This is the original 1978 source code of Microsoft BASIC for 6502 with all original comments, documentation and easter eggs:
M6502.MAC (1978-07-27, 6955 lines, 161,685 bytes)
This is currently the oldest publicly available piece of source written by Bill Gates.
Like the 8080 version, the 6502 version was developed on a PDP-10, using the MACRO-10 assembler. A set of macros developed by Paul Allen allowed MACRO-10 to understand and translate 6502 assembly, albeit in a modified format to fit the syntax of macros, for example:
MACRO-10 did not support hex numbers, which is why most numbers are in decimal format. In the floating point code, all numbers are octal. The RADIX statement switches between the two. Octal can also be forced with a ^O prefix.
Conditional translation is done using the IFE and IFN statements, which test whether the argument is zero. The following only adds the string to the binary if REALIO is equal to 4:
IFE REALIO-4,<DT"APPLE BASIC V1.1">
The source defines many macros that make development easier. There are some examples:
|Get the next character and make sure it’s Q, otherwise SYNTAX ERROR. This pattern is used a lot.
|Most 16 bit constants are loaded into A/Y with this macro, but macros for A/X and X/Y also exist.
|This loads an immediate constant into A/Y.
|This pushes a 16 bit value from memory (absolute or zero page) onto the stack.
|A compact way to express out-of-bounds branches. Macros exist for all branches.
||This emits a byte value of 0x2C (BIT absolute), which skips the next instruction. (The ^O1000 part wraps the byte in a PDP-10 instruction – see below.)
The BASIC source supports several compile-time configuration options:
||FOR ADDITIONAL PRECISION
||40 bit (9 digit) vs 32 bit (7 digit) float
||LONG ERROR MESSAGES
||Error message strings instead of two-character codes
||CAPABILITY TO SET AND READ A CLK
||TI and TI$ support
||PRINT#, INPUT#, CMD, SYS (!), OPEN and CLOSE support
||SAVE AND LOAD COMMANDS
||LOAD, SAVE (and on Commodore: VERIFY) support
||FOR THE "NULL" COMMAND
||NULL support, a command to configure the number of NUL characters to print to the terminal after each line break
||If 1, the ROR instruction is not used
||TELLS IF THIS IS ON ROM
||The RAM version can optionally jetison the SIN, COS, TAN and ATN commands at startup
||Column width for TAB()“
||LONG INITIALIZATION SWITCH
||The top of stack at startup
||Page of the input buffer; if 0, the buffer uses parts of the zero page
||INPUT BUFFER SIZE
||TERMINAL LINE LENGTH
||ADDRESS OF START OF PURE SEGMENT
||KIM-specific smaller config
The constant REALIO is used to configure what computer system to generate the binary for. It has one of the following values:
||PDP-10 SIMULATING 6502
||SIMULATED BASIC FOR THE 6502 V1.1
||Paul Allen’s Simulator on PDP-10
||KIM BASIC V1.1
||OSI 6502 BASIC VERSION 1.1
||OSI Model 500
||### COMMODORE BASIC ###
||Commodore PET 2001
||APPLE BASIC V1.1
||STM BASIC V1.1
All versions except Commodore also print “COPYRIGHT 1978 MICROSOFT” in a new line.
The target defines the setting of the configuration constants, but some code is also conditionally compiled depending on a specific target.
What is interesting is that initially it was Microsoft adapting their source for the different computers, instead of giving source to the different vendors and having them adapt it. Features like file I/O and time support seem to have been specifically developed for Commodore, for example. Later, the computer companies would get the source from Microsoft and develop themselves – source code of the Apple and Commodore derivatives is available; they both contain Microsoft comments.
By the way, the numbering of these targets probably indicated in which order Microsoft signed contracts with computer manufacturers. MOS was first (for the KIM), then OSI, then Commodore/MOS again (this time for the PET), then Apple.
The PDP-10 Target
Paul Allen’s additional macros for 6502 development made the MACRO-10 assembler output one 36 bit PDP-10 instruction word for every 6502 byte. When targeting a real 6502 machine, the 6502 binary could be created by simply extracting one byte from every PDP-10 word.
In the case of targeting the simulator, the code created by the assembler could just be run without modification, since every emitted PDP-10 instruction was constructed so that it would trap – the linked-in simulator would then extract the 6502 opcode from the instruction and emulate the 6502 behavior.
While this trick was mostly abstracted by the (unreleased) macro package, its workings can be seen in a few cases in the BASIC source. Here, it defines SKIP1 and SKIP2. Instead of just emitting 0x24 or 0x2C, respectively, it combines it with the octal value of 01000 to make it a PDP-10 instruction that traps:
DEFINE SKIP1, <XWD ^O1000,^O044> ;BIT ZERO PAGE TRICK.
DEFINE SKIP2, <XWD ^O1000,^O054> ;BIT ABS TRICK.
In the initialization code, it writes a JMP instruction into RAM. On the simulator, it has to patch up the opcode of JMP (0x4C, decimal 76) to be the correct PDP-10 instruction:
LDAI 76 ;JMP INSTRUCTION.
IFE REALIO,<HRLI 1,^O1000> ;MAKE AN INST.
With this information, we can reconstruct what the set of 6502 macros, which is not part of this source, probably looked like. Here is LDAI (LDA immediate):
DEFINE LDAI (Q),<
XWD ^O1000,^O251 ;EMIT OPCODE
XWD ^O1000,<Q> ;EMIT OPERAND
You can also see native TJSR PDP-10 assembly instructions for character I/O:
TJSR INSIM##> ;GET A CHARACTER FROM SIMULATOR
TJSR OUTSIM##> ;CALL SIMULATOR OUTPUT ROUTINE
The DDT command, which breaks into the PDP-10’s DDT debugger, only exists in this config:
DDT: PLA ;GET RID OF NEWSTT RETURN.
The KIM and OSI Targets
The KIM target is meant for the MOS KIM-1 and Ohio Scientific OSI Model 500 single-board computers. These are the first ports to specific computers, and also the cleanest, i.e. except for the character I/O interface and the very simple LOAD/SAVE implementation for the KIM, there is nothing specific about these targets.
The Commodore Target
The Commodore target is meant for the Commodore PET 2001. It includes LOAD/SAVE/VERIFY (the commands jump directly to outside “KERNAL” ROM code), the I/O commands (SYS, PRINT#, OPEN etc.), the GET command and the π, ST, TI and TI$ symbols. CLEAR is renamed to CLR, "OK" is renamed to "READY.", the BEL character is not printed, and character I/O code behaves differently to account for the more featureful screen editor of the PET.
Oh, and the Commodore version of course includes the Bill Gates WAIT 6502,1 easter egg! This is the WAIT instruction:
; THE WAIT LOCATION,MASK1,MASK2 STATEMENT WAITS UNTIL THE CONTENTS
; OF LOCATION IS NONZERO WHEN XORED WITH MASK2
; AND THEN ANDED WITH MASK1. IF MASK2 IS NOT PRESENT, IT
; IS ASSUMED TO BE ZERO.
FNWAIT: JSR GETNUM
JSR COMBYT ;GET MASK2.
STORDO: STX EORMSK
WAITER: LDADY POKER
ZERRTS: RTS ;GOT A NONZERO.
Note how the BEQ instruction references ZSTORDO, not STORDO – execution sneaks out of this function here.
Well, on non-Commodore machines, ZSTORDO is assigned to be the same as STORDO, so everything is fine:
But on Commodore, we have this code hidden near the top of the floating point math package – close enough so the BEQ can reach it, but inside code that is least likely to get touched:
ZSTORD:! LDA POKER
MRCHKR: LDXI 12
MRCHR: LDA 60000,X,>
MRCHR: LDA SINCON+36,X,>
(IF1 and IF2 are true on the first and the second assembler pass, respectively, so the conditional there is to hint to the assembler in the first pass that SINCON+36 is not a zero page address. Also note that all numbers here are octal, since this code is in the floating point package.)
First of all, the final line here removes ZSTORD from the list of symbols after the second pass, so that Commodore would not notice it in a printout of all symbols – very smart!
As has been discussed before, this code writes the string “MICROSOFT!” into the PET’s screen RAM if the argument to WAIT is “6502”. The encoded string is hidden as two extra 40 bit floating point numbers appended to the coefficients used by the SIN function:
SINCON: 5 ;DEGREE-1.
204 ; -14.381383816
206 ; 42.07777095
207 ; -76.704133676
207 ; 81.605223690
206 ; -41.34170209
203 ; 6.2831853070
241 ; 7.2362932E7
217 ; 73276.2515
These last ten bytes, nicely disguised as octal values of floating point constants, spell out “MICROSOFT!” backwards after clearing the upper two bits. What’s interesting is that the floating point values next to them are actually incorrect: They should be 7.12278788E9 and 26913.7691 instead.
Also note that these constants are not conditionally assembled! All versions built since the Commodore easter egg was introduced also contained these 10 bytes – including BASIC for the Motorola 6800!
The Apple Target
The Apple target is meant for the Apple II, and contains no customizations other than some changes around I/O handling (which calls into the monitor ROM). Note that this is not yet the “AppleSoft” version of BASIC, which was a more customized version modified by Apple later.
The STM Target
“STM” most likely stands for “Semi-Tech Microelectronics” – a company that never shipped a 6502-based computer. Their first machine was the “Pied Piper”, a Z80-based system, and they later made a PC clone. It seems they had a 6502-based computer in development that never shipped – or at least they were considering making one, and Microsoft added the target; this target doesn’t actually change any of the defaults.
Organization of the Source
The source uses the PAGE and SUBTTL keywords for organization. Here are the headings:
SUBTTL INTRODUCTION AND COMPILATION PARAMETERS.
SUBTTL SOME EXPLANATION.
SUBTTL PAGE ZERO.
SUBTTL RAM CODE.
SUBTTL DISPATCH TABLES, RESERVED WORDS, AND ERROR TEXTS.
SUBTTL GENERAL STORAGE MANAGEMENT ROUTINES.
SUBTTL ERROR HANDLER, READY, TERMINAL INPUT, COMPACTIFY, NEW, REINIT.
SUBTTL THE "LIST" COMMAND.
SUBTTL THE "FOR" STATEMENT.
SUBTTL NEW STATEMENT FETCHER.
SUBTTL LOAD AND SAVE SUBROUTINES.
SUBTTL "IF ... THEN" CODE.
SUBTTL "ON ... GO TO ..." CODE.
SUBTTL LINGET -- READ A LINE NUMBER INTO LINNUM
SUBTTL "LET" CODE.
SUBTTL PRINT CODE.
SUBTTL INPUT AND READ CODE.
SUBTTL THE NEXT CODE IS THE "NEXT CODE"
SUBTTL DIMENSION AND VARIABLE SEARCHING.
SUBTTL MULTIPLE DIMENSION CODE.
SUBTTL INTEGER ARITHMETIC ROUTINES.
SUBTTL FRE FUNCTION AND INTEGER TO FLOATING ROUTINES.
SUBTTL SIMPLE-USER-DEFINED-FUNCTION CODE.
SUBTTL STRING FUNCTIONS.
SUBTTL PEEK, POKE, AND FNWAIT.
SUBTTL FLOATING POINT ADDITION AND SUBTRACTION.
SUBTTL NATURAL LOG FUNCTION.
SUBTTL FLOATING MULTIPLICATION AND DIVISION.
SUBTTL FLOATING POINT MOVEMENT ROUTINES.
SUBTTL SIGN, SGN, FLOAT, NEG, ABS.
SUBTTL COMPARE TWO NUMBERS.
SUBTTL GREATEST INTEGER FUNCTION.
SUBTTL FLOATING POINT INPUT ROUTINE.
SUBTTL FLOATING POINT OUTPUT ROUTINE.
SUBTTL EXPONENTIATION AND SQUARE ROOT FUNCTION.
SUBTTL EXPONENTIATION FUNCTION.
SUBTTL POLYNOMIAL EVALUATOR AND THE RANDOM NUMBER GENERATOR.
SUBTTL SINE, COSINE AND TANGENT FUNCTIONS.
SUBTTL ARCTANGENT FUNCTION.
SUBTTL SYSTEM INITIALIZATION CODE.
Paul Allen vs. Bill Gates
The source of the 8080 version states:
PAUL ALLEN WROTE THE NON-RUNTIME STUFF.
BILL GATES WROTE THE RUNTIME STUFF.
MONTE DAVIDOFF WROTE THE MATH PACKAGE.
People have since wondered what runtime vs. non-runtime meant, especially since Paul Allen’s recent debate on whether the company’s ownership was faily split.
The BASIC for 6502 source sheds some light on this:
THE CODE TO INPUT A LINE, CRUNCH IT, GIVE ERRORS,
FIND A SPECIFIC LINE IN THE PROGRAM,
PERFORM A "NEW", "CLEAR", AND "LIST" ARE
ALL IN THIS AREA. [...]
So by “runtime” they just literally mean “at run time”: all code that is active when the program runs, as opposed to non-runtime, which is all code that assists editing the program.
By this understanding, we can assume this:
- Paul Allen wrote the macro package for the MACRO-10 assembler, the 6502 simulator, the tokenizer, the detokenizer, as well as finding, inserting and deleting BASIC lines.
- Bill Gates implemented all BASIC statements, functions, operators, expression evaluation, stack management for FOR and GOSUB, the memory manager, as well as the array and string library.
- Monte Davidoff wrote the floating point math package.
Version and Date
The last entry in the change log has a date of 1978-07-27. Both the comment in the first line of the file and the message printed at startup call it version 1.1.
What does this say about the version of the source? Is it the last version? Let’s look at the last bug fix and compare which BASIC binaries contain this fix, and let’s see whether there are fixes in BASIC binaries that are not in the source.
I have previously compared binaries of derivatives of BASIC for 6502 and compiled the information at github.com/mist64/msbasic. The last entry in the log of this source is about a bug that failed to correctly invalidate a pointer in the RETURN statement. According to my analysis of BASIC 6502 versions, this is fixed in the BASIC binaries for AIM-65, SYM-1, Commodore v2, KBD BASIC and MicroTAN, i.e. on everything my previous analysis calls CONFIG_2A and higher.
The same analysis also came to the conclusion that there were two successors, CONFIG_2B and CONFIG_2C. At least the two CONFIG_2B fixes exist in two BASIC binaries: KBD BASIC and MicroTAN, but they don’t exist in this source. It’s very unlikely that both these bugs (and only these!) got fixed by the two computer manufacturers independently, so it’s safe to assume that this source is not the final version – but pretty close to it!
- This code is comparing a keyboard input character to the BEL code. Bob Albrecht is a computer educator that “was instrumental in helping bring about a public-domain version of Basic (called Tiny Basic) for early microcomputers.”.
CMPI 7 ;IS IT BOB ALBRECHT RINGING THE BELL
;FOR SCHOOL KIDS?
- External documentation usually calls the conversion of ASCII BASIC text into the compressed format “tokenizing”. The source calls this “crunching”.
- Microsoft is still spelled “Micro-Soft”.
- Apparently the multiplication function could use some performance improvements:
BNE MLTPL2 ;SLOW AS A TURTLE !
- The NEW command is actually called SCRATCH in labels and comments – maybe other BASIC dialects called it that, and they decided to rename it to NEW later?
- The math package documentation says:
THE MATH PACKAGE CONTAINS FLOATING INPUT (FIN),
FLOATING OUTPUT (FOUT), FLOATING COMPARE (FCOMP)
... AND ALL THE NUMERIC OPERATORS AND FUNCTIONS.
THE FORMATS, CONVENTIONS AND ENTRY POINTS ARE ALL
DESCRIBED IN THE MATH PACKAGE ITSELF.
Commodore’s derived source changes this to:
; MATH PACKAGE
; THE MATH PACKAGE CONTAINS FLOATING INPUT FIN, OUTPUT
; FOUT, COMPARE FCOMP...AND ALL THE NUMERIC OPERATORS
; AND FUNCTIONS. THE FORMATS, CONVENTIONS AND ENTRY
; POINTS ARE ALL DESCRIBED IN THE MATH PACKAGE ITSELF.
- CHRGET is a central piece of BASIC for 6502. Here it is in its entirety:
; THIS CODE GETS CHANGED THROUGHOUT EXECUTION.
; IT IS MADE TO BE FAST THIS WAY.
; ALSO, [X] AND [Y] ARE NOT DISTURBED
; "CHRGET" USING [TXTPTR] AS THE CURRENT TEXT PNTR
; FETCHES A NEW CHARACTER INTO ACCA AFTER INCREMENTING [TXTPTR]
; AND SETS CONDITION CODES ACCORDING TO WHAT'S IN ACCA.
; NOT C= NUMERIC ("0" THRU "9")
; Z= ":" OR END-OF-LINE (A NULL)
; [ACCA] = NEW CHAR.
; THE FOLLOWING EXISTS IN ROM IF ROM EXISTS AND IS LOADED
; DOWN HERE BY INIT. OTHERWISE IT IS JUST LOADED INTO THIS
; RAM LIKE ALL THE REST OF RAM IS LOADED.
CHRGET: INC CHRGET+7 ;INCREMENT THE WHOLE TXTPTR.
CHRGOT: LDA 60000 ;A LOAD WITH AN EXT ADDR.
CMPI " " ;SKIP SPACES.
QNUM: CMPI ":" ;IS IT A ":"?
BCS CHRRTS ;IT IS .GE. ":"
SBCI "0" ;ALL CHARS .GT. "9" HAVE RET'D SO
SBCI 256-"0" ;SEE IF NUMERIC.
;TURN CARRY ON IF NUMERIC.
;ALSO, SETZ IF NULL.
CHRRTS: RTS ;RETURN TO CALLER.
Did you ever wonder why all versions have $EA60 encoded into the LDA instruction that later gets overwritten? Because it’s 60000 decimal. That’s why! The source actually uses 60000 as a placeholder for 16 bit values in several places.
- The handling of π, ST, TI and TI$ (all Commodore-specific) looks wonky: Instead of making them tokens, they are special cased in several places. I always assumed it was Commodore adding this without understanding (or wanting to disrupt) the existing code, but it was Microsoft adding these features. Maybe they were added by someone other than the original developers?
Origin of the File
The source was posted on the Korean-language blog 6502.tistory.com without further comment, in a marked-up format:
FILE: "david mac g5 b:m6502.asm"
000001 TITLE BASIC M6502 8K VER 1.1 BY MICRO-SOFT
006955 END $Z+START
End of File -- Lines: 6955 Characters: 154740
Total number of files : 1
Total file lines : 6955
Total file characters : 154740
This formatting was created by an unpublished tool by David T. Craig, who published a lot of Apple-related soure code (Apple II, Apple III, Lisa) in this format in as early as 1993, first anonymously, later with his name).
The filename “david mac g5 b:m6502.asm” (disk name “david mac g5 b”, file name “m6502.asm”, since it was a classic Mac OS tool) confirms David Craig’s involvement, and it means the line numbers were added no earlier than 2003.
Given all this, it is safe to assume the file with the Microsoft BASIC for 6502 source originated at Apple, and was given to David Craig together with the other source he published.
The version I posted is a reconstruction of the original file, with the header, the footer and the line numbers removed, and the spaces converted back into tabs. I chose the name “M6502.MAC” to be consistent with the MACRO-10 file extension used by the Microsoft BASIC for 8080 sources.