Coco/R for C and C++ ==================== Version 1.00 - Apr-13-1994 - First released version Version 1.02 - Sep-15-1994 - Minor bug repaired in SCANNER.FRM file. - Some bugs found in CRA.C (incorrect deterministic automata on transitions with character sets with some elements in common; not a complete inclusion). Algorithm corrected. - Minor bug fixed in the CONTEXT management of CRA.C (reported by Universidad Rafael Landivar's students). - Minor bug repaired in State0 vector generation (thank to Jean Carlo Solis of the Universidad Rafael Landivar for the fix). - Improved portability in the source code and generated scanners still further (thank to Martin Mueller of the University of New Mexico for changes to NextStep). - Improved internal table dump routines. Version 1.03 - Oct-12-1994 - Fixed minor bug in SCANNER.FRM file when using MS-DOS. (O_BINARY added to the open mode options. Thanks to L. Barrera and L. Bocalleti of the Universidad del Valle de Guatemala for the fixes). Version 1.04 - Jan-15-1995 - Minor bug repaired in the documentation for Cocol (the grammar for productions had semantic actions before "=", needed only for Modula or Pascal versions). - Simple bug fixed in the generation of literal token generation. - Corrected some of the .FRM files when checking the number of arguments to the compiler. - Corrected the management of TAB characters in reporting an error's column number. (TAB character is assumed to be 8 characters long, same as UNIX TAB). Thank to Markus Klingspor of Wuerzburg University, Germany, for reporting the bugs and his suggested corrections. Version 1.05 - Mar-10-1995 - Corrected compiler.frm: (Included some function header definitions to work with GNU g++, thanks to Markus Klinsgpor). - Improved performance of the generated scanner. - Corrected a bug in CRA.C in the CONTEXT management. - Corrected an interesting bug in GenCompiler(), which had remained unnoticed for so long, because it works in many DOS and UNIX compilers (it did not work in Watcom C++ - thanks again for discovering the problem to Markus Klingspor and his students) Version 1.06 - Mar-29-1996 - Support provided for C++. The scanner and parser may optionally be generated as classes. This allows you to have multiple scanners and parsers in the same program. (Thanks to Pat Terry, Markus Klinsgpor, and some of our students for comments, testing and for fixing some bugs.) - Fixed a bug in the parser's ExpectWeak function. (Thanks to Pat Terry.) - Special thanks to Pat Terry of Rhodes University for testing and improvements to many of the new object classes, source code and examples. - Support added for customizing file extensions in the generated parsers and scanners (CRCEXT and CRHEXT environment variables). - Improved source debugging (-d option) on the ".atg" file. With these new fixes, you can set breakpoints on any part of the ".atg" file. Previous versions only supported breakpoints on semantic actions. Version 1.061 - Aug-27-1996 - Fixed a bug in the comments generation. (Pat Terry) Version 1.062 - Nov-25-1996 - Fixed a bug in the expression graph construction. Function crt.c:LinkGraph. It did not work fine when using meaningless "(" ... ")" constructs like: ("A" "B") "C". Thanks to Joe Armstrong for reporting Version 1.07 - Oct-10-1997 - Coco/R did not report an error when using a CHARACTER SET name in the productions. Thanks to Barrie Stott (UK) for discovering the bug. - Incorrect token names were generated for token strings starting with a digit character (for example "2bOrNot2b"), or for names that became too long. Thanks to Soteri Panagou (RU) for reporting these bugs. - Fixed a bug that sometimes allowed undefined character set names to cause the system to loop - Fix all main drivers frames (compiler.frm, cr.frm, etc.) to support relative and absolute paths in filenames. In the Unix version, only the "-" character is valid to supply the command line options. /cplus2/cr_error.[hc]pp was also fixed. All other frame files kept the same as in version 1.06. - Improved documentation compatible with that in Pat Terry's book "Compilers and Compiler Generators" (International Thomson, 1997, ISBN 1-85032-298-8) Version 1.08 - Mar-31-1998 - Support for MS Visual C++ 5.0 - Support for defining variables in the command line using the -Dvarname=value option. For example: cocor -DCRHEXT=hxx -DCRCEXR=cxx -DCRFRAMES=..\frm taste.atg NOTE: a simple -D without a variable means DEBUG! - Reorganized the code to support user defined file extensions. Version 1.09 - Aug-24-98 - Minor cosmetic changes to adapt code to allow for greater consistency with Java version; a few minor bugs fixed in comment handling. - Taste package slightly improved - Fixed a bug in the scanner generator (ConvertToStates function) - Fixed bug in use of high ASCII characters in strings - Literal token strings of zero length no longer allowed Version 1.10 - Oct-06-98 - Fixed elusive bug in CRA module in AddMeltSets that caused problems with very complex scanners - Fixed bug in CRX.C that occasionally resulted in first sets being incorrectly initialized in parser. - Added debugging assert macros to help detect obscure memory leaks; small changes in CRA, COLLECT and CRT to handle this. Version 1.11 - Nov-05-98 - Applied Ken Stout's fix to allow long file pointers - Eliminated allowing incomplete strings within attributes and actions - Minor improvements to documentation Version 1.12 - Jul-09-99 - Attributes may now be demarcated by matching <. and .> or by < and >. This extension allows for the > character to appear in an actual parameter. - Omitting the closing .) in an action can lead to errors that are hard to find, so the digraph (. appearing within a semantic action is now flagged as an error (Thanks to Hans Ellenberger for this suggestion). - Small additions to C++ interface to allow position of token to be extracted, and text to be extracted from a known position Version 1.13 - Sep-02-99 - Fixed minor bug in parser that caused a crash when the TOKENS section had (illegally) supplied tokens with semantic attributes Version 1.14 - Nov-18-99 - Improved the scanner generator (LeftContext) - Fixed minor bug in comment handling when they were not properly sepcified - Changed hidden token names to EOF_Sym and No_Sym to avoid name clashes - Fixed errors in string handling with embedded quotes like 'xxx\'xxx' - Fixed nasty bug whereby EOF_Sym sometimes went missing from AllSyncs - Added CHR('A') as an alternative to CHR(65) in CHARACTER specification - Improved function header generation for void functions in C version - Better support for ANSI C compilers, WIN32 compilers and latest EGCS - for example, function prototypes use (void) instead of (), main returns an int and so on. - Added /q option to support interaction with editors (for exqample QEdit) - Most frame files changed to correspond to these changes. - Added -q directive to provide listing output more suited to interfacing with editors that can interrogate error messages by line, column. Version 1.15 - Apr-30-00 - Insistence that IGNORE CASE appear first if it appears at all - Improved the scanner generator (LeftContext No longer needed) Version 1.16 - June-19-01 - Minor bug fixed in scanner that caused trouble with tokens of the silly form ttt = hhh [ { hhh } ] . - Definition of string changed as below CHARACTERS cntl = CHR(0)..CHR(31). back = CHR(92) . noQuote1 = ANY - '"' - cntl - back . noQuote2 = ANY - "'" - cntl - back . graphic = ANY - cntl . TOKENS string = '"' {noQuote1 | back graphic } '"' | "'" {noQuote2 | back graphic } "'" . (This may break some existing source. The effect is that \" \' and \\ in strings are all treated as in C. Previously this was not so consistent. Version 1.17 - Sep-17-02 - Removed test for bad context clauses which has been proven wrong - Minor changes to CRA.C - Provided Borland C++ 5.5 32 bit make files for use with their free release compiler - Extra checks incorporated for bad comment description - Error message text altered =end=