%if 0 ; docs{ A BOOTABLE FORTH STYLE "OS" This file represents a bootable readonly forth-like system. It uses a byte-code approach to creating a simple forth dictionary. Its overall aim is to create a portable and minimal booting interactive system. All commands entered interactively (in the SHELL loop) are 'compiled' into bytecode either into an anonymous code buffer or to the dictionary and then executed. Also, the general aim of the code is to get to a source interpreter and compiler in the minimum code size possible. This is by no means a standard forth! It is an experiment. MOTIVATION "our civilization will collapse under the weight of its own complexity" charles moore. This code is attempt to answer a question: "what is the smallest bootable, interactive, portable system that can add to itself from source code, in a structured, orderly and understandable way?" To elaborate... "Smallest" The aim of any developer should be to reduce the size and complexity of the core system (not increase it) while satisfying the conditions below. Currently (2018) about 7K core system. "Portable" from microcontroller to mainframe using a rigorous practical virtual machine and bytecodes. "interactive": actually this condition is not so important because a system that can add to itself from source code can easily compile and execute an interactive REPL interpreter or shell. But it is important that the user can choose what bits to add to the system, and what not to add. "Add to itself" via disk, serial connection, or tcp, using globally unique and locatable object names (words). So each word has a domain name which, when combined with the word itself, forms a unique name. "from source code" This condition implies that a compiler must be present "Bootable" doesnt require any operating system on any architecture. "Structured" must add to itself in manageable, contained units of code that can be debugged individually. "Understandable" must strive to be simple enough for an average coder to understand after reading for a few days. Actually forth contains partial but powerful solutions to some of these criteria. The question above is important because it has implications for our consumption of resources, impact on the environment, beholden-ness to obsolescence and relationship with technology. In particular power consumption is related to software complexity. Simple software can run on small low-power consumption computers (such as a coin-cell powered bicycle computer). Another important aspect is digital security and the "knowability" of code. Opaque massive compiled code is inherently insecure because it is unknown and unknowable. At some point governments will need to acknowledge this, because the trojan horses they implant in various systems will be compromised by the trojan horses that others place in their systems. It is a futile arms race of insecurity. This question also has implications for how human beings interact with technology and whether they are in control of those interactions or are victims of them. The divide between "programmer" and "user" is artificial and serves commercial purposes not human ones. Just as the divide between operating system and software is also artificial. To expand a little, the code uses bytecode and a simple 2 stack virtual machine to hopefully allow portability between microcontrollers and modern laptops/desktops. By attempting to port to very disparate chip architectures (avr micros, x86, z80, arm), the designer is encouraged to think about what is essential for usability and what is just fluff. This may seem to contradict one of charles moore's principles which is "dont code for what you may need in the future, dont try to generalise your code". But it is more an attempt to harness the power of the commonality of the internet and the power of virtual machines to cheat obsolescence. Tinyness and minimalism will hopefully not be sacrificed. The system tries to make the core as tiny as possible while not sacrificing "understandability". So the aim is to get to a source code compiler in the minimum amount of code. Forth ideas facilitate this. Another important idea is that of universal naming. All objects (in this case forth words) should have a resolvable, locatable, unambiguous universal name eg 4th.core.dup Since almost all code is source, then each object (forth word) consists of a minimal syntax (space delimited words) and a series of named-objects (other forth words). This has the simple but powerful consequence that, given the name of a word, all "dependencies" can can be located and obtained, and the given word can be compiled and run. This also applies to data structures. On top of this system we could code a parsing machine see bumble.sf.net/books/gh/gh.c (nearly but incomplete) which would allow the forth machine to recognise and assemble more expressive syntaxes for code and data structures. JOURNEY copying machine code or peeks and pokes into a vic20 (?) qbasic as a teenager. Nice help screens on an IBM XT? First had to learn x86 assembly which I had been meaning to do since the age of 12. Learn about x86 bios calls. Then had to think about forth for a number of years. First heard of forth from D. McBain. Then had to think about parsing and compiling etc. Coding php scripts, in Wagga Wagga NSW. html-form code editor in php, writing text converters in #!/usr/bin/sed and thinking about the sed "virtual machine" in Almetlla de Mar, Catalunya. learning [TCL] and its simple bracket syntax (almost as simple as forth). Learning a little Java in Bogota, but never achieving anything useful. Linux shell scripting and bash. c coding. Some avr assembly. Markdown and code that produces code. FEATURES Case sensitive, and all core words are lower case, but in documentation I may make them upper case so that it is obvious that I am referring to the forth WORD and not the English word. Unlike some forths, this version compiles commands entered interactively to an anonymous buffer. The advantage of this is that flow control words like if, fi, else, begin, until... can be used interactively, not just in colon : definitions. Bytecode. Not a standard forth: eg, THEN is FI "execute" is "pcall" to match with "fcall" opcode Even "proceedures" (everything after the no-op procedure) are written in pseudo forth. That means that they are just a series of calls - to opcodes and other proceedures. The idea is to make porting to another architecture simpler. DIFFERENCES FROM (94?) STANDARD FORTH I have attempted to keep the idea of a "compiling" forth. This means that there is really no such thing as "interpreting" in this forth. The only interpreting opcode is PCALL and I may remove it. My motivation for this is to stick as closely to the operation of real machines. Chips general can only execute machine code that exists at some location in memory. I want to adhere to this principle to make the idea of a "virtual machine" more coherent. * All defining words will probably have to be immediate. * All standard words are lower case. postpone is post RECURSE is just the name of the word. For example : gcd ( a b -- gcd) ?dup if tuck mod gcd fi ; just works and finds the greatest common divisor. IF/THEN is IF/FI because I find THEN just too confusing. EXECUTE is PCALL WORD is WPARSE because the word word is used for way too many things in Forth. eg function in dictionary, 2 bytes of data, space delimited text ...Also wparse only works for whitespace (tab, newline, space ) not just space character. VARIABLE is VAR because I dont need to type any more than I already do. IMMEDIATE is IMM COMPILE, is ITEM, and works slightly differently from COMPILE, (it takes a flag that indicates if the argument is an opcode, literal or procedure) But that may change. I is ii J is jj CHAR is an immediate word. to use CHAR in a colon def do "char * literal" or just use [char] accept does not take a char limit yet. COMPARE is ( a A n -- flag ) rather than ( a n A N -- a' ) so this compare assumes that the 2 strings are of equal length. Need to do things like "[ create this ] " to test create interactively (but not in a : def) because this is a "compiling" forth. STATUS The system is edging towards a usable forth-style system. It can read and write to usb from an x86 system in real mode, but there is no easy way to edit buffers yet (apart from 'accept') so is probably not usable as a standalone system yet. PROBLEMS defining words do not update the dictionary code pointer but 'does>' does, so we can use does at the end of a defining word to make sure the code pointer is updated PERFORMANCE A virtual machine is normally considered to run slower than native machine code, but some optimisation techiques are possible. We can replace x86 'calls' to opcodes with jumps and we can keep the top of the stack in ax. Also it should not be too difficult to compile bytecode to native machine code, if necessary. * 16 june 2018 performance testing: With exec.x using calls to opcodes and pop/push return pointer in the opcodes, performance is as follows using the word 'timeloop' to measure looping speed With qemu on asus e402m machine 500 000 loops took 495 milliseconds 1000 000 loops took 935 milliseconds Booting from usb 1 million loops took 110 ms 2 million loops took 220 ms 10 million loops took 1100 ms So as expected, much faster on the actually machine rather than in qemu. But is this fast enough to play chess? * 17 june 2018 On Ibm thinkpad r52, using exec/call version, real mode using word 'timeloop'. booting from usb. Strangely this seems twice as fast as when I converted to using jmps and having top of stack held in dx. 1 million loops took 55 milliseconds 10 million loops took 275 ms 40 million loops took 1155 ms So surprisingly the old r52 thinkpad was 4 times faster than the more modern asus machine in real mode. * 18 june 2018 system converted from call/ret (in exec.x) to jmp ... With qemu on asus e402m machine. Using word 'timeloop' 500 000 loops took 275 milliseconds 1000 000 loops took 605 milliseconds So by converting from call/ret to jmp there is about a 30% improvement in qemu. On asus eeepc (seashell), using jmp in exec/opcode, real mode booting from usb. 1 million loops took 55 milliseconds 10 million loops took 605 ms 16 million loops took 1045 ms asus e402m machine Booting from usb, jmp (not call/ret) 10 million loops took 990 ms So not much improvement on modern machine. Perhaps modern chips do realmode very badly. With qemu on asus e402m machine, jmps and top of stack in DX real mode. 1 million loops took 385 milliseconds 2 million loops took 825 milliseconds So there seems to be a significant improvement with top of stack in DX. VIRTUAL MACHINE AND OPCODES This forth system tries to emphasize portability, knowability and interoperability. The idea is that each opcode should be standard and have a defined semantic operation on the machine. Also, periferals are considered part of the machine. But periferals (sensors, actuator, transducers etc) are myriad and pluggable and unpluggable. So how do we handle this situation: The answer is that the set of available opcodes defines the machine- actually forms the machines "signature". Since possible periferals and devices attached to the core 2-stack machine are infinite, we need at least 2 bytes to describe them. Or a utf8 type of variable length encoding. So: "1234567" may be the opcode to read a 3 axis accelerometer. The number 1234567 must be unique, universal and defined in some public document. But in the actual machine this opcode will be probably mapped to a much smaller number (say 66) so that it can be encoded in one byte or less (depending on how many periferals/sensors the actual machine has). So the signature of the stack machine includes all standard opcodes expressed as a list of ranges 1-4, 7-9, 11, 13 etc as well as any mappings eg 1-4,7,9,66:1234567 Also there are further subtleties here, which need attention. Eg: Absent opcodes can be "emulated" or in some cases just keyed to NOP. For example an opcode which moves a robotic arm could be emulated as a video monitor display of the robotic arm moving. An opcode which sets the text colour for a monitor may not be essential for the operation of the software and may just be keyed to NOP. It would be nice to be able to create missing opcodes using "high level" forth (source code) as well as assembler appropriate to the current chip architecture. Some opcodes, even if absent, can always be constructed from core stack machine operations. For example double precision arithmetic (eg D+ D- ) can be constructed from 16 bit forth opcodes (+ - dup swap etc). But if speed and efficiency is important for the given application, then these operations should be coded as opcodes written in assembler. The general aim is for the structure of the target machine to be knowable and analysable from within code. STRUCTURE OF THE DICTIONARY One of the key ideas of a forth-like system is the use of a linked list with each item in the list being a data structure with the name and code for a particular function. This linked list is called the dictionary. The current code uses the following structure for the dictionary: [link to previous word] 2 bytes [name of word] [count of name] 1 byte || IMMEDIATE FLAG [code] [data or "parameters"] ... next word structure This uses "reverse" name counts. So the byte containing the length of the name is after the name, not before as in the majority of forths. This allows us to decompile byte code by getting a list of pointers to code and then looking up the name. But it may have other unknown disadvantages. eg db 'minus', 5 dw exec ; link to previous Another refinement is to hold the top element of the stack in ax, which simplifies a lot of stack manipulation. eg 1+ becomes "inc ax" etc. But I found this also introduced complications. OBJECT ORIENTATION The code below may contain a hint as to how to emulated object orientation with forth. Each of the function pointers could be like a method on an object. create buttons ' ring , ' open , ' laugh , ' cry , : button ( nth --) 0 max 3 min cells buttons + @ execute ; CULTURE Forth has its own culture. It uses a set of words and concepts which are completely different from the main stream coding world. It has a set of naming conventions which are not related to normal coding naming conventions. These days forth is a forgotten backwater with little or no mainstream relevance apart from embedded systems, but its ideas remain powerful. TOS top of stack, displayed as rightmost element. NOS next element on stack. . display something (print to screen and remove from stack) , compile something (store executable code or data in memory) : define a new word or data buffer @ fetch something from memory (and push to data stack) ! store something in memory. ' get the execution address of something. (something) is the runtime behaviour of "something" [something] is an immediate version of the word "something" That means that it executes "immediately" or at compile-time. XT means "execution token" which is just the address of code that can be run by the virtual machine. (either byte code or machine-code). INSPIRATION The code was original inspired by the helpful and simple instructions from MikeOS on how to get a minimal booting x86 "operating system" working. The code was also inspired by the incredible simplicity of forth-like systems, and also, the ease of implementing a bytecode system using indirect jumps. In general, simplicity appears to be much more fertile than complexity. Linus Torvalds was inspired by the simplicity of Tannenbaum's minix system to start the Linux journey. Tannenbaum may have been inspired by the simplicity of early unix systems. Donald Knuths "literate programming" where the source code includes the documentation and also self-documents. The idea is that the source code is a document which describes itself, both at compile-time and at run-time IMMEDIATE WORDS One of the main semantic problems of forth systems is the idea of immediate vs non-immediate words and "run-time" versus "compile-time". Immediate words execute at compile-time and non-immediate words execute at run-time. These ideas occur because forth is a compiling system. eg char " puts the character " (integer 34) on the stack at run-time but [char] " or [ char ] " does the same thing at compile time, no not correct... FORTH BOOKS "Threaded Interpretive Languages" by RG Loeliger Apparently some info on how to build a forth system. "A Complete Forth" (1983) A good general overview of early forth "Thinking Forth": Brodie An introduction to forth. Available free online. "Forth Programming Handbook" 3rd Edition, Conklin, Rather (2007) A good reference and explainer for the 1994 standard language. Also reasonable recent. "Forth: The next step" by Ron Geere This is a simple and useful book which defines some handy new words in forth (such as squareroot etc). "Forth Application", S.D.Roberts Strange and unreadable, at least to me. But no doubt with some good ideas. Scientific Forth: Julian V. Noble A vey well regarded forth book but out of print. Finite State Machines in Forth, an article by J.V.Noble http://galileo.phys.virginia.edu/classes/551.jvn.fall01/fsm.html Good articles by Noble http://galileo.phys.virginia.edu/classes/551.jvn.fall01/primer.htm http://galileo.phys.virginia.edu/classes/551.jvn.fall01/primer.htm Realtime forth FORTH HISTORY AND VERSIONS Figforth, forth 79 standard, forth 94 standard LMI-Forth for the IBM pc MVP-Forth not sure where it ran Open Firmware, by Mitch Bradley. Still used at OLPC for booting laptops. A major project. cforth: based on mitch bradley olpc forth gforth: a gnu forth probably still being developed. amforth: for avr chips flashforth: for microcontrollers using flash memory Multos - stack based vm os Colour vision systems, bacchus marsh, victoria, au Keycorp, chatswood, KYC www.keycorp.net www.forthos.org http://www.vsta.org/contact/andy.html A bootable x86 forth in protected mode with oo extension Example of forthos oo syntax Set -> new constant mySet 1 mySet -> add 123 mySet -> add * Add a method to a class Set -> :method addRange ( high low self -- ) -rot do i over -> add loop drop method; TO DO (june 2018): immediate tasks: * asci animals. An asci drawer, write a line word that takes a 'bearing' (north, south, east, west) and a length. Then define a 'turn' word which creates asci corners, based on current bearing and turn angle eg 90, -90, 180, -180 etc So the 'line' word should also leave the current bearing on the stack. * make create set the domain for each word. * list words with their fully qualified name. * fix create bug for defining words by writing (create) and create as an immediate word, and adding code to semicolon to see if def bit is set and compiling semicolon code there. So we check the def bit and either do FCALL (;) or LIT, FCALL C, LITW, (;), , etc depending on whether def is set. Bug has been fixed for does> and (does>) * allow simple editing of blocks to test writing blocks * parse includes 1 space before text, which is not right. * incorporate michael petch code for reading sector. * change 'decomp' so that it displays the address as well as the name of the fcall functions. * write search to find all words containing text * use wsize in see see, etc to decompile whole word. * write DEPS ( xt -- ) create a list of xts which are dependencies of word represented by xt. * make a 'buffer:' word * write a 'dependency' compiler. Find a word on disk, leave its block number + offset and then find all sub-words and leave their block numbers + offset on the stack or in some kind of buffer. Check that no duplicates are added. This is tricky, we need offsets and lengths of each word in the block or file, otherwise we need 1 file per word. A dependency compiler needs a set type and an add word is there any reason to loop on the return stack as opposed to the data stack. It would be more convenient for rloop.x to decrement the top data stack item!!! * a "search bit" in the name count field (next to the immediate bit, perhaps. This makes the word invisible in the dictionary)... not sure if necessary. * make READ do a number of attempts to read disk blocks (1K) * rewrite some words as "source" now that the : compiler more or less works. * find out what words are really necessary for the : word * it would be good to be able to add new "opcodes" dynamically to allow for the situation where new hardware becomes or is available. For example, if a gyroscope is available, then extended opcodes should read data from it. Yes the machine should be able to build apon itself/ add new opcodes to itself. * make a "document" dictionary, or shadow comments to other diskblocks * "dictionary full" check, ie. is there any more memory in which to place new words? 'unused' word or 'free' * stack underflow check? * implement block and buffer. Buffer writes to disk and frees a buffer and block reads from disk. block call buffer * all defining words need to be immediate, so 'does>' could just do that automatically. see mike gonta: "all you wanted to know about usb booting but were afraid to ask" for important disk geometry info, fat12 info and read/write usb memory. complete source example MEMORY MAP The way that the code and data is laid out in memory is important. The bootloading segment (512 bytes) loads more code into memory (a few K) and then jumps to the entry point. When the code dictionary grows (by the use of new colon : definitions) the new dictionary entries should be placed in memory after the end of the dictionary. CHALLENGES How to write new code words back to disk? We can either write compiled code to disk as part of the dictionary, or just write source code to disk in forth-style "blocks". This is potentially dangerous, if we accidentally write to the computer hard disk we may corrupt the file system or even the operating system! Some kind of fat12 file system would be convenient, so that I could edit source code on someother operating system. Or at least copy it. HOW TO BUILD AND RUN THIS CODE We can either run the code in a virtual machine like qemu or else actually write it to a usb or cd and boot it! The simulator is good for testing, but the usb or cd boot shows you how it works on real hardware! tools: nasm, qemu, dd, bash shell (or any other shell for compiling etc) linux (makes things easier), sed, date, vim, htmldoctor, enscript - for formatting and printing source code * create a printable pdf of the code in 2 columns landscape format enscript -o - -2r -f Courier7 os.asm | ps2pdf - os.pdf * compile with nasm into a bootable executable nasm -fbin -o os.bin os.asm; * make a 1.4Meg floppy image and insert the executable into it. sudo rm os.flp; sudo mkdosfs -C os.flp 1440; sudo dd status=noxfer conv=notrunc if=os.bin of=os.flp' * run the executable floppy image with qemu simulator (VM) sudo qemu-system-i386 -fda os.flp' To "burn" to usb or cd try: * use dmesg to see what your usb flash drive is called * unmount the usb flash drive umount /dev/sdc * write the bootable floppy image to flash drive: WARNING! sudo dd if=os.flp of=/dev/sdc !! The code above deletes all other files on the memory stick Be very very careful that you dont do this to your hard-disk or you will end up with an unbootable computer. !! The light should flash on the usb stick indicating that data is being written. Then just reboot the computer and choose the boot device. I think I had to "enable csm" or "choose floppy" mode in my bios to get the usb stick to boot. Some bash aliases for compiling and running the code. # make qemu open maximised but not full screen alias os='sudo qemu-system-i386 -fda os.flp & sleep 1; wmctrl -r QEMU -b add,maximized_vert,maximized_horz' Below is a bash function which includes preprocessing of forth source code in %if 0; %endif; blocks. The preprocessing code is in os.sed. This is convenient because NASM doesnt seem to have any kind of multiline DB or DW syntax which means for defining forth source code in a nasm file we have to put "db ' " around every line. The sed line deletes the marker lines (lines ending in "code{" and "}code" ) as well as comment and empty lines and put the "db '" syntax around each line of text. # run the forth os os() { if [ -z "$1" ]; then name="os" else name=$1; fi echo " running ${name}.flp " sudo qemu-system-i386 -fda ${name}.flp & sleep 1; wmctrl -r QEMU -b add,maximized_vert,maximized_horz } # compile the forth style byte code system with some sed preprocessing # of the nasm file (to allow multiline text source code) ccos() { if [ -z "$1" ]; then name="os" else name=$1; fi echo " compiling ${name}.asm " cat ${name}.asm | sed -f os.sed > ${name}.pre.asm nasm -fbin -o ${name}.bin ${name}.pre.asm; sudo rm ${name}.flp; sudo mkdosfs -C ${name}.flp 1440; sudo dd status=noxfer conv=notrunc if=${name}.bin of=${name}.flp } cos() { cat os.asm | sed -f os.sed > os.pre.asm nasm -fbin -o os.bin os.pre.asm; sudo rm os.flp; sudo mkdosfs -C os.flp 1440; sudo dd status=noxfer conv=notrunc if=os.bin of=os.flp } * The sed script used to preprocess the nasm source file #!/bin/sed /code{ *$/,/}code *$/ { # ignore lines starting with "times" in code blocks # 't' jumps to end of script. s/^ *times/times/;t; /^ *$/d; /^ *;/d; /code{/d; /}code/d; # need to 'escape' double quotes s/"/",'"',"/g; s/^ */db " /; # put newline on the end of each line #s/ *$/ "/ s/ *$/ ",13,10/ } HISTORY 16 july 2018 might be good to have a begin/until which has a counter with no limit. wrote 'println' which is like type but only prints one line. Hopefully fixed a bug in l.0 which sets a list to zero probably will put code in c, and , to update >code if necessary. 11 july 2018 made some progress on : deps. fixed a bug in list.addu more 'create' bugs... The following line doesnt work [ create pawn 1 c, 2 c, 3 c, ] This is because the >code pointer is not updated after the c, words. So, 'here' is updated, I think I need to change the way 'here' and >code work. 6 july 2018 wrote 'within'. ppm format easiest to display, or pgm as a glyph a random number series. an editable buffer, test.tonumber and test.type removed from bytecode 4 july 2018 also accept can be put in source. wrote a source 'accept' with a maximum number of chars. used a helper variable. wrote a 'continue' which jumps out of an if clause and back to begin. But will fail for 2 nested ifs tried to write an accept with arrow keys but got lost. simplest random seed using the clock, but need a series. eg var R : rnd clock drop 16 mod ; : r' R @ getnext R ! ; 2 july 2018 Would like to write 'accept' as source with some command history. cleaning up. removed old some bytecode words like 'seecomp' and 'one' which were used to debug the system. Left dotstack.p as bytecode because it could be useful to debug bytecode words. 1 july 2018 Moved [ and ] into source. wrote a new shell in source. opcodes: 1410 bytes ( 62 opcodes) bytecode: 3689 bytes ( 89 bytecode words ) total: 9427 bytes ( 221 total words ) 30 june 2018 wrote a source .S version of .s should add some command line history and write that history to blocks, so that work is not lost. write accept as source and allow arrow keys to edit. Multiline accept. Make an asci sperm whale 29 june 2018 asci animals: could do sperm whale, python. 28 june 2018 Writing a little chess: code. Also fixed a cbw sign extension problem for c@ c@+ probably caused in the transition to the top dx engine. Tidying up lib.p block0 now loads all the other blocks. Made numbers and opcodes work in immediate state. The ops get compiled to a short buffer and executed immediately. working on the buffer editor ed: so as to try to make this forth self-supporting. That is, to be able to edit and develop the system with no OS. Need lots more 'line arithmetic' to make it work. Also, how to edit over rs323 connection. eg vi etc opcodes: 1404 bytes ( 62 opcodes) bytecode: 3852 bytes ( 212 words ) total: 9059 bytes 27 june 2018 An x86 assembler can work like this code: doit ax inc, ax bx sub, ;code So the operand comes first and is pushed onto the stack, then the inc, word compiles the mnemonic with its operand. This raises the questions of wordlists and vocabs. ax is a word in the assembler word list. Recursive factorial, and arithmetic, calculate e exponent. Put domain fields in some words. The recursive factorial is very succint, like gcd. 26 june 2018 All this needs reconsideration. But the domain table idea seems sound at the moment This is claytons object orientation. Domain table is like this domain.table: db 1, 0, 3, '4th' db 2, 1, 1, 4, 'set' ; extra space here for new domains the 1st number is the domain symbolic constant, used in the domain field of the compiled word. The second is a count of how many prefix domains (none for 4th) then comes a list of prefix domains as symbolic constants, then the counted string name. This is compact for compilation but allows all names in the dictionary (and outside of it) to be unique. Claytons oo is like this * define the object and just return a reference to the obj at runtime : set create allot blah blah does> blah, not much ; imm :dom the word ':dom' compiles a new domain for the last compiled word. The new domain 'set' with possible prefix domains will be appended to the domain.table if there is sufficient room. So defining words should automatically set the domain of defined words to the same as themselves? * create an instance of the object 20 set ss When set create 'ss' it probably should sets its domain automatically to 'set'? * create some words which operate on a ref to the instance eg : additem ( address -- ) blah blah ; * set the domain of the new word to set dom set. So additem has a full name of '4th.set.additem' but it could still be used by a different object. * use the new method on the instance ss m> additem 'm>' can use the reference to ss (left on the stack by ss) to get its xt and therefore its domain. Then use that domain to do a special find only searching for that domain. m> could change the domain used by 'find' when searching for a word (eg 'additem') in the dictionary. So find will only find 'methods' of 'set' because they are the only words with that domain. But normally these words/methods will be invisible to find because the domain will be '4th' or somesome. Write a word that prints fully qualified names using the domain field in the compiled word eg 4th.set.additem etc Apart from using this as a claytons OO system, it is a useful step towards universal unique locatable names in source code. Thinking about domain fields and how to use them. Have got this schema so far: domain field is 1 byte in each word. the 'find' word has an implicit or explict domain during its search. Or a list of domains. So when we do a normal 'find' we wont find words with different domains unless we explicitly search for them. This may be parallel to forths 'wordlist' concept, same same but different. Also, we can take one step towards an object system by hiding object methods in their own domain. 25 june 2018 working on the ed: edit buffer object. This seems promising. Once we can edit a buffer then the system becomes almost self sustaining. then do chess. Started to put in domain fields. wrote a list: data type that has an addu "method" for adding unique elements as well as a normal add method. I think the code is more readable Wrote allot0, dump and dumpw wrote !1+ but not !1- That is, decrement the value at the address and store new value there. Modified list: so that object reference is no longer left on stack. Wrote 'thru' to load a series of blocks. ." prints in what word an address is found, and the offset from the start of the word. Thinking about how to code data structures. Put one reference on the stack and make 'methods' that access and manipulate part of the structure. Also, make the methods use a simple reference to the object. Eg an editing buffer has capacity, insertion point and pointer to buffer. 24 june 2018 Found a bug in r@. Needed to dig under the word return pointer (of r@ itself). One extra ';' causes fatal errors. made the 'unloop' word which is needed for exiting from loops before the counter has finished. wrote erase and fill. Wrote set.ls and set.add and set: which appear to create a set that only contains unique items (16bit elements). opcodes: 1302 bytes bytecode: 3847 bytes total: 8274 bytes 23 june 2018 made atxy and getxy as opcodes to obtain and set the current cursor position. wrote 'arrow' to move cursor around. Now the trick is to move the insertion point around in a text buffer. This is a map. Cant do an exit from a do/loop, need an unloop or 'leave' word which will get the loop parameters of the return stack. The create bug still exists but can be avoided by putting does> at the end of defining words. Will try to write ed which allows basic editing of a buffer (1024bytes) which can then be saved to disk with 'save'. opcodes: 1270 bytes bytecode: 3839 bytes total: 7770 bytes 22 june 2018 made write.x appear to work. Wrote 'save' which saves the first buffer to disk block n. This maybe enough to use the os to develop itself. Only tested on qemu. Need to test on real hardware. But need some kind of editor to edit buffers/blocks. opcodes: 1245 bytes bytecode: 3829 bytes total: 7322 bytes included a "sanity-check" in the block writing routine which checks that the number 31415 is the 8th byte of the boot sector 21 june 2018 Since the aim is to achieve a minimum size and simplicity while maintaining portability, it is useful to record the system size. opcodes: 1231 bytes bytecode: 3829 bytes total: 7280 bytes Regarding the create runtime bug (dictionary code pointer not advanced etc): Wrote 'def' to set the defining control bit in a word. After consideration, we see that defining words, when they exit at run-time need to do essentially what ; semicolon does at run-time. However, when a defining word runs, there is no semicolon. This arises because the system compiles even code which is entered interatively. When a defining word stops defining, the compile point needs to be switched back to the 'anon' buffer and the >code point needs to be updated. One solution, is to make CREATE set a 'def' control bit in the count byte of the name. So create becomes an immediate word and compiles a call to (create) which is just the same as the current create. Then when semicolon executes it checks def control bit, and if set, compiles a context switch. This bug also effects does>. The does> bug seems fixed by essentially putting ; semicolon code into (does>) We could write (;) a runtime semicolon. 20 june 2018 Struggling with this code>here bug in defining words. One hack is to define var like : var create 0 , code>here ; imm but that is inelegant. When I put code>here into c, I seem to get a complete meltdown, maybe a stack overflow, or infinite loop. Not sure why yet. Discovered a bug in the way that code>here is called at the end of a new defining word, so that space is not allocated for the data field of the new word. So when we execute a defining word the defined word doesnt get any data space. This is tricky, maybe c, , and allot etc also need to update >code or call code>here. Or in here>anon call code>here first so that the code compile point is set correctly wrote 'mname' which shows the machine name and signature, and maybe opcodes. Edited the bash script cos.sh (and .bashrc) so that todays date is inserted in the machine name when the code is compiled. This helps to know what version we are running. 19 june 2018 in' is a useful word. Finds the word in which a memory address occurs (this is not its standard name). Wrote a new bash function ccos so that I can compile different versions of the machine easily. 18 june 2018 Realised that a unique machine name is useful for installing or compiling new opcodes. The name indicates the underlying hardware and the 2 stack machine implementation strategy. Converted to holding TOS (top-of-stack) in the dx register. This should improve performance, especially for calculation intensive operations. But the performance difference is not very noticable at the moment. Need better error reporting in IN, or item, so we can work out where something is going wrong. If exec.x was an opcode and could call itself, then using call/ret for each opcode may have some useful and interesting consequences. But real machines dont seem to do this (an opcode which can call itself), and I cant think of any immediate benefit. Converted call/ret in exec.x/opcodes to jmp [] etc. This appears to be working. Thinking about using jit just-in-time opcodes which would allow the compilation to machine code of the most used words. thus improving performance. Conversion appears initially successful. 17 june 2018 will try to convert 'calls' in exec.x to jumps. which should save code-space and time. Old version saved to os.call.asm 16 june 2018 fixed do/loop to make it standard. fixed sed preprocessing for 'times' lines. wrote write.x but havent tested. wrote timeloop to do simple performance testing. appeared to fix read.x by using int 13h ah=8 function to get disk parameters. On the asus computer, sectors=63 and sides=32 drive=0 15 june 2018 Could write a 'set' type, with a capacity and length, with an 'add' method, which would only add a 16bit value if it wasnt already in the set. This would be useful for building a set of unique xts (execution tokens) when finding the compiled dependencies for a given word. Also, could adapt this technique to a 'point set' where each element in the set is 2 16bit values, which could be used for making a list of block number + offsets for a dependency compile from source (only the required word and its dependencies are compiled). Realised that the exec.x function could just jump to the opcodes, rather than calling them, which would simplify the coding of each opcode. made 'splash' show opcode and byte codes sizes. changed ' to make it more standard. rewrote ." to simplify trying to write a new read.x to calculate chs for floppy emulation, but need to print off and analyse. Thought of a dependency compiler creating a list of block numbers and offsets. 13 june 2018 realised that if fi begin again etc can all be coded in source. 12 june 2018 made a splash screen 'splash'. Saw how the scale operator */ is useful for multiplying by fractions without losing precision eg: 100 2 3 */ gives 2/3rds of 100. The intermediate result is a double number. could make 'state' have 3 states, not 2. ie immediate/delegate/compile in immediate, everything is compiled, in delegate, the word itself decides whether to compile or execute, and 'compile' where everything gets compiled even immediate words. 11 june 2018 can define [char] like this. : [char] postpone char postpone literal ; defined ." Also postpone could have a syntax like [: ... :] which basically would force even immediate words to be compiled. This would just be a branch in the item,/compile, word. fixed parse so that it doesnt include final delimiter started a sed preprocessor in a separate file 'os.sed' Need to stop os.sed from mangling labels like 'block1:' etc. Realised that postpone is an important word. It just compiles an FCALL to the word even if the word is immediate. Can be defined in source. Thought about how to emulate opcodes in high level forth. The opcode calls a word similar to FCALL that pushes address of emulating word on call stack. But need to do some very simple assembling to get the write address into the machine code. Does 'call,' need to be immediate? 10 june 2018 made 'postpone' which compiles an fcall to immediate words like sliteral (otherwise they would execute immediatly). This is because we want sliteral to execute when parse has finished (at run time) not at compile time. made s" using postpone. did sliteral and (does>) the runtime behaviour of does> and wrote does> . Made call, ( xt -- ) to compile an FCALL to xt realised the importance of (does>) and other (...) words. Otherwise we would have to compile lots of code while executing 'does>' 9 june 2018 Made the preprocessor put a 13,10 at the end of each line because plain text source will have them. This allows a comment-to-end of line syntax which is handy. Fixed a problem in HERE>CODE. This word executes the ANON buffer but was not resetting it, so words like IMM were executing multiple times. FIB stopped working today, not sure why Made a LITERAL word to use with CHAR in colon defs more or less fixed PARSE. so comments like ( a b --) are easy now. eg: : ( [char] ) parse ; imm 8 june 2018 wrote a very basic sed preprocessor and a bash function to use it (see comments at top of file about how to compile and run). But need a forth comment syntax to make it worthwhile, and to get that I need PARSE to work, either in source or bytecode. rewrote INWORDS and FINDWORDS as source. These words are only really useful as a testing mechanism. 7 june 2018 wrote r.v.nobles recursive euclid "gcd" function. Recursive functions work on this forth!! Copied a recursive FIB fibonacci word. And I didnt even think about them when I was coding! Rewrote keycode as source. When loading source code into a buffer, it might be good to get rid of extra white space... Defined char but realised that there is a problem with the use of char in a colon def, because the character code is pushed onto the stack at compile time, instead of compiling a literal to push onto stack at run time 6 june 2018 Thought that PCALL could be modified to execute opcodes as well. Not so easy. ACCEPT uses the rstack to save the buffer address, but 2DUP makes this unnecessary. Also a char limit would be good In this system, it seems all defining words must be immediate??? So we must do : variable create 0 , ; immediate : var create 0 , ; imm need to fix other parse now. using CREATE in : not so good ? Thinking about how to implement immediate execution of opcodes and literal numbers (to push onto the stack). Could compile to a small anonymous immediate buffer and execute immediately. Wrote CREATE. Appears to be working. Made : COLON and ; use CREATE which appears to be working. 5 june 2018 fixed WPARSE to use the >IN stream and got : COLON to use it. Thought of the idea of a machine "signature". Which is just a list of opcodes plus mappings. eg 1-5,7-33,34:2101 This means that the given machine has standard opcodes 1-5 and 7-33 and the standard opcode 2101 which is mapped to 34. Made a byte code version of [ and ] for testing. Started CREATE. wrote WPARSE for parsing whitespace delimited words, skipping initial whitespace. opcodes cant execute immediately! nor numbers! fix? WPARSE advances the input stream with IN+ >CODE is a pointer but HERE is a value and >IN is 2 values This is a bit confusing and not consistent. Could make >CODE a value, and CODE a pointer. wrote a HERE>CODE word which sets the here var to the next available location in dictionary. It is be called by CREATE and thus all defining words. It also writes an EXIT to end of ANON and executes anon with FCALL This ensures that what is in the anonymous compile buffer ANON will always get executed before new code is written to the dictionary. We can call this a "context switch". The ANON compile buffer is where non-defining interactive commands are compiled to. 4 june 2018 Wrote a whitespace word WSPACE that returns a flag indicating if a char is tab=9 cr=13 space=32 or 0 etc This can be used in WPARSE Wrote LOAD which just loads the first block. A better LOAD would be : load 2* block1 + 2 first read first 1 K source ; which should load any block number eg 3 load. But we should also implement block & buffer. And started to move code into that source code block from lib.p rewrote >IN and IN0 so that they would use a length variable for input streams. Realised that SOURCE and IN, are different, because SOURCE establishes a new input stream. We dont seem to have to save the current one because, by the time it executes the current stream has been compiled... but, if we execute SOURCE immediately then we may get serious problems. 3 june 2018 made an word IMM which sets the last word in the dictionary to immediate. Thinking about the exec function which executes bytecode. should this be able to call itself? I cant think of any analogy in a real machine, so I dont think so. Could be good to have a word which pushes the address of the opcode table onto the stack to allow manipulation of opcodes from within forth 2 june 2018 Changed LAST to be a pointer. Made a STATE variable Put state test into ITEM, WORD. But [ and ] words must be immediate themselves! or they dont work. Made the READ opcode sort of working (up to sector 18 ?) but need to grapple with the idea of emulated disk geometries and convert sector number to head+cylinder+sector number etc Want to make this opaque, so that the opcode handles this mess and the code can just treat the disk as an enormous array of sectors/blocks. After that can code buffer and block words. Maybe use last byte of block to store "update" bit and block number. This info is used by buffer and block words to determine if a read and/or write to disk is necessary. Will just have one source buffer initially for simplicity. Think I am close to achieving a robust, compact, and powerful system. Once LOAD/BUFFER/BLOCK are working well, we can rewrite many words as source code, thus reducing the core even further. Still need to fix DO and LOOP And write VAR/VARIABLE, CON/CONSTANT etc. Also could try to wrap core and source blocks in a super basic fat12 file system so that source code can be edited on another operating system? But even FAT12 looks complicated! 29 may 2018 writing out some standard forth words in %if0 block Will need to preprocess %if0 block below to put "db ' " etc in front of every line of some forth source. squareroot word 28 may 2018 thinking about CREATE and DOES>. realised that the implementation of these words is not that difficult. DOES> needs to compile an FCALL in new defined words to the code immediately after it, as well as an EXIT for the defining word. Probably will do this by jumping over parameter field. 23 may 2018 Made EKEY work on x86 architecture. Made a LOOP immediate word. but its not like the standard forth LOOP at the moment. 15 May 2018 Finally got a : compiler working, so new words can be added to the dictionary. Wrote inputcompile (IN,) to compile the input buffer to the HERE pointer. Made a basic foreground colour changer just for fun FG opcode and a video mode changer VID for colour vga style monitors 11 May 2018 separated this forth-like system into a new file. Up until now, have been developing as part of the osdev booklet. 10 june 2017 made a return stack with es:di and made fcall.x and exit.x use the return stack, apparently successfully which allows nested procedures. %endif ; }docs BITS 16 [ORG 0] jmp 07C0h:bootload ; Goto segment 07C0 db 'pi:' dw 31415 ; magic number drive: db 0 ; a variable to hold boot drive number db 'boot:now' bootload: mov ax, cs ; the code segment is already correct (?!) mov ds, ax ; set up data and extended segments mov es, ax mov [drive], dl ; save the boot drive number mov ax, 07C0h ; Set up 4K stack space after this bootloader add ax, 288 ; (4096 + 512) / 16 bytes per paragraph mov ss, ax ; with a 4K gap between stack and code mov sp, 4096 ; save the DL register or else dont modify it ; it contains the number of the boot medium (hard disk, ; usb memory stick etc) ; The 'floppy' Drive is NOT necesarily 0!!! reset: ; Reset the virtual floppy drive (usb) mov ax, 0 ; mov dl, [drive] ; the boot drive number (eg for usb 128) int 13h ; jc reset ; ERROR => reset again readdisk: mov ax, 1000h ; ES:BX = 1000:0000 mov es, ax ; es:bx determines where data loaded to mov bx, 0 ; mov ah, 2 ; Load disk data to ES:BX ;mov al, 8 ; Load 8 sectors 512 bytes * 8 == 4K mov al, 16 ; Load 16 sectors 512 bytes * 16 == 8K mov ch, 0 ; Cylinder=0 mov cl, 2 ; start sector=2 (sector 1 is the boot sector) mov dh, 0 ; Head=0 mov dl, [drive] ; int 13h ; Read! jc readdisk ; ERROR => Try again or exit jmp 1000h:0000 ; Jump to the loaded code times 510-($-$$) db 0 ; pad out the boot sector ; (512 bytes) dw 0AA55h ; end with standard boot signature ; **** ; this below is the magic line to make the new memory offsets ; work. Or compile the 2 files separately ; https://forum.nasm.us/index.php?topic=2160.0 section stage2 vstart=0 jmp start ; aliases for each bytecode, these aliases need to be in the ; same order as the pointer table below ; The nasm code below gives values of 1,2,3,4,5 etc to each ; bytecode alias. A new opcode can be inserted without having ; to update all the following opcodes. DUP equ 1 DROP equ DUP+1 SWAP equ DROP+1 OVER equ SWAP+1 TWODUP equ OVER+1 FLAGS equ TWODUP+1 DEPTH equ FLAGS+1 RDEPTH equ DEPTH+1 RON equ RDEPTH+1 ROFF equ RON+1 STORE equ ROFF+1 STOREPLUS equ STORE+1 FETCH equ STOREPLUS+1 FETCHPLUS equ FETCH+1 CSTORE equ FETCHPLUS+1 CSTOREPLUS equ CSTORE+1 CFETCH equ CSTOREPLUS+1 CFETCHPLUS equ CFETCH+1 COUNT equ CFETCHPLUS ; count is an alias for c@+ EQUALS equ CFETCHPLUS+1 NOTEQUALS equ EQUALS+1 LESSTHAN equ NOTEQUALS+1 ULESSTHAN equ LESSTHAN+1 LIT equ ULESSTHAN+1 LITW equ LIT+1 EMIT equ LITW+1 KEY equ EMIT+1 EKEY equ KEY+1 GETXY equ EKEY+1 ATXY equ GETXY+1 PLUS equ ATXY+1 MINUS equ PLUS+1 INCR equ MINUS+1 DECR equ INCR+1 NEGATE equ DECR+1 DPLUS equ NEGATE+1 LOGOR equ DPLUS+1 ; logical or LOGXOR equ LOGOR+1 LOGAND equ LOGXOR+1 DIVMOD equ LOGAND+1 TIMESTWO equ DIVMOD+1 MULT equ TIMESTWO+1 UMULT equ MULT+1 FCALL equ UMULT+1 PCALL equ FCALL+1 EXIT equ PCALL+1 LJUMP equ EXIT+1 JUMP equ LJUMP+1 JUMPZ equ JUMP+1 JUMPF equ JUMPZ ; jumpf (false) is an aliase for jumpz JUMPNZ equ JUMPZ+1 JUMPT equ JUMPNZ ; jump-true alias for jump-not-zero RLOOP equ JUMPNZ+1 DIVTWO equ RLOOP+1 READ equ DIVTWO+1 ; loads sectors from disk WRITE equ READ+1 ; writes sectors to disk (usb etc) FG equ WRITE+1 ; foreground colour BG equ FG+1 ; background colour CLS equ BG+1 ; clear screen VID equ CLS+1 ; video mode PIX equ VID+1 ; one pixel on screen at xy GLYPH equ PIX+1 ; display glyph on screen RTC equ GLYPH+1 ; real time clock CLOCK equ RTC+1 ; number of clock ticks since midnight NOOP equ CLOCK+1 ; no operation & end marker ;*** control bits and mask IMMEDIATE equ 0b10000000 MASK equ 0b00011111 ; Machine signature goes here ; machine name: eg x86.v1 opcodes and mappings ; the machine name can be used to install or compile new ; opcodes for this particular architecture ; but should this be an opcode or a forth word? machine.doc: ; db ' allows access to the machine name, signature and opcodes ' ; dw $-machine.doc machine: dw 0 db 'machine', 7 machine.p: db LITW dw machine.name db EXIT machine.name: ; the name of the machine, including on what chip it is ; implemented and a mnemonic about how it is implemented ; maybe include compile date in this machine name. db machine.sig-machine.name-1, '4th.x86.topdx' ; maybe a machine description here ; the machine signature which is like an opcode map machine.sig: db 4, '1-55' ; a list of domains, or tiered domains, to provide universal ; unique locatable names machine.dom: ; structure of domain table record: ; constant, prefix count, prefix list, name count, name ; the code to print a domain name will probably be recursive. ; A list of symbolic constants for domain name prefixes LANG equ 1 ; language name CORE equ 2 ; core ops or primitives MON equ 3 ; monitor TIME equ 4 ; time and date DISK equ 5 ; storage INOUT equ 6 ; input output db 0, 0, 9, 'undef' db 1, 0, 3, '4th' db 2, 1, 1, 4, 'core' db 3, 1, 1, 3, 'mon' db 4, 1, 1, 4, 'time' db 5, 1, 1, 4, 'disk' db 6, 1, 1, 5, 'inout' ; should objects have their own domain? times 20 db 0 ; need extra space for new doms, but where? ; a table of code pointers. The pointers have the ; same offset in the table as the value of the opcode op.table: dw 0, dup.x, drop.x, swap.x, over.x, twodup.x dw flags.x, depth.x, rdepth.x dw ron.x, roff.x dw store.x, storeplus.x, fetch.x, fetchplus.x dw cstore.x, cstoreplus.x dw cfetch.x, cfetchplus.x dw equals.x, notequals.x, dw lessthan.x, ulessthan.x, lit.x, litw.x dw emit.x, key.x, ekey.x, getxy.x, atxy.x dw plus.x, minus.x, incr.x, decr.x, neg.x dw dplus.x dw logor.x, logxor.x, logand.x dw divmod.x, timestwo.x, mult.x, umult.x dw fcall.x, pcall.x, exit.x dw ljump.x, jump.x, jumpz.x, jumpnz.x, rloop.x dw divtwo.x dw read.x, write.x dw fg.x, bg.x, cls.x, vid.x, pix.x dw glyph.x dw rtc.x, clock.x ; could put just-in-time opcodes here. dw noop.x, -1 ; this is the function which executes the byte codes ; takes a pointer to the code. Jumps are relative to the first ; byte of the jump instruction. This is not an opcode because ; it represents the machine itself... exec: dw machine.p db 'exec', 4 exec.x: ; get exec.x return address out of the way. ; Actually exec.x never returns! (since SHELL is an infinite loop) pop word [returnexec] ; save return ip pop si ; get pointer to code exec.nextopcode: xor ax, ax ; set ax := 0 lodsb ; al := [si]++ cmp al, 0 ; zero marks end of code je .exit .opcode: mov bx, ax ; get opcode (1...nop etc) into bx shl bx, 1 ; double bx because its a word pointer ; each opcode jumps back to exec.nextopcode. This improves speed ; compared to doing call/ret. ; Can hold top of stack in dx "sub dx, bx " etc jmp [op.table+bx] ; use opcode as offset into opcode table ; check for stack underflow here ??? ; Actually this is never executed because each opcode ; jumps straight back to exec.nextopcode: directly jmp exec.nextopcode .exit: ;This never exits. shell is infinite loop, but it could exit... push word [returnexec] ; restore fn return ip ret returnexec dw 0 plus.doc: ; db 'add the top 2 elements of the stack.' ; db ' ( n1 n2 -- n1+n2 ) ' ; db ' This opcode is agnostic about whether the two 16 bit ' ; db ' numbers are signed or unsigned. The flags opcode can be ' ; db ' checked to see if there is an overflow or carry. ' ; dw $-plus.doc plus: db LANG ; domain name prefix dw exec.x db '+', 1 plus.x: pop bx add dx, bx jmp exec.nextopcode dplus.doc: ; db 'add the 2 double numbers ' ; db ' ( d D -- d+D ) ' ; db ' This opcode is agnostic about whether the two 16 bit ' ; db ' numbers are signed or unsigned. The flags opcode can be ' ; db ' checked to see if there is an overflow or carry. ' ; dw $-dplus.doc dplus: db LANG ; domain name prefix dw plus.x db 'd+', 2 dplus.x: ; TOS (dx) has high word of first double mov ax, dx shl eax, 16 pop ax ; lower word here pop dx shl edx, 16 pop dx add edx, eax push dx ; lsw become NOS shr edx, 16 ; msw becomes TOS (dx) ; a stack method to convert singles to doubles ; push word [wHigh] ; push word [wLow] ; pop dword [dwResult] jmp exec.nextopcode minus.doc: ; db 'subtract the top element of stack from next top' ; db ' ( n1 n2 -- n1-n2 ) ' ; dw $-minus.doc minus: db LANG ; domain name prefix dw dplus.x db '-', 1 minus.x: pop ax sub ax, dx mov dx, ax ; tos=dx jmp exec.nextopcode incr.doc: ; db ' ( n -- n+1 ) ; db 'Increment the top element of the data ; db 'stack by one. ' ; dw $-incr.doc incr: db LANG ; domain name prefix dw minus.x db '1+', 2 incr.x: inc dx jmp exec.nextopcode decr.doc: ; db ' ( n -- n-1 ) ; db 'Decrement top element of the data stack by one. ' ; dw $-decr.doc decr: db LANG ; domain name prefix dw incr.x db '1-', 2 decr.x: dec dx jmp exec.nextopcode neg.doc: ; db ' ( n -- -n ) ; db 'Negates the top item of the stack' ; dw $-neg.doc neg: db LANG ; domain name prefix dw decr.x db 'neg', 3 neg.x: neg dx jmp exec.nextopcode logor.doc: ; db ' ( n1 n2 -- n1 V n2 ) ; db 'the logical OR of n1 and n2' ; dw $-logor.doc logor: db LANG ; domain name prefix dw neg.x db 'or', 2 logor.x: pop ax or dx, ax jmp exec.nextopcode logxor.doc: ; db ' ( n1 n2 -- n1 V n2 ) ; db 'the logical or of n1 and n2' ; dw $-logxor.doc logxor: db LANG ; domain name prefix dw logor.x db 'xor', 3 logxor.x: pop ax xor dx, ax ; top of stack == dx jmp exec.nextopcode logand.doc: ; db ' ( n1 n2 -- n1 && n2 ) ; db 'the logical and of n1 and n2' ; dw $-logand.doc logand: db LANG ; domain name prefix dw logxor.x db 'and', 3 logand.x: pop ax and dx, ax jmp exec.nextopcode divtwo.doc: ; db '(n1 - n1/2) ' ; db ' divide n1 by 2 ' ; dw $-divtwo.doc divtwo: db LANG ; domain name prefix dw logand.x db '/2', 2 divtwo.x: shr dx, 1 ; do dx := (dx+1)/2 /tos=dx jmp exec.nextopcode divmod.doc: ; db '(n1 n2 - remainder quotient) ' ; db ' divide n1 by n2 and provide remainder and quotient. ' ; db ' n2 is the top item on the stack ' ; dw $-divmod.doc divmod: db LANG ; domain name prefix dw divtwo.x db '/mod', 4 divmod.x: ; maybe could optimise by doing div dx ?? mov bx, dx ; divisor is top of stack. tos=dx, xor dx, dx ; set dx := 0 pop ax ; dividend is next element div bx ; does dx:ax / bx remainder->dx; quotient->ax push dx ; put remainder on stack mov dx, ax ; put quotient on top of stack jmp exec.nextopcode timestwo.doc: ; db '(n1 -- n1*2 ) ' ; db ' double n1. This basically performs a ' ; db ' left shift on the bits in n1 ' ; dw $-timestwo.doc timestwo: db LANG ; domain name prefix dw divmod.x db '2*', 2 timestwo.x: shl dx, 1 ; tos=dx jmp exec.nextopcode mult.doc: ; db '(n1 n2 -- n1*n2 ) ' ; db ' signed multiplication ' ; dw $-mult.doc mult: db LANG ; domain name prefix dw timestwo.x db '*', 1 mult.x: pop ax imul dx ; do dx:ax := ax*dx mov dx, ax ; result in top of stack, tos=dx jmp exec.nextopcode umult.doc: ; db '(n1 n2 -- n1*n2 ) ' ; db ' unsigned multiplication ' ; dw $-umult.doc umult: db LANG ; domain name prefix dw mult.x db 'u*', 2 umult.x: ;mov bx, dx ; tos=dx ; we dont use the result in dx here so could not do this ;xor dx, dx ; set dx := 0 pop ax mul dx ; do dx:ax := ax*dx mov dx, ax ; result in top of stack, tos=dx jmp exec.nextopcode fcall.doc: ; db 'Call a virtual proceedure on the bytecode stack machine' ; db 'The current code pointer (in the SI register)' ; db 'is saved - pushed onto the return stack [es:di] and the address ; db 'of the virtual proc to execute is loaded into SI. ' ; dw $-fcall.doc fcall: db LANG ; domain name prefix dw umult.x db 'fcall', 5 fcall.x: lodsw mov [es:di], si add di, 2 mov si, ax ; adjust the si code pointer jmp exec.nextopcode pcall.doc: ; db ' ( xt -- ) ' ; db ' Call a procedure using the top element on ' ; db ' the data stack as the execution address. This ' ; db ' allows the implementation of function pointers' ; db ' In standard forths this is called "execute". ' ; dw $-pcall.doc pcall: db LANG ; domain name prefix dw fcall.x db 'pcall', 5 ; or call it exec/ex/execute pcall.x: mov [es:di], si ; save ip to return stack add di, 2 mov si, dx ; get proc exec address from stack pop dx jmp exec.nextopcode ; Need to think about this stack op call, is it really necessary? ; Not working! ocall doesnt return. Real machines dont really do ; this so probably dont need to have it ocall.doc: ; db ' ( opcode -- ) ' ; db ' Remove opcode from stack and execute it on machine. ' ; db ' This may allow immediate execution of opcodes....' ; dw $-pcall.doc ocall: dw pcall.x db 'ocall', 5 ; or call it exec/ex/execute ocall.x: ; experimental! ; add to optable, EQUs. ;pop word [ocall.return] ; save return ip ; top of stack == dx ; pop bx ; get opcode (1...NOP etc) into bx ; shl bx, 1 ; double bx because its a word pointer ; call [op.table+bx] ; use opcode as offset into ;push word [ocall.return] ; restore fn return ip jmp exec.nextopcode ocall.return dw 0 exit.doc: ; db 'exit a virtual procedure by restoring si ' ; db 'code pointer' ; dw $-exit.doc exit: db LANG ; domain name prefix dw ocall.x db 'exit', 4 exit.x: sub di, 2 mov si, [es:di] ; restore si from rstack jmp exec.nextopcode dup.doc: ; db 'Duplicates the top item on the stack.' ; dw $-dup.doc dup: db LANG ; domain name prefix for universal naming dw exit.x ; link to previous word db 'dup', 3 ; strings are 'counted' dup.x: push dx ; dup TOS jmp exec.nextopcode drop.doc: ; db 'removes the top item on the stack.' ; dw $-drop.doc drop: db LANG ; domain name prefix for universal naming dw dup.x ; link to previous word db 'drop', 4 ; name with reverse count drop.x: pop dx ; remove top element of stack jmp exec.nextopcode swap.doc: ; db 'swaps the top 2 items on the stack.' ; dw $-swap.doc swap: dw drop.x ; link to previous word db 'swap', 4 swap.x: pop ax ; get top stack item push dx ; put them back on in reverse order mov dx, ax ; tos=dx jmp exec.nextopcode over.doc: ; db ' ( n1 n2 -- n1 n2 n1 ) ' ; db ' Puts a copy of 2nd stack item on top of stack. ' ; db ' dont use this, will probably remove. ' ; dw $-over.doc over: db LANG ; domain name prefix for universal naming dw swap.x ; link to previous word db 'over', 4 over.x: pop ax ; get NOS (next-on-stack) stack item push ax ; push dx ; mov dx, ax ; add copy of 2nd item on top of stack jmp exec.nextopcode twodup.doc: ; db ' ( n1 n2 -- n1 n2 n1 n2 ) ' ; db ' copies 2 stack items onto stack ' ; dw $-twodup.doc twodup: db LANG ; domain name prefix for universal naming dw over.x db '2dup', 4 twodup.x: pop ax ; get NOS next stack item push ax ; push dx ; push ax ; n1 n2 n1 n2 jmp exec.nextopcode flags.doc: ; db ' ( -- n ) ' ; db ' push flag register onto the stack ' ; db ' execution flags such as carry overflow negate etc ' ; db ' are pushed onto the stack ' ; dw $-flags.doc flags: db LANG ; domain name prefix for universal naming dw twodup.x db 'flags', 5 flags.x: push dx ; save NOS on stack pushf pop dx ; get flags into TOS (dx) jmp exec.nextopcode depth.doc: ; db ' ( -- n ) ' ; db ' Puts on the stack the number of stack items ' ; db ' before this word was executed ' ; dw $-depth.doc depth: db LANG ; domain name prefix for universal naming dw flags.x db 'depth', 5 depth.x: push dx mov bx, sp mov dx, 4096 ; 4K stack (but could change!) sub dx, bx ; shr dx, 1 ; div by 2 (2 byte stack cell) ;causing problems ??? ;dec ax ; the exec.x call doesnt count jmp exec.nextopcode rdepth.doc: ; db ' ( -- n ) ' ; db ' Puts on the stack the number of stack items ' ; db ' on the return stack before this word ' ; db ' was executed ' ; dw $-rdepth.doc rdepth: db LANG ; domain name prefix for universal naming dw depth.x db 'rdepth', 6 rdepth.x: push dx mov dx, di shl dx, 1 jmp exec.nextopcode ron.doc: ; db '( S: n -- )( R: -- n ) ' ; db ' put the top item of the data stack onto the return stack.' ; dw $-ron.doc ron: db LANG ; domain name prefix dw rdepth.x db '>r', 2 ron.x: mov ax, dx ; value to store at address (tos=dx) stosw ; [es:di] := ax, di+2 pop dx jmp exec.nextopcode roff.doc: ; db '( S: -- n)( R: n -- ) ' ; db ' put the top item of the return stack onto the data stack.' ; dw $-roff.doc roff: db LANG ; domain name prefix dw ron.x db 'r>', 2 roff.x: push dx ; push new nos on stack sub di, 2 ; mov dx, [es:di] ; get top item off return stack jmp exec.nextopcode store.doc: ; db '( w adr -- ) ' ; db ' place 2 byte value w at address "adr" ' ; dw $-store.doc store: db LANG ; domain name prefix dw roff.x db '!', 1 store.x: mov bx, dx ; pointer to address pop ax ; value to store at address mov [bx], ax ; 2 byte is stored pop dx jmp exec.nextopcode storeplus.doc: ; db '( w adr -- adr+2 ) ' ; dw $-storeplus.doc storeplus: db LANG ; domain name prefix dw store.x ; link to previous word db '!+', 2 storeplus.x: mov bx, dx ; pointer to address pop ax ; value to store at address mov [bx], ax ; 2 byte value is stored inc dx ; advance address and put on stack inc dx ; advance address and put on stack jmp exec.nextopcode fetch.doc: ; db '( adr -- n ) ' ; db ' Replace the top element of the stack with the ' ; db ' value of the 16bites at the given memory address ' ; dw $-fetch.doc fetch: db LANG ; domain name prefix dw storeplus.x ; link to previous word db '@', 1 fetch.x: mov bx, dx ; address in tos mov dx, word [bx] jmp exec.nextopcode fetchplus.doc: ; db '( adr -- adr+2 n ) ' ; db ' Replace the top element of the stack with the ' ; db ' value of the 16bites at the given memory address ' ; db ' and increment the address by 2 bytes. ' ; dw $-fetchplus.doc fetchplus: db LANG ; domain name prefix dw fetch.x ; link to previous word db '@+', 2 fetchplus.x: mov bx, dx mov dx, word [bx] ; value on top of stack add bx, 2 ; increment address by 1 word (2 bytes) push bx ; save address on stack jmp exec.nextopcode cstore.doc: ; db '( n adr -- ) store the byte value n at address adr.' ; db ' eg: 10 myvar ! ' ; db ' puts the value 10 at the address specified by "myvar" ' ; db ' The address is the top value on the stack. ' ; dw $-cstore.doc cstore: db LANG ; domain name prefix dw fetchplus.x ; link to previous word db 'c!', 2 cstore.x: mov bx, dx ; pointer to address pop ax ; value to store at address mov [bx], al ; only the low value byte is stored pop dx jmp exec.nextopcode cstoreplus.doc: ; db '( n adr -- adr+1 ) store the byte value n at address adr.' ; db ' And increment the address ' ; dw $-cstoreplus.doc cstoreplus: db LANG ; domain name prefix dw cstore.x ; link to previous word db 'c!+', 3 cstoreplus.x: mov bx, dx ; pointer to address pop ax ; value to store at address mov [bx], al ; only the low value byte is stored inc dx ; advance address and put on stack (tos=dx) jmp exec.nextopcode cfetch.doc: ; db '( adr -- n ) Replace the top element of the stack with the value ' ; db ' of the byte at the given memory address.' ; db ' eg: myvar @ . ' ; db ' displays the value at the address given by "myvar" ' ; dw $-cfetch.doc cfetch: db LANG ; domain name prefix dw cstoreplus.x ; link to previous word db 'c@', 2 cfetch.x: mov bx, dx ; tos=dx xor dx, dx ; set dx := 0 mov al, byte [bx] cbw ; convert signed byte to signed word ax mov dx, ax ; put literal value on stack (tos=dx) jmp exec.nextopcode ; need to sign extend !! cfetchplus.doc: ; db '( adr -- adr+1 n ) ' ; db ' Replace the top element of the stack with the value ' ; db ' of the byte at the given memory address and increment the ' ; db ' address . This is exactly the same as "count"' ; dw $-fetchplus.doc cfetchplus: db LANG ; domain name prefix dw cfetch.x ; link to previous word db 'c@+', 3 cfetchplus.x: mov bx, dx xor dx, dx ; set dx := 0 mov al, byte [bx] cbw ; convert signed byte to signed word ax mov dx, ax ; put literal value on stack (tos=dx) inc bx ; increment address by 1 push bx ; save new address on stack jmp exec.nextopcode equals.doc: ; db ' ( n1 n2 -- flag ) ' ; db 'Puts -1 (true) on the stack if n1==n2 ' ; db 'otherwise puts zero (false) on the stack. ' ; dw $-equals.doc equals: db LANG ; domain name prefix dw cfetchplus.x ; link to previous word db '=', 1 equals.x: pop bx ; 2nd stack item cmp dx, bx ; if tos==nos je .true .false: mov dx, 0 jmp .exit .true: mov dx, -1 .exit: jmp exec.nextopcode notequals.doc: ; db ' ( n1 n2 -- flag ) ' ; db 'Puts 0 (false) on the stack if n1==n2 ' ; db 'otherwise puts -1 (true) on the data stack' ; dw $-notequals.doc notequals: db LANG ; domain name prefix dw equals.x ; link to previous word db '<>', 2 notequals.x: pop bx ; 2nd stack item cmp dx, bx jne .true .false: mov dx, 0 jmp .exit .true: mov dx, -1 .exit: jmp exec.nextopcode lessthan.doc: ; db ' ( n1 n2 -- flag ) ' ; db 'Puts 0 (false) on the stack if n1 mov ah, 0 ; set ah = 0 push ax ; save asci code onto stack, high byte zero mov dx, 0 ; false flag jmp exec.nextopcode .extended: mov al, ah ; set ah = al xor ah, ah ; set high byte to zero push ax ; save event code onto stack, high byte zero mov dx, 1 ; true flag (is extended char) jmp exec.nextopcode getxy.doc: ; db ' ( -- x y ) ; db ' Return the x and y position of the cursor. ' ; dw $-getxy.doc getxy: db MON ; monitor domain name prefix dw ekey.x ; link to previous db 'getxy', 5 ; reverse counted string getxy.x: ; dl=x/col dh=y/row mov ah, 03h ; bios function: get cursor position into dx int 10h ; invoke bios mov bx, dx mov dl, dh and dx, 0x00FF and bx, 0x00FF push bx jmp exec.nextopcode atxy.doc: ; db ' ( x y -- ) ; db ' set the x and y position of the cursor. ' ; dw $-setxy.doc atxy: db MON ; monitor domain name prefix dw getxy.x db 'atxy', 4 atxy.x: ; dl=col dh=row mov dh, dl pop bx mov dl, bl mov ah, 02h ; bios function: set cursor position specified in dx int 10h ; invoke bios pop dx jmp exec.nextopcode ljump.doc: ; db 'jumps to a relative virtual instruction.' ; db ' The jump is given in the next 2 bytes' ; dw $-ljump.doc ljump: dw atxy.x ; link to prev db 'ljump', 5 ; reverse count ljump.x: xor ax, ax ; set ax := 0 lodsw ; al := [si]++ get relative jump target into AL sub si, 2 ; realign si to JUMP instruction, add si, ax ; adjust the si code pointer by offset jmp exec.nextopcode jump.doc: ; db 'jumps to a relative virtual instruction.' ; db ' The relative jump is given in the next byte.' ; db ' eg: JUMP, -2, jumps back 2 instructions in the bytecode' ; db ' eg: LIT, '*', EMIT, JUMP, -3, ' ; db ' prints a never-ending list of asterixes ' ; dw $-jump.doc jump: dw ljump.x ; link to prev db 'jump', 4 ; reverse count jump.x: ; jumps can be handled in the exec routine ; handle jumps by modifying virtual ip (in this case SI) xor ax, ax ; set ax := 0 lodsb ; al := [si]++ get relative jump target into AL cbw ; convert signed byte al to signed word ax (neg offset) sub si, 2 ; realign si to JUMP instruction, add si, ax ; adjust the si code pointer by jump offset ; do we need to decrement si ?? yes, more logical jmp exec.nextopcode jumpz.doc: ; db ' ( n -- ) ; db 'jumps to a relative virtual instruction if top ' ; db 'stack element is zero. The flag value is removed ' ; db 'from the stack ; db ' The relative jump is given in the next byte.' ; db ' eg: JUMPZ, -2, jumps back 2 instructions in the bytecode' ; db ' eg: KEY, DUP, EMIT, LIT, '0', MINUS, JUMPNZ, -6 ' ; db ' allows the user to type until zero is pressed. ' ; dw $-jump.doc ; handle jumps by modifying virtual ip (in this case SI) jumpz: dw jump.x ; link to prev db 'jumpz', 5 ; reverse count jumpz.x: xor ax, ax ; set ax := 0 lodsb ; al := [si]++ get relative jump target into AL ; check stack for zero, if not continue with next instruction cmp dx, 0 ; if dx != 0 continue (tos==dx) jne .exit cbw ; convert signed byte al to signed word ax (neg offset) sub si, 2 ; realign si to JUMP instruction, add si, ax ; adjust the si code pointer by jump offset .exit: pop dx jmp exec.nextopcode ; should jumps take top stack element off ? yes jumpnz.doc: ; db 'jumps to a relative virtual instruction if top stack element ' ; db ' is not zero. ; db ' The relative jump is given in the next byte.' ; db ' eg: JUMPNZ, -2, jumps back 2 instructions in the bytecode' ; db ' eg: KEY, DUP, EMIT, LIT, 'q', MINUS, JUMPNZ, -6 ' ; db ' allows the user to type until "q" is pressed. ' ; dw $-jumpnz.doc jumpnz: dw jumpz.x ; link to prev db 'jumpnz', 6 ; reverse count jumpnz.x: ; handle jumps by modifying virtual ip (in this case SI) xor ax, ax ; set ax := 0 lodsb ; al := [si]++ get relative jump target into AL ; check stack for zero, if so continue with next ; instruction (dont jump) cmp dx, 0 ; if bx != 0 continue je .exit ; the only difference with jumpz ! cbw ; convert signed byte al to signed word ax (neg offset) sub si, 2 ; realign si to JUMP instruction, add si, ax ; adjust the si code pointer by jump offset .exit: pop dx jmp exec.nextopcode rloop.doc: ; db ' ( R: n -- n-1 ) ' ; db ' Decrements loop counter on return stack and jumps to ' ; db ' target if counter > 0 ' ; db ' like the x86 loop instruction this is a pre-decrement ' ; db ' so a loop counter of 2 will loop twice. The disadvantage ' ; db ' is that a loop counter of 0 will loop 2^16 times. ' ; dw $-rloop.doc rloop: dw jumpnz.x ; link to prev db 'rloop', 5 ; reverse count rloop.x: ; handle loops by modifying virtual ip (in this case SI) xor ax, ax ; set ax := 0 lodsb ; al := [si]++ get relative loop target into AL ; check return stack for zero, if so continue with next ; instruction (dont jump/loop) mov bx, [es:di-2] ; get top return stack item into bx dec bx ; decrement the loop counter on the return stack cmp bx, 0 ; if bx != 0 continue mov [es:di-2], bx ; update the counter je .exit ; the only difference with jumpz ! cbw ; convert signed byte al to signed word ax (neg offset) sub si, 2 ; realign si to JUMP instruction, add si, ax ; adjust the si code pointer by jump offset .exit: jmp exec.nextopcode ; sectors/track and sides are set by bootup code int 13h, ah=8 ; SectorsPerTrack: dw 18 ; standard floppy config ; Sides: dw 2 ; floppies only have one 'platter' read.doc: ; db ' ( 1st-sector n addr -- flag=T/F ) ; db ' reads n sectors from disk starting at sector to memory addr ' ; db ' returns 0 on failure, 1 on success. ' ; dw $-read.doc read: db DISK ; domain name prefix dw rloop.x db 'read', 4 read.x: ; 1st-sect n address mov fs, dx ; destination memory address in this segment .reset: ; Reset the virtual floppy drive (usb) mov ax, 0 ; mov dl, byte [drive.d] ;boot drive number (eg for usb 128) int 13h ; jc .reset ; read error => reset again .read: ; need to save es since rstack points with it!! mov [saven.es], es mov ax, ds ; set es:=ds because we read into this segment mov es, ax ; es:bx determines where data loaded to pop gs ; how many sectors ( -> AL) pop ax ; start sector ; -- thanks to mike gonta for this code .logical: ; Will there be issues with multi-track reads? ; Also, sectors start at one not zero !! ; So sector 1 is the boot sector on disk. ; Calculate track (cylinder), head and sector (chs) settings ; for the int 13h (ah=2) read disk function ; IN: logical sector in AX, OUT: correct registers for int 13h ; cl := start sector ; ch := track (or cylinder) ; dh := head (or side) ; dl := drive or boot device ; ah := 2 (Int 13h read sectors from disk function) ; al := how many sectors to read ; es:bx := destination address in memory ; ;push bx ;push ax ; ax has how many sectors to read mov bx, ax ; Save logical sector mov dx, 0 div word [sectorspertrack.d] ; First the sector add dl, 01h ; Physical sectors start at 1 mov cl, dl ; Sectors belong in CL for int 13h mov ax, bx mov dx, 0 ; Now calculate the head div word [sectorspertrack.d] mov dx, 0 div word [sides.d] mov dh, dl ; Head/side mov ch, al ; Track ;pop ax ;pop bx mov dl, byte [drive.d] ; Set correct device mov bx, fs ; es:bx is destination address in ram for read mov ax, gs ; ax (al) := how many sectors to read mov ah, 2 ; int 13H function ah:2 "Load disk data to ES:BX" int 13h ; Read! jc .readerror ; ERROR => Try again ;*** restore es since rstack points with it!! mov es, [saven.es] mov dx, 1 ; return true flag for success jmp exec.nextopcode .readerror: ;*** restore es since rstack points with it!! mov es, [saven.es] mov dx, 0 ; return false flag for read error jmp exec.nextopcode saven.es: dw 0 ; this has the potential to 'brick' a computer so we need to ; get it right before doing it. write.doc: ; db ' ( first-sector n addr -- flag=T/F/2 ) ; db ' write sectors to disk ' ; dw $-write.doc write: db DISK ; domain name prefix dw read.x db 'write', 5 write.x: ; to do: test, very, very, carefully ; 1st-sect n address mov fs, dx ; source memory address in this segment .reset: ; Reset the virtual floppy drive (usb) mov ax, 0 ; mov dl, byte [drive.d] ;boot drive number (eg for usb 128) int 13h ; jc .reset ; read error => reset again .testread: ; below is a sanity check read to make sure we are going ; to write to the correct disk. ; cl := (1) start sector, in this case boot sector ; ch := (0) track (or cylinder) number ; dh := (0) head (or side) ; dl := drive or boot device 0-3=diskette; 80H-81H=hard disk ; ah := 3 (Int 13h write sectors to disk function) ; al := (1) how many sectors to read ; es:bx := where to read data. In this case a testbuffer ; which is 'first' + 512 bytes mov [saven.es], es mov ax, ds ; at the moment 64K segment mov es, ax ; es:bx determines where data is read from mov dh, 0 ; first side mov dl, byte [drive.d] ; Set correct device mov bx, fs ; es:bx is destination address in ram for read add bx, 1024 ; read to next block mov ch, 0 ; first track/cylinder mov cl, 1 ; read boot sector (1) mov ax, 1 ; ax (al) := how many sectors to read mov ah, 2 ; int 13H function ah:2 "read data from disk from ES:BX" int 13h ; read it! jc .readerror ; if error should try again, but we dont... mov dx, word [es:bx+8] ; address of magic number ; if the magic number (pi*10000) is not there then this ; maybe the wrong disk, so dont write to it! cmp dx, 31415 jne .notpi ;mov es, [saven.es] ;jmp exec.nextopcode .write: ;*** need to save es since rstack points with it!! ;mov [saven.es], es mov ax, ds ; at the moment 64K segment mov es, ax ; es:bx determines where data is written from pop gs ; how many sectors ( will go in AL) pop ax ; logical start sector .logical: ; Will there be issues with multi-track write? probably ; So sector 1 is the boot sector on disk. ; Calculate track (cylinder), head and sector (chs) settings ; for the int 13h (ah=3) write to disk function ; IN: logical sector in AX, OUT: correct registers for int 13h, ah=3 ; cl := start sector (1-n) ; ch := track (or cylinder) number (0-n) ; dh := head (or side) ; dl := drive or boot device 0-3=diskette; 80H-81H=hard disk ; ah := 3 (Int 13h write sectors to disk function) ; al := how many sectors to read ; es:bx := source address in memory of data to write ; ; ax has how many sectors to write mov bx, ax ; Save logical sector mov dx, 0 div word [sectorspertrack.d] ; First the sector add dl, 01h ; Physical sectors start at 1 mov cl, dl ; Sectors belong in CL for int 13h mov ax, bx mov dx, 0 ; Now calculate the head div word [sectorspertrack.d] mov dx, 0 div word [sides.d] mov dh, dl ; Head/side mov ch, al ; Track ; setting the drive correctly is pretty ; important, see sanity check above mov dl, byte [drive.d] ; Set correct device mov bx, fs ; es:bx is destination address in ram for write mov ax, gs ; ax (al) := how many sectors to read mov ah, 3 ; int 13H function ah:3 "write data to disk from ES:BX" int 13h ; Write it! jc .writeerror ; if error should try again, but we dont... ;*** restore es since rstack points with it!! mov es, [saven.es] mov dx, 1 ; return true flag for success jmp exec.nextopcode .readerror: .writeerror: ;*** restore es since rstack points with it!! mov es, [saven.es] mov dx, 0 ; return false flag for read error jmp exec.nextopcode .notpi: mov es, [saven.es] mov dx, -1 ; notpi flag jmp exec.nextopcode ; This opcode and vid.x are obviously not going to be ; available on all hardware. So we need to think about ; how to configure plugable opcodes. Eg: what if a device ; has a gyroscope. We want opcodes to read from that gyroscope ; fg.doc: ; db ' ( n -- ) ' ; db ' set foreground colour for emit' ; dw $-fg.doc fg: db MON ; domain name prefix dw write.x db 'fg', 2 fg.x: mov [fg.d], dl ; pop dx jmp exec.nextopcode fg.d: db 5 ; colour cyan ; not working bg.doc: ; db ' ( n -- ) ' ; db ' set the background colour for emit' ; dw $-bg.doc bg: db MON ; domain name prefix dw fg.x db 'bg', 2 bg.x: mov [bg.d], dl ; pop dx jmp exec.nextopcode ;ret bg.d: db 5 ; cls.doc: ; db ' ( -- ) ' ; db ' clear screen' ; dw $-cls.doc cls: db MON ; domain name prefix dw bg.x db 'cls', 3 cls.x: push dx mov ah, 06h ; ah=function number for int10 (06) mov al, 00h ; al=number of lines to scroll (00=clear screen) ;mov bx, 700h ; bh=color attribute for new lines (grey) mov bx, 00h ; bh=color attribute for new lines xor cx, cx ; ch=upper left hand line number of window (dec) ; cl=upper left hand column number of window (dec) mov dx, 184fh ; dh=low right hand line number of window (dec) ; dl=low right hand column number of window (dec) int 10h pop dx jmp exec.nextopcode vid.doc: ; db ' ( n -- ) ' ; db ' set video mode to n' ; db ' on x86 try 13h mode' ; dw $-vid.doc vid: db MON ; domain name prefix dw cls.x db 'vid', 3 vid.x: mov ax, dx ; get video mode mov ah, 0 ; ah=0 set video mode function, al=mode int 10h pop dx jmp exec.nextopcode pix.doc: ; db ' ( x y -- ) ' ; db ' display one pixel at position xy' ; dw $-pix.doc pix: db MON ; domain name prefix dw vid.x db 'pix', 3 pix.x: ; using interrupts to draw pixels is very slow but ok ; for playing around pop cx ; get x-coordinate ; y-coordinate in dx mov al, 15 ; white mov ah, 0ch ; put pixel int 10h ; draw pixel pop dx jmp exec.nextopcode glyph.doc: ; db ' ( a -- ) ' ; db ' display a colour glyph at pixel position xy' ; dw $-glyph.doc glyph: db MON ; domain name prefix dw pix.x db 'glyph', 5 glyph.x: jmp exec.nextopcode ; this may be useful for timing code ; clock updates at 1193180/65536 (about 18.2) ticks per second. ; counts per second 18 ; counts per minute 1092 ; counts per hour 65543 ; counts per day 1573040 ; clock incremented approx every 55ms clock.doc: ; db ' ( -- D ) ' ; db ' number of clock ticks since midnight ' ; dw $-clock.doc clock: db TIME ; domain name prefix dw glyph.x db 'clock', 5 clock.x: push dx mov ah, 00h ; interrupt to get clock ticks since midnight int 1Ah ; cx:dx now holds number of clock ticks since midnight push dx ; low byte on lower stack position mov dx, cx ; high byte on higher stack position jmp exec.nextopcode ; There are "issues" here. It is not always possible to ; set format for rtc time, so need to check format from status ; register B and convert if necessary. Also, should check for ; 2 values the same in a loop, so overcome updating problems rtc.doc: ; db ' ( -- secs mins hours days months years ) ' ; db 'return 6 values on stack representing real time and date. ' ; db 'called time&date in standard forths. ' ; dw $-rtc.doc rtc: db TIME ; domain name prefix dw clock.x db 'rtc', 3 rtc.x: push dx ; save top of stack xor ax, ax ; set ah, al == 0 ; status register B controls format of rtc values but ; cant always be set cli mov al, 0x0B ; try to set date format out 0x70, al ; address reg ; set bit 1=24hour, bit 2=binary/bcd mov al, 6 ; set 2 low bits on date register B out 0x71, al ; set register sti .updating: ; not used at the moment mov al, 0x0A ; check if an update in progress out 0x70, al ; address reg in al, 0x71 ; get data from cmos data reg test al, 0x80 ; is high bit set? ; in theory we are not supposed to read the real time clock ; data if an update is in progress, but we wont worry at the ; moment about this cli ; disable interrupts mov al, 0x00 ; select seconds out 0x70, al ; address reg in al, 0x71 ; get seconds data from data reg sti push ax cli mov al, 0x02 ; select minutes out 0x70, al ; address rtc minute register in al, 0x71 ; get data from cmos data reg sti push ax cli mov al, 0x04 ; select Hour out 0x70, al ; address rtc minute register in al, 0x71 ; get hour data from cmos data reg sti push ax mov al, 0x07 ; Day of month out 0x70, al ; cmos select reg in al, 0x71 ; cmos data reg push ax mov al, 0x08 ; Month out 0x70, al ; cmos select reg in al, 0x71 ; cmos data reg push ax mov al, 0x09 ; Year out 0x70, al ; cmos select reg in al, 0x71 ; cmos data reg mov dx, ax ; tos=dx jmp exec.nextopcode noop.doc: ; db 'Does nothing. For some reason most machines ' ; db 'include this instruction. Also it is a good ' ; db 'end marker for the opcodes ' ; dw $-noop.doc noop: db LANG ; domain name prefix dw rtc.x db 'nop', 3 noop.x: jmp exec.nextopcode ; ******************************* ; end of opcodes ; ******************************* ; ************ ; some immediate words ; ************ hello.doc: ; db ' ( -- ) ; db ' Just testing immediate procs ' hello: db LANG ; domain name prefix for universal naming dw noop.x db 'hello', IMMEDIATE | 5 hello.p: db LIT, '!', EMIT db LIT, 'h', EMIT db LIT, 'i', EMIT db EXIT literal.doc: ; db ' (comp: n -- )(run: -- n) ; db ' Push the TOS onto the data stack at run-time ' ; db ' This advances compile point by 3 bytes ' literal: db LANG ; domain name prefix for universal naming dw noop.x db 'literal', IMMEDIATE | 7 literal.p: db LIT, LITW ; n op=LITW db FCALL dw ccomma.p ; n /compile LITW opcode at HERE db FCALL ; tos is 2bytes so use , not c, dw comma.p ; tos will be pushed onto stack at runtime db EXIT ; create hangs on no input, but wparse doesnt. ; important word!! probably used by all defining words ; including colon : and VAR & CONSTANT ; create is not immediate funnily enough but colon is. ; there is a bug with defining words. actually we need a (create) ; which is this code below, and then an immediate create which ; sets a defining bit. create.doc: ; db ' ( -- ) ; db ' create a new word header in the dictionary ' ; db ' CREATE also compiles code to push the address of ' ; db ' the parameter field onto the stack at runtime. ' ; db ' As often pointed out, CREATE does not allocate any space ' ; db ' for the parameter field. The coder can do that with ALLOT or ' ; db ' C, or , etc .' ; db ' In this implementation, the start of the parameter field ' ; db ' is just the next available byte in the dictionary. ' ; db ' ** A standard definition of VARIABLE (VAR) is: ' ; db ' : CREATE 0 , ; immediate ' create: db LANG ; domain name prefix linked to domain table dw literal.p db 'create', 6 create.p: ; here>code does lots of important stuff, see above. ; such as executing the anon buffer db FCALL dw heretocode.p ; now compile the name to the dictionary. Use code in colon ; to see how. ; before any new words have been compiled with colon : ; then "last" points to "last" ! After new words have been ; added then "last" will point to the last word added db FCALL dw last.p ; /get pointer to last word in dictionary db FETCH ;*** insert link to previous last word in dict at HERE db FCALL dw comma.p ;*** get the name & length of the new word db FCALL dw wparse.p ; a n ;*** if wparse returns 0 then no name, only whitespace db DUP ; a n n db JUMPZ, .noname-$ ; a n db DUP, RON ; a n r: n ; compile string to current compile position db FCALL dw scompile.p ; r: n db ROFF ; n ; compile count|control byte after name db FCALL dw ccomma.p ; set last pointer to new word execution address db FCALL dw here.p ; adr /xt for new word db FCALL dw last.p ; adr last /pointer to last word db STORE ; now compile some code that puts the parameter field address ; on the stack db FCALL dw here.p ; adr / address of next byte in dict (current xt) db LIT, 4, PLUS ; adr+3 / adjust for LITW, , EXIT == 4 bytes db LIT, LITW ; adr+3 op=LITW db FCALL dw ccomma.p ; adr+3 / compile LITW opcode db FCALL ; address is 2bytes so use , not c, dw comma.p ; /compile address of next byte in dictionary db LIT, EXIT ; op=EXIT db FCALL dw ccomma.p ; / compile EXIT opcode db FCALL dw codetohere.p ; set >code to here db EXIT .noname: db DROP, DROP ; clear data stack db LIT, '?', EMIT ; db LIT, '?', EMIT ; db EXIT callcomma.doc: ; db ' ( xt -- ) ; db ' compile an fcall to xt ' callcomma: dw create.p db 'call,', IMMEDIATE | 5 callcomma.p: db LIT, FCALL ; xt op=fcall db FCALL dw ccomma.p ; xt db FCALL dw comma.p db EXIT ; This can be tested with : d drop ; : var go ; find d (does>)' ; should reset the code field of "go" to behaviour of "d" ; all defining words need to be immediate, so 'create' could ; just do that automatically rundoes.doc: ; db ' ( -- ) ; db ' perform run-time behaviour of does> ' rundoes: dw callcomma.p db '(does>)', 7 rundoes.p: ; xt /xt -> code after does> in defining word ;db EXIT ; when (does>) runs then 'here' should be pointing just after ; the parameter field of the new word. That is, just after any ; data space that has been allotted with 'allot' or ',' etc db FCALL dw here.p ; xt here db DUP ; xt H H db FCALL dw last.p ; xt H H L* db FETCH ; xt H H last db DUP ; xt H H L L ; set new compile point to last xt (word just created) db FCALL dw ishere.p ; xt H H L ; compile jump opcode db LIT, JUMP ; xt H H L op=jump db FCALL dw ccomma.p ; xt H H L ; calculate jump offset. The jump must jump over the ; parameter field of the defined word (this field is ; allocated with 'allot' or , or c, etc) db MINUS ; xt H (H-L) ; compile offset db FCALL dw ccomma.p ; xt H ; reset "here" to end of new word (after parameter field) db FCALL dw ishere.p ; xt ; compile code to push parameter field of new word onto stack ; at the run-time of the new word its param field goes onto ; the stack, so that the code after does> can use it. db FCALL dw last.p ; xt last* db FETCH ; xt last db LIT, 4, PLUS ; xt last+4 db FCALL dw literal.p ; xt ; compile call to xt (code just after does> in defining word) db FCALL dw callcomma.p ; ; the code below is just ; semicolon ; could write ; db FCALL ; dw semicolon.p ; compile an exit for the new word being defined db LIT, EXIT ; op=exit db FCALL dw ccomma.p ; update the dictionary compile point and reset compile ; point to anon buffer (these are tasks that semicolon normally ; does) db FCALL dw codetohere.p ; do code>here db FCALL dw heretoanon.p ; db EXIT ; there are 3 "times" here. Compile-time for the defining word ; run-time for defining word (which is also compile-time for ; the defined word). And run-time for the defined word. ; These three "times" can make thinking about defining words ; tricky does.doc: ; db ' ( -- ) ; does: dw rundoes.p db 'does>', IMMEDIATE | 5 does.p: ; compile current 'here' (compile point) so that it will ; be pushed onto stack at run-time of defining word ; (which is compile-time of defined word) db FCALL dw here.p ; H ; adjust the literal to account for the code following ; which will be compiled db LIT, 7, PLUS ; H+7 db FCALL dw literal.p ; ; compile "fcall (does>)" ; but this is the same as call, db LIT, FCALL ; H op=fcall db FCALL dw ccomma.p ; H db LITW dw rundoes.p ; xt db FCALL dw comma.p ; compile an exit (for does> at run-time) db LIT, EXIT ; op=exit db FCALL dw ccomma.p db EXIT ; things to do in CREATE (called by : COLON) ; Set HERE to next code space with HERE>CODE ; create a header. first link back using LAST ; and set LAST to new word ; then compile name, with S, ; compile count, then just exit and let IN, handle the rest ; (which is actually calling : COLON anyway). ; In semicolon: ; when ; compile "exit" set >code to here ; colon.doc: ; db ' ( -- ) ; db ' Creates a new word in the dictionary.' ; db ' The only difference with the CREATE word is that ' ; db ' no code is appended after the word name and count ' colon: dw does.p db ':', IMMEDIATE | 1 colon.p: ; HERE>CODE in CREATE will execute stuff in the ANON buffer ; (as with all defining words which use create.) db FCALL dw create.p ; HERE and >CODE @ should both be pointing just after ; code to push parameter field onto stack, so we need to ; repoint these back to the xt (just after the name count byte) db FCALL dw tocode.p ; >code db DUP ; >code >code db FETCH, LIT, 4 ; >code code* 4 db MINUS ; >code code*-4 db SWAP ; code*-4 >code db STORE ; db FCALL dw heretocode.p ; set here to point to >code db EXIT ;.noname: ; db DROP, DROP ; clear data stack ; db LIT, '?', EMIT ; ; db LIT, '?', EMIT ; ; db EXIT semicolon.doc: ; db ' ( -- ) semicolon: dw colon.p db ';', IMMEDIATE | 1 semicolon.p: ; compile an exit ; set >code to here ; set ishere to anon buffer db LIT, EXIT db FCALL dw ccomma.p ; compile opcode exit db FCALL dw codetohere.p ; do code>here db FCALL dw heretoanon.p ; compile all to anon, not dictionary db EXIT begin.doc: ; db ' ( -- ) ; db ' marks a jump back address for until/again etc' ; db ' the jump address is left on the data stack ' ; db ' forth- : begin here ; immediate ' begin: dw semicolon.p db 'begin', IMMEDIATE | 5 begin.p: db FCALL dw here.p ; ad /current compilation adr db EXIT again.doc: ; db ' ( comp: jb -- ) ( run: -- ) ; db ' jumps back to begin at runtime. At compile time ' ; db ' gets the jump-back address from the data stack an compiles ' ; db ' it in. ' again: dw begin.p db 'again', IMMEDIATE | 5 again.p: db LIT, JUMP db FCALL dw ccomma.p ; compile to current position (here) ;db LIT, 2 ; op 2 ;db FCALL ;dw itemcompile.p ; compile to current position (here) db FCALL dw here.p ; jb here db DECR ; jb here-1 /align to jump db MINUS ; jb-here-1 db FCALL dw ccomma.p ; compile to current position (here) db EXIT until.doc: ; db ' ( n -- ) ; db ' at run-time: jumps back to begin if n is true' ; db ' at compile-time: get begin address from data ' ; db ' stack and compiles a conditional relative jump ' ; db ' back to begin. ' until: dw again.p db 'until', IMMEDIATE | 5 until.p: db LIT, JUMPF db LIT, 2 ; jb op 2 db FCALL dw itemcompile.p ; compile to current position (here) db FCALL dw here.p ; jb here db DECR ; jb here-1 /align to jump db MINUS ; jb-here-1 db FCALL dw ccomma.p ; compile to current position (here) db EXIT do.doc: ; db ' (run: limit start -- r: start limit ) ; db ' (comp: -- 'here' ) ; db ' At run-time: puts start and limit on return stack ' ; db ' at compile time: marks a jump back address for loop etc' ; db ' the jump address is left on the data stack ' do: dw until.p db 'do', IMMEDIATE | 2 do.p: ; compile >r >r to get loop parameters on the ; return stack. db LIT, RON db FCALL dw ccomma.p db LIT, RON db FCALL dw ccomma.p ; mark code compile point to be used by loop to jump back db FCALL dw here.p ; ad /current compilation adr db EXIT loop.doc: ; db ' (comp: 'here' -- ) (run: -- ) ; db ' at run-time: takes loop parameters off stack ' ; db ' increments the iterator, checks if it is = to limit ' ; db ' and jumps back to DO if not. If = then it removes loop ' ; db ' parameters from the stack and discards them. ' ; db ' at compile-time: get begin address from data ' ; db ' stack and compiles a rloop jump ' ; db ' back to begin (address on data stack). ' loop: dw do.p db 'loop', IMMEDIATE | 4 loop.p: ; jump-back /left by DO on stack. ; could be better to write (loop) to call at run-time ; compile r> r> to get loop parameters off the stack db LIT, ROFF ; jb op db FCALL dw ccomma.p ; jb ; counter db LIT, ROFF ; jb op db FCALL dw ccomma.p ; jb ; increment the counter by 1 db LIT, INCR ; jb op db FCALL dw ccomma.p ; jb ; 2dup to save rstack values db LIT, TWODUP ; jb op db FCALL dw ccomma.p ; jb ; put them back on rstack db LIT, RON ; jb op db FCALL dw ccomma.p ; jb ; db LIT, RON ; jb op db FCALL dw ccomma.p ; jb ; check if counter==limit db LIT, EQUALS ; jb op db FCALL dw ccomma.p ; jb ; db LIT, JUMPF ; jb op=jumpt db FCALL dw ccomma.p ; compile opcode to current position (here) db FCALL dw here.p ; jb here db DECR ; jb here-1 /align to jump db MINUS ; jb-here-1 db FCALL dw ccomma.p ; compile to current position (here) ; now get rid of the loop parameters db LIT, ROFF ; op db FCALL dw ccomma.p ; db LIT, DROP ; op db FCALL dw ccomma.p ; db LIT, ROFF ; op db FCALL dw ccomma.p ; db LIT, DROP ; op db FCALL dw ccomma.p ; db EXIT if.doc: ; db ' run-time: ( n -- ) ' ; db ' compile-time: ( -- adr ) ' ; db ' if the value n on the stack is zero ' ; db ' skip statements after this until next "fi" ' ; db ' compiles a jumpzero and puts the current ' ; db ' address on the data stack. The "fi" command consumes ' ; db ' that address ' if: dw loop.p db 'if', IMMEDIATE | 2 if.p: db LIT, JUMPF ; op db FCALL dw ccomma.p ; compile to current position (here) db FCALL dw here.p ; ad /current compilation adr ; we compile a 2 byte jump which will do nothing. ; eg JUMPF, 2 which just goes to the next instruction. ; but this will be replaced by the real target when "fi" ; compiles db LIT, 2 ; compile temporary jump target (2) db FCALL dw ccomma.p ; db EXIT else.doc: ; db ' compile.time: ( ad -- jad ) ; db ' at compile time, get the "if" jump address ' ; db ' from the data stack and completes the if jump ' ; db ' using offset from here. Then it leaves the ' ; db ' else jump address on stack for "fi" to deal with ' else: dw if.p db 'else', IMMEDIATE | 4 else.p: ; ja db LIT, JUMP ; ja op db FCALL dw ccomma.p ; compile opcode to current position (here) db FCALL dw here.p ; ja ad /current compilation adr db LIT, 2 ; compile temporary jump target 2 db FCALL dw ccomma.p ; db SWAP ; ad ja db DUP ; ad ja ja db FCALL dw here.p ; ad ja ja target db SWAP ; ad ja target ja db MINUS ; ad ja t-ja db INCR ; .... adjust jump target db SWAP ; ad n ja db CSTORE ; ad db EXIT ; better to use the data stack for compile time ; behaviors fi.doc: ; db ' run-time: ( -- ) ; db ' compile-time: ( ad -- ) ' ; db ' at compile time obtains the correct jump ' ; db ' address from the data stack and compiles the correct ' ; db ' target address into a previously compile "if" clause. ' ; db ' This word is called "then" in traditional forths. ' fi: dw else.p db 'fi', IMMEDIATE | 2 fi.p: ;*** we use the data stack for compile time calculations ; "ja" means jump address db DUP ; ja ja db FCALL dw here.p ; ja ja target db SWAP ; ja target ja db MINUS ; ja t-ja db INCR ; .... adjust jump target db SWAP ; a1 a2 n ja db CSTORE ; a1 a2 db EXIT ; this can be written as source but is useful for debugging ; bytecode words and so will be left. dotstack.doc: ; db ' ( -- ) ; db ' display the items on the data stack without ' ; db ' altering it. The top (or most recent) item ' ; db ' is printed rightmost ' dotstack: dw fi.p db '.s', 2 dotstack.p: ; below we need a copy of the stack depth ; because it gets decremented by the rloop ; need to sieve stack items onto rstack ; with >r, swap, r>, swap, >r etc db DEPTH, DUP ; n n db JUMPNZ, 4 ; n db DROP db EXIT ;*** put all items on return stack db DUP ; n n db RON ; n r: n db SWAP, ROFF ; n a n-x db SWAP, RON ; n n-x r: a db RON ; n r: a n-x db RLOOP, -5 ; n r: a n-x-1 db ROFF ; n 0 r: a b c ... db DROP ; n r: a b c ... ;*** print all stack items db RON ; r: a b c ... n db ROFF ; n r: a b c ... db ROFF ; n c r: a b db DUP ; n c c r: a b db FCALL dw udot.p ; n c r: a b db LIT, ' ', EMIT ; n c r: a b db SWAP, RON ; c r: a b n db RLOOP, -11 ; c r: a b n-1 db ROFF ; a b c ... 0 db DROP ; a b c ... db EXIT dotrstack.doc: ; db ' ( -- ) ; db ' display the top 2 items on the return stack ' dotrstack: dw dotstack.p db '.r2', 3 dotrstack.p: db LIT, ' ', EMIT db ROFF, ROFF ; n m db DUP ; n m m db FCALL dw udot.p ; n m db LIT, ' ', EMIT db RON ; n db DUP ; n n db FCALL dw udot.p ; n db RON db EXIT rcount.doc: ; db ' ( adr -- adr-n n ) ' ; db ' count a reverse counted string, ignoring control bit(s). ' ; db ' Given a pointer to the count byte of a ' ; db ' reverse counted string, return the address of the 1st byte ' ; db ' of the string and its length. This proceedure ' ; db ' may also handle the anding out of the immediate ' ; db ' control bit which is stored in the msb of the ' ; db ' count for execution tokens ' ; dw $-rcount.doc rcount: dw dotrstack.p db 'rcount', 6 rcount.p: ; adr db DUP, CFETCH ; a n / get the count db LIT, 0b00011111 ; a n mask db LOGAND ; a n&mask db DUP ; a n n db RON, MINUS ; adr-n db ROFF ; adr-n n db EXIT def.doc: ; db ' ( xt -- ) ' ; db ' Set the "defining" control bit in the count byte ' ; dw $-def.doc def: dw rcount.p db 'def', 3 def.p: ; adr db DECR ; adr-1 db DUP, CFETCH ; a n / get the count ; the def bit is the second bit in count byte db LIT, 0b01000000 ; a n mask db LOGOR ; a n.v.mask db SWAP ; m a db CSTORE ; db EXIT dotxt.doc: ; db ' ( adr - ) ' ; db ' given a valid execution token on the stack ' ; db ' the name of the procedure' ; db ' The definition might be ; db ' : .xt -1 rcount type ; ' ; dw $-dotxt.doc dotxt: dw def.p db '.xt', 3 dotxt.p: ; xt db DECR ; adr-1 db FCALL dw rcount.p db FCALL dw type.p db EXIT nextxt.doc: ; db ' ( xt -- XT ) ' ; db ' get next execution address in the dictionary ' nextxt: dw dotxt.p db 'xt+', 3 nextxt.p: ; xt db DECR ; xt-1 /points to count|control byte db FCALL dw rcount.p ; adr n /start of name db DROP ; adr db DECR, DECR ; adr-2 db FETCH ; XT / get word pointer db EXIT opcode.doc: ; db ' ( adr -- n ) ' ; db ' Given the address of an execution token ' ; db ' or procedure on the stack provides the numeric ' ; db ' opcode for that procedure or else 0 for ' ; db ' an address which does not correspond to a bytecode.' ; db ' This is used to compile text to bytecode ' ; db ' if not an opcode, then compile FCALL etc ' opcode: dw nextxt.p db 'opcode', 6 opcode.p: ; adr db DUP ; adr adr db LITW ; dw op.table ; a a T db FETCHPLUS ; a a T+2 [T] db SWAP, RON ; a a [T] r: T+2 db EQUALS ; a flag r: T+2 db JUMPT, 19 ; a r: T+2 db DUP, ROFF ; a a T+2 ;**** check if end of table db DUP ; a a T+2 T+2 db FETCH ; a a T+2 [T+2] db LIT, -1 ; a a T+2 [T+2] -1 db EQUALS ; a a T+2 flag db JUMPF, .moreops-$ ; a a T+2 db DROP, DROP, DROP ; db LIT, 0 ; 0 / return false if not opcode db EXIT ; .moreops: db JUMP, -21 ; a a T+2 ;** opcode found, calculate offset db DROP, ROFF ; T+2 db DECR, DECR ; T db LITW dw op.table db MINUS ; T - optable db DIVTWO ; opcode db EXIT dotcode.doc: ; db ' ( opcode - ) ' ; db ' given a valid opcode on the stack, print ' ; db ' the textual version of the opcode. ' ; dw $-dotcode.doc dotcode: dw opcode.p db '.code', 5 dotcode.p: ; op ;*** check for invalid code, nop always last db DUP ; op op db LIT, NOOP, INCR ; op op nop+1 db ULESSTHAN ; op flag db JUMPT, 12 ; op ;*** invalid code db FCALL dw udot.p db LIT, '?', EMIT db LIT, '?', EMIT db EXIT ;*** valid opcode db DUP, PLUS ; op*2 db LITW dw op.table ; op*2 op.table db PLUS ; op*2+op.table db FETCH ; [op*2+op.table] / get execution adr db DECR ; a-1 db FCALL dw rcount.p ; adr n ;*** does the same as rcount ;db DUP, CFETCH ; adr n / get the count ;db DUP ; adr n n ;db RON, MINUS ; adr-n ;db ROFF ; adr-n n db FCALL dw type.p db EXIT udot.doc: ; db ' ( n -- ) ' ; db ' display top stack element as unsigned ' ; db ' number in the current base. ' ; dw $-udot.doc udot: dw dotcode.p db 'u.', 2 udot.p: db LIT, 0, RON ; n r: 0 db LITW dw base.d ; n adr r: 0 db CFETCH ; n base db DIVMOD ; rem quotient db ROFF, INCR, RON ; rem quotient r: 0+1 db DUP, JUMPNZ, -9 ; rem rem rem ... 0 db DROP ; rem rem ... r: x db LITW ; use digit lookup table dw digits.d ; r r r ... adr r: x db PLUS ; r r ... adr+r r: x db CFETCH ; r r ... d r: x db EMIT ; rem ... print asci digit db RLOOP, -6 ; rem ... r: x-1 db ROFF, DROP ; clear rstack db EXIT ddot.doc: ; db ' ( n -- ) ' ; db ' display top stack element as unsigned ' ; db ' number in decimal ' ; dw $-ddot.doc ddot: dw udot.p db 'd.', 2 ddot.p: ; n db LIT, 0, RON ; n r: 0 db LIT, 10 ; n 10 r: 0 db DIVMOD ; rem quotient db ROFF, INCR, RON ; rem quotient r: 0+1 db DUP, JUMPNZ, -7 ; rem rem rem ... 0 r: x db DROP ; rem rem ... r: x db LIT, '0', PLUS, EMIT ; rem ... print remainder db RLOOP, -4 ; rem ... r: x-1 db ROFF, DROP ; clear rstack db EXIT dothex.doc: ; db ' ( n -- ) ' ; db ' display top stack element as an unsigned ' ; db ' number in hexadecimal ' ; dw $-dothex.doc dothex: dw ddot.p db '.hex', 4 dothex.p: ; n db LITW dw base.d ; n adr db CFETCH ; n base db SWAP ; base n /save original base db LIT, 16 ; base n 16 db LITW dw base.d ; base n 16 adr db CSTORE ; base n /set new base to 16 db FCALL dw udot.p ; base /print number db LITW dw base.d ; base adr db CSTORE ; /restore original base db EXIT dot.doc: ; db ' ( n -- ) ' ; db ' display top stack element as a signed number ' ; db ' in the current base (if u. does so) ' ; dw $-dot.doc dot: dw dothex.p db '.', 1 dot.p: ; n db DUP ; n n db LIT, 0 ; n n 0 db LESSTHAN ; n flag / n<0 ? db JUMPF, 6 ; n db NEGATE ; -n db LIT, '-', EMIT ; -n /print negative sign db FCALL dw udot.p db EXIT cdot.doc: ; db ' ( n -- ) ' ; db ' display top stack element as a signed 8 bit ' ; db ' number in the current base (if u. does so) ' ; db ' This is useful for displaying relative jumps ' ; db ' which are 1 byte signed numbers. ' ; db ' eg: 255 = -1, 254 = -2, 128 = -127 ; dw $-cdot.doc cdot: dw dot.p db 'c.', 2 cdot.p: ; n db DUP ; n n db LITW dw 128 ; n n 128 db LESSTHAN ; n flag / n < 128 db JUMPT, 6 ; n db LITW dw 256 db MINUS ; n-256 db FCALL dw dot.p db EXIT todigit.doc: ; db ' ( c -- n flag ) ' ; db ' converts the ascii character of a digit ' ; db ' on the stack to its corresponding integer ' ; db ' using the base (1digit', 6 todigit.p: ; c db DUP ; c c db LITW dw digits.d ; c c adr db LITW dw base.d ; c c adr adr db CFETCH ; c c a n db RON ; c c a r: n db CFETCHPLUS ; c c a+1 C r: n db SWAP, RON ; c c C r: n a+1 db EQUALS ; c flag r: n a+1 db JUMPT, 10 ; c r: n a+1 db DUP ; c c r: n a+1 db ROFF ; c c a+1 r: n db RLOOP, -8 ; c c a+1 r: n-1 ;*** not a valid digit db DROP, DROP ; c r: 0 db ROFF ; c 0 db EXIT ;*** valid asci digit, convert to number db DROP ; r: n-x a+1 db ROFF, DROP ; r: n-x db ROFF ; n-x db LITW dw base.d ; n-x adr db CFETCH ; n-x n db SWAP ; n n-x db MINUS ; x db LIT, -1 ; x -1 / -1 is true flag db EXIT digits.d: db '0123456789abcdefghijklmnopqrstuvwxyz' ; base is 16bits in all forths. so we can do: 10 base ! base.doc: ; db ' ( -- adr ) ' ; db ' puts on the stack the address of the current ' ; db ' numeric base ' ; db ' which is used for displaying and parsing ' ; db ' numbers. The base should be 1 < base < 37 ' ; db ' since these are the digits which can be ' ; db ' displayed using numerals and letters ' ; db ' eg: base c@ . ' ; db ' displays the current base ' ; dw $-base.doc base: dw todigit.p db 'base', 4 base.p: db LITW dw base.d db EXIT base.d: db 10 bin.doc: ; db ' ( -- ) ' ; db ' sets the numeric base to binary ' ; dw $-bin.doc bin: dw base.p db 'bin', 3 bin.p: db LIT, 2 db LITW dw base.d db CSTORE db EXIT hex.doc: ; db ' ( -- ) ' ; db ' sets the numeric base to hexadecimal ' ; dw $-hex.doc hex: dw bin.p db 'hex', 3 hex.p: db LIT, 16 db LITW dw base.d db CSTORE db EXIT deci.doc: ; db ' ( -- ) ' ; db ' sets the numeric base to 10 ' ; dw $-deci.doc deci: dw hex.p db 'deci', 4 deci.p: db LIT, 10 db LITW dw base.d db CSTORE db EXIT tonumber.doc: ; db ' ( adr n -- adr/n flag ) ' ; db ' Given a pointer to string adr with length "n" ' ; db ' attempt to convert the string to ' ; db ' a number. If successful put number and true flag' ; db ' on the stack, if not put pointer to incorrect digit' ; dw $-tonumber.doc tonumber: dw deci.p db '>number', 7 tonumber.p: ; a n db LIT, 0, RON ; a n r: 0 /false neg flag db RON ; a r: 0 n ;*** check for +/- at first char db DUP, CFETCH ; a c r: 0 n db LIT, '+' ; a c '+' ... db EQUALS ; a flag ... db JUMPF, 8 ; a ... db ROFF, DECR, RON ; a r: n-1 db INCR ; a+1 r: 0 n-1 db JUMP, 16 ; a+1 r: 0 n-1 db DUP, CFETCH ; a c r: 0 n db LIT, '-' ; a c '-' r: 0 n db EQUALS ; a flag r: 0 n db JUMPF, 9 ; a r: 0 n ;*** set a +/- flag on the rstack db ROFF, DECR ; a n-1 r: 0 db ROFF, DECR ; a n-1 -1 db RON, RON ; a r: -1 n-1 db INCR ; a+1 r: -1 n-1 ;*** exit if zero length string or just +/- db ROFF ; a n r: -1 db DUP ; a n n r: -1 db JUMPNZ, 5 ; a n r: -1 db ROFF, DROP ; a 0 db EXIT db RON ; a r: -1 n db LIT, 0 ; a 0 r: n /initial sum db SWAP ; 0 a r: n db CFETCHPLUS ; 0 a+1 d r: n ;*** check if digit db FCALL dw todigit.p ; 0 a+1 D flag r: n db JUMPF, 24 ; 0 a+1 D r: n ;*** db RON ; 0 a+1 r: n 0-9 db SWAP ; a+1 s r: n digit /s is sum db LITW dw base.d db CFETCH ; a+1 s base r: n digit db UMULT ; a+1 s*base r: n digit db ROFF ; a+1 s*base digit r: n db PLUS ; a+1 s r: -flag n db SWAP ; s a+1 r: -flag n db RLOOP, -16 ; s a+1 r: -flag n-1 /back to c@+ db ROFF, DROP ; s a+1 r: -flag db DROP ; s r: -flag db ROFF ; s -flag db JUMPF, 3 ; s / skip if + db NEGATE ; -s / negate if flag set db LIT, -1 ; s -1 /value and true flag db EXIT ;*** non digit ; sum a+1 d db DROP, SWAP ; a+1 sum r: 0 n db DROP ; a+1 r: 0 n db LIT, 0 ; a+1 0 db ROFF, DROP ; clear return stack db ROFF, DROP ; clear return stack db EXIT toword.doc: ; db ' ( -- adr n ) ' ; db ' put on the stack a pointer to the current ' ; db ' word and its length. ' toword: dw tonumber.p db '>word', 5 toword.p: ; handle zero case (no word found) db LITW dw toword.d ; adr db FETCH ; aw db DUP ; aw aw db LITW dw toin.d ; aw aw a db FETCH ; aw aw ap db SWAP ; aw ap aw db MINUS ; aw n db EXIT toword.d dw 0 ; pointer to start of current word wspace.doc: ; db ' ( c -- flag ) ' ; db ' return true if character c is whitepace (tab, space, 0 etc) ' wspace: dw toword.p db 'wspace', 6 wspace.p: ; stack diagram no good ; c ; space db DUP ; c c db LIT, ' ' ; c c space db EQUALS ; c flag db SWAP ; flag c ; carriage return db DUP ; c c db LIT, 10 ; c c space db EQUALS ; c flag db SWAP ; flag c ; newline db DUP ; flag c c db LIT, 13 ; flag c c cr db EQUALS ; flag c flag db SWAP ; f f c ; zero db DUP ; f f c c db LIT, 0 ; f f c c 0 db EQUALS ; f f c f db SWAP ; f f f c ; tab db DUP ; f f f c c db LIT, 9 ; f f f c c tab db EQUALS ; f f f c f db SWAP ; f f f f c ; combine flags db DROP ; f f f f db LOGOR ; f f f db LOGOR ; f f db LOGOR ; f db LOGOR ; f db EXIT wparse.doc: ; db ' ( -- adr n ) ' ; db ' return next word and length in the current input stream.' ; db ' the input stream is parsed for whitespace delimited words. ' ; db ' WPARSE skips initial whitespace. It is an important word ' ; db ' because it is the standard way to get the next word in ' ; db ' the current input stream, so things like CHAR and CREATE ' ; db ' rely on it. ; db ' For a more traditional forth "parse" see "parse.p" ' ; dw $-parse.doc wparse: dw wspace.p db 'wparse', 6 wparse.p: ; >in db FCALL dw toin.p ; adr n db DUP ; a n n ;*** no more text so nothing to parse db JUMPZ, .notext-$ ; a n ;*** get the name of the new word db RON ; adr r: n .nextspace: db CFETCHPLUS ; a+1 [a] r: n ; check for other whitespace, eg 0 tab, cr etc ; write a "whitespace" word db FCALL dw wspace.p ; a+1 flag r: n /true if whitespace ;db LIT, ' ' ; a+1 c space r: n ;db EQUALS ; a+1 flag ... db JUMPF, .nonspace-$ ; a+1 ... db RLOOP, .nextspace-$ ; a+1 r: n-1 ;*** no char found, so return 0 db DECR ; a r: 0 db ROFF ; a 0 db EXIT ; ;*** char found .nonspace: db DECR ; a r: n-m ;*** for debug ; db DUP, CFETCH, EMIT db DUP ; a a ... ;*** check rstack==0 and exit if so ;*** scan for next whitespace .nextchar: db CFETCHPLUS ; a a+1 [a] r: n-m ;*** check for whitespace character. db FCALL dw wspace.p ; a a+1 flag db JUMPT, .space-$ ; a a+1 ... db RLOOP, .nextchar-$ ; a a+1 r: n-m-1 db INCR ; a a+2 ... /balance decr ;*** .space: db DECR ; a a+p-1 r: 0 db ROFF, DROP ; a a+p-1 /clear rstack ;*** now calculate length of word db RON, DUP, ROFF ; a a a+p-1 db SWAP, MINUS ; a p-1 ; do 2dup, in+ at this point to advance the parse point ; in the input stream. db TWODUP ; a n a n db FCALL dw inplus.p ; a n db EXIT .notext: ; a 0 db EXIT ; the length returned may be +1 (?) ; this allows comments eg ; : ( 41 parse ; imm parse.doc: ; db ' ( char -- adr n ) ' ; db ' scan input stream for char and return length and address.' ; db ' The current input stream (IN) is scanned for the next char' ; db ' matching char. The parse position of the stream is advanced ' ; db ' after this call. If not found, then parse position and 0' ; db ' is returned.' ; dw $-parse.doc parse: dw wparse.p db 'parse', 5 parse.p: db RON ; r: char db FCALL dw toin.p ; adr n r: char db DUP ; a n n r: char ;* no more text so nothing to parse db JUMPZ, .notext-$ ; a n r: char ;* make the input stream length the counter db RON ; adr r: char n db DUP ; a a ... ;*** scan for next char .nextchar: db CFETCHPLUS ; a a+1 [a] r: char n ;*** check for scan character. db ROFF, ROFF ; a a+1 [a] n char db TWODUP ; a a+1 [a] n char n char db RON, RON ; a a+1 [a] n char r: char n db SWAP, DROP ; a a+1 [a] char r: char n db EQUALS ; a a+1 T/F r: char n ;db FCALL ;dw wspace.p ; a a+1 flag db JUMPT, .found-$ ; a a+1 r: char n db RLOOP, .nextchar-$ ; a a+1 r: char n .notfound: ; a a+n r: char 0 db DROP ; a r: char 0 db ROFF, ROFF ; a 0 char db DROP ; a 0 db EXIT ;db INCR ; a a+2 ... /balance the next decr ;*** .found: ;db DECR ; a a+p-1 r: char 0 db ROFF, DROP ; a a+p-1 r: char /clear rstack db ROFF, DROP ; a a+p-1 /clear rstack ; calculate the length until the character db RON, DUP, ROFF ; a a a+p-1 db SWAP, MINUS ; a p-1 ; Advance the parse point in the input stream. db TWODUP ; a n a n db FCALL dw inplus.p ; a n ; decrement text length to ignore final delimiter character db DECR ; a n-1 db EXIT .notext: ; a 0 r: char db ROFF, DROP ; a 0 db EXIT ; this is almost identical to "source" inputcompile.doc: ; db ' ( -- ) ' ; db ' compiles the input stream starting from the current ' ; db ' input parse position until the end of the stream ' ; dw $-inputcompile.doc inputcompile: dw parse.p db 'in,', 3 inputcompile.p: .nextword: db FCALL dw wparse.p ; a+x n /address+length of next word ;*** if parse returns 0 then no more words (all space) db DUP ; a+x n n db JUMPZ, .endofstream-$ ; a+x n ;*** convert name to number/opcode/fpointer + flag db FCALL dw tick.p ; n/op/xt flag ;db FCALL ;dw dotstack.p ;*** if flag==0 abort, unknown word/number db DUP ; m flag flag db JUMPZ, .error-$ ; n/op/xt flag ;*** immediate words like if/fi begin will leave ; parameters on the data stack db FCALL dw itemcompile.p ; db JUMP, .nextword-$ .endofstream: ;*** no more words (>in returned zero) ;*** compile a final exit even if no semi-colon was ; given db DROP, DROP ; clear stack db LIT, EXIT db FCALL dw ccomma.p ; compile opcode EXIT db EXIT .error: ;db FCALL ;dw dotstack.p ; n/op/xt flag db LIT, 13, EMIT db LIT, 10, EMIT db DROP, DROP ; db FCALL dw toword.p ; ad n db LIT, 5, FG ; set word colour to cyan db FCALL dw type.p db LIT, 4, FG ; set word colour to red db LIT, ' ', EMIT db LIT, '?', EMIT db LIT, '?', EMIT db LIT, 7, FG ; set word colour to cyan db LIT, 13, EMIT db LIT, 10, EMIT ;*** compile an exit even when an error occurs ; db LIT, EXIT db FCALL dw ccomma.p ; compile opcode EXIT db EXIT ; now, should adr be a counted string? should we compile only one ; block (1K), or have another paramter to determine the length? ; this should be (adr length -- ). And source should be just the ; same word as inputcompile.p ; One problem: words outside of : defs are compiled to the anon ; buffer, and then need to be executed. Eg the immediate word ; source.doc: ; db ' ( adr n -- ) ' ; db ' compile forth source code from address adr ' ; db ' This is called "load" in many forths.' ; dw $-source.doc source: dw inputcompile.p db 'source', 6 source.p: ; need to save the current input stream and position ; eg in.d in.length toin.d and toword.d ; but where to save ? ;*** set "in" var to adr db FCALL dw resetin.p ; set in = adr = >in = >word db LITW dw anon.d db FCALL dw ishere.p ;*** rub out anon db LIT, EXIT db FCALL ; compile opcode exit dw ccomma.p db LITW dw anon.d db FCALL dw ishere.p ;*** compile input buffer db FCALL dw inputcompile.p .run: ; now run the compiled stuff ; but only run anon if it has something in it... db LITW dw anon.d db PCALL db FCALL dw ok.p db EXIT ok.doc: ; db ' ( -- ) ' ; db ' just prints ok ' ok: dw source.p db 'ok', 2 ok.p: db LIT, 13, EMIT db LIT, 10, EMIT db LIT, 2, FG ; set to green db LIT, 'o', EMIT db LIT, 'k', EMIT db LIT, 7, FG ; set to white db LIT, 13, EMIT db LIT, 10, EMIT db EXIT %if 0 %endif find.doc: ; db ' ( adr n -- xt ) ' ; db ' return the execution address for the given word function. ' ; db ' Given a pointer to a string "adr" with length n ' ; db ' return the execution token (address) for the word' ; db ' or else zero if the word was not found. ' ; dw $-find.doc find: dw ok.p db 'find', 4 find.p: ; a n db FCALL ; pointer to last word in dictionary dw last.p ; a+1 n A db FETCH ; a+1 n last .again: db DECR ; point to count|control byte of name db FCALL dw rcount.p ; a n A N ;*** ;db DUP ; a+1 n A-1 A-1 ;db CFETCH ; .. A-1 N / get the count ;db DUP ; .. A-1 N N ;db RON, MINUS ; .. adr-N r: N ;db ROFF ; .. adr-N N ; a n A N ;db FCALL ;dw print.p ;*** now compare the 2 string lengths n & N db SWAP ; a n N A db RON, RON ; a n r: A N db DUP, ROFF ; a n n N r: A db EQUALS ; a n flag r: A db JUMPF, .lengthsnotequal-$ ; a n r: A db ROFF ; a n A db SWAP ; a A n ;*** save values on rstack, clumsy db RON, RON, RON; r: n A a db ROFF, DUP ; a a r: n A db ROFF, DUP ; a a A A r: n db ROFF, DUP ; a a A A n n db RON, SWAP, RON ; a a A n r: n A ;*** compare two strings db FCALL ; are strings equal? dw compare.p ; a flag r: n A ; db JUMPF, .notequal-$ ; a r: n A ; if strings same clear stacks db DROP ; r: n A db ROFF, ROFF ; A n db PLUS, INCR ; A+n+1 ; A+n+1 is the execution address. found, so exit now db EXIT ; if false, balance stacks and jump down .notequal: ; a r: n A db ROFF, ROFF ; a A n db SWAP ; a n A db JUMP, 3 ; a n A /get next pointer .lengthsnotequal: db ROFF ; a n A db DECR, DECR ; a n A-2 /A-2 points to previous db FETCH ; a n [A-2] db DUP ; a n [A-2] [A-2] db JUMPNZ, .again-$ ; a n [A-2] db DROP, DROP, DROP ; clear stack db LIT, 0 ; zero means not found db EXIT immediate.doc: ; db ' ( -- ) ' ; db ' make the last word immediate. ' ; db ' Set the immediate control bit' immediate: dw find.p db 'imm', 3 immediate.p: db FCALL dw last.p ; addr db FETCH ; xt db DECR ; xt-1 db DUP ; xt-1 xt-1 db CFETCH ; xt-1 [xt-1] /get the count/control byte db LIT, IMMEDIATE ; xt-1 count|control 0b10000000 db LOGOR ; xt-1 nVimm / zero or non zero db SWAP ; m xt-1 db CSTORE ; trap! the count|control is only one byte db EXIT isimmediate.doc: ; db ' ( xt -- flag ) ' ; db ' returns 0 if word is not immediate. <>0 otherwise ' ; db ' given the execution address for a procedure ' ; db ' return a flag indicating if the procedure ' ; db ' is immediate or not. Immediate procedures are ' ; db ' executed at compile time, not compiled ' ; db ' so, essentially, they compile themselves. ' ; db ' this allows the compiler to be extended by ' ; db ' procedures. The immediate control bit is the ' ; db ' most significant bit of the count byte in the ' ; db ' name. Immediate words have both a compile-time and ' ; db ' a run-time behaviour, whereas non-immediate words ' ; db ' have only a run-time behaviour. isimmediate: dw immediate.p db 'imm?', 4 isimmediate.p: ; xt db DECR ; xt-1 db CFETCH ; [xt-1] /get the count byte db LIT, IMMEDIATE ; n 0b10000000 db LOGAND ; n & imm / zero or non zero db EXIT tick.doc: ; db ' ( p n -- n flag) ' ; db ' given a pointer "p" to a string of length n ' ; db ' attempt to convert the name to either ' ; db ' an opcode, procedure execution code, or number/integer ' ; db ' the flag indicates the type of token returned ' ; db ' a flag of zero means that the name is neither ' ; db ' number nor opcode nor xt' ; db ' flag=0 not number nor word ' ; db ' flag=1 if number, 2 if opcode, 3 if procedure ' ; dw $-tick.doc tick: dw isimmediate.p db 'tick', 4 tick.p: ; a n db TWODUP ; a n a n db FCALL dw find.p ; a n xt/0 db DUP ; a n xt xt db JUMPNZ, .opcode-$ ; a n xt ;*** not found, try to convert to number db DROP ; a n db FCALL dw tonumber.p ; adr/n flag db JUMPNZ, .number-$ ; adr/n ;*** not a number, push false flag and exit db LIT, 0 ; adr 0 db EXIT ;*** is a number .number: ; n db LIT, 1 ; n 1 /flag=1 means number db EXIT .opcode: ;*** check if opcode ; a n xt db SWAP, DROP ; a xt db SWAP, DROP ; xt db DUP ; xt xt db FCALL ; does this address correspond to an opcode dw opcode.p ; xt op|0 db DUP ; xt op|0 op|0 db JUMPZ, .proceedure-$ ; xt op ;*** is an opcode db SWAP, DROP ; op /drop execution token address db LIT, 2 ; op 2 /flag=2 means opcode db EXIT .proceedure: ;*** must be procedure, ; xt 0=op db DROP ; xt db LIT, 3 ; xt 3 /flag=3 means procedure db EXIT args.doc: ; db ' ( op -- flag ) ' ; db ' given a valid opcode return flag=1 if the ' ; db ' the opcode requires a one byte ' ; db ' argument or return flag=2 if the opcode requires ' ; db ' a 2byte argument or flag=0 if no arguments required. ' ; dw $-args.doc args: dw tick.p db 'args', 4 args.p: ; lit, jump, jumpz, jumpnz, rloop db DUP ; op op db LIT, LIT ; op op op2 db EQUALS ; op flag db JUMPT, .one-$ ; op db DUP ; op op db LIT, JUMP ; op op op2 db EQUALS ; op flag db JUMPT, .one-$ ; op db DUP ; op op db LIT, JUMPZ ; op op op2 db EQUALS ; op flag db JUMPT, .one-$ ; op db DUP ; op op db LIT, JUMPNZ ; op op op2 db EQUALS ; op flag db JUMPT, .one-$ ; op db DUP ; op op db LIT, RLOOP ; op op op2 db EQUALS ; op flag db JUMPT, .one-$ ; op ; litw, fcall, ljump, db DUP ; op op db LIT, LITW ; op op op2 db EQUALS ; op flag db JUMPT, .two-$ ; op db DUP ; op op db LIT, FCALL ; op op op2 db EQUALS ; op flag db JUMPT, .two-$ ; op db DUP ; op op db LIT, LJUMP ; op op op2 db EQUALS ; op flag db JUMPT, .two-$ ; op .zero: db DROP db LIT, 0 ; 0 db EXIT .one: db DROP db LIT, 1 ; 1 db EXIT .two: db DROP db LIT, 2 ; 2 db EXIT ccomma.doc: ; db ' ( n -- ) ' ; db ' compile byte value n at next available byte ' ; b ' as given by here.p ' ; dw $-ccomp.doc ccomma: dw args.p db 'c,', 2 ccomma.p: ;*** compile single byte db FCALL ; /get compile point dw here.p ; n adr db CSTOREPLUS ; a+1 db FCALL dw ishere.p ; /update compile point ; need to update >code if here is > than >code ; because custom defining words allocate data space in the ; dictionary. .end: db EXIT comma.doc: ; db ' ( n -- ) ' ; db ' compile 2 byte value at next available space ' ; db ' as given by the "here" variable ' ; db ' code: : , here !+ ishere ; ' ; dw $-comma.doc comma: dw ccomma.p db ',', 1 comma.p: db FCALL ; /get compile point dw here.p ; n adr db STOREPLUS ; a+2 db FCALL dw ishere.p ; /set compile point to new address (a+2) db EXIT ; trying to get dictionary compile point to advance db FCALL ; /get dict compile point dw tocode.p ; code* db FETCH ; >code db FCALL ; /get here compile point dw here.p ; >code here db INCR ; >code here+1 ; check if tocode < here+1 db ULESSTHAN ; flag=T/F db JUMPF, .end-$ ; ;db FCALL ;dw dotstack.p ;db KEY ;db EMIT ;db FCALL ;dw codetohere.p .end: scompile.doc: ; db ' ( adr n -- ) ' ; db ' compile the string at adr with length n to the ' ; b ' current compile position as given by here ' ; dw $-wordcompile.doc scompile: dw comma.p db 's,', 2 scompile.p: ; ad n ; if text length is 0 then do nothing. db DUP ; ad n n db JUMPNZ, .notzero-$ ; ad n db DROP, DROP ; db EXIT .notzero: ;*** n > 0 db RON ; ad r: n /n is loop counter .nextchar: db CFETCHPLUS ; ad+1 c r: n db FCALL dw ccomma.p ; ad+1 r: n db RLOOP, .nextchar-$ ; ad+1 r: n-1 db DROP ; r: 0 db ROFF, DROP ; /get rid of rloop counter db EXIT ; standard core forth word ; probably could write this as source sliteral.doc: ; db ' (comp: adr n -- ) (run: -- A n ) ' ; db ' compile the string at adr with length n to the ' ; db ' current definition. At run-time push the address ' ; db ' and length of the compiled string onto the stack ' ; dw $-sliteral.doc sliteral: dw scompile.p db 'sliteral', IMMEDIATE | 8 sliteral.p: ; ad n ; if text length is 0 then do nothing. db DUP ; ad n n db JUMPNZ, .notzero-$ ; ad n db DROP, DROP db EXIT .notzero: ; do other stuff here like compile ; code to push new A and n on stack at runtime ; ... compile jump db FCALL dw here.p ; ad n here ; adjust address by length of code about to be compiled db LIT, 8, PLUS ; ad n H+5 db FCALL dw literal.p ; ad n db DUP ; ad n n db FCALL dw literal.p ; ad n ; compile a jump over the string db LIT, JUMP ; ad n op=jump db FCALL dw ccomma.p ; ad n ; calculate and compile jump offset (jump over string) db DUP ; ad n n db LIT, 2, PLUS ; ad n n+2 db FCALL dw ccomma.p ; ad n ; copy the string to the current compile position db FCALL dw scompile.p db EXIT ; this doesnt always actually compile something (if it is ; an immediate word etc) so should be called something else ; like doword.p doitem.p itemcompile.doc: ; db ' ( n flag -- ) ' ; db ' compile item "n" at next availabe position ' ; b ' as given by "here" variable where flag indicates ' ; db ' the type of item. ' ; db ' flag=1 literal, 2 opcode, 3 procedure' ; dw $-compile.doc itemcompile: dw sliteral.p db 'item,', 5 itemcompile.p: ;*** 1=literal number ; n flag db DUP ; n flag flag db LIT, 1, EQUALS ; n flag 0/-1 db JUMPZ, .notnumber-$ ; n flag ;*** check if state is immediate db FCALL dw state.p ; n flag adr db FETCH ; n flag state db JUMPT, .immediateNumber-$ ; n flag ;*** compile number db DROP ; n db LIT, LITW ; n op db FCALL ; /get compile point dw here.p ; n op adr db CSTOREPLUS ; n a+1 db STOREPLUS ; a+3 db FCALL dw ishere.p ; /update compile point db EXIT .immediateNumber: ; n flag ; do almost nothing because the number is already ; on the stack. db DROP ; n db EXIT .notnumber: ;*** check if is opcode ; n flag db DUP ; n flag flag db LIT, 2, EQUALS ; n flag 0/-1 db JUMPZ, .notopcode-$ ; n flag ;*** check if state is immediate db FCALL dw state.p ; op flag adr db FETCH ; op flag state db JUMPT, .immediateOp-$ ; n flag ;*** 2 is opcode ;*** compile the bytecode to given address db DROP ; op ; why not just use ccomma.p ?? db FCALL ; /get compile point dw here.p ; op adr db CSTOREPLUS ; adr+1 db FCALL ; /set compile point dw ishere.p ; db EXIT .immediateOp: ; op flag db DROP ; op db LITW ; dw obuff.d ; op adr db CSTOREPLUS ; adr+1 db LIT, EXIT ; a+1 op=exit db SWAP ; op a+1 db CSTORE ; ; now actually execute the opcode db LITW ; dw obuff.d ; adr db PCALL db EXIT .notopcode: ;*** check if procedure ; n flag db DUP ; n flag flag db LIT, 3, EQUALS ; n flag 0/-1 db JUMPF, .error-$ ; n flag ;*** 3 procedure ; xt flag db DROP ; xt ;*** check if word is immediate db DUP ; xt xt db FCALL dw isimmediate.p ; xt flag db JUMPT, .immediate-$ ; xt ; bug! bug! This only applies to procedures, not ; opcodes, so even if state is immediate, opcodes and ; numbers will not be executed immediately. I think this ; is a problem. ;*** check if state is immediate ; xt db FCALL dw state.p ; xt adr db FETCH ; xt state db JUMPT, .immediate-$ ; xt ;*** neither word nor state is immediate, so compile db LIT, FCALL ; xt op db FCALL ; /get compile point dw here.p ; xt op adr db CSTOREPLUS ; xt adr+1 db STOREPLUS ; adr+3 db FCALL ; /set compile point dw ishere.p ; db EXIT ;*** immediate proc, execute, dont compile .immediate: db PCALL db EXIT .error: ; n flag db DROP, DROP db EXIT headdot.doc: ; db ' ( xt -- ) ' ; db ' show the dictionary header for a word function. ' ; db ' The word header is shown in a form suitable for ' ; db ' debugging. ' ; dw $-headdot.doc headdot: dw itemcompile.p db 'head.', 5 headdot.p: ; xt db DUP ; xt xt db DECR ; xt xt-1 /points to count|control byte db FCALL dw rcount.p ; xt adr n /start of name db DROP ; xt adr db DECR, DECR ; xt adr-2 ;db FETCH ; XT / get word pointer ;*** print head memory address db LIT, 13, EMIT db LIT, 10, EMIT db DUP ; xt a a db FCALL dw udot.p ; xt a db LIT, ':', EMIT db LIT, ' ', EMIT ;*** print the link address to next word in dictionary db LIT, '[', EMIT db FETCH ; xt [a] db DUP ; xt [a] [a] db FCALL dw udot.p ; xt [a] db LIT, ']', EMIT db LIT, ' ', EMIT db LIT, '-', EMIT db LIT, '>', EMIT db LIT, ' ', EMIT ;*** print the name of the linked word db DECR ; xt A-1 /point to count|control byte db FCALL dw rcount.p ; xt A-n n db FCALL dw type.p ; xt db LIT, 13, EMIT ; db LIT, 10, EMIT ; db DECR ; xt-1 /point to count|control byte db FCALL dw rcount.p ; a-n n db RON, DUP ; a-n a-n R: n db FCALL dw udot.p ; a-n R: n db LIT, ':', EMIT db LIT, ' ', EMIT db ROFF ; a-n n db DUP, RON ; a-n n R: n ;*** print word name in quotes db LIT, '"', EMIT db FCALL dw type.p ; R: n db LIT, '"', EMIT db ROFF ; n db LIT, ' ', EMIT ;*** print word length|control in quotes ;!! need to handle immediate words which have msb set db FCALL dw udot.p ; db LIT, 13, EMIT ; db LIT, 10, EMIT ; db EXIT worddump.doc: ; db ' ( xt -- ) ' ; db ' shows how a word is compiled in the dictionary ' ; db ' The dictionary header and compiled code is displayed ' ; db ' for the word corresponding to the execution address. ' ; dw $-worddump.doc worddump: dw headdot.p db '..word', 6 worddump.p: db DUP ; xt xt db FCALL dw headdot.p ; xt db LIT, 50 ; xt n /decompile up to n bytes db FCALL dw decomp.p db EXIT ; : word.. dup head. 20 decomp ; decomp.doc: ; db ' ( adr n -- a2 ) ' ; db '' ; db ' decompiles n bytes starting at address n ' ; db ' returns the next address after the decompiled ' ; db ' bytes. Need to handle jumps too. This cannot decompile' ; db ' machine code, only bytecodes ' ; dw $-decomp.doc decomp: dw worddump.p db 'un,', 3 decomp.p: ; adr n db RON ; adr r: n .again: db LIT, 13, EMIT db LIT, 10, EMIT db DUP ; a a db FCALL dw udot.p ; a db LIT, ':', EMIT db LIT, ' ', EMIT db CFETCHPLUS ; a+1 c r: n db DUP ; a+1 c c db JUMPZ, .invalid-$ ; a+1 c db DUP, LIT, NOOP, INCR ; a+1 c c op+1 r: n db ULESSTHAN ; a+1 c flag r: n db JUMPT, .valid-$ ; a+1 c r: n ;*** invalid opcode .invalid: ; a+x c db FCALL dw udot.p db LIT, '?', EMIT ;*** clear rstack db ROFF, DROP db EXIT .valid: ;*** valid opcode db DUP ; a+1 c c r: n db FCALL dw dotcode.p ; a+1 c r: n db LIT, ' ', EMIT ;*** is fcall db DUP ; a+1 c c r: n db LIT, FCALL ; a+1 c c op r: n db EQUALS ; a+1 c flag ... db JUMPF, 21 ; a+1 c ... ;*** fcall, get xt db DROP ; a+1 ... db FETCHPLUS ; a+3 xt ... db LIT, '<', EMIT db FCALL dw dotxt.p ; a+3 r: n db LIT, '>', EMIT db LIT, ' ', EMIT db RLOOP, .again-$ ; a+3 r: n-1 db ROFF, DROP ; clear rstack db EXIT ;*** check if 1 or 2 byte argument or non db DUP ; a+1 c c r: n db FCALL dw args.p ; a+1 c args r: n db DUP ; a+1 c args args r: n db JUMPZ, .zerobytes-$ ; a+1 c args ... db LIT, 1, EQUALS ; a+1 c flag db JUMPT, .onebyte-$ ; a+1 c ;*** opcode takes 2 byte argument .twobytes: db DROP ; a+1 ... db FETCHPLUS ; a+3 xt ... db LIT, '<', EMIT db FCALL dw dot.p ; a+3 r: n db LIT, '>', EMIT db LIT, ' ', EMIT db RLOOP, .again-$ ; a+3 r: n-1 db ROFF, DROP ; clear rstack db EXIT ;*** opcode takes 1 byte argument .onebyte: db DROP ; a+1 ... db CFETCHPLUS ; a+2 n ... db LIT, '<', EMIT ;*** cdot prints 8 bit number as signed db FCALL dw cdot.p ; a+2 r: n db LIT, '>', EMIT db LIT, ' ', EMIT db RLOOP, .again-$ ; a+2 r: n-1 db ROFF, DROP ; clear rstack db EXIT ;*** opcode takes no argument .zerobytes: ; a+1 c args ... db DROP, DROP db RLOOP, .again-$ ; a+1 r: n-1 db ROFF, DROP ; clear rstack db EXIT tocode.doc: ; db ' ( -- adr ) ' ; db ' puts on the stack a pointer to the next available byte' ; db ' in the dictionary. ' ; dw $-tocode.doc tocode: dw decomp.p db '>code', 5 tocode.p: db LITW dw tocode.d ; db FETCH / maybe not db EXIT tocode.d: dw dictionary codetohere.doc: ; db ' ( -- ) ' ; db ' sets the dict compile point to the here var' ; dw $-codetohere.doc codetohere: dw tocode.p db 'code>here', 9 codetohere.p: db FCALL dw here.p ; a db LITW dw tocode.d ; a A db STORE ; db EXIT ; this is an important word because it is called by ; create, and therefore all defining words. In this implementation ; of forth it writes an exit to end of anon (here) and ; then executes the anon buffer. This is because this forth=froth ; is a "compiling" forth, rather than an "interpreted" forth, ; so the default behavior is to compile even interactive commands heretocode.doc: ; db ' ( -- ) ' ; db ' sets the here var to the dict compile point' ; dw $-codetohere.doc heretocode: dw codetohere.p db 'here>code', 9 heretocode.p: db LIT, EXIT ; op=exit db FCALL dw ccomma.p ; compile exit op to end of anon (at "here") db FCALL ; execute everything in the anon buffer dw anon.d ; need to "empty" the anon buffer after executing ; to avoid multiple unwanted calls db FCALL ; set here to start dw heretoanon.p db LIT, EXIT ; op=exit db FCALL dw ccomma.p ; compile EXIT op to start of anon (at "here") db LITW dw tocode.d ; >code* db FETCH ; [>code] db LITW dw here.d ; [>code] a db STORE ; db EXIT heretoanon.doc: ; db ' ( -- ) ' ; db ' sets the compile point to the start of the anon buffer' ; dw $-heretoanon.doc heretoanon: dw heretocode.p db 'here>anon', 9 heretoanon.p: db LITW dw anon.d ; anon* db LITW dw here.d ; anon* here* db STORE ; db EXIT here.doc: ; db ' ( -- adr ) ' ; db ' puts on the stack the current compile position' ; db ' in the code data space ' ; dw $-here.doc here: dw heretoanon.p db 'here', 4 here.p: db LITW dw here.d db FETCH db EXIT here.d: dw 0 ; pointer to next byte ishere.doc: ;db ' ( adr -- ) ' ;db ' set position of next available byte' ;db ' for compilation ("here" variable) to address adr' ;dw $-ishere.doc ishere: dw here.p db 'here!', 5 ishere.p: ; adr ;db DUP ; A A db LITW dw here.d ; A A b db STORE ; A ; this seems the right place to update the dictionary ; compile point, if necessary, but it is not working. ;db FCALL ; ;dw tocode.p ; A C* ;db FETCH ; A C ;db ULESSTHAN ; t/f ;db JUMPT, .end-$ ;db FCALL ;dw codetohere.p .end: db EXIT ; according to standard forth docs, this should not be modified by ; forth words except [ ] etc. ; [ and ] are be defined in source as ; : [ 1 state ! ; immediate ; : ] 0 state ! ; immediate state.doc: ; db ' ( -- adr ) ' ; db ' pushes a pointer to current state (immediate or compile)' ; db ' modified by [ and ] only (not by colon) ' ; db ' state 0=normal/compile state 1=immediate ' ; dw $-state.doc state: dw ishere.p db 'state', 5 state.p: db LITW dw state.d db EXIT state.d: dw 0 first.doc: ; db ' ( -- adr ) ' ; db ' address of first buffer (for loading source from disk)' ; dw $-first.doc first: dw state.p db 'first', 5 first.p: db LITW dw first.d db FETCH db EXIT ; near the end of this segment, just for testing. ; in reality should be just after the dictionary or stack ; remember distinction between disk map and ram memory map first.d: dw 50*1024 blockone.doc: ; db ' ( -- adr ) ' ; db ' sector number of first source block (1K) on disk' ; db ' each sector is 512 bytes in length. ' ; dw $-blockone.doc blockone: dw first.p db 'block1', 6 blockone.p: db LITW ; the 2+ is a hack because this is 2 short for some ; reason dw 1+(diskcode-$$)/512 db EXIT ; "a" is top of stack. ie rightmost is last-on first-off copy.doc: ; db ' ( A n a -- ) ' ; db ' copy n bytes from address A to address a ' ; db ' this cannot deal with overlapping memory areas. ' ; dw $-copy.doc copy: dw blockone.p db 'copy', 4 copy.p: ; A n a db SWAP ; A a n db RON ; A a r: n /n loop counter db SWAP ; a A .again: db CFETCHPLUS ; a A+1 [A] ; db DUP, EMIT ; debug db SWAP ; a [A] A+1 db RON ; a [A] r: n A+1 db SWAP ; [A] a r: n A+1 db CSTOREPLUS ; a+1 r: n A+1 db ROFF ; a+1 A+1 r: n db RLOOP, .again-$ ; a+1 A+1 r: n-1 db ROFF ; a+n A+n 0 db DROP, DROP, DROP ; db EXIT compare.doc: ; db ' ( a A n -- flag) ' ; db ' given 2 pointers to strings a and A ' ; db ' compare the 2 strings for n bytes ' ; db ' and put -1 on stack as flag if the strings are ' ; db ' the same or flag=0 on stack if the strings ' ; db ' are different. ' ; dw $-compare.doc compare: dw copy.p db 'compare', 7 compare.p: ; a A n db RON ; a A r: n /n loop counter db CFETCHPLUS ; a A+1 [A] ; db DUP, EMIT ; debug db SWAP ; a [A] A+1 db RON, RON ; a r: n A+1 [A] db CFETCHPLUS ; a+1 [a] r: n A+1 [A] ; db DUP, EMIT ; debug db ROFF ; a+1 [a] [A] r: n A+1 db EQUALS ; a+1 flag r: n A+1 db JUMPT, 10 ; a+1 r: n A+1 db ROFF, ROFF ; a+1 A+1 n db DROP, DROP, DROP ; clear stacks db LIT, 0 ; flag=0 (false) db EXIT db ROFF ; a+1 A+1 r: n db RLOOP, -18 ; a+1 A+1 r: n-1 db ROFF ; a+n A+n 0 db DROP, DROP, DROP ; clear stacks db LIT, -1 db EXIT type.doc: ; db ' ( adr n -- ) ' ; db ' Prints out n number of characters starting ' ; db ' at address adr. ' ; dw $-type.doc type: dw compare.p db 'type', 4 type.p: ; adr n ;*** if count zero, do nothing db DUP ; adr n n db JUMPNZ, .sometext-$ ; adr n db DROP, DROP ; clear data stack db EXIT .sometext: db RON ; adr r: n .nextchar: db CFETCHPLUS ; adr+1 c r: n db EMIT ; adr+1 r: n db RLOOP, .nextchar-$ ; adr+1 r: n-1 db ROFF, DROP, DROP ; clear stacks db EXIT accept.doc: ; db ' ( buffer -- ) ; db ' receive a line of input from the terminal ' ; db ' and store it as a counted string in the buffer. ' ; db ' should have a character limit. backspaces are mainly ; db ' handled. ' ; dw $-accept.doc accept: dw type.p db 'accept', 6 accept.p: ;*** to elimate backspace problems, need to ; emit character after finding out what it is ; not before ; a db DUP, RON ; a r: a db INCR ; a+1 r: a .nextkey: db KEY, DUP ; a+1 c c r: a db DUP, EMIT ; a+1 c c r: a ;*** enter terminates input db LIT, 13, EQUALS ; a+1 c flag r: a /enter press db JUMPT, .exit-$ ; a+1 c r: a ;*** handle backspace db DUP ; a+1 c c r: a db LIT, 8, EQUALS ; a+1 c flag r: a /backspace db JUMPF, .notback-$ ; a+1 c r: a db DROP ; a+1 r: a ;*** test for at 1st char db DUP ; a+1 a+1 r: a db ROFF, DUP, RON ; a+1 a+1 a r: a db MINUS ; a+1 n r: a db LIT, 1 ; a+1 n 1 r: a db EQUALS ; a+1 flag r: a db JUMPT, .nextkey-$ ;*** not 1st char so go back 1 space db LIT, ' ', EMIT ; a+n r: a db LIT, 8, EMIT ; go back db DECR ; a+n-1 r: a db JUMP, .nextkey-$ ; a+n-1 r: a /get next char ; put the character limit test here. .notback: db SWAP ; c a+1 r: a db CSTOREPLUS ; a+2 r: a db JUMP, .nextkey-$ ; a+2 r: a .exit: db LIT, 10, EMIT ; print newline if enter pressed db LIT, 13, EMIT ; db DROP ; a+n r: a db ROFF ; a+n a db DUP, RON ; a+n a r: a db MINUS ; n r: a db DECR ; n-1 r: a db ROFF ; n-1 a db CSTORE db EXIT in.doc: ; db ' ( -- adr ) ; db 'Puts on the stack the address of the ' ; db 'current input source/ buffer ' ; dw $-in.doc in: dw accept.p db 'in', 2 in.p: db LITW dw in.d db EXIT in.d: dw 0 ; need to initialize in.length: dw 0 ; need to initialize term.doc: ; db ' ( -- adr ) ; db 'Puts on the stack the address of the ' ; db 'user input buffer (terminal buffer). This' ; db 'is a common source for interpreting and ' ; db ' compiling ' ; dw $-term.doc term: dw in.p db 'term', 4 term.p: db LITW dw term.d db EXIT term.d: times 128 db 0 ; counted buffer for user input dotin.doc: ; db ' ( -- ) ; db 'display the contents of the current input stream ' ; db 'or buffer. ' ; dw $-dotin.doc dotin: dw term.p db '.in', 3 dotin.p: ; change this because the input stream is not always a ; counted string db LITW dw in.d ; adr db COUNT ; a+1 n db DUP ; a+1 n n db FCALL dw udot.p ; a+1 n db LIT, ':', EMIT ; a+1 n db FCALL dw type.p db EXIT ; : in.. in count dup u. sp type ; toin.doc: ; db ' ( -- adr n ) ' ; db ' put on stack parse position in input stream' ; db ' and number of characters remaining in stream. ' ; db ' This is used with parse etc' ; dw $-toin.doc ; was a strange bug with this link toin: dw dotin.p db '>in', 3 toin.p: ;** need to remember that in.length, toin.d and in.d are ; pointers and need to be fetched before use... *p ; db LITW dw toin.d ; adr db FETCH ; >in db DUP ; >in >in db LITW dw in.d ; >in >in in.d db FETCH ; >in >in in db MINUS ; >in offset db LITW dw in.length ; >in offset adr db FETCH ; >in offset [in.length] db SWAP ; >in length offset db MINUS ; >in remainder ;db FCALL ;dw dotstack.p db EXIT toin.d: dw 0 ; maybe call this setin or in! instore or in+ ; why not just get adr and n from >word? inplus.doc: ; db ' ( adr n -- ) ' ; db ' update word and parse position in input' ; db ' where the start of the word is given by pointer' ; db ' adr and the length of the word is n. The parse ' ; db ' position will be adr+n after this call ' ; db ' eg: pad resetin pad accept >in parse 2dup type ' ; db ' atin >in .s etc' ; dw $-inplus.doc inplus: dw toin.p db 'in+', 3 inplus.p: ; probably should check here that the new ; parse position is not beyond the end ; of the input stream (length) ; adr n db SWAP, DUP ; n adr adr db LITW dw toword.d ; n adr adr a2 db STORE ; n adr db PLUS ; n+adr db LITW dw toin.d ; n+adr ap db STORE ; db EXIT ; we dont actually use the toword.p proceedure. resetin.doc: ; db ' ( adr n -- ) ' ; db ' set word and parse position to 0 in input' ; db ' and set the input buffer to point to address ' ; db ' adr and the stream length to n. ' ; db ' set vars in.d in.length toin.d toword.d ' ; dw $-resetin.doc resetin: dw inplus.p db 'in0', 3 resetin.p: ; adr n db LITW dw in.length ; adr n in.length db STORE ; adr db DUP ; adr adr db LITW dw in.d ; adr adr in.d db STORE ; adr db DUP ; adr adr db LITW dw toin.d ; adr adr toin.d db STORE ; adr db LITW dw toword.d ; adr word.d db STORE ; db EXIT anon.doc: ; db ' ( -- adr ) ; db 'Puts on the stack the address of the ' ; db 'buffer to hold anonymous definitions. This ' ; db 'contains compiled byte code for user input ' ; dw $-anon.doc anon: dw resetin.p db 'anon', 4 anon.p: db LITW dw anon.d db EXIT anon.d: times 128 db 0 ; compiled byte code obuff.doc: ; db ' ( -- adr ) ; db ' A buffer used to execute opcodes which have been ' ; db ' called in immediate state. ; dw $-opbuff.doc obuff: dw anon.p db 'obuff', 5 obuff.p: db LITW dw obuff.d db EXIT ; this will contain one opcode and EXIT obuff.d: times 4 db 0 ; buff.doc: ; db ' ( -- adr ) ; db ' a testing buffer' ; dw $-buff.doc buff: dw obuff.p db 'buff', 4 buff.p: db LITW dw buff.d db EXIT buff.d: times 64 db 0 ; compiled byte code pad.doc: ; db ' ( -- adr ) ; db 'Puts on the stack the address of the general purpose ' ; db 'text buffer ' ; dw $-pad.doc pad: dw buff.p db 'pad', 3 pad.p: db LITW dw pad.d db EXIT pad.d: times 128 db 0 drive.doc: ; db ' ( -- adr ) ; db ' a variable to hold virtual drive number ' dw pad.p db 'drive', 5 drive.p: db LITW dw drive.d ; ad db EXIT drive.d: db -1 sides.doc: ; db ' ( -- adr ) ; db ' how many sides or platters a disk has. ' ; db ' This information may be needed for disk reads. ' dw drive.p db 'sides', 5 sides.p: db LITW dw sides.d db EXIT sides.d: dw -1 sectorspertrack.doc: ; db ' ( -- adr ) ; db ' How many sectors each track has. eg 18 for floppy ' ; db ' This information may be needed for disk reads. ' dw sides.p db 'sectors', 7 sectorspertrack.p: db LITW dw sectorspertrack.d db EXIT sectorspertrack.d: dw -1 lib.doc: ; db ' ( -- adr ) ; db ' some source code to load' ; dw $-lib.doc lib: dw sectorspertrack.p db 'lib', 3 lib.p: db LITW dw lib.d db EXIT lib.d: db lib.end-$-1 ; block1 is the sector (512 byte block) of the first ; source code block (forth block=1K bytes) on disk ; actually parse and compile this source code ; use clock to time compile time db ' clock ' ; load the 0 block, which loads all the others db ' block1 2 first read drop first 1024 source ' db ' clock .s ' ;db ' timer cr .s ' lib.end: last.doc: ; db ' ( -- adr ) ; db 'Puts on the stack a pointer to address of the last word ' ; db 'in the dictionary. This changes when new words are ' ; db 'added via colon : definitions or other defining words ' last: dw lib.p db 'last', 4 last.p: db LITW dw last.d ; ad db EXIT last.d: dw last.p ; testing multisector stack machine byte code code: db JUMP, .system-$ .system: ; lib.p just loads the 1st block (block 0) which loads the ; other disk source code blocks and then compiles and runs ; 'shell' which is the repl interpreter (normally called 'quit' ; in forth) db FCALL dw lib.p db COUNT ; source takes address + length db FCALL dw source.p ;db FCALL ;dw shell.p db 0 start: mov ax, cs ; cs is already correct (?!) mov ds, ax ; data segment ;*** save the (virtual) drive we have loaded ; code from. Handy for disk writes later ; have to set data segment DS first ; get the disk geometry parameters and save for use by ; read.x ; code from mike gonta. This fixed read problems where there ; are more than 18 sectors on a track etc. mov [drive.d], byte dl ; mov ah, 8 int 13h ; jc .diskerror and cx, 3Fh mov [sectorspertrack.d], cx movzx dx, dh add dx, 1 mov [sides.d], dx ; point es:di directly after the code and data segment ; i.e. after the 8 sectors (8 * 512 bytes) ; which contain code and data. We will use es:di ; as the return stack pointer. When ; a value is pushed on the return stack, the value ; is written to [es:di] and di is incremented by 2 ; add ax, 256 ; 256 * 16 = 4096, 4K (8 sectors) ; put a gap of 4K between code and stacks for ; dictionary entries etc ;*** 8K code + 8K gap then rstack. But are the data and ; return stacks growing towards each other??? ;add ax, 512 ; 512 * 16 = 8K add ax, 1024 ; 1024 * 16 = 16K mov es, ax ; using es:di as return stack pointer mov di, 0 ; the calculations are as follows ; we have loaded 16 sectors = 16 * 512 bytes = 9162 bytes == 8K ; we want a data stack of size 4K ; (which is big) = 4094 bytes ; also we want a return stack of size 4K/8K ; for hefty recursive functions, although these ; huge sizes are not necessary. ; x86 hardware stack grows up or down? ... ; divide by 16 because that is how segment ; addressing works ; That is: if we multiply the number in ss or es ; or ds by 16 ; we get a absolute memory address ;*** ax is pointing to start of the rstack so ;*** add 4K more for data stack add ax, 256 ; 256*16=4K mov ss, ax ; a 4K stack here mov sp, 4096 ; set up the stack pointer push code call exec.x stayhere: jmp stayhere ;*** new words can be compiled here ; dictionary: dw 0 ; Pad remainder of n( = n/2 K) sectors with 0s ; The number below (6,7,8 etc) only has to be as big ; as the dictionary. times (6*1024)-($-$$) db 0 ; MEMORY MAP (may 2018) ; To avoid confusion, remember the clear distinction between ; the disk map (code and data on disk) and the ram memory map. ; ; need to clarify this memory map. ; This may change as code grows, but the idea is to keep code small ; The addresses below are segmented, so multiply the first part by ; 16 to get the real address. ; ; address contents ; ------- -------- ; 1000:0000 8K of code and data, including the dictionary ; and any new words defined by colon : ; ??:0000 8K return stack pointed to by es:di ; ??:0000 4K data stack pointed to by ss:sp ; ; this is silly, can erase memory without writing to disk ; in fact, is there any need to erase stack memory? ; times 4096 db 0 ;*** some text/forth code at sector ? which we can load ;*** with source.p or with load opcode diskcode: ; can load this with "block1 2 first read first 1024 source" block1: ; see os.sed for a proprocessing sed script. ; This script allows us to write multiline ; forth source code without the "db ' " guff. %if 0 ; code{ ; switch to video mode 16, not portable 16 vid ; standard ' tick (?). There is another word TICK which works ; a bit differently since it returns a flag as well indicating ; if the word is a number, opcode or fcall. : ' wparse find ; imm ; a ' to use in colon defs ;: ['] wparse find post literal ; imm ; if call, is not immediate then this gets simpler ; does it need to be immediate? ; ans forth 'postpone' word. Compiles a call to a word, even if ; it is immediate. This is used in words like [char] but is tricky ; to think about somethimes. : post wparse find ' call, call, ; imm ; makes all words execute immediately until ] ' ; This word sets the state variable to true (immediate) ' ; so that all words parsed will execute immediately ' ; without being compiled. ' : [ 1 state ! ; imm ; This word sets the state variable to false (normal) ; so that all words parsed will be compiled unless they ; have their immediate control bit set. : ] 0 state ! ; imm ; Put the value of the following character on the stack : char wparse drop c@ ; imm ; standard [char] : [char] post char post literal ; imm ; comments like ( -- ) : ( [char] ) parse drop drop ; imm ; increments value stored at address a ; or call this ++ : !1+ ( a -- ) dup @ 1+ swap ! ; ; ans forth word, add n to [A] and store at A : +! ( n A -- ) 2dup @ + swap ! drop ; ; a modulus operator : mod /mod drop ; ; a divide word : / /mod swap drop ; ; type a space : space ( -- ) 32 emit ; ; n spaces : spaces ( n -- ) begin space 1- dup 0 = until drop ; ; a non standard type of comment that reads until a new line : ./ 10 parse drop drop ; imm : count c@+ ; ; classic forth variables. ; The does> is a hack because there is a bug in using create in ; defining words. does> forces the >code point to be updated to 'here' : var create 0 , does> ; imm ; definition of CONSTANT : con create , does> @ ; imm ; the current default domain. This may have to be in bytecode ; because create needs to use it to set the domain var dom 1 dom ! ; a simple 'block' word that just reads block n into first buffer ; in a proper implementation this would use 'buffer' to flush ; code to disk and then read the block. Also, the block would only ; be read if it wasnt already loaded to ram (at 'first') ; ans blk variable var blk : block ( n -- adr ) dup blk ! 2* block1 + 2 first read drop first ; : load ( n -- ) block 1024 source ; ; get the loop counter, needed for thru, called I in ansi forth : ii r> r> r> dup >r swap >r swap >r ; ; "forth inc" word : thru swap do ii load loop ; ;: compile> wparse tick item, ; imm ; lookup a word and return the opcode or zero : >op wparse find opcode ; imm ; use this >op in : defs : [>op] post >op post literal ; imm ; parse, compile and execute words. This is the interpreter. ; called 'quit' in ansi forth : shell begin here>anon [>op] exit c, here>anon term dup accept count in0 in, anon pcall ok again ; 1 24 thru clock shell ; pad remainder of 1st disk block with zeros times 1*1024-($-block1) db 0 ; block2: ; put a space on the stack : bl 32 ; ; type a newline : cr 13 emit 10 emit ; : K 1024 u* ; ; duplicate TOS if non-zero : ?dup dup 0 = if exit else dup fi ; ; logical not! or try xor ... : not ( t/f -- not ) if 0 else 1 fi ; ; ans standard tuck. This can be a handy way to save a value ; for later processing. Could be an opcode : tuck ( a b -- b a b ) swap 2dup drop ; ; get a copy of topmost return stack item. We need to dig ; under the word return pointer : r@ ( -- n ) ( R: n -- n ) r> r> dup >r swap >r ; ; a standardish word to see a decompilation: check! : see wparse tick drop ..word ; imm ; determine the used size of a block by searching for the 1st ; zero byte. Use like this: "first bsize" : bsize dup begin c@+ 0 = until swap - ; ; list all words in the dictionary : ww. last @ begin ; dup .xt space xt+ dup 0 = dup imm? ; make red if immediate word if 12 fg else 11 fg fi dup .xt space xt+ dup 0 = until drop ; ; an alias for ww. : ls ww. ; ; an implementation of the standard forth word 'I' which gets ; a copy of the loop counter onto the data stack. We have to ; 'dig' under the current word return pointer which is the ; current top of the return stack and also under the loop limit ; parameter : ii++ r> r> r> 1+ >r >r >r ; ; use this before 'exiting' from a loop. : unloop r> r> r> drop drop >r ; ; discard loop parameters and continue execution after next 'loop' : leave ( -- ) ; ; Display asci codes of keys pressed ; could recode this to use ekeycode : keycode begin key dup u. space dup emit cr [char] q = until ; ; just testing quote preprocessing by os.sed ; : quote char " char ' char " ; ; deletes the last word in the dictionary by regressing 'last' ; and moving back the >code compile pointer to the end of the ; previous word : del ( -- ) last @ 1- rcount drop 2 - >code ! last @ xt+ last ! ; ; a recursive greatest common divisor word ; from r.v.noble : gcd ( a b -- gcd ) ?dup if tuck mod gcd fi ; ; list all opcodes (=< NOP which is the last in the table) : opcodes [>op] nop 1+ 0 do ii dup u. [char] : emit .code space loop ; ; handy to compile an opcode by name eg: op, fcall ; this can be handy for opcodes that arent normally compiled. ; eg jump, jumpz, rloop etc ; we may also need code that compiles an opcode at runtime : op, wparse find opcode c, ; imm ;: [op,] wparse find opcode literal c, ; imm ; list colours : color 0 begin dup fg space dup u. 1+ dup 33 = until drop ; ; print n coloured blocks : glow ( n -- ) begin dup fg 219 emit 1- dup 1 = until drop ; ; pad remainder of 2nd block with zeros times 2*1024-($-block1) db 0 ; block3 ; 3rd disk block ; parse and find words, more or less working ; : findwords term accept term count in0 ; begin ; wparse 2dup type space find dup u. space ; 0 = ; until ; ; frees a buffer (eg first) and writes old contents to disk if ; necessary. ;: buffer ( n -- adr //to do! ) ; ; breaks out of one level of if/begin etc and jumps back to ; last begin. Preserves jump-back address for 'again' to use later. ; ( comp: jb JB -- jb JB ) ( run: -- ) ; this is not elegant because it fails if there are 2 levels of nesting. : continue [>op] jump c, 2dup drop ./ jb JB jb here 1- ./ jb JB jb here-1 - c, ; imm ; print the stack, source version ; display the items on the data stack without ; altering it. The top (or most recent) item ; is printed rightmost : .S ( -- ) depth 0 = if exit fi depth ./ ... n n begin swap >r depth 1 = until ./ ... n r: .... begin r> dup ./ n a a u. space swap ./ a n 1- dup 0 = ./ a n-1 n-1 until drop ; ; saves the current buffer (first) to block number n on disk : save ( n -- flag=-1/0/1 ) 2* block1 + 2 first write ; ; display all words in buffer ; : inwords term accept term count in0 ; begin ; wparse 2dup type cr ; 0 = ; until ; ; standard s" word : s" [char] " parse post sliteral ; imm ; An fcall will get compiled when ." is run, ; not when it is compiled. This is an important technique ; It is like a "double postpone" ; ; more elaborate versions... ;: ." post s" [ wparse type find ] literal post call, ; imm ;: ." post s" [ ' type ] literal post call, ; imm : ." post s" ' type literal post call, ; imm ; algorithm to convert BCD encoded values to binary (normal) ; binary = ((bcd / 16) * 10) + (bcd & 0xf) ; draw a line starting a pixel position xy of length n ; ( x y n -- ) ; : line 1 do 2dup pix 1+ swap 1+ swap loop ; ; an asci table : asci 1 begin dup u. space dup emit space 1+ dup 255 = until drop ; ; a colourful asci list. call with eg: 4 asc ;: asc 0 ; begin dup fg dup emit 1+ dup 255 = until drop ; ; a convenience to get rid of stuff on the stack : dd drop drop drop ; ; prints out time taken given 2 clock-ticks values (18/sec) on stack ; this has a limit of +/- 52 minutes. takes 2 double values ; ( D:t1 D:t2 -- ) : timer ; get rid of high clock value on stack drop swap drop swap - 55 u* u. ." milliseconds." cr ; ; do not change this word because it is being used to ; measure performance across different machines and different ; exec.x versions. : timeloop ( n -- ) ; juggle number of loops >r clock r> dup u. ." 000 loops" begin 1000 begin 1- dup 1 = until drop 1- dup 1 = until drop ." took " clock timer ; times 3*1024-($-block1) db 0 ; start block4: ; Display keyboard event codes of keys pressed : ekeycode begin ekey 2dup swap u. space u. drop cr [char] q = until ; ; fill u bytes of memory with 0 starting at addr : erase ( addr u -- ) 0 do 0 c!+ loop drop ; ; fill u bytes of memory with byte b starting at addr : fill ( addr u b -- ) swap 0 do dup c!+ loop drop drop ; ; display u bytes of data : dump ( addr u -- ) swap dup u. ." : " swap ./ print mem address 0 do c@+ u. space loop drop ; ; display u words of data : dumpw ( addr u -- ) swap dup u. ." : " swap ./ print mem address 0 do @+ u. space loop drop ; ; just display the current disk geometry parameters which ; will be used for reads and writes from and to disk (usb etc) : geom ( -- ) ." boot drive: " drive c@ u. cr ." sectors per track: " sectors @ u. cr ." sides or platters: " sides @ u. cr ; ; allocate u bytes of data-space, beginning at the next available ; location. Normally used immediately after "create". : allot ( n -- ) here + here! ; ; allot data space and initialize to 0 : allot0 ( n -- ) here swap 2dup erase + here! ; ; convert from n to cells (in this case 2 bytes) : cells ( n -- ) 2* ; ; find out how many bytes are free in source block n : bfree ( n -- free ) block dup begin c@+ 0 = until swap - 1024 swap - swap drop ; : df ( -- ) 16 0 ." Free bytes in disk blocks: " cr do ." block " ii dup u. ." : " bfree u. cr loop ; ; print some machine info ;: mach ( -- ) ; machine count 2dup ." name: " type cr ; + count ." signature: " type cr ; ; print the machine name : mname ( -- ) machine count type ; ; list execution tokens and their names : listxt ( -- ) last @ begin dup u. space dup .xt space xt+ dup 0 = until drop ; times 4*1024-($-block1) db 0 ; block5: ; move vectors, need to rethink ; [ create knight 14 c, 10 c, 25 c, 23 c, -14 c, -10 c, -25 c, -23 c, ] ; create a board data structure and game state : chess: create 0 c, ./ the turn white or black 0 c, ./ the current piece/square pointer ./ here a move vector pointer. ; The board structure, a 12x8 grid of squares ; 33 are the side squares (off the board) used to make finding ; legal moves easier. The actual board is empty ie 0 8 0 do 33 c, 33 c, 0 c, 0 c, 0 c, 0 c, 0 c, 0 c, 0 c, 0 c, 33 c, 33 c, loop ; a move stack ( for depth search ) ; piece vectors eg pawn +8 +7 +9 which gives possible ; legal moves does> 2 + ; imm ; the depth search will probably be recursive. : here+ ( n -- ) here + here! ; ; return the absolute value of n : abs ( n -- ) dup 0 < if 0 swap - fi ; ; A pointer to the board structure, for speed. ; set up a board for a new game. Pieces are just numbers. ; piece values are scaled by 2 for accuracy. eg ; rook=10 bishop=7 knight=6 ... ; can just use allot instead of here+ : cc.new ( A -- ) ; also reset move stack, turn, square/piece vector etc ; white pieces here! 2 here+ 10 c, 6 c, 7 c, 18 c, 100 c, 7 c, 6 c, 10 c, 4 here+ 2 c, 2 c, 2 c, 2 c, 2 c, 2 c, 2 c, 2 c, ; now black, all negative 52 here+ -2 c, -2 c, -2 c, -2 c, -2 c, -2 c, -2 c, -2 c, 4 here+ -10 c, -6 c, -7 c, -18 c, -100 c, -7 c, -6 c, -10 c, ; ; make some visual rep for a piece, try asci chars ... : .piece ( n -- ) dup 0 < if abs 4 fg else 2 fg fi dup 100 = if [char] K emit drop exit fi dup 18 = if [char] Q emit drop exit fi dup 10 = if [char] R emit drop exit fi dup 7 = if [char] B emit drop exit fi dup 6 = if [char] N emit drop exit fi dup 2 = if [char] p emit drop exit fi 15 fg . ; times 5*1024-($-block1) db 0 ; do defined as source. ; : do [>op] >r dup c, c, here ; imm ; need to compile code which checks : ?do [>op] >r dup c, c, here ; imm ; : (?do) ... ; ; try a word that types one line ; eg ltype ( A n -- A' n' ) : Type ( A n -- ) dup 0 = if drop drop exit fi 0 do c@+ emit loop drop ; ; Testing >number by accepting input and ; displaying the number : try.>number begin cr term accept term count dup 0 = if exit fi >number space . space dup u. space . cr again ; ; testing accept and type : try.type begin [char] > emit pad dup accept count dup 0 = if exit fi dup u. [char] : emit type cr again ; ; simplest random : rnd clock drop 16 mod ; : junk 0 do rnd fg 224 rnd + emit loop ; : asc 255 0 do ii dup 32 mod not if cr 15 fg ii u. space fi dup 14 mod 1+ fg emit loop ; ; shows the board. A is a reference to a game or board ; structure : cc.bo ( A -- ) 8 0 do 12 0 do c@+ .piece space loop cr loop ; ; just shows square numbers : cc.sq ( -- ) 96 0 do ii dup 12 mod if u. else cr u. fi space loop ; ; puts a piece P on board at position x y ; eg 0 0 9 game puts a white queen at A1 : cc.put ( x y P A -- ) ; ; show whos turn it is. : cc.turn ( A -- ) c@ if ." black to play" else ." white to play" fi ; ; find the next piece on the board of the ; current colour by scanning the squares. : cc.pp+ ( A -- ) nop ; ; this is the hardest to code. Needs to find the next legal ; move on the board. : cc.move+ ( A -- ) ; ; just for testing chess: C C cc.new ;C cc.bo times 6*1024-($-block1) db 0 ; compute the nth fibonacci using recursion : fib ( n -- fib[n] ) dup 1 swap < if 1- dup 1- fib swap fib + fi ; ; Testing tick by accepting input and ; db ' displaying the found execution token : try.tick begin term accept term count dup 0 = if exit fi tick ./ 0/n/op.xt flag=0/1/2/3 cr .s cr again ; ; the opposite of xt+, finds the previous definition in ; the dictionary, so goes in the opposite direction from ; the linked list. : xt- ( xt -- prev.xt ) last @ ./ xt last begin ./ xt a dup ./ xt a a >r ./ xt a r: a xt+ ./ xt a+ r: a 2dup ./ xt a+ xt a+ r: a = ./ xt a+ T/F r: a if ./ xt a+ r: a drop ./ xt drop ./ r> ./ a exit ./ a fi r> ./ xt a+ a drop ./ xt a+ dup ./ xt a+ a+ 0 = ./ xt a+ true/false until ; ; show the compiled size in bytes of a word : wsize ./ xt dup xt- ./ xt xt- 1- rcount drop ./ xt head 2 - swap ./ head xt - ; ./ head-2-xt times 7*1024-($-block1) db 0 ; the escape keycode : esc ( -- n ) 27 ; ; backspace key : backspace 8 ; ; move cursor up one row (check for zero) : c.up getxy 1- atxy ; ; move cursor down one row : c.down getxy 1+ atxy ; ; move cursor left one (need to check for zero) : c.left getxy swap 1- swap atxy ; ; move cursor right one : c.right getxy swap 1+ swap atxy ; ; move the cursor around with arrow keys : arrow ." arrow keys!! q to exit " cr begin ekey 1 = if dup 72 = if c.up fi dup 75 = if c.left fi dup 77 = if c.right fi dup 80 = if c.down fi drop else [char] q = if exit fi fi again ; ; make an asci box n wide, eg: 7 box ; : ascline begin 196 emit loop : box ( n -- ) 218 emit dup 1 do 196 emit loop 191 emit cr 192 emit dup 1 do 196 emit loop 217 emit ; ; returns the lowest number of "a" and "b" : min ( a b -- min ) 2dup < if drop else swap drop fi ; times 8*1024-($-block1) db 0 ; returns the maximum number of "a" and "b". : max ( a b -- max ) 2dup < if swap drop else drop fi ; ; return true if n=b or n=a otherwise return false : of2 ( n a b -- t/f ) >r >r dup r> = ./ f r: b if r> drop drop -1 exit fi r> = ; ; An editable buffer with capacity, insertion point offset from the ; start of the buffer and pointer to buff ; takes a capacity and pointer to buffer at compile time, and ; returns a pointer to object (capacity) at run time : ed: ( comp: buff* cap -- ) ( run: -- a ) create , ./ name of object and capacity dup , , ./ insertion point = (start) and pointer to buffer does> ; imm ; return capacity : ed.cap ( a -- ) @ ; ; return start buffer address : ed.start ( a -- ) 4 + @ ; ; get last address in buffer : ed.end ( a -- ) dup ed.start ed.cap + ; ; return insertion point : ed.ins ( a -- ) 2 + @ ; ; set insertion point to n : ed.ins! ( n a -- ) 2 + ! ; ; add n to the insertion point : ed.ins+ ( n a -- ) 2 + +! ; ; we need insertion point algebra like line-down line-up ; which will return new buffer points and can be used to ; set the insertion point or for other things (like highlighting) ; buffer character algebra. ; beginning of line containing p or no change if p is already : ed.b : limit ( n a b -- n/a/b ) >r min r> max ; ; display part of the edit buffer with the cursor (at insertion point) ; a different colour : ed.show ( a -- ) 0 4 atxy ./ position cursor dup dup ed.ins ./ a a i 100 - swap ./ a i-100 a ed.start max ./ a max(i-100,start) ;.s cr 300 0 ./ how many chars to show do ./ a A >r dup ed.ins ./ a ins r: A r@ = ./ a t/f r: A if 4 fg r> c@+ emit 2 fg else r> c@+ emit fi loop drop drop ; first 1024 ed: ee times 10*1024-($-block1) db 0 ; edit block number n : edit ( n -- ) block cls 0 1 atxy 80 glow ./ header 25 2 atxy 15 fg ." Editing Block " blk @ u. 2 fg begin ee ed.show ;30 25 atxy ." Edit Buffer: q to quit " ekey 2dup swap 10 23 atxy ." stack: " .s space ." keycode: " u. space u. if ./ k dup 77 = ./ k t/f if 1 ee ed.ins+ fi dup 75 = ./ k t/f if -1 ee ed.ins+ fi fi [char] q = until ; ; get a copy of the loop limit : iimax r> r> dup >r swap >r ; ; a word that types one line : println ( A n -- A' n' /type one line of input) 0 do c@+ dup emit 13 = if ii iimax swap - unloop exit fi loop 0 ; : page ( A n -- /print one page at a time) 10000 0 do println dup 0 = if drop drop unloop exit fi ii 22 mod 21 = if 10 fg ." any key ..." key emit 7 fg fi loop ; ; displays block n on the screen. : list ( n -- /display block n on screen ) block 1024 page ; times 11*1024-($-block1) db 0 ; a simple approx * 10K : pi 31415 ; ; recursive factorial. very succint in forth ; 16bit forth can only hold 8! no more : fact dup 1 = if exit fi dup 1- fact u* ; ; recursive arithmetic series : arith dup 1 = if exit fi dup 1- arith + ; : 10K 10000 ; ; calculate E exponent using infinite series and factorial ; we can only loop upto 8 because this is a 16 bit forth. ; accurate to 2 decimal places. All is scaled by 10K : Ecc 10K 8 1 do 10K ii fact / + loop ; ; exponential e * 10K : E 27182 ; ; e == 1 + 1 + 1/1*2 + 1/1*2*3 + ... ; golden ratio phi (1 + 5^1/2)/2 * 10K : phi 16180 ; ; This is very cool and gets a good approximation within about ; 6 or so iterations. ; gets the next approximation to the square root, from ron geere. ; based on newtons method : approx ( n x -- n x' ) over over / + 2 / ; ; return b, the approximate square root of a (maximum 32767 if the / ; division operator is signed, or else 64K if not), ; this just does the iterations of the "approx" word. : sqrt ( a -- b ) 60 5 0 do approx loop swap drop ; ; test 'find' : try.find listxt begin cr term accept term count dup 0 = if exit fi find space u. cr again ; times 12*1024-($-block1) db 0 ; create a list data type (an array of 16bit cells with length ; and capacity). At run time, put reference on stack to the first ; byte of the data structure : list: ( comp: n -- ) ( run: -- addr ) create dup , 0 , cells allot0 does> ; imm ; addr is a reference to a list: object ; check if the list is full (i.e. size==capacity) : l.full ( addr -- flag=t/f ) @+ swap @ = ; ; check if the list is empty : l.empty ( addr -- flag=t/f ) 2 + @ 0 = ; ; increment list size by 1. But all these words may degrade ; performance. : l.incr ( addr -- ) 2 + dup !1+ ; ; delete last element by decreasing list size : l.del ( addr -- ) 2 + dup ./ a+2 a+2 @ 1- swap ! ; ./ a ; delete all elements, set size to 0 : l.0 ( addr -- ) 2 + 0 swap ! ; ; return the list size : l.size ( addr -- n ) 2 + @ ; ; add a new element to the list and increment size ; use: 5 obj l.add : l.add ( n addr -- ) dup l.full if drop drop exit fi tuck 2 + @+ ./ a n a+4 len 2* + ! ./ store new element. 2 + !1+ ; ./ increment size times 13*1024-($-block1) db 0 ; list elements : l.ls ( addr -- ) dup l.empty if ." list empty" drop exit fi dup l.size ./ a size swap 4 + swap 0 ./ a+4 size 0 do @+ u. space loop drop ; ; list elements as opcodes : l.ops ( addr -- ) dup l.empty if ." list empty" drop exit fi dup l.size ./ a size swap 4 + swap 0 ./ a+4 size 0 do @+ .code space loop drop ; times 14*1024-($-block1) db 0 ; see if list has element n : l.has ( n addr -- flag=t/f ) dup l.empty if drop drop 0 exit fi swap >r ./ a r: n dup r> swap ./ a n a dup l.size ./ a n a size swap 4 + swap 0 ./ a n a+4 size 0 do ./ a n a+4 @+ swap >r ./ a n m r: a+x >r dup r> ./ a n n m r: a+x = if r> drop drop drop 1 unloop exit fi r> ./ a n a+x loop ./ drop drop drop 0 ; ; add an element only if list doesnt already contain element : l.addu ( n addr -- ) 2dup l.has ./ n A t/f if drop drop exit fi l.add ; times 15*1024-($-block1) db 0 ; get a unique list of dependencies for a word ; list: and l.addu to build this unique list ; will have to follow unconditional jumps to their target. ;: op? ; dup 0 < if drop 0 exit fi [>op] nop swap < if 0 exit fi 1 ; 10 list: dep.oo 10 list: dep.cc : deps ( xt -- ) dup dup wsize + swap ./ xt+size xt begin c@+ dup ./ end A+1 n n dep.oo l.addu space dup .code ./ end A+n op dup [>op] fcall = if drop dup 2 + swap ./ end A+n+2 A+n @ ./ get fcall xt ;dup ." <" .xt ." >" ;deps ./ recursive! else args + ./ end A+x fi 2dup < until cr dep.oo l.ops cr dep.oo l.ls ; times 16*1024-($-block1) db 0 : wc ( xt -- n count words, not opcodes from word at xt ) 5000 0 ./ assume <5000 words do ; assume all ops at top of dict, bad assumption! dup opcode if drop ii unloop exit fi xt+ loop drop ; : allwords last @ wc ; ; count words defined in bytecode : bytewords last 4 - wc ; : splash ( -- ) ( produces a colourful splash screen) 31 spaces 8 glow cr 30 spaces 10 glow cr 29 spaces 12 glow cr 28 spaces 14 glow cr 28 spaces 14 glow cr 29 spaces 12 glow cr 30 spaces 10 glow cr 31 spaces 8 glow cr cr 7 fg 23 spaces ." Teatree, A 'Forthish' System " cr cr 13 fg 20 spaces 6 fg ." Machine: " 13 fg mname cr 20 spaces 6 fg ." Opcodes: " 13 fg ' nop literal ' exec literal - u. ." bytes" 10 fg ." ( " [>op] nop u. ." opcodes )" 13 fg cr 20 spaces 6 fg ." Byte codes: " 13 fg ' last literal ' nop literal - u. ." bytes" 10 fg ." ( " bytewords u. ." bytecode words ) " 13 fg cr 20 spaces 6 fg ." Total Size: " 13 fg last @ u. ." bytes" 10 fg ." ( " allwords u. ." total words ) " 13 fg cr ; times 17*1024-($-block1) db 0 ; like accept but with no character limit : gett ( a -- ) dup 1+ dup ./ a a+1 a+1 begin key dup ./ a a+1 a+n k k 13 = if drop swap - swap c! exit fi dup 8 = if ./ backspace ./ start of buffer? drop 2dup < if 8 emit space 8 emit 1- fi continue fi dup emit ./ a a+1 a+n k swap c!+ ./ a a+1 a+n+1 again ; times 18*1024-($-block1) db 0 ; an accept with a limit var accept.max : accept ( a n -- ) accept.max ! dup 1+ dup ./ a a+1 a+1 begin key dup ./ a a+1 a+n k k 13 = if drop swap - swap c! exit fi dup 8 = if ./ start of buffer? drop 2dup < if 8 emit space 8 emit 1- fi else >r 2dup swap - ./ a a+1 a+n diff r: k accept.max @ < if r> dup emit ./ a a+1 a+n k swap c!+ ./ a a+1 a+n+1 else r> drop fi fi again ; times 19*1024-($-block1) db 0 ; an accept with a limit and arrow keys ; This is a work in progress and has some tricky bits to do. ; return true if a=A and b=B : 2= ( a b A B -- t/f ) swap >r = not if r> drop drop 0 exit fi r> = ; ; returns true if ekey is a left arrow : arrow.l ( n 0/1 -- flag ) 75 1 2= ; : arrow.r ( n 0/1 -- flag ) 77 1 2= ; var accept.buff ./ pointer to buffer var accept.max ./ maximum chars var accept.ins ./ insertion point var accept.line ./ display line times 20*1024-($-block1) db 0 ; : accept.show ( A -- ) ; : accepta ( a n -- ) accept.max ! dup accept.buff ! dup accept.ins ! dup 1+ dup ./ a a+1 a+1 begin ekey 2dup ./ a a+1 a+n k 0/1 k 0/1 arrow.l if ." <-" continue fi 2dup arrow.r if ." ->" continue fi if continue fi ./ a a+1 a+n k dup 13 = if drop swap - swap c! exit fi dup 8 = if ./ start of buffer? drop 2dup = not if 8 emit space 8 emit 1- fi else >r 2dup swap - ./ a a+1 a+n diff r: k accept.max @ < if r> dup emit ./ a a+1 a+n k swap c!+ ./ a a+1 a+n+1 else r> drop fi fi again ; times 21*1024-($-block1) db 0 ; ans 94 forth word ; return true != 0 if a <= n < b, otherwise return false==0 ; notice r >r dup r> 1- swap ./ n a-1 n < swap ./ t/f n r> < ./ t/f t/f and ; splash times 22*1024-($-block1) db 0 %endif ; }code ; the %if 0 trick below allows including forth source code without ; hundreds of "db" lines. But this source needs to be preprocessed ; to insert the dbs before every line ; below contains lots of standardish forth words. ; This is just a guide to naming conventions and to avoid name clashes ; with standard words. This is also an attempt to create a "literate" ; style of forth coding, where the explanation and documentation of ; the forth word is clearly visible next to the words definition. %if 0 A word to tell if a character is a digit. : ?digit ( c -- flag ) [char] 0 [char] 9 1+ within words ( -- ) list all the words in the dictionary in search order quit this is the standard forth repl loop. ie read evaluate print loop. It is a strange name and I will call it "shell" instead. The quit loop allows the user to type and execute commands. and compile new words. d. ( d -- ) display the top 2 stack items as a signed double precision integer. d+ d2* d2/ double precision integer operations The high cell is the high order byte of the double number. m* ( n1 n2 -- d ) multiple n1 by n2 and leave double precision result d on stack var creates a new variable : var ( -- ) create we could define colon : with an immediate create. eg [create] : ... source ; : bl 32 ; put a space character on the stack (20H) ; ideas for words of3 ( a b c n -- n/0 ) if n == a or b or c return that value, otherwise return 0. : of3 dup >r = if drop drop r> exit fi r> dup >r = if drop r> exit fi r> dup >r = if r> else r> drop 0 fi ; better to use r@ here, more readable than r> dup >r of2 ( a b n -- n/0 ) if n == a or b return that value, otherwise return 0. : of2 dup >r = if drop r> exit fi r> dup >r = if r> else r> drop 0 fi flags ( -- flag-reg ) put the overflow/carry etc flags register on the stack. I always wonder why standard forths dont have this. How do you know when an overflow or carry occurs? How can you call forth a virtual machine if it does not have this? ; input stream >in ( -- a-addr ) return address containing offset in characters from the start of the current input source to the start of the current parse point. In the current forth, ">in" works differently. Need to think about this design. evaluate ( ... addr u -- ... ) set input source to addr with length u. set >in to zero etc The current forth calls this in0 or setin and works differently A similar word might be "source" or "load". parse ( char -- addr n ) parse text at starting at current parse point ( >in ) using char as the delimiter. Also copy to temp location. But current forth is different. Word is not copied. Also we need a wparse which parses to any whitespace (eg newline is a word delimiter too!). Maybe parse should just update the >in variable as well which should make things simple. ( a comment) brackets for multiline comments a definition of "(" but I may parse bracket comments until a dot . so as to be able to write more "literate" forth. : ( char ) parse/word ... word ( char - c-adr n ) the same as parse but skip all leading occurences of char. My implementation is call WPARSE and works slightly differently (only parses on whitespace) ;*** stack over ( a b -- a b a ) nip ( a b -- b ) rot ( a b c -- b c a ) tuck ( a b -- b a b ) ; characters bl ( -- 20H ) return asci char 32 which is a space. char n ( -- char ) put the asci value of the first character in the next word in the input stream on the stack. In the current forth, this will be an immediate word. Not working in : COLON defs, needs to push a literal In many forths [char] must be used in : definitions. This is because the current forth compiles even words entered interactively. ;*** Strings ; we can implement string functions by compiling right in the ; midst of the current word and compile a JUMP over the string ; if necessary ," compile a string at current position in data-space (at "here") s" compile a string and put its address and length on stack at runtime ." compile a string and print it out at runtime. search find one string in another ; text display space display one space on terminal spaces ( n -- ) display n spaces on terminal cr emit one newline/ carriage return to terminal ; time and date ms ( u -- ) wait for u milliseconds time&date ( -- secs mins hours days months years ) return a structure representing time and date ; double numbers, which are 2 stack item numbers ; high (top) stack position is high order value ; eg D == a b where "b" is high value and a low d+ ( a b c d -- b:a + d:c ) d- d2* d2/ d>s convert double number ; arithmetic : */ ( a b c -- a*b/c). multiply 3TOS by NOS, keeping precision (double?) and then divide by TOS. ; mathematics : min ( a b -- min ) returns the lowest number of "a" and "b" (but signs ?). : max ( a b -- max ) returns the maximum number of "a" and "b". : mod ( m n -- p ) the modulus of m with n. Euclids greatest common divisor (after R.V.Noble) This is amazingly succinct. And recursion works! : gcd ( a b -- gcd) ?dup if tuck mod gcd fi ; ; sum of 1..n arithmetic series ( n -- ); eg 100 asum = 1+2+3...+100 : sum 0 do ii 0 = if 0 fi ii + loop ." sum = " u. ; ; or eg sum = n*(n-1)/2 : sum dup 1- u* 2 / ; The following gives pi to 2 decimal places, scaled by 100 : pi 35500 113 / ; Better, if double arithmetic is available : pi 103993. 33102. / ; pi.approx ( ) pi=4/1-4/3+4/5-4/7+4/9-4/11 an infinite series, but this may take 500000 iterations to get to 4 decimal places In forth we can scale by 10000 to get 5 decimal places See also the inverse tangent function. : 40K 40000 ; : pi.approx ( -- pi-ish) 0 1 begin dup 40K swap / swap >r + r> 2 + dup 40K swap / swap >r + r> 2 + dup 64001 = until drop ; The code above does 64K iterations quickly but only gets to 2 decimal places accurately ie 31407 (3.1407). Also we scale the result by 10K because forth doesnt have floating arithmetic. We really need double arithmetic (32 bit, 2 cells) to get a better approximation ; ans 94 forth word : within ( n a b ) return true != 0 if a <= n < b, otherwise return false==0 notice r min max r> = ;" I have used the reverse names compared to those used by Ron Geere, since "within" is now a fairly standard word with its parameter order. squareroot ( n -- n^0.5 ) use (n/x + x)/2 for successive approximations ; This is very cool and gets a good approximation within about ; 6 or so iterations. : approx ( n x -- n x' ) gets the next approximation to the square root, from ron geere. over over / + 2 / ; : sqrt ( a -- b ) return b, the approximate square root of a (maximum 32767 if the / division operator is signed, or else 64K if not), this just does the iterations of the "approx" word. 60 5 0 do approx loop swap drop ; ;*** execution execute ( ... xt -- ... ) execute word specified at address xt. Called 'pcall' in this forth recurse ( -- ) append execution behaviour to current definition to allow for recursive functions. In this implementation words can just call themselves eg: : gcd ( a b -- gcd) ?dup if tuck mod gcd fi ; ; dictionary and defining words state ( -- addr ) return address of state variable (either compile or run...) This is a way to implement the [ ] words which in this version of forth will make all words immediate when between "[" and "]". [ ( -- ) Enter "immediate" state. All words are executed, not compiled In standard forths, these are called "compile" and "interpret" states. ] leave "immediate" state. All words are compiled, unless they are marked immediate in their control bits (1st bit of the name count). literal ( n -- ) compile value n so that value n is put on the data stack at run-time. In this forth there is also and opcode LITERAL which could be confusing although they do almost the same thing. compile, ( xt -- ) append execution behaviour of xt to current definition. But because I am using opcodes as well, I have modified this word. Was going to call "item," because my version also compiles literal numbers. "opcode," is just "c," : unused ( -- u ) return number of bytes of memory remaining for new dictionary entries. : free first here - . ." bytes cr ; an older name for "unused". where first is the address of the first disk buffer. >body ( xt -- addr ) given execution token for a word, return the start of the parameter field (which is just after code). But current forth doesnt maintain a link to parameter field normally. : ' ( -- xt ) search dictionary for name and return execution token example: A table of function pointers [create] buttons ' start , ' slow , ' fast , ' finish : allot ( n -- ) allocate u bytes of data-space, beginning at the next available location. Normally used immediately after "create". here + here! ; create compile "name" in the dictionary and put address on stack at runtime (no data space is allocated). Because all words are first compiled in this forth we need an immediate version [create] to use this outside of a : definition. Or use [ create ] name ... This was called " really important word. Allows the creation of new defining words (eg "constant") that have a special behaviour. Needs to be used with create. Implentation is slightly tricky because no room in new word code field for a fcall to the does> code. So need to code a jump after the parameter field. Compiles a call to following words in defined words. * create a constant defining word : constant create 1 allot does> @ ; buffer: ( n -- ) create a dictionary entry for "name" associated with n bytes of data-space. create allot ; dump ( adr +n -- ) display the contents of a memory region of length n. print address on left and 8 values per line in hex values or current base. ;*** blocks and buffers list ( n -- ) display the contents of disk block "n" (1K of source code). blocks are good because no file system is required. buffer ( n -- addr ) return address where block n may be loaded. No read is done. A disk write is done if necessary. Buffer manages a list of buffers and blocks and finds a suitable place to put the requested disk block, but it doesnt actually read it into memory. block ( n -- addr ) return buffer address containing disk block n. Data is read if necessary. No write is done update ( -- ) mark the last disk block accessed as having changed data (needs to be written to disk). This could be stored in the last byte of the block itself load ( n -- ) load disk block (1024bytes) "n" and interpret the contents of the block as forth source code. flush ensure all updated buffers are written to disk and free all buffers. // create a dictionary entry for name associated with 1 cell // eg: variable data 6 data ! : variable ( -- ) create 1 allot ; immediate // create a new variable and assign a value to it. : value ( n -- ) create , ; immediate : to ( newval to name ) // a constant value : constant create , does> @ ; immediate %endif ; 0