Kelbt 0.12 - May 1, 2007
========================
 -Don't emit empty switch blocks for trial, undo and final actions.
 -Ignore write statements that are not at the root of the include stack.
 -Added an option (-l) for turning off the writing of line directives.
 -Improved error messages.
 -A commit is now always a full commit of the stack. It induces any remaining
  final actions and cleans up the stack. Motivation: it should be possible to
  reliably control when final actions are executed. In the previous scheme
  reliable control was not possible because if retries existed ahead of the
  scoped commit then the number of retries would not drop to zero and the final
  actions would not be executed. Further, no important practical uses for the
  scoped commit have surfaced. That is, there are no examples of needing to
  commit a production while preserving earlier alternatives. Therefore the added
  complexity of the scoped commit feature is not justified.
 -The malloc used by the final action execution/commit code has been eliminated
  and replaced with a pointer mechanism.
 -Added a class keyword for nonterminals and the token type. The keyword goes
  directly in front of the nonterm/type/token name. Constructors and
  destructors are invoked for nonterminal class types and the token type when
  it is a class.
      nonterm class Foo {};
      type class Foo {};
      token uses class MyToken;
 -Added the "write finish" statement which invokes destructors for the items
  still on the stack. 
 -Blocks of LangEls are now tracked and freed in the finish code.
 -Kelbt now uses Ragel's import feature instead of repeating the definition of
  tokens inside the section parser.
 -The branch point printing function (-i) now prints dot sets to give you a
  sense of where in the grammar the conflicts are.
 -Added the shortest statement. This changes the default action ordering from
  a longest-match strategy to a shortest-match strategy. To make this work
  reliably the nonterminal which is to be made into a shortest match should be
  wrapped in a fresh nonterminal which is then specified using the shortest
  statement. This syntax will likely be improved in the future.
      shortest lifted_list;
      lifted_list: list;
      list: list item;
      list: ;
 -Did some cleanup of the syntax highlighting. Removed junk left over from the
  port of the ragel syntax file a long time ago.
 -Fixed an missing variable initialization. This eliminates errors reported by
  valgrind. 
 -Added an install target to the makefiles.

Kelbt 0.11 - Feb 12, 2007
=========================
 -The commit and final block execution code now uses malloc to allocate an
  array of pointers instead of using a variable-sized array on the stack,
  which is not portable C.
 -Applied more fixes to cxxprep from Gilles J. Seguin.
 -Updated to the latest aapl.

Kelbt 0.10 - Jan 27, 2007
=========================
 -Applied fixes to cxxprep from Gilles J. Seguin.
 -Adopted Ragel's write statement style of generating code. The types of
  statements are:
      write instance_data;   - the runtime data used by the parser.
      write token_defs;      - pound defines for the tokens.
      write types;           - data structures representing language elements.
      write data;            - static data used by the parser.
      write init;            - initialization code.
      write exec;            - the parse loop.
 -The access statement was added. It can be used to specify where the instance
  data is located. For example, inside a structure:
      access parser->;
 -Removed return statements from the exec loop. Errors and regular exits now
  just drop out at the bottom of exec.
 -The generated code has been made compatible with C.
 -Added a C test.
 -The generated data structures and static data elements now have their names
  prefixed with the parser name.
 -The "token uses" statement was added. This allows the user to change the
  name of the data structure which contains token data. It defaults to
  "Token".
      token uses ParserToken;
 -Implemented single-line parser specifications. This is useful mostly for
  write statements.
 -Fixes to line directive writing were made.

Kelbt 0.9 - Jan 22, 2007
========================
 -Fix to initialization code: the sentinel at the top of the stack needs
  to have its pointers initialized.

Kelbt 0.8 - Jan 16, 2007
========================
 -Added a priority mechanism for tweaking control of the parser. Priorities
  are associated with characters in productions and these take precedence over
  the default ordered choice strategy. The ordered choice strategy is also an
  'innermost' strategy where all the possibilities within a non-terminal take
  precedence over any characters which follow the non-terminal. Priorities let
  one override this in particular cases.
 -Line directives are now generated for inline code blocks which are passed
  through to the output file.
 -Access to the non-terminal element and the right hand side elements of the
  production being reduced now works properly in final actions.

Kelbt 0.7 - Dec 8, 2006
=======================
 -Added a credits file.
 -Kelbt no longer generates code which writes 'parse error' when parsing
  fails. This can be done by the caller of parseLangEl after looking at the
  return value.
 -The scanner is now a class. It tracks the current line and guards against
  passing tokens to the parser if no parser name has been given.
 -Separated the parsing of the 'parser' and 'include' statements from the
  scanner.

Kelbt 0.6 - Dec 2, 2006
=======================
 -Replaced the Flex scanner with a Ragel scanner.
 -Replaced the Bison parser with a Kelbt-0.5 parser. Kelbt is now
  self-hosting. Error handling is now broken and must be added back.
 -The code was cleaned up: a large amount of code left over from the initial
  fork from Ragel was removed.
 -Bug fix: when the retry point is at the root of the stack, meaning it is the
  only real element, we cannot consult the transition that the sentinel
  element below it took because that transition does not exist. There is no
  state and no type associated with the sentinel. This case is now caught and
  uses the start state directly.

Kelbt 0.5 - Nov 18, 2006
========================
 -Fixed production-based commits which became broken when the state table
  construction was improved in the previous version.
 -The commit, execFinal and final functions were moved into the main
  processing loop. The execFinal that was called in finish() is now called
  from the main loop by setting a commit on the final shift of the eof
  transition which completes the parse.
 -The specified parser name is now used as the name of parser class when the
  functions are generated.
 -The startState variable was moved to the static table data section. The
  mutable parser data is no longer static. 
 -The struct block was removed, as it is now depreciated.
 -Can no longer compile test cases with -pedantic and -ansi due to variable
  length arrays on stack.

Kelbt 0.4 - Oct 30, 2006
========================
 -A considerable amount of code cleanup and was done. This includes removal of
  unused code, simplification of data structures and rewrites of LR state
  table construction functions. The unnecessary complexity was because kelbt
  was derived from a version of a keller which supported right-regular
  parsing. Along with the cleanup comes a speedup in LR table construction.
 -An improvement was made to the ordering algorithm. When it's time to assign
  an ordering to a reduction after recursing down a production, the old way
  was to find the expand-to state of the production and assign an ordering to
  any transition which had a reduction of the production. The problem here is
  that it assumes that every transition that has the reduction can follow that
  instance of the production. In reality this is not true. There may be some
  transitions that have the reduction, but which come from a different
  instance of the production. Now we make sure that we assign a time only to
  reductions on characters which can follow the production instance.  We do
  this by computing the characters on the fly. This gives a more accurate
  ordering.
  
Kelbt 0.3 - Aug 8, 2006
=======================
 -Fixed a bug in LALR(1) state table construction which caused reductions
  actions for some middle-recursive grammars to not be added. For example this
  grammar fails to set a reduce of A-2 on the ')' character.
     S: A;
     S: B;
     A: '(' A ')'; 
     A: '(' 'x' ')';
     B: '(' B ')'; 
     B: '(' 'x' ')';

Kelbt 0.2 - Aug 1, 2006
=======================
 -Fixed the rejection command: It now immediately shifts the nonterminal being
  rejected before jumping to the parse error handling. Previously this
  nonterminal would get lost and would muck things up.
 -Added a test showing variable length list processing.
 -The cxxprep example can use the aapl in the root kelbt package.

Kelbt 0.1 - Jun 26, 2006
========================
 -Initial release.
