bionic (3) Marpa::R2::Changes.3pm.gz

NAME

       Marpa::R2::Changes - Differences between Marpa::R2 and Marpa::XS

About this document

       This document describes the incompatible differences between Marpa::XS and Marpa::R2.  (Differences that
       do not give rise to incompatibility are outside of its scope.)  It is intended for readers already
       familiar with Marpa::XS, who are writing new applications for Marpa::R2, and for readers migrating
       Marpa::XS applications and tools to Marpa::R2.

Changes

   Additional reserved symbol names
       Marpa::XS reserved, for its internal use, all symbol names ending with the right square bracket (""]"").
       In addition, Marpa::RS reserved symbols ending with the right parenthesis ("")""), the right angle
       bracket ("">""), and the right curly bracket (""}"").  Any other valid Perl string remains an acceptable
       symbol name.

   The return value of the read() method has changed
       The return value of the Marpa::R2 recognizer's "read()" method differs from its Marpa::XS equivalent.  In
       Marpa::XS it returned the number of distinct terminals (by symbol ID) allowed in the next "read()".  In
       Marpa::R2 it returns the number of recognizer events that occurred during the read.  Examples of
       recognizer events are exhaustion, the Earley sets exceeding a designated "warning" level, and other
       circumstances settable by the user.  For more detail, see the documentation of recognizer's "read"
       method.

   Rule LHS's are no longer a source of action names
       In Marpa::XS, if there was no explicit action name for a rule, Marpa would try to find a closure that had
       the same name as the rule's LHS.  The use of rule LHS's as action names had a potential for unpleasant
       surprises.  A surprise could occur if the rule's LHS coincided with a function name without the
       prorgrammer realizing or intending it.  This kind of 'action at a distance' bug can be very hard to
       detect and trace.

       It was originally thought that implicitly using the LHS as the name of an action would be convenient
       enough to outweigh the dangers.  But in fact, this feature wound up being little used.  And accidental
       resolution via a rule LHS was a danger for all users, whether they used the feature or not.  For these
       reasons, as well as potential optimization and efficiency considerations, Marpa::R2 no longer does
       implicit action resolution using a rule LHS.

   Different rules with the same rank now appear in arbitrary order
       In ranking parse trees, if two rule instances are for different rules but have the same rule rank, they
       will now appear in arbitrary order.  This is probably the behavior that programmers have always expected.

       In Marpa::XS, when the "null_ranking" named argument of rules was in use for one of the rules, specific
       guarantees were made for the order in some of the cases.  The intent was to be orthogonal with the
       guarantees made for the ranking of null variants within the same rule.  These additional guarantees
       proved useless in practice, cumbersome to implement, and, when documented, opaque and unintuitive.  In
       Marpa::R2 they have been dropped.

   Null actions now come from the rules
       In Marpa::XS null actions were specified by symbol.  This created a dual semantics -- one for non-nulled
       rules, and another for nulled rules.   The conventions and behaviors of the two semantics were quite
       dissimilar.  The rules for their coordination were complicated, and it was possible for a programmer
       expecting one semantics, to be surprised by a result from the other.

       In Marpa::R2 the semantics of nulled rules is the same as that of non-nulled rules, and the semantics of
       nulled symbols comes from the semantics of the nulled rules.  This requires rule evaluation closures to
       be aware they might be called for nulled rules.  But it greatly simplifies the semantics conceptually.
       For more detail, see Marpa::R2::NAIF::Semantics::Null.

   Actions can now be constants
       If an action name resolves to a constant, that constant is the action.  The effect is the same as if the
       action name resolved to a function that returned that constant, except that it is more efficient.

       Perl cannot reliably distinguish between non-existent symbols and symbols whose value is "undef", so
       constants whose value is "undef" are not allowed.  The "::undef" reserved action name can be used
       instead.

   Actions names beginning with ""::"" are reserved
       Action names which start with ""::"" are reserved.  ""::undef"" is a safe way of specify a constant whose
       value is "undef".  Use of a reserved name which has not yet been defined causes an exception to be
       thrown.

   The "default_null_value" named argument for grammars has been removed
       Symbols no longer have null values, so the "default_null_value" named argument of grammars has been
       removed.

   The "null_value" symbol property has been removed
       Symbols no longer have null values.  Use of the "null value" symbol property now causes an exception.

   The token value argument of read() has changed
       The Marpa::R2 recognizer's "read()" method differs from its Marpa::XS equivalent.  In Marpa::R2, If
       "read()"'s token value argument is omitted, then the value of the token will be a Perl "undef".  If
       "read()"'s token value is given explicitly, then that explicit value will be the value of the token.  In
       particular, an explicit "undef" token value argument will behave differently from an omitted token value
       argument.  For details, see the documentation of recognizer's "read" method.

   The token value argument of "alternative()" has changed
       The Marpa::R2 recognizer's "alternative()" method differs from its Marpa::XS equivalent.  Its token value
       argument must now be a reference to the token value, not the token value itself, as in Marpa::XS.  If
       alternative's token value argument is omitted or a Perl "undef", then the value of the token will be a
       Perl "undef".  If alternative's token value argument is reference to "undef", then the value of the token
       is a Perl "undef".  For details, see the documentation of the "alternative" method.

   Marpa::R2::Recognizer::value() does not accept named arguments
       In the Marpa::XS recognizer, the "new()", "set()" and value() methods all accepted named arguments.  As
       of Marpa::R2, the "value()" method will no longer do so.

       Allowing named arguments for the "value()" was a holdover from a previous interface, which also seemed
       like it might be a convenience.  But, since it was even more important that the "value()" method be
       convenient as the termination test controlling a loop over the parse results, a lot of special logic was
       added to deal with arguments which only made sense before the first pass of the loop, etc., etc.

       Eliminating named arguments from the "value()" method eliminates a variety of special cases and, as a
       result, the documentation of the "value()" method is now simpler, shorter and clearer.  Anything that
       could be done by providing named arguments to the "value()" method can be done more using the
       recognizer's "set()" method, and the code will be clearer for it.

   Marpa's grammar rewriting is now invisible
       Internally, Marpa rewrites its grammars.  In Marpa::XS, most details of these rewrites were invisible,
       but not all.  In Marpa::R2, all internal rules and symbols are now completely invisible to the user, even
       in the tools for debugging grammars.

   By default, the non-LHS symbols are the terminals
       Traditionally, a symbol has been a terminal if it is not on the LHS of any rule, and vice versa.  This is
       now the default in Marpa::R2, replacing the more complicated, and less intuitive, scheme that was in
       Marpa::XS.  Marpa::R2 still allows the user to use any non-nulling symbol as a terminal, including those
       symbols that appear on the LHS of a rule, but this is now an option, and never the default.  For more,
       see "Terminal symbols" in Marpa::R2::NAIF::Grammar.

   The lhs_terminals grammar named argument has been eliminated
       The lhs_terminals named argument of grammar objects implemented what is now the default behavior.  Since
       it no longer performs a function, its use is now a fatal error.

   Nulling symbols cannot be terminals
       In Marpa::XS, it was possible for a symbol to be both nulling and a terminal.  In practice that meant
       that the symbol was nulling, but that, on input, that property could be overriden, and a specific
       instance of the nulling symbol could be made non-nulling.  This behavior was worse than useless and non-
       intuitive -- it was dangerous and logically inconsistent.

       Marpa::R2 will not allow a nulling symbol to be used as a terminal.  To the extent that the Marpa::XS
       behavior made sense, it can be duplicated by creating a symbol which is the LHS of two rules, one empty,
       and the other rule with a RHS consisting of exactly one terminal symbol.

   A sequence must have a unique LHS
       The LHS of a sequence rule may not be on the LHS of any other rule, whether another sequence rule, or a
       BNF rule.  This is not as severe a restriction as it might sound -- while sequences cannot share the same
       LHS with other rules directly, they can do so indirectly.  For details, see "Duplicate rules" in
       Marpa::R2::NAIF::Grammar.

       In Marpa::XS, the definition of when a sequence was a duplicate was more liberal, but it was also
       complicated and non-intuitive.  The new definition is simpler and more intuitive, and its greater
       restrictiveness is easy to work around.

   The terminal status of a symbol is locked once set
       Once a symbol is marked as a terminal or a non-terminal, its terminal status cannot be changed.  We doubt
       this will affect any actual applications.  It would only affect an application that changes symbols from
       their default status to non-terminal, and then only if they attempted to mark the same symbol as a
       terminal at another point.  Few Marpa::R2 applications change symbols from their default terminal status,
       and none to my knowledge mark symbols as non-terminals.

   Evaluation of infinite loops has been changed
       Infinite loops (cycles) are still, by default, fatal errors.  For those considering programming with
       them, and evaluating parses from grammars with cycles, the semantics of cycles is now more closely
       specified.  For details of the new semantics, see Marpa::R2::NAIF::Semantics::Infinite.

   The range of values allowed for ranks has been clarified
       Symbols and rules have numeric ranks.  Previously, no mention was made of range of values allowed.  This
       is implemented-defined, except that the magnitudes of the ends of the range will always be at least the
       28th power of 2, less 1.  That is, numbers in the range between -134,217,727 and 134,217,727 will always
       be allowed as ranks.

Copyright and License

         Copyright 2014 Jeffrey Kegler
         This file is part of Marpa::R2.  Marpa::R2 is free software: you can
         redistribute it and/or modify it under the terms of the GNU Lesser
         General Public License as published by the Free Software Foundation,
         either version 3 of the License, or (at your option) any later version.

         Marpa::R2 is distributed in the hope that it will be useful,
         but WITHOUT ANY WARRANTY; without even the implied warranty of
         MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
         Lesser General Public License for more details.

         You should have received a copy of the GNU Lesser
         General Public License along with Marpa::R2.  If not, see
         http://www.gnu.org/licenses/.