by propagating a failure from another method it invoked) before it has invoked commit itself, then the parsing position restored and the next choice will be attempted. If it calls commit then any subsequent failure for that choice will cause the entire any_of to fail, propagating that to the caller and no further choices will be attempted. If none of the choices match then a simple failure message is printed: Found nothing parseable As this is unlikely to be helpful to users, a better message can be provided by the final choice instead. Don't forget to commit before printing the failure message, or it won't count. $self->any_of( 'token_int', 'token_string', ..., sub { $self->commit; $self->fail( "Expected an int or string" ) } ); commit $parser->commit; Calling this method will cancel the backtracking behaviour of the innermost maybe, list_of, sequence_of, or any_of structure forming method. That is, if later code then calls fail, the exception will be propagated out of maybe, no further list items will be attempted by list_of or sequence_of, and no further code blocks will be attempted by any_of. Typically this will be called once the grammatical structure alter has been determined, ensuring that any further failures are raised as real exceptions, rather than by attempting other alternatives. sub parse_statement { my $self = shift; $self->any_of( ... sub { $self->scope_of( "{", sub { $self->commit; $self->parse_statements; }, "}" ), }, ); } Though in this common pattern, "committed_scope_of" may be used instead. TOKEN PARSING METHODS The following methods attempt to consume some part of the input string, to be used as part of the parsing process. expect $str = $parser->expect( $literal ); $str = $parser->expect( qr/pattern/ ); @groups = $parser->expect( qr/pattern/ ); Expects to find a literal string or regexp pattern match, and consumes it. In scalar context, this method returns the string that was captured. In list context it returns the matching substring and the contents of any subgroups contained in the pattern. This method will raise a parse error (by calling fail) if the regexp fails to match. Note that if the pattern could match an empty string (such as for example qr/\d*/), the pattern will always match, even if it has to match an empty string. This method will not consider a failure if the regexp matches with zero-width. maybe_expect $str = $parser->maybe_expect( ... ); @groups = $parser->maybe_expect( ... ); Since version 0.10. A convenient shortcut equivalent to calling expect within maybe, but implemented more efficiently, avoiding the exception-handling set up by maybe. Returns undef or an empty list if the match fails. substring_before $str = $parser->substring_before( $literal ); $str = $parser->substring_before( qr/pattern/ ); Since version 0.06. Expects to possibly find a literal string or regexp pattern match. If it finds such, consume all the input text before but excluding this match, and return it. If it fails to find a match before the end of the current scope, consumes all the input text until the end of scope and return it. This method does not consume the part of input that matches, only the text before it. It is not considered a failure if the substring before this match is empty. If a non-empty match is required, use the fail method: sub token_nonempty_part { my $self = shift; my $str = $parser->substring_before( "," ); length $str or $self->fail( "Expected a string fragment before ," ); return $str; } Note that unlike most of the other token parsing methods, this method does not consume either leading or trailing whitespace around the substring. It is expected that this method would be used as part a parser to read quoted strings, or similar cases where whitespace should be preserved. nonempty_substring_before $str = $parser->nonempty_substring_before( $literal ); $str = $parser->nonempty_substring_before( qr/pattern/ ); Since version 0.20. A variant of "substring_before" which fails if the matched part is empty. The example above could have been written: sub token_nonempty_part { my $self = shift; return $parser->nonempty_substring_before( "," ); } This is often useful for breaking out of repeating loops; e.g. sub token_escaped_string { my $self = shift; $self->expect( '"' ); my $ret = ""; 1 while $self->any_of( sub { $ret .= $self->nonempty_substring_before( qr/%|$/m ); 1 } sub { my $escape = ( $self->expect( qr/%(.)/ ) )[1]; $ret .= _handle_escape( $escape ); 1 }, sub { 0 }, ) return $ret; } generic_token $val = $parser->generic_token( $name, $re, $convert ); Since version 0.08. Expects to find a token matching the precompiled regexp $re. If provided, the $convert CODE reference can be used to convert the string into a more convenient form. $name is used in the failure message if the pattern fails to match. If provided, the $convert function will be passed the parser and the matching substring; the value it returns is returned from generic_token. $convert->( $parser, $substr ) If not provided, the substring will be returned as it stands. This method is mostly provided for subclasses to define their own token types. For example: sub token_hex { my $self = shift; $self->generic_token( hex => qr/[0-9A-F]{2}h/, sub { hex $_[1] } ); } token_int $int = $parser->token_int; Expects to find an integer in decimal, octal or hexadecimal notation, and consumes it. Negative integers, preceded by -, are also recognised. token_float $float = $parser->token_float; Since version 0.04. Expects to find a number expressed in floating-point notation; a sequence of digits possibly prefixed by -, possibly containing a decimal point, possibly followed by an exponent specified by e followed by an integer. The numerical value is then returned. token_number $number = $parser->token_number; Since version 0.09. Expects to find a number expressed in either of the above forms. token_string $str = $parser->token_string; Expects to find a quoted string, and consumes it. The string should be quoted using " or ' quote marks. The content of the quoted string can contain character escapes similar to those accepted by C or Perl. Specifically, the following forms are recognised: \a Bell ("alert") \b Backspace \e Escape \f Form feed \n Newline \r Return \t Horizontal Tab \0, \012 Octal character \x34, \x{5678} Hexadecimal character C's \v for vertical tab is not supported as it is rarely used in practice and it collides with Perl's \v regexp escape. Perl's \c for forming other control characters is also not supported. token_ident $ident = $parser->token_ident; Expects to find an identifier, and consumes it. token_kw $keyword = $parser->token_kw( @keywords ); Expects to find a keyword, and consumes it. A keyword is defined as an identifier which is exactly one of the literal values passed in. EXAMPLES Accumulating Results Using Variables Although the structure-forming methods all return a value, obtained from their nested parsing code, it can sometimes be more convenient to use a variable to accumulate a result in instead. For example, consider the following parser method, designed to parse a set of name: "value" assignments, such as might be found in a configuration file, or YAML/JSON-style mapping value. sub parse_dict { my $self = shift; my %ret; $self->list_of( ",", sub { my $key = $self->token_ident; exists $ret{$key} and $self->fail( "Already have a mapping for '$key'" ); $self->expect( ":" ); $ret{$key} = $self->parse_value; } ); return \%ret } Instead of using the return value from list_of, this method accumulates values in the %ret hash, eventually returning a reference to it as its result. Because of this, it can perform some error checking while it parses; namely, rejecting duplicate keys. TODO * Make unescaping of string constants more customisable. Possibly consider instead a parse_string_generic using a loop over substring_before. * Easy ability for subclasses to define more token types as methods. Perhaps provide a class method such as __PACKAGE__->has_token( hex => qr/[0-9A-F]+/i, sub { hex $_[1] } ); * Investigate how well from_reader can cope with buffer splitting across other tokens than simply skippable whitespace AUTHOR Paul Evans