Version 1.00 08/11/2000 This is the initial release version of the interpreter. Version 1.01 14/11/2000 - Fixed bug in code that reads programs tokenised with Acorn's tokens that affected the SUM' token. - Fixed problem with SYS code in RISC OS version of program to do with returning the processor flags. Only 'carry' was being returned. - Fixed bug in SYS code where SWIs that failed were always causing the Basic program to be halted with an error. Execution now continues if an 'X' form SWI is used. - 'QUIT=1' was not being flagged as a syntax error. Fixed. - If a line containing an error such as: 100 abcd was the last line of the program, the line number of the line in error was not being given in the error message. Fixed. Version 1.02 20/11/2000 - Fixed bug in NetBSD and Linux versions where the type of a parameter in a call to read() was wrong. This showed up on big endian machines. - Fixed bug where a statement such as DEF PROC1FORa%=1 TO 10 was being handled incorrectly. The 'FOR' toke was being taken as part of the procedure name. - Fixed bug with array references where the first array index started with an unary indirection operator. This was giving a 'type mismatch' error. - The program does not work very well on big endian machines. 1.02 should hopefully be better in this respect. - Compilation has been tidied up under NetBSD and Linux. The same makefile is now used for both. Version 1.03 22/11/2000 - Fixed 'PRINT ""1' bug where '"1 1' was output instead of just ' 1'. - More big endian problems sorted out. Version 1.04 26/11/2000 - Fixed bug where hex constants such as &FFFFFFFF were being stored incorrectly (only the least significant byte was being stored so &FFFFFFFF was showing up as &FF). - Yet more big endian and alignment issues sorted out. Version 1.05 01/12/2000 - Libraries can now have variables that are private to those libraries. - LIBRARY and INSTALL can now handle lists of libraries. - BPUT can now be followed by a list of values instead of just one. - OSCLI has been extended to allow the program to capture the output from commands. Version 1.06 15/12/2000 - Bug where functions could be used as procedures (and vice versa) fixed. - Fixed bug in the VAL function where it would sometimes return the wrong result as the string being converted was not terminated properly. - Fixed bug in BPUT introduced in 1.05. - HIMEM can now be changed. - Blank lines in files are no longer discarded when loading a program. - The initial value and step used if the lines of a program are renumbered when loading a program have been changed to 1 and 1 respectively, the idea being that they will match the numbers of the lines of the program in the file in an editor. - Under RISC OS, the 'edit' command can now invoke editors such as StrongED and Zap. - Added limited support for making BBC MOS calls via USR and CALL so that OS_Byte 0 could be used to determine the type of the machine on which the interpreter is being run. Version 1.07 01/01/2001 - Fixed bug in function SUM when the argument was a string array. The function should return all of the strings in the array concatenated together but it was producing rubbish. - Fixed bug in function SUM LEN. This was failing with the error message 'type mismatch: array wanted'. - Fixed bugs in the EOF and EXT functions that show up when the file is not a disk file but, for example, a serial port. - Sorted out some more portability issues, but at a cost of making the program about 10% slower. - The code for the OSCLI ... TO statement has been improved. - The RISC OS version of the program has been changed to use direct OS calls in fileio.c instead of C library functions. This allows Basic programs to carry out I/O operations on a file using a mixture of Basic statements and SWIs. - The INPUT# code has been tidied up. - Typing in token values directly on the command line is now handled correctly, for example, typing in the hex value &F1 no longer gives the error 'The interpreter has gone wrong'. - Under NetBSD and Linux, the handling of I/O redirection on the program's command line has been sorted out. It is now possible to invoke the program so that it takes input from or directs output to a file instead of using the keyboard for input and screen for output. - The QUIT command can now optinally be followed by a value that the interpreter passes back to the underlying operating system as a return or status code. Version 1.08 19/04/2001 - Fixed 'big endian' bug in code that checks if a file contains a tokenised Basic program. The test was failing on big endian machines. - Changed code that writes the four byte start marker at the start of a Basic program so that it is always written in the same order, that is, fixed another endian bug. - Extended the WAIT statement so that the time to wait can be supplied. The time interval is in centiseconds. Version 1.09 29/04/2001 - Fixed bug in INSTR where the end of the string being searched was missed when the first character of the wanted string occured a number of times in the search string. - Tidied up STR function so that STR$~ produced the same results as PRINT~, for 'STR$~255' now produced 'FF' instead of 'ff'. Also changed the format for floating point values so that an exponent is marked with an 'E' instead of an 'e'. - Fixed bug in INPUT statement where INPUT ' cleared the screen instead of skipping to the next line. - Fixed bug in INPUT statement where 'INPUT TAB() ' printed a '?' prompt when it should not have done so. - Fixed a problem in EVAL where the pointer to the expression that contained the EVAL function was being corrupted if the string being evaluated contained a reference to a function that had not been called before. 'EVAL(EVAL(a$))' now works as well. Version 1.10 28/05/2001 - Fixed bug in EDIT introduced fixing EVAL in version 1.09. A 'bad token' error message was being produced when editing a single line with EDIT under DOS and Unix. The amended line was being saved correctly but the error message was then being displayed. - Tidied up handling of @% in PRINT and STR$ when the number of digits to print is zero. '@%=0: PRINT PI' now produces 3.141592654 instead of 3, as per the Acorn interpreter. - Added function XLATE$. This either translates a string using a user-supplied translate table or translates it to lower case if no translate table is supplied. - Added function VERIFY. This is used to check that a string contains only specific characters. - Changed EDIT and EDITO so that EDIT uses the last EDITO value instead of LISTO when converting the program to text when it is edited. If EDITO has not been used, the LISTO value is used instead. Version 1.11 17/09/2001 Darren Salt (ds@youmustbejoking.demon.co.uk) contributed a number of the changes in this version of the program. - Added support for reading Acorn five byte floating point values (Darren Salt). - Improved support for reading and writing Acorn eight byte floating point values on machines that use a different format for this sort of number to what Acorn use (Darren Salt). - Added support for reading gzip'ed files under Linux (Darren Salt). - The Linux version of the program can be installed using 'configure' (Darren Salt). - Added code to detect the format of eight byte floating point numbers on the platform on which the interpreter is running. This is used when reading and writing such values (DD, based on original code by Darren Salt). - The first line of a program in text form is now ignored if it starts with a '#' in all versions of the interpreter. This only used to be supported in the NetBSD and Linux versions. - Changed handling of syntax errors detected when a line is tokenised. The interpreter now puts out a slightly more meaningful 'syntax error' message if a program with such an error in it is run instead of 'STOP'. - DIM statements of the form 'DIM abc%!8 100' are now dealt with correctly. Previously the code always wanted a variable as the location at which to store the address of the byte array and did not allow the use of an indirected expression. This was an oversight in the code and has now been fixed. - Fixed bug where the string memory management was not being reset under certain conditions when a program was edited. The program was being corrupted if a string expression was evaluated immediately prior to editing the program and then again immediately afterwards. - Fixed bug in ERROR statement code where ERROR 0, was not being treated as a fatal error, that is, the error could be trapped by ON ERROR. Also fixed problem where ERR was not returning the user-supplied error number after an ERROR statement. - Fixed bug in handling of ON ERROR LOCAL where a storage overwrite was causing the interpreter to abort with a 'interpreter has gone wrong' message when restoring LOCAL variables to their original values when dealing with an ENDPROC after branching to the statements after the ON ERROR LOCAL. - Fixed bug in READ statement that affected the reading of strings in cases such as 'READ a$,b$: DATA abcd,', that is, there is nothing after the last comma in the DATA statement. It should return a null string, but the code was skipping to the next DATA statement instead. Version 1.12 01/12/2001 - Split interpreter.c into four files and moved some stuff around. Renamed some files. Version 1.13 28/11/2002 - Included new RND code written by Crispian Daniels. - Included fixes for various graphics problem written by Crispian Daniels. - Incorporated new MacOS target written by Crispian Daniels. - Incorporated new FreeBSD target written by Dirk Engling. - Removed limit on the number of command line arguments returned by ARGV$. - Fixed problem in Unix version of program where screen output was not being displayed before using OSCLI. - Added new VDU function. This returns the value of the specified RISC OS VDU variable. - Added new function COLOUR. This returns a value that represents the closest colour to the colour with the specified colour components in the current screen mode. - Added missing GCOL ,, statement - Added extended form of GCOL statement: GCOL OF ... ON - Added extended form of COLOUR statement: COLOUR OF ... ON - Added extended MODE statement: MODE ,,, - Extended the DIM statement to support the DIM LOCAL version of the statement. - Checked program compiled and worked using the Castle C compiler. Version 1.14 28/07/2003 - Quick hack to fix DIM ... LOCAL problem. The problem was that the code to check that writes using indirection operators were to legal addresses blew away writes to byte arrays on the stack. I have removed this check so that the stack can written to. Version 1.15 16/08/2003 - Added '*key' to allow strings to be defined for function keys as on a BB Micro or RISC OS machine. Version 1.16 25/01/2004 - Added OpenBSD target (changes supplied by Julian Smith). - Added Amiga target (changes supplied by Stefan Haubenthal). - Fixed bug in tokenisation code where any keyword immediately followed by a '.' was taken as abbreviated and the '.' always skipped. This should not have been happening when the full keyword was supplied. The '.' in this case is the start of a floating point number. - Changed the handling of the function 'TOP'. Brandy used to have a token for this function but it is not how the RISC OS Basic interpreter treats it. It uses the token for 'TO' followed by the letter 'P'. This means that the interpreter determines from the context whether this is the function 'TOP' or the keyword 'TO' followed by a variable name starting with 'P'. Brandy now works the way the RISC OS interpreter does. - Fixed bug in handling of 'ERR'. The number of the last error was not being saved so that it could be referenced using ERR except in the case of errors raised via 'ERROR'. This bug was introduced in 1.11 when the *opposite* problem was fixed! - Added '*' command emulation to versions running Unix-type operating system. Previously it was DOS only. - Changed handling of OS commands to that leading blanks and '*' are removed first. - Added new command line option '-nostar' to stop program checking all '*' commands to see if they are emulated by the interpreter. - Tidied up the output on 'help' which shows the where the Basic program and variables are. Version 1.17 21/03/04 - Fixed problem when reading a BBC Basic program that has been crunched. Lines are expanded into their text form before being retokenised. Lines can be up to 253 bytes long in their tokenised form but can expand to much more than this. The buffer that held the expanded form allowed for the expansion but could still be too small. Fixed. - Fixed bug where the top level operator stack was being removed from the Basic stack when ON ERROR was used. The program would crash horribly. Version 1.18 21/01/05 - Incorporated changes to the Amiga target from Stefan Haubenthal. - Incorporated changes to the program by 'Sprow'. Version 1.19 09/10/05 - Included fix for bug in print_screen() supplied by Mark de Wilde. Version 1.20 31/12/06 - Replaced obsolete graphics library jlib with SDL - Resolve minor warnings and missing echo_off echo_on in RISC OS version. (patch from Sprow) (Subsequent changes as part of the Matrix Brandy fork are documented in ChangeLog.)