An external "tokenizer" for the TRS-80 Model 100 BASIC language. Converts BASIC programs in ASCII text (.DO) to executable BASIC (.BA) on a host machine. Useful for large programs which the Model 100 cannot tokenize due to memory limitations.

$ cat FOO.DO
10 ?"Hello!"

$ tokenize FOO.DO 
Tokenizing 'FOO.DO' into 'FOO.BA'

$ hd FOO.BA
00000000  0f 80 0a 00 a3 22 48 65  6c 6c 6f 21 22 00        |....."Hello!".|
0000000e

This program creates an executable BASIC file that works on the Model 100, the Tandy 102, Tandy 200, Kyocera Kyotronic-85, and Olivetti M10. Those five machine have identical tokenization. It does not (yet) work for the NEC PC-8201/8201A/8300 whose N82 BASIC has a different tokenization.

Additionally, this project provides a decommenter and cruncher (whitespace remover) to save bytes in the tokenized output. This allows one to have both well-commented, easy to read source code and a small executable size.

The Tandy/Radio-Shack Model 100 portable computer can save its BASIC files in ASCII (plain text) or in a “tokenized” format where the keywords — such as FOR, IF, PRINT, REM — are converted to a single byte. The Model 100 automatically tokenizes an ASCII program when it is LOADed so that it can be RUN or SAVEd. The tokenized format saves space and loads faster, but where it is tokenized can matter.

Programs for the Model 100 are generally distributed in ASCII format which is good for portability and easy transfer. However, ASCII files have downsides:

1. RUNning an ASCII program is quite slow because the Model 100 must tokenize it first.

2. Large programs can run out of memory (?OM Error) when tokenizing on the Model 100 because both the ASCII and the tokenized versions must be in memory simultaneously.

This program tokenizes on a host computer before downloading to the Model 100.

Tokenized BASIC files use the extension .BA. ASCII formatted BASIC files should be given the extension .DO so that the Model 100 will see them as text documents, although people often misuse .BA for ASCII BASIC.

• tokenize: A shell script which ties together all the following tools. Most people will only run this program directly.

• m100-tokenize: Convert M100 BASIC program from ASCII (.DO) to executable .BA file.

• m100-sanity: Clean up the BASIC source code (sort lines and remove duplicates).

• m100-jumps: Analyze the BASIC source code and output a list of lines that begin with a comment and that are referred to by other lines in the program. Input should have been run through m100-sanity.

• m100-decomment: Modify the BASIC source code to remove REM and ' comments except it keeps lines mentioned on the command line (output from m100-jumps). This can save a lot of space if the program was well commented.

• m100-crunch: Modify the BASIC source code to remove all whitespace in an attempt to save even more space at the expense of readability.

Presuming you have flex installed, just run make to compile.

$ git clone https://github.com/hackerb9/tokenize
$ make
$ make install

 flex m100-tokenize.lex  &&  gcc lex.tokenize.c

One can either use the tokenize wrapper or run the executables manually.

The "tokenize" script is easiest. By default, the output will be exactly the same, byte for byte, as a .BA file created on actual hardware.

tokenize INPUT.DO [ OUTPUT.BA ]
tokenize [ -d | --decomment ] INPUT.DO [ OUTPUT.BA ]
tokenize [ -c | --crunch ] INPUT.DO [ OUTPUT.BA ]

The -d option decomments before tokenizing.

The -c option decomments and removes all optional whitespace before tokenizing.

$ tokenize PROG.DO
Output file 'PROG.BA' is newer than 'PROG.DO'. 
Overwrite [yes/No/rename]? R
Old file renamed to 'PROG.BA~'

$ wc -c M100LE.DO 
17630 M100LE.DO

$ tokenize M100LE.DO 
Tokenizing 'M100LE.DO' into 'M100LE.BA'

$ wc -c M100LE.BA
15667 M100LE.BA

$ tokenize -c M100LE.DO M100LE-crunched.BA
Decommenting, crunching, and tokenizing 'M100LE.DO' into 'M100LE-crunched.BA'

$ wc -c M100LE-crunched.BA
6199 M100LE-crunched.BA

In this case, using tokenize -c reduced the BASIC executable from 16 to 6 kibibytes, which is quite significant on a machine that might have only 24K of RAM. However, this is an extreme example from a well commented program. Many Model 100 programs have already been "decommented" and "crunched" by hand to save space.

Tip: When distributing a crunched program, it is a good idea to also include the original source code to make it easy for people to learn from, debug, and improve it.

Certain programs should usually be run to process the input before the final tokenization step, depending upon what is wanted. m100-sanity is strongly recommended. (See Abnormal code below.)

flowchart LR;
m100-sanity ==> m100-tokenize
m100-sanity ==> m100-jumps ==> m100-decomment
m100-decomment --> m100-crunch --> m100-tokenize
m100-decomment --> m100-tokenize

m100-sanity INPUT.DO | m100-tokenize > OUTPUT.BA

10 REM This line would normally be removed
20 GOTO 10   ' ... but now line 10 should be kept.

jumps=$(m100-sanity INPUT.DO | m100-jumps)
m100-sanity INPUT.DO | 
    m100-decomment - - $jumps | 
	m100-tokenize > OUTPUT.BA

10 REM This line would normally be removed
20 GOTO 10

jumps=$(m100-sanity INPUT.DO | m100-jumps)
m100-sanity INPUT.DO | 
    m100-decomment - - $jumps | 
	m100-crunch | 
	m100-tokenize > OUTPUT.BA

10REM
20GOTO10

Across the eight Kyotronic-85 sisters, there are actually only two different tokenized formats: "M100 BASIC" and "N82 BASIC".

The three Radio-Shack portables (Models 100, 102 and 200) all share the same tokenized BASIC. While less commonly seen, the Kyocera Kyotronic-85 and Olivetti M10 also use that tokenization, so one .BA program can work for any of them. However, the NEC family of portables -- the PC-8201, PC-8201A, and PC-8300 -- run N82 BASIC which has a different tokenization format.

The .BA files generated by tokenize aim to be exactly the same, byte for byte, as the output from tokenizing on a Model 100 using LOAD and SAVE. There are some bytes, however, which can and do change but do not matter.

An artifact of the .BA file format is that the saved file contains the pointer locations of where the program happened to be in memory on the computer from which it was saved. The pointers are never used as they are recalculated when the program is loaded into RAM.

Specifically, the output of this program is intended to be identical to:

• A Model 100
• that has been freshly reset
• with no other BASIC programs on it
• running LOAD "COM:88N1" and SAVE "FOO" while a host computer sends the ASCII BASIC program over the serial port.

I (hackerb9) believe the Tandy 102, Kyotronic-85, and M10 also output byte identical files, but the Tandy 200 does not. The 200 has more ROM than the other Model T computers, so it stores the first BASIC program at a slightly different RAM location (0xA000 instead of 0x8000). This has no affect on compatibility between machines, but it does change the line number pointers in the .BA file.

Since two .BA files can be the identical program despite having different checksums, this project includes the bacmp program, described below.

This program is written in Flex, a lexical analyzer, because it made implementation trivial. The tokenizer itself, m100-tokenize, is mostly just a table of keywords and the corresponding byte they should emit. Flex handles special cases, like quoted strings and REMarks, easily.

The downside is that one must have flex installed to modify the tokenizer. Flex is not necessary to compile on a machine as flex generates portable C code. See the tokenize-cfiles.tar.gz in the github release or run make cfiles.

The tokenize script always uses the m100-sanity program to clean up the source code, but one can run m100-tokenize directly to purposefully create abnormal, but valid, .BA files. These programs cannot be created on genuine hardware, but will run.

Here is an extreme example.

Source code for "GOTO 10" by hackerb9

1 I=-1
10 I=I+1: IF I MOD 1000 THEN 10 ELSE I=0
10 PRINT: PRINT
10 PRINT "This is line ten."
10 PRINT "This is also line ten."
5  PRINT "Line five runs after line ten."
10 PRINT "Where would GOTO 10 go?"
15 PRINT "  (The following line is 7 GOTO 10)"
7 GOTO 10
8 ERROR "Line 8 is skipped by GOTO 10."
10 PRINT: PRINT "It goes to the *next* line ten!"
10 FOR T=0 TO 1000: NEXT T
10 PRINT "Exceptions: Goes to *first* line ten"
10 PRINT "  if the current line is ten, or"
10 PRINT "  if a line number > 10 is seen."
10
10 'Shouldn't 10 GOTO 10 go to itself?
10 'It shouldn't have to search at all.
10 'Maybe it's a bug in Tandy BASIC?
10 'Perhaps it checks for (line >= 10)
10 'before it checks for (line == 10)?
10
10 'BTW, >= is one less op than >.
10 'On 8085, >= is CMP then check CY==0.
10 'But > also requires checking Z==0.
10
10 PRINT
10 PRINT "The next line is 9 GOTO 10. This time"
10 PRINT "it goes to the FIRST line ten, because"
10 PRINT "line 20 comes before the next line ten."
9 GOTO 10
20 ERROR "Line 20 is never reached, but it has an effect because 20>10."
10 ERROR "This is the final line ten. The previous GOTO 10 won't find it because line 20 comes first."
10
15
20 
0 PRINT "This program examines how"
1 PRINT "Model T computers run"
2 PRINT "degenerate tokenized BASIC."
3 PRINT "Trying to load it as a .DO"
4 PRINT "file will not work as"
5 PRINT "Tandy BASIC corrects issues"
6 PRINT "such as duplicate line numbers"
7 PRINT "and out of order lines."
8 PRINT "Please use hackerb9's"
9 PRINT "pre-tokenized GOTO10.BA."

To run this on a Model 100, download GOTO10.BA which was created using m100-tokenizer.

• Tokenized BASIC files are binary files, not ASCII and as such cannot be transferred easily using the builtin TELCOM program or the LOAD "COM:" or SAVE "COM:" commands. Instead, one must use a program such as TEENY which can transfer 8-bit data.

• Line endings in the input file can either be CRLF (standard for a Model 100 text document) or simply LF (UNIX style).

• Even when the tokenized format is the same, there exist a multitude of filename extensions to signify to users that a program is intended to run on a certain platform and warn that it may use system specific features.

  Conventions for filename extensions vary. Here are just some of them:

  • .DO This is the extension the Model 100 uses for plain text BASIC files, but in general can mean any ASCII text document with CRLF line endings.

  • .BA All tokenized BASIC programs are named .BA on the Model 100, but note that most .BA files found on the Internet are in ASCII format. Before the existence of tokenizers like this, one was expected to know that ASCII files had to be renamed to .DO when downloading to a Model 100.

  • Although the BASIC language and tokenization is the same, some programs use POKEs or CALLs which work only one one model of portable computer and will cause others to crash badly, possibly losing files. To avoid this, some filename extensions are used:

  • .100 An ASCII BASIC file that includes POKEs or CALLs specific to the Model 100/102.

  • .200 An ASCII BASIC file specific to the Tandy 200.

  • .BA1 A tokenized BASIC file specific to the Model 100/102.

  • .BA2 A tokenized BASIC file specific to the Tandy 200.

  • The .BA0 and .NEC.BA extension signify a tokenized BASIC file specific to the NEC portables. This is a different tokenization format than any of the above and is not yet supported.

• To save in ASCII format on the Model 100, append , A:

  save "FOO", A

Run make check to try out the tokenizer on some sample Model 100 programs and some strange ones designed specifically to exercise peculiar syntax. The program bacmp is used to compare the generated .BA file with one created on hackerb9's Tandy 200.

Note that without m100-sanity, the SCRAMB.DO test, whose input is scrambled and redundant, would fail.

20 GOTO 10
10 GOTO 10
10 PRINT "!dlroW ,olleH"

The included bacmp.c program may be useful for others who wish to discover if two tokenized BASIC files are identical. Because of the way the Model 100 works, a normal cmp test will fail. There are pointers from each BASIC line to the next which change based upon where the program happens to be in memory. The bacmp program handles that by allowing line pointers to differ as long as they are offset by a constant amount.

Note that generating correct line pointers is actually unnecessary in a tokenizer; they could be any arbitrary values. The Model 100 always regenerates the pointers when it loads a .BA file. However, some emulators insist on having them precisely correct, so this tokenizer has followed suit.

• Hackerb9 has documented the file format of tokenized BASIC at http:https://fileformats.archiveteam.org/wiki/Tandy_200_BASIC_tokenized_file

• None known. Reports are gratefully accepted.

Here are some other ways that you can tokenize Model 100 BASIC when the source code and the tokenized version won't both fit in the M100's RAM.

• Perhaps the simplest is to load the ASCII BASIC file to the M100 over a serial port from a PC host. On the Model 100, type:

      LOAD "COM:88N1"
  

  or for a Tandy 200, type

      LOAD "COM:88N1ENN"
  

  On the host side, send the file followed by the ^Z ('\x1A') character to signal End-of-File. See hackerb9's sendtomodelt for a script to make this easier.

• Robert Pigford wrote a Model 100 tokenizer that runs in Microsoft Windows. Includes PowerBASIC source code. http:https://www.club100.org/memfiles/index.php?&direction=0&order=nom&directory=Robert%20Pigford/TOKENIZE

• Mike Stein's entoke.exe is a tokenizer for Microsoft DOS. http:https://www.club100.org/memfiles/index.php?&direction=0&order=nom&directory=Mike%20Stein

• The Model 100's ROM disassembly shows that the tokenization code is quite short. http:https://www.club100.org/memfiles/index.php?action=downloadfile&filename=m100_dis.txt&directory=Ken%20Pettit/M100%20ROM%20Disassembly

• The VirtualT emulator can be scripted to tokenize a program through telnet. https://sourceforge.net/projects/virtualt/