2 \ A sometimes minimal FORTH compiler and tutorial for Linux / i386 systems. -*- asm -*-
3 \ By Richard W.M. Jones <rich@annexia.org> http://annexia.org/forth
4 \ This is PUBLIC DOMAIN (see public domain release statement below).
5 \ $Id: jonesforth.f,v 1.15 2007-10-11 07:39:51 rich Exp $
7 \ The first part of this tutorial is in jonesforth.S. Get if from http://annexia.org/forth
9 \ PUBLIC DOMAIN ----------------------------------------------------------------------
11 \ I, the copyright holder of this work, hereby release it into the public domain. This applies worldwide.
13 \ In case this is not legally possible, I grant any entity the right to use this work for any purpose,
14 \ without any conditions, unless such conditions are required by law.
16 \ SETTING UP ----------------------------------------------------------------------
18 \ Let's get a few housekeeping things out of the way. Firstly because I need to draw lots of
19 \ ASCII-art diagrams to explain concepts, the best way to look at this is using a window which
20 \ uses a fixed width font and is at least this wide:
22 \<------------------------------------------------------------------------------------------------------------------------>
24 \ Secondly make sure TABS are set to 8 characters. The following should be a vertical
25 \ line. If not, sort out your tabs.
31 \ Thirdly I assume that your screen is at least 50 characters high.
33 \ START OF FORTH CODE ----------------------------------------------------------------------
35 \ We've now reached the stage where the FORTH system is running and self-hosting. All further
36 \ words can be written as FORTH itself, including words like IF, THEN, .", etc which in most
37 \ languages would be considered rather fundamental.
39 \ Some notes about the code:
41 \ I use indenting to show structure. The amount of whitespace has no meaning to FORTH however
42 \ except that you must use at least one whitespace character between words, and words themselves
43 \ cannot contain whitespace.
45 \ FORTH is case-sensitive. Use capslock!
47 \ The primitive word /MOD (DIVMOD) leaves both the quotient and the remainder on the stack. (On
48 \ i386, the idivl instruction gives both anyway). Now we can define the / and MOD in terms of /MOD
49 \ and a few other primitives.
53 \ Define some character constants
55 : BL 32 ; \ BL (BLank) is a standard FORTH word for space.
57 \ CR prints a carriage return
60 \ SPACE prints a space
63 \ NEGATE leaves the negative of a number on the stack.
66 \ Standard words for booleans.
71 \ LITERAL takes whatever is on the stack and compiles LIT <foo>
74 , \ compile the literal itself (from the stack)
77 \ Now we can use [ and ] to insert literals which are calculated at compile time. (Recall that
78 \ [ and ] are the FORTH words which switch into and out of immediate mode.)
79 \ Within definitions, use [ ... ] LITERAL anywhere that '...' is a constant expression which you
80 \ would rather only compute once (at compile time, rather than calculating it each time your word runs).
82 [ \ go into immediate mode (temporarily)
83 CHAR : \ push the number 58 (ASCII code of colon) on the parameter stack
84 ] \ go back to compile mode
85 LITERAL \ compile LIT 58 as the definition of ':' word
88 \ A few more character constants defined the same way as above.
89 : ';' [ CHAR ; ] LITERAL ;
90 : '(' [ CHAR ( ] LITERAL ;
91 : ')' [ CHAR ) ] LITERAL ;
92 : '"' [ CHAR " ] LITERAL ;
93 : 'A' [ CHAR A ] LITERAL ;
94 : '0' [ CHAR 0 ] LITERAL ;
95 : '-' [ CHAR - ] LITERAL ;
96 : '.' [ CHAR . ] LITERAL ;
98 \ While compiling, '[COMPILE] word' compiles 'word' if it would otherwise be IMMEDIATE.
100 WORD \ get the next word
101 FIND \ find it in the dictionary
102 >CFA \ get its codeword
106 \ RECURSE makes a recursive call to the current word that is being compiled.
108 \ Normally while a word is being compiled, it is marked HIDDEN so that references to the
109 \ same word within are calls to the previous definition of the word. However we still have
110 \ access to the word which we are currently compiling through the LATEST pointer so we
111 \ can use that to compile a recursive call.
113 LATEST @ \ LATEST points to the word being compiled at the moment
114 >CFA \ get the codeword
118 \ CONTROL STRUCTURES ----------------------------------------------------------------------
120 \ So far we have defined only very simple definitions. Before we can go further, we really need to
121 \ make some control structures, like IF ... THEN and loops. Luckily we can define arbitrary control
122 \ structures directly in FORTH.
124 \ Please note that the control structures as I have defined them here will only work inside compiled
125 \ words. If you try to type in expressions using IF, etc. in immediate mode, then they won't work.
126 \ Making these work in immediate mode is left as an exercise for the reader.
128 \ condition IF true-part THEN rest
129 \ -- compiles to: --> condition 0BRANCH OFFSET true-part rest
130 \ where OFFSET is the offset of 'rest'
131 \ condition IF true-part ELSE false-part THEN
132 \ -- compiles to: --> condition 0BRANCH OFFSET true-part BRANCH OFFSET2 false-part rest
133 \ where OFFSET if the offset of false-part and OFFSET2 is the offset of rest
135 \ IF is an IMMEDIATE word which compiles 0BRANCH followed by a dummy offset, and places
136 \ the address of the 0BRANCH on the stack. Later when we see THEN, we pop that address
137 \ off the stack, calculate the offset, and back-fill the offset.
139 ' 0BRANCH , \ compile 0BRANCH
140 HERE @ \ save location of the offset on the stack
141 0 , \ compile a dummy offset
146 HERE @ SWAP - \ calculate the offset from the address saved on the stack
147 SWAP ! \ store the offset in the back-filled location
151 ' BRANCH , \ definite branch to just over the false-part
152 HERE @ \ save location of the offset on the stack
153 0 , \ compile a dummy offset
154 SWAP \ now back-fill the original (IF) offset
155 DUP \ same as for THEN word above
160 \ BEGIN loop-part condition UNTIL
161 \ -- compiles to: --> loop-part condition 0BRANCH OFFSET
162 \ where OFFSET points back to the loop-part
163 \ This is like do { loop-part } while (condition) in the C language
165 HERE @ \ save location on the stack
169 ' 0BRANCH , \ compile 0BRANCH
170 HERE @ - \ calculate the offset from the address saved on the stack
171 , \ compile the offset here
174 \ BEGIN loop-part AGAIN
175 \ -- compiles to: --> loop-part BRANCH OFFSET
176 \ where OFFSET points back to the loop-part
177 \ In other words, an infinite loop which can only be returned from with EXIT
179 ' BRANCH , \ compile BRANCH
180 HERE @ - \ calculate the offset back
181 , \ compile the offset here
184 \ BEGIN condition WHILE loop-part REPEAT
185 \ -- compiles to: --> condition 0BRANCH OFFSET2 loop-part BRANCH OFFSET
186 \ where OFFSET points back to condition (the beginning) and OFFSET2 points to after the whole piece of code
187 \ So this is like a while (condition) { loop-part } loop in the C language
189 ' 0BRANCH , \ compile 0BRANCH
190 HERE @ \ save location of the offset2 on the stack
191 0 , \ compile a dummy offset2
195 ' BRANCH , \ compile BRANCH
196 SWAP \ get the original offset (from BEGIN)
197 HERE @ - , \ and compile it after BRANCH
199 HERE @ SWAP - \ calculate the offset2
200 SWAP ! \ and back-fill it in the original location
203 \ UNLESS is the same as IF but the test is reversed.
205 \ Note the use of [COMPILE]: Since IF is IMMEDIATE we don't want it to be executed while UNLESS
206 \ is compiling, but while UNLESS is running (which happens to be when whatever word using UNLESS is
207 \ being compiled -- whew!). So we use [COMPILE] to reverse the effect of marking IF as immediate.
208 \ This trick is generally used when we want to write our own control words without having to
209 \ implement them all in terms of the primitives 0BRANCH and BRANCH, but instead reusing simpler
210 \ control words like (in this instance) IF.
212 ' NOT , \ compile NOT (to reverse the test)
213 [COMPILE] IF \ continue by calling the normal IF
216 \ COMMENTS ----------------------------------------------------------------------
218 \ FORTH allows ( ... ) as comments within function definitions. This works by having an IMMEDIATE
219 \ word called ( which just drops input characters until it hits the corresponding ).
221 1 \ allowed nested parens by keeping track of depth
223 KEY \ read next character
224 DUP '(' = IF \ open paren?
225 DROP \ drop the open paren
228 ')' = IF \ close paren?
232 DUP 0= UNTIL \ continue until we reach matching close paren, depth 0
233 DROP \ drop the depth counter
237 From now on we can use ( ... ) for comments.
239 STACK NOTATION ----------------------------------------------------------------------
241 In FORTH style we can also use ( ... -- ... ) to show the effects that a word has on the
242 parameter stack. For example:
244 ( n -- ) means that the word consumes an integer (n) from the parameter stack.
245 ( b a -- c ) means that the word uses two integers (a and b, where a is at the top of stack)
246 and returns a single integer (c).
247 ( -- ) means the word has no effect on the stack
250 ( Some more complicated stack examples, showing the stack notation. )
251 : NIP ( x y -- y ) SWAP DROP ;
252 : TUCK ( x y -- y x y ) DUP ROT ;
253 : PICK ( x_u ... x_1 x_0 u -- x_u ... x_1 x_0 x_u )
254 1+ ( add one because of 'u' on the stack )
255 4 * ( multiply by the word size )
256 DSP@ + ( add to the stack pointer )
260 ( With the looping constructs, we can now write SPACES, which writes n spaces to stdout. )
263 DUP 0> ( while n > 0 )
265 SPACE ( print a space )
266 1- ( until we count down to 0 )
271 ( Standard words for manipulating BASE. )
272 : DECIMAL ( -- ) 10 BASE ! ;
273 : HEX ( -- ) 16 BASE ! ;
276 PRINTING NUMBERS ----------------------------------------------------------------------
278 The standard FORTH word . (DOT) is very important. It takes the number at the top
279 of the stack and prints it out. However first I'm going to implement some lower-level
282 U.R ( u width -- ) which prints an unsigned number, padded to a certain width
283 U. ( u -- ) which prints an unsigned number
284 .R ( n width -- ) which prints a signed number, padded to a certain width.
288 will print out these characters:
289 <space> <space> - 1 2 3
291 In other words, the number padded left to a certain number of characters.
293 The full number is printed even if it is wider than width, and this is what allows us to
294 define the ordinary functions U. and . (we just set width to zero knowing that the full
295 number will be printed anyway).
297 Another wrinkle of . and friends is that they obey the current base in the variable BASE.
298 BASE can be anything in the range 2 to 36.
300 While we're defining . &c we can also define .S which is a useful debugging tool. This
301 word prints the current stack (non-destructively) from top to bottom.
304 ( This is the underlying recursive definition of U. )
306 BASE @ /MOD ( width rem quot )
307 ?DUP IF ( if quotient <> 0 then )
308 RECURSE ( print the quotient )
311 ( print the remainder )
313 '0' ( decimal digits 0..9 )
315 10 - ( hex and beyond digits A..Z )
323 FORTH word .S prints the contents of the stack. It doesn't alter the stack.
324 Very useful for debugging.
327 DSP@ ( get current stack pointer )
331 DUP @ U. ( print the stack element )
338 ( This word returns the width (in characters) of an unsigned number in the current base )
339 : UWIDTH ( u -- width )
340 BASE @ / ( rem quot )
341 ?DUP IF ( if quotient <> 0 then )
342 RECURSE 1+ ( return 1+recursive call )
351 UWIDTH ( width u uwidth )
352 -ROT ( u uwidth width )
353 SWAP - ( u width-uwidth )
354 ( At this point if the requested width is narrower, we'll have a negative number on the stack.
355 Otherwise the number on the stack is the number of spaces to print. But SPACES won't print
356 a negative number of spaces anyway, so it's now safe to call SPACES ... )
358 ( ... and then call the underlying implementation of U. )
363 .R prints a signed number, padded to a certain width. We can't just print the sign
364 and call U.R because we want the sign to be next to the number ('-123' instead of '- 123').
370 1 ( save a flag to remember that it was negative | width n 1 )
379 SWAP ( flag width u )
380 DUP ( flag width u u )
381 UWIDTH ( flag width u uwidth )
382 -ROT ( flag u uwidth width )
383 SWAP - ( flag u width-uwidth )
388 IF ( was it negative? print the - character )
395 ( Finally we can define word . in terms of .R, with a trailing space. )
398 ( The real U., note the trailing space. )
401 ( ? fetches the integer at an address and prints it. )
402 : ? ( addr -- ) @ . ;
404 ( c a b WITHIN returns true if a <= c and c < b )
420 ( DEPTH returns the depth of the stack. )
423 4- ( adjust because S0 was on the stack when we pushed DSP )
427 ALIGNED takes an address and rounds it up (aligns it) to the next 4 byte boundary.
429 : ALIGNED ( addr -- addr )
430 3 + 3 INVERT AND ( (addr+3) & ~3 )
434 ALIGN aligns the HERE pointer, so the next word appended will be aligned properly.
436 : ALIGN HERE @ ALIGNED HERE ! ;
439 STRINGS ----------------------------------------------------------------------
441 S" string" is used in FORTH to define strings. It leaves the address of the string and
442 its length on the stack, (length at the top of stack). The space following S" is the normal
443 space between FORTH words and is not a part of the string.
445 This is tricky to define because it has to do different things depending on whether
446 we are compiling or in immediate mode. (Thus the word is marked IMMEDIATE so it can
447 detect this and do different things).
449 In compile mode we append
450 LITSTRING <string length> <string rounded up 4 bytes>
451 to the current word. The primitive LITSTRING does the right thing when the current
454 In immediate mode there isn't a particularly good place to put the string, but in this
455 case we put the string at HERE (but we _don't_ change HERE). This is meant as a temporary
456 location, likely to be overwritten soon after.
458 ( C, appends a byte to the current compiled word. )
460 HERE @ C! ( store the character in the compiled image )
461 1 HERE +! ( increment HERE pointer by 1 byte )
464 : S" IMMEDIATE ( -- addr len )
465 STATE @ IF ( compiling? )
466 ' LITSTRING , ( compile LITSTRING )
467 HERE @ ( save the address of the length word on the stack )
468 0 , ( dummy length - we don't know what it is yet )
470 KEY ( get next character of the string )
473 C, ( copy character )
475 DROP ( drop the double quote character at the end )
476 DUP ( get the saved address of the length word )
477 HERE @ SWAP - ( calculate the length )
478 4- ( subtract 4 (because we measured from the start of the length word) )
479 SWAP ! ( and back-fill the length location )
480 ALIGN ( round up to next multiple of 4 bytes for the remaining code )
481 ELSE ( immediate mode )
482 HERE @ ( get the start address of the temporary space )
487 OVER C! ( save next character )
488 1+ ( increment address )
490 DROP ( drop the final " character )
491 HERE @ - ( calculate the length )
492 HERE @ ( push the start address )
498 ." is the print string operator in FORTH. Example: ." Something to print"
499 The space after the operator is the ordinary space required between words and is not
500 a part of what is printed.
502 In immediate mode we just keep reading characters and printing them until we get to
503 the next double quote.
505 In compile mode we use S" to store the string, then add TELL afterwards:
506 LITSTRING <string length> <string rounded up to 4 bytes> TELL
508 It may be interesting to note the use of [COMPILE] to turn the call to the immediate
509 word S" into compilation of that word. It compiles it into the definition of .",
510 not into the definition of the word being compiled when this is running (complicated
513 : ." IMMEDIATE ( -- )
514 STATE @ IF ( compiling? )
515 [COMPILE] S" ( read the string, and compile LITSTRING, etc. )
516 ' TELL , ( compile the final TELL )
518 ( In immediate mode, just read characters and print them until we get
519 to the ending double quote. )
523 DROP ( drop the double quote character )
524 EXIT ( return from this function )
532 CONSTANTS AND VARIABLES ----------------------------------------------------------------------
534 In FORTH, global constants and variables are defined like this:
536 10 CONSTANT TEN when TEN is executed, it leaves the integer 10 on the stack
537 VARIABLE VAR when VAR is executed, it leaves the address of VAR on the stack
539 Constants can be read but not written, eg:
543 You can read a variable (in this example called VAR) by doing:
545 VAR @ leaves the value of VAR on the stack
546 VAR @ . CR prints the value of VAR
547 VAR ? CR same as above, since ? is the same as @ .
549 and update the variable by doing:
551 20 VAR ! sets VAR to 20
553 Note that variables are uninitialised (but see VALUE later on which provides initialised
554 variables with a slightly simpler syntax).
556 How can we define the words CONSTANT and VARIABLE?
558 The trick is to define a new word for the variable itself (eg. if the variable was called
559 'VAR' then we would define a new word called VAR). This is easy to do because we exposed
560 dictionary entry creation through the CREATE word (part of the definition of : above).
561 A call to WORD [TEN] CREATE (where [TEN] means that "TEN" is the next word in the input)
562 leaves the dictionary entry:
567 +---------+---+---+---+---+
568 | LINK | 3 | T | E | N |
569 +---------+---+---+---+---+
572 For CONSTANT we can continue by appending DOCOL (the codeword), then LIT followed by
573 the constant itself and then EXIT, forming a little word definition that returns the
576 +---------+---+---+---+---+------------+------------+------------+------------+
577 | LINK | 3 | T | E | N | DOCOL | LIT | 10 | EXIT |
578 +---------+---+---+---+---+------------+------------+------------+------------+
581 Notice that this word definition is exactly the same as you would have got if you had
584 Note for people reading the code below: DOCOL is a constant word which we defined in the
585 assembler part which returns the value of the assembler symbol of the same name.
588 WORD ( get the name (the name follows CONSTANT) )
589 CREATE ( make the dictionary entry )
590 DOCOL , ( append DOCOL (the codeword field of this word) )
591 ' LIT , ( append the codeword LIT )
592 , ( append the value on the top of the stack )
593 ' EXIT , ( append the codeword EXIT )
597 VARIABLE is a little bit harder because we need somewhere to put the variable. There is
598 nothing particularly special about the user memory (the area of memory pointed to by HERE
599 where we have previously just stored new word definitions). We can slice off bits of this
600 memory area to store anything we want, so one possible definition of VARIABLE might create
603 +--------------------------------------------------------------+
606 +---------+---------+---+---+---+---+------------+------------+---|--------+------------+
607 | <var> | LINK | 3 | V | A | R | DOCOL | LIT | <addr var> | EXIT |
608 +---------+---------+---+---+---+---+------------+------------+------------+------------+
611 where <var> is the place to store the variable, and <addr var> points back to it.
613 To make this more general let's define a couple of words which we can use to allocate
614 arbitrary memory from the user memory.
616 First ALLOT, where n ALLOT allocates n bytes of memory. (Note when calling this that
617 it's a very good idea to make sure that n is a multiple of 4, or at least that next time
618 a word is compiled that HERE has been left as a multiple of 4).
620 : ALLOT ( n -- addr )
621 HERE @ SWAP ( here n )
622 HERE +! ( adds n to HERE, after this the old value of HERE is still on the stack )
626 Second, CELLS. In FORTH the phrase 'n CELLS ALLOT' means allocate n integers of whatever size
627 is the natural size for integers on this machine architecture. On this 32 bit machine therefore
628 CELLS just multiplies the top of stack by 4.
630 : CELLS ( n -- n ) 4 * ;
633 So now we can define VARIABLE easily in much the same way as CONSTANT above. Refer to the
634 diagram above to see what the word that this creates will look like.
637 1 CELLS ALLOT ( allocate 1 cell of memory, push the pointer to this memory )
638 WORD CREATE ( make the dictionary entry (the name follows VARIABLE) )
639 DOCOL , ( append DOCOL (the codeword field of this word) )
640 ' LIT , ( append the codeword LIT )
641 , ( append the pointer to the new memory )
642 ' EXIT , ( append the codeword EXIT )
646 VALUES ----------------------------------------------------------------------
648 VALUEs are like VARIABLEs but with a simpler syntax. You would generally use them when you
649 want a variable which is read often, and written infrequently.
651 20 VALUE VAL creates VAL with initial value 20
652 VAL pushes the value (20) directly on the stack
653 30 TO VAL updates VAL, setting it to 30
654 VAL pushes the value (30) directly on the stack
656 Notice that 'VAL' on its own doesn't return the address of the value, but the value itself,
657 making values simpler and more obvious to use than variables (no indirection through '@').
658 The price is a more complicated implementation, although despite the complexity there is no
659 performance penalty at runtime.
661 A naive implementation of 'TO' would be quite slow, involving a dictionary search each time.
662 But because this is FORTH we have complete control of the compiler so we can compile TO more
663 efficiently, turning:
667 and calculating <addr> (the address of the value) at compile time.
669 Now this is the clever bit. We'll compile our value like this:
671 +---------+---+---+---+---+------------+------------+------------+------------+
672 | LINK | 3 | V | A | L | DOCOL | LIT | <value> | EXIT |
673 +---------+---+---+---+---+------------+------------+------------+------------+
676 where <value> is the actual value itself. Note that when VAL executes, it will push the
677 value on the stack, which is what we want.
679 But what will TO use for the address <addr>? Why of course a pointer to that <value>:
681 code compiled - - - - --+------------+------------+------------+-- - - - -
682 by TO VAL | LIT | <addr> | ! |
683 - - - - --+------------+-----|------+------------+-- - - - -
686 +---------+---+---+---+---+------------+------------+------------+------------+
687 | LINK | 3 | V | A | L | DOCOL | LIT | <value> | EXIT |
688 +---------+---+---+---+---+------------+------------+------------+------------+
691 In other words, this is a kind of self-modifying code.
693 (Note to the people who want to modify this FORTH to add inlining: values defined this
694 way cannot be inlined).
697 WORD CREATE ( make the dictionary entry (the name follows VALUE) )
698 DOCOL , ( append DOCOL )
699 ' LIT , ( append the codeword LIT )
700 , ( append the initial value )
701 ' EXIT , ( append the codeword EXIT )
704 : TO IMMEDIATE ( n -- )
705 WORD ( get the name of the value )
706 FIND ( look it up in the dictionary )
707 >DFA ( get a pointer to the first data field (the 'LIT') )
708 4+ ( increment to point at the value )
709 STATE @ IF ( compiling? )
710 ' LIT , ( compile LIT )
711 , ( compile the address of the value )
713 ELSE ( immediate mode )
714 ! ( update it straightaway )
718 ( x +TO VAL adds x to VAL )
720 WORD ( get the name of the value )
721 FIND ( look it up in the dictionary )
722 >DFA ( get a pointer to the first data field (the 'LIT') )
723 4+ ( increment to point at the value )
724 STATE @ IF ( compiling? )
725 ' LIT , ( compile LIT )
726 , ( compile the address of the value )
727 ' +! , ( compile +! )
728 ELSE ( immediate mode )
729 +! ( update it straightaway )
734 PRINTING THE DICTIONARY ----------------------------------------------------------------------
736 ID. takes an address of a dictionary entry and prints the word's name.
738 For example: LATEST @ ID. would print the name of the last word that was defined.
741 4+ ( skip over the link pointer )
742 DUP C@ ( get the flags/length byte )
743 F_LENMASK AND ( mask out the flags - just want the length )
746 DUP 0> ( length > 0? )
748 SWAP 1+ ( addr len -- len addr+1 )
749 DUP C@ ( len addr -- len addr char | get the next character)
750 EMIT ( len addr char -- len addr | and print it)
751 SWAP 1- ( len addr -- addr len-1 | subtract one from length )
753 2DROP ( len addr -- )
757 'WORD word FIND ?HIDDEN' returns true if 'word' is flagged as hidden.
759 'WORD word FIND ?IMMEDIATE' returns true if 'word' is flagged as immediate.
762 4+ ( skip over the link pointer )
763 C@ ( get the flags/length byte )
764 F_HIDDEN AND ( mask the F_HIDDEN flag and return it (as a truth value) )
767 4+ ( skip over the link pointer )
768 C@ ( get the flags/length byte )
769 F_IMMED AND ( mask the F_IMMED flag and return it (as a truth value) )
773 WORDS prints all the words defined in the dictionary, starting with the word defined most recently.
774 However it doesn't print hidden words.
776 The implementation simply iterates backwards from LATEST using the link pointers.
779 LATEST @ ( start at LATEST dictionary entry )
781 ?DUP ( while link pointer is not null )
783 DUP ?HIDDEN NOT IF ( ignore hidden words )
784 DUP ID. ( but if not hidden, print the word )
787 @ ( dereference the link pointer - go to previous word )
793 FORGET ----------------------------------------------------------------------
795 So far we have only allocated words and memory. FORTH provides a rather primitive method
798 'FORGET word' deletes the definition of 'word' from the dictionary and everything defined
799 after it, including any variables and other memory allocated after.
801 The implementation is very simple - we look up the word (which returns the dictionary entry
802 address). Then we set HERE to point to that address, so in effect all future allocations
803 and definitions will overwrite memory starting at the word. We also need to set LATEST to
804 point to the previous word.
806 Note that you cannot FORGET built-in words (well, you can try but it will probably cause
809 XXX: Because we wrote VARIABLE to store the variable in memory allocated before the word,
810 in the current implementation VARIABLE FOO FORGET FOO will leak 1 cell of memory.
813 WORD FIND ( find the word, gets the dictionary entry address )
814 DUP @ LATEST ! ( set LATEST to point to the previous word )
815 HERE ! ( and store HERE with the dictionary address )
819 DUMP ----------------------------------------------------------------------
821 DUMP is used to dump out the contents of memory, in the 'traditional' hexdump format.
823 Notice that the parameters to DUMP (address, length) are compatible with string words
826 : DUMP ( addr len -- )
827 BASE @ ROT ( save the current BASE at the bottom of the stack )
828 HEX ( and switch to hexadecimal mode )
831 ?DUP ( while len > 0 )
833 OVER 8 U.R ( print the address )
836 ( print up to 16 words on this line )
837 2DUP ( addr len addr len )
838 1- 15 AND 1+ ( addr len addr linelen )
840 ?DUP ( while linelen > 0 )
842 SWAP ( addr len linelen addr )
843 DUP C@ ( addr len linelen addr byte )
844 2 .R SPACE ( print the byte )
845 1+ SWAP 1- ( addr len linelen addr -- addr len addr+1 linelen-1 )
849 ( print the ASCII equivalents )
850 2DUP 1- 15 AND 1+ ( addr len addr linelen )
852 ?DUP ( while linelen > 0)
854 SWAP ( addr len linelen addr )
855 DUP C@ ( addr len linelen addr byte )
856 DUP 32 128 WITHIN IF ( 32 <= c < 128? )
861 1+ SWAP 1- ( addr len linelen addr -- addr len addr+1 linelen-1 )
866 DUP 1- 15 AND 1+ ( addr len linelen )
867 DUP ( addr len linelen linelen )
868 ROT ( addr linelen len linelen )
869 - ( addr linelen len-linelen )
870 ROT ( len-linelen addr linelen )
871 + ( len-linelen addr+linelen )
872 SWAP ( addr-linelen len-linelen )
875 DROP ( restore stack )
876 BASE ! ( restore saved BASE )
880 CASE ----------------------------------------------------------------------
882 CASE...ENDCASE is how we do switch statements in FORTH. There is no generally
883 agreed syntax for this, so I've gone for the syntax mandated by the ISO standard
886 ( some value on the stack )
894 The CASE statement tests the value on the stack by comparing it for equality with
895 test1, test2, ..., testn and executes the matching piece of code within OF ... ENDOF.
896 If none of the test values match then the default case is executed. Inside the ... of
897 the default case, the value is still at the top of stack (it is implicitly DROP-ed
898 by ENDCASE). When ENDOF is executed it jumps after ENDCASE (ie. there is no "fall-through"
899 and no need for a break statement like in C).
901 The default case may be omitted. In fact the tests may also be omitted so that you
902 just have a default case, although this is probably not very useful.
904 An example (assuming that 'q', etc. are words which push the ASCII value of the letter
910 'q' OF 1 TO QUIT ENDOF
911 's' OF 1 TO SLEEP ENDOF
913 ." Sorry, I didn't understand key <" DUP EMIT ." >, try again." CR
916 (In some versions of FORTH, more advanced tests are supported, such as ranges, etc.
917 Other versions of FORTH need you to write OTHERWISE to indicate the default case.
918 As I said above, this FORTH tries to follow the ANS FORTH standard).
920 The implementation of CASE...ENDCASE is somewhat non-trivial. I'm following the
921 implementations from here:
922 http://www.uni-giessen.de/faq/archiv/forthfaq.case_endcase/msg00000.html
924 The general plan is to compile the code as a series of IF statements:
926 CASE (push 0 on the immediate-mode parameter stack)
927 test1 OF ... ENDOF test1 OVER = IF DROP ... ELSE
928 test2 OF ... ENDOF test2 OVER = IF DROP ... ELSE
929 testn OF ... ENDOF testn OVER = IF DROP ... ELSE
930 ... ( default case ) ...
931 ENDCASE DROP THEN [THEN [THEN ...]]
933 The CASE statement pushes 0 on the immediate-mode parameter stack, and that number
934 is used to count how many THEN statements we need when we get to ENDCASE so that each
935 IF has a matching THEN. The counting is done implicitly. If you recall from the
936 implementation above of IF, each IF pushes a code address on the immediate-mode stack,
937 and these addresses are non-zero, so by the time we get to ENDCASE the stack contains
938 some number of non-zeroes, followed by a zero. The number of non-zeroes is how many
939 times IF has been called, so how many times we need to match it with THEN.
941 This code uses [COMPILE] so that we compile calls to IF, ELSE, THEN instead of
942 actually calling them while we're compiling the words below.
944 As is the case with all of our control structures, they only work within word
945 definitions, not in immediate mode.
948 0 ( push 0 to mark the bottom of the stack )
952 ' OVER , ( compile OVER )
954 [COMPILE] IF ( compile IF )
955 ' DROP , ( compile DROP )
959 [COMPILE] ELSE ( ENDOF is the same as ELSE )
963 ' DROP , ( compile DROP )
965 ( keep compiling THEN until we get to our zero marker )
974 DECOMPILER ----------------------------------------------------------------------
976 CFA> is the opposite of >CFA. It takes a codeword and tries to find the matching
977 dictionary definition. (In truth, it works with any pointer into a word, not just
978 the codeword pointer, and this is needed to do stack traces).
980 In this FORTH this is not so easy. In fact we have to search through the dictionary
981 because we don't have a convenient back-pointer (as is often the case in other versions
982 of FORTH). Because of this search, CFA> should not be used when performance is critical,
983 so it is only used for debugging tools such as the decompiler and printing stack
986 This word returns 0 if it doesn't find a match.
989 LATEST @ ( start at LATEST dictionary entry )
991 ?DUP ( while link pointer is not null )
993 2DUP SWAP ( cfa curr curr cfa )
994 < IF ( current dictionary entry < cfa? )
995 NIP ( leave curr dictionary entry on the stack )
998 @ ( follow link pointer back )
1000 DROP ( restore stack )
1001 0 ( sorry, nothing found )
1005 SEE decompiles a FORTH word.
1007 We search for the dictionary entry of the word, then search again for the next
1008 word (effectively, the end of the compiled word). This results in two pointers:
1010 +---------+---+---+---+---+------------+------------+------------+------------+
1011 | LINK | 3 | T | E | N | DOCOL | LIT | 10 | EXIT |
1012 +---------+---+---+---+---+------------+------------+------------+------------+
1015 Start of word End of word
1017 With this information we can have a go at decompiling the word. We need to
1018 recognise "meta-words" like LIT, LITSTRING, BRANCH, etc. and treat those separately.
1021 WORD FIND ( find the dictionary entry to decompile )
1023 ( Now we search again, looking for the next word in the dictionary. This gives us
1024 the length of the word that we will be decompiling. (Well, mostly it does). )
1025 HERE @ ( address of the end of the last compiled word )
1026 LATEST @ ( word last curr )
1028 2 PICK ( word last curr word )
1029 OVER ( word last curr word curr )
1030 <> ( word last curr word<>curr? )
1031 WHILE ( word last curr )
1033 DUP @ ( word curr prev (which becomes: word last curr) )
1036 DROP ( at this point, the stack is: start-of-word end-of-word )
1037 SWAP ( end-of-word start-of-word )
1039 ( begin the definition with : NAME [IMMEDIATE] )
1040 ':' EMIT SPACE DUP ID. SPACE
1041 DUP ?IMMEDIATE IF ." IMMEDIATE " THEN
1043 >DFA ( get the data address, ie. points after DOCOL | end-of-word start-of-data )
1045 ( now we start decompiling until we hit the end of the word )
1049 DUP @ ( end start codeword )
1052 ' LIT OF ( is it LIT ? )
1053 4 + DUP @ ( get next word which is the integer constant )
1056 ' LITSTRING OF ( is it LITSTRING ? )
1057 [ CHAR S ] LITERAL EMIT '"' EMIT SPACE ( print S"<space> )
1058 4 + DUP @ ( get the length word )
1059 SWAP 4 + SWAP ( end start+4 length )
1060 2DUP TELL ( print the string )
1061 '"' EMIT SPACE ( finish the string with a final quote )
1062 + ALIGNED ( end start+4+len, aligned )
1063 4 - ( because we're about to add 4 below )
1065 ' 0BRANCH OF ( is it 0BRANCH ? )
1067 4 + DUP @ ( print the offset )
1071 ' BRANCH OF ( is it BRANCH ? )
1073 4 + DUP @ ( print the offset )
1077 ' ' OF ( is it ' (TICK) ? )
1078 [ CHAR ' ] LITERAL EMIT SPACE
1079 4 + DUP @ ( get the next codeword )
1080 CFA> ( and force it to be printed as a dictionary entry )
1083 ' EXIT OF ( is it EXIT? )
1084 ( We expect the last word to be EXIT, and if it is then we don't print it
1085 because EXIT is normally implied by ;. EXIT can also appear in the middle
1086 of words, and then it needs to be printed. )
1087 2DUP ( end start end start )
1088 4 + ( end start end start+4 )
1089 <> IF ( end start | we're not at the end )
1094 DUP ( in the default case we always need to DUP before using )
1095 CFA> ( look up the codeword to get the dictionary entry )
1096 ID. SPACE ( and print it )
1104 2DROP ( restore stack )
1108 EXECUTION TOKENS ----------------------------------------------------------------------
1110 Standard FORTH defines a concept called an 'execution token' (or 'xt') which is very
1111 similar to a function pointer in C. We map the execution token to a codeword address.
1113 execution token of DOUBLE is the address of this codeword
1116 +---------+---+---+---+---+---+---+---+---+------------+------------+------------+------------+
1117 | LINK | 6 | D | O | U | B | L | E | 0 | DOCOL | DUP | + | EXIT |
1118 +---------+---+---+---+---+---+---+---+---+------------+------------+------------+------------+
1121 There is one assembler primitive for execution tokens, EXECUTE ( xt -- ), which runs them.
1123 You can make an execution token for an existing word the long way using >CFA,
1124 ie: WORD [foo] FIND >CFA will push the xt for foo onto the stack where foo is the
1125 next word in input. So a very slow way to run DOUBLE might be:
1128 : SLOW WORD FIND >CFA EXECUTE ;
1129 5 SLOW DOUBLE . CR \ prints 10
1131 We also offer a simpler and faster way to get the execution token of any word FOO:
1135 (Exercises for readers: (1) What is the difference between ['] FOO and ' FOO?
1136 (2) What is the relationship between ', ['] and LIT?)
1138 More useful is to define anonymous words and/or to assign xt's to variables.
1140 To define an anonymous word (and push its xt on the stack) use :NONAME ... ; as in this
1143 :NONAME ." anon word was called" CR ; \ pushes xt on the stack
1144 DUP EXECUTE EXECUTE \ executes the anon word twice
1146 Stack parameters work as expected:
1148 :NONAME ." called with parameter " . CR ;
1150 10 SWAP EXECUTE \ prints 'called with parameter 10'
1151 20 SWAP EXECUTE \ prints 'called with parameter 20'
1153 Notice that the above code has a memory leak: the anonymous word is still compiled
1154 into the data segment, so even if you lose track of the xt, the word continues to
1155 occupy memory. A good way to keep track of the xt and thus avoid the memory leak is
1156 to assign it to a CONSTANT, VARIABLE or VALUE:
1159 :NONAME ." anon word was called" CR ; TO ANON
1163 Another use of :NONAME is to create an array of functions which can be called quickly
1164 (think: fast switch statement). This example is adapted from the ANS FORTH standard:
1166 10 CELLS ALLOT CONSTANT CMD-TABLE
1167 : SET-CMD CELLS CMD-TABLE + ! ;
1168 : CALL-CMD CELLS CMD-TABLE + @ EXECUTE ;
1170 :NONAME ." alternate 0 was called" CR ; 0 SET-CMD
1171 :NONAME ." alternate 1 was called" CR ; 1 SET-CMD
1173 :NONAME ." alternate 9 was called" CR ; 9 SET-CMD
1180 0 0 CREATE ( create a word with no name - we need a dictionary header because ; expects it )
1181 HERE @ ( current HERE value is the address of the codeword, ie. the xt )
1182 DOCOL , ( compile DOCOL (the codeword) )
1183 ] ( go into compile mode )
1187 ' LIT , ( compile LIT )
1191 EXCEPTIONS ----------------------------------------------------------------------
1193 Amazingly enough, exceptions can be implemented directly in FORTH, in fact rather easily.
1195 The general usage is as follows:
1197 : FOO ( n -- ) THROW ;
1200 25 ['] FOO CATCH \ execute 25 FOO, catching any exception
1202 ." called FOO and it threw exception number: "
1204 DROP \ we have to drop the argument of FOO (25)
1207 \ prints: called FOO and it threw exception number: 25
1209 CATCH runs an execution token and detects whether it throws any exception or not. The
1210 stack signature of CATCH is rather complicated:
1212 ( a_n-1 ... a_1 a_0 xt -- r_m-1 ... r_1 r_0 0 ) if xt did NOT throw an exception
1213 ( a_n-1 ... a_1 a_0 xt -- ?_n-1 ... ?_1 ?_0 e ) if xt DID throw exception 'e'
1215 where a_i and r_i are the (arbitrary number of) argument and return stack contents
1216 before and after xt is EXECUTEd. Notice in particular the case where an exception
1217 is thrown, the stack pointer is restored so that there are n of _something_ on the
1218 stack in the positions where the arguments a_i used to be. We don't really guarantee
1219 what is on the stack -- perhaps the original arguments, and perhaps other nonsense --
1220 it largely depends on the implementation of the word that was executed.
1222 THROW, ABORT and a few others throw exceptions.
1224 Exception numbers are non-zero integers. By convention the positive numbers can be used
1225 for app-specific exceptions and the negative numbers have certain meanings defined in
1226 the ANS FORTH standard. (For example, -1 is the exception thrown by ABORT).
1228 0 THROW does nothing. This is the stack signature of THROW:
1231 ( * e -- ?_n-1 ... ?_1 ?_0 e ) the stack is restored to the state from the corresponding CATCH
1233 The implementation hangs on the definitions of CATCH and THROW and the state shared
1236 Up to this point, the return stack has consisted merely of a list of return addresses,
1237 with the top of the return stack being the return address where we will resume executing
1238 when the current word EXITs. However CATCH will push a more complicated 'exception stack
1239 frame' on the return stack. The exception stack frame records some things about the
1240 state of execution at the time that CATCH was called.
1242 When called, THROW walks up the return stack (the process is called 'unwinding') until
1243 it finds the exception stack frame. It then uses the data in the exception stack frame
1244 to restore the state allowing execution to continue after the matching CATCH. (If it
1245 unwinds the stack and doesn't find the exception stack frame then it prints a message
1246 and drops back to the prompt, which is also normal behaviour for so-called 'uncaught
1249 This is what the exception stack frame looks like. (As is conventional, the return stack
1250 is shown growing downwards from higher to lower memory addresses).
1252 +------------------------------+
1253 | return address from CATCH | Notice this is already on the
1254 | | return stack when CATCH is called.
1255 +------------------------------+
1256 | original parameter stack |
1258 +------------------------------+ ^
1259 | exception stack marker | |
1260 | (EXCEPTION-MARKER) | | Direction of stack
1261 +------------------------------+ | unwinding by THROW.
1265 The EXCEPTION-MARKER marks the entry as being an exception stack frame rather than an
1266 ordinary return address, and it is this which THROW "notices" as it is unwinding the
1267 stack. (If you want to implement more advanced exceptions such as TRY...WITH then
1268 you'll need to use a different value of marker if you want the old and new exception stack
1269 frame layouts to coexist).
1271 What happens if the executed word doesn't throw an exception? It will eventually
1272 return and call EXCEPTION-MARKER, so EXCEPTION-MARKER had better do something sensible
1273 without us needing to modify EXIT. This nicely gives us a suitable definition of
1274 EXCEPTION-MARKER, namely a function that just drops the stack frame and itself
1275 returns (thus "returning" from the original CATCH).
1277 One thing to take from this is that exceptions are a relatively lightweight mechanism
1282 RDROP ( drop the original parameter stack pointer )
1283 0 ( there was no exception, this is the normal return path )
1286 : CATCH ( xt -- exn? )
1287 DSP@ 4+ >R ( save parameter stack pointer (+4 because of xt) on the return stack )
1288 ' EXCEPTION-MARKER 4+ ( push the address of the RDROP inside EXCEPTION-MARKER ... )
1289 >R ( ... on to the return stack so it acts like a return address )
1290 EXECUTE ( execute the nested function )
1294 ?DUP IF ( only act if the exception code <> 0 )
1295 RSP@ ( get return stack pointer )
1297 DUP R0 4- < ( RSP < R0 )
1299 DUP @ ( get the return stack entry )
1300 ' EXCEPTION-MARKER 4+ = IF ( found the EXCEPTION-MARKER on the return stack )
1301 4+ ( skip the EXCEPTION-MARKER on the return stack )
1302 RSP! ( restore the return stack pointer )
1304 ( Restore the parameter stack. )
1305 DUP DUP DUP ( reserve some working space so the stack for this word
1306 doesn't coincide with the part of the stack being restored )
1307 R> ( get the saved parameter stack pointer | n dsp )
1308 4- ( reserve space on the stack to store n )
1309 SWAP OVER ( dsp n dsp )
1310 ! ( write n on the stack )
1311 DSP! EXIT ( restore the parameter stack pointer, immediately exit )
1316 ( No matching catch - print a message and restart the INTERPRETer. )
1335 ( Print a stack trace by walking up the return stack. )
1337 RSP@ ( start at caller of this function )
1339 DUP R0 4- < ( RSP < R0 )
1341 DUP @ ( get the return stack entry )
1343 ' EXCEPTION-MARKER 4+ OF ( is it the exception stack frame? )
1345 4+ DUP @ U. ( print saved stack pointer )
1350 CFA> ( look up the codeword to get the dictionary entry )
1351 ?DUP IF ( and print it )
1352 2DUP ( dea addr dea )
1353 ID. ( print word from dictionary entry )
1354 [ CHAR + ] LITERAL EMIT
1355 SWAP >DFA 4+ - . ( print offset )
1358 4+ ( move up the stack )
1365 C STRINGS ----------------------------------------------------------------------
1367 FORTH strings are represented by a start address and length kept on the stack or in memory.
1369 Most FORTHs don't handle C strings, but we need them in order to access the process arguments
1370 and environment left on the stack by the Linux kernel, and to make some system calls.
1372 Operation Input Output FORTH word Notes
1373 ----------------------------------------------------------------------
1375 Create FORTH string addr len S" ..."
1377 Create C string c-addr Z" ..."
1379 C -> FORTH c-addr addr len DUP STRLEN
1381 FORTH -> C addr len c-addr CSTRING Allocated in a temporary buffer, so
1382 should be consumed / copied immediately.
1383 FORTH string should not contain NULs.
1385 For example, DUP STRLEN TELL prints a C string.
1389 Z" .." is like S" ..." except that the string is terminated by an ASCII NUL character.
1391 To make it more like a C string, at runtime Z" just leaves the address of the string
1392 on the stack (not address & length as with S"). To implement this we need to add the
1393 extra NUL to the string and also a DROP instruction afterwards. Apart from that the
1394 implementation just a modified S".
1397 STATE @ IF ( compiling? )
1398 ' LITSTRING , ( compile LITSTRING )
1399 HERE @ ( save the address of the length word on the stack )
1400 0 , ( dummy length - we don't know what it is yet )
1402 KEY ( get next character of the string )
1405 HERE @ C! ( store the character in the compiled image )
1406 1 HERE +! ( increment HERE pointer by 1 byte )
1408 0 HERE @ C! ( add the ASCII NUL byte )
1410 DROP ( drop the double quote character at the end )
1411 DUP ( get the saved address of the length word )
1412 HERE @ SWAP - ( calculate the length )
1413 4- ( subtract 4 (because we measured from the start of the length word) )
1414 SWAP ! ( and back-fill the length location )
1415 ALIGN ( round up to next multiple of 4 bytes for the remaining code )
1416 ' DROP , ( compile DROP (to drop the length) )
1417 ELSE ( immediate mode )
1418 HERE @ ( get the start address of the temporary space )
1423 OVER C! ( save next character )
1424 1+ ( increment address )
1426 DROP ( drop the final " character )
1427 0 SWAP C! ( store final ASCII NUL )
1428 HERE @ ( push the start address )
1432 : STRLEN ( str -- len )
1433 DUP ( save start address )
1435 DUP C@ 0<> ( zero byte found? )
1440 SWAP - ( calculate the length )
1443 : CSTRING ( addr len -- c-addr )
1444 SWAP OVER ( len saddr len )
1445 HERE @ SWAP ( len saddr daddr len )
1448 HERE @ + ( daddr+len )
1449 0 SWAP C! ( store terminating NUL char )
1451 HERE @ ( push start address )
1455 THE ENVIRONMENT ----------------------------------------------------------------------
1457 Linux makes the process arguments and environment available to us on the stack.
1459 The top of stack pointer is saved by the early assembler code when we start up in the FORTH
1460 variable S0, and starting at this pointer we can read out the command line arguments and the
1463 Starting at S0, S0 itself points to argc (the number of command line arguments).
1465 S0+4 points to argv[0], S0+8 points to argv[1] etc up to argv[argc-1].
1467 argv[argc] is a NULL pointer.
1469 After that the stack contains environment variables, a set of pointers to strings of the
1470 form NAME=VALUE and on until we get to another NULL pointer.
1472 The first word that we define, ARGC, pushes the number of command line arguments (note that
1473 as with C argc, this includes the name of the command).
1480 n ARGV gets the nth command line argument.
1482 For example to print the command name you would do:
1485 : ARGV ( n -- str u )
1486 1+ CELLS S0 @ + ( get the address of argv[n] entry )
1487 @ ( get the address of the string )
1488 DUP STRLEN ( and get its length / turn it into a FORTH string )
1492 ENVIRON returns the address of the first environment string. The list of strings ends
1493 with a NULL pointer.
1495 For example to print the first string in the environment you could do:
1496 ENVIRON @ DUP STRLEN TELL
1498 : ENVIRON ( -- addr )
1499 ARGC ( number of command line parameters on the stack to skip )
1500 2 + ( skip command line count and NULL pointer after the command line args )
1501 CELLS ( convert to an offset )
1502 S0 @ + ( add to base stack address )
1506 SYSTEM CALLS AND FILES ----------------------------------------------------------------------
1508 Miscellaneous words related to system calls, and standard access to files.
1511 ( BYE exits by calling the Linux exit(2) syscall. )
1513 0 ( return code (0) )
1514 SYS_EXIT ( system call number )
1519 UNUSED returns the number of cells remaining in the user memory (data segment).
1521 For our implementation we will use Linux brk(2) system call to find out the end
1522 of the data segment and subtract HERE from it.
1524 : GET-BRK ( -- brkpoint )
1525 0 SYS_BRK SYSCALL1 ( call brk(0) )
1529 GET-BRK ( get end of data segment according to the kernel )
1530 HERE @ ( get current position in data segment )
1532 4 / ( returns number of cells )
1536 MORECORE increases the data segment by the specified number of (4 byte) cells.
1538 NB. The number of cells requested should normally be a multiple of 1024. The
1539 reason is that Linux can't extend the data segment by less than a single page
1540 (4096 bytes or 1024 cells).
1542 This FORTH doesn't automatically increase the size of the data segment "on demand"
1543 (ie. when , (COMMA), ALLOT, CREATE, and so on are used). Instead the programmer
1544 needs to be aware of how much space a large allocation will take, check UNUSED, and
1545 call MORECORE if necessary. A simple programming exercise is to change the
1546 implementation of the data segment so that MORECORE is called automatically if
1547 the program needs more memory.
1549 : BRK ( brkpoint -- )
1553 : MORECORE ( cells -- )
1558 Standard FORTH provides some simple file access primitives which we model on
1559 top of Linux syscalls.
1561 The main complication is converting FORTH strings (address & length) into C
1562 strings for the Linux kernel.
1564 Notice there is no buffering in this implementation.
1567 : R/O ( -- fam ) O_RDONLY ;
1568 : R/W ( -- fam ) O_RDWR ;
1570 : OPEN-FILE ( addr u fam -- fd 0 (if successful) | c-addr u fam -- fd errno (if there was an error) )
1572 CSTRING ( fam cstring )
1573 SYS_OPEN SYSCALL2 ( open (filename, flags) )
1575 DUP 0< IF ( errno? )
1582 : CREATE-FILE ( addr u fam -- fd 0 (if successful) | c-addr u fam -- fd errno (if there was an error) )
1586 CSTRING ( fam cstring )
1587 420 ROT ( 0644 fam cstring )
1588 SYS_OPEN SYSCALL3 ( open (filename, flags|O_TRUNC|O_CREAT, 0644) )
1590 DUP 0< IF ( errno? )
1597 : CLOSE-FILE ( fd -- 0 (if successful) | fd -- errno (if there was an error) )
1602 : READ-FILE ( addr u fd -- u2 0 (if successful) | addr u fd -- 0 0 (if EOF) | addr u fd -- u2 errno (if error) )
1603 ROT SWAP -ROT ( u addr fd )
1607 DUP 0< IF ( errno? )
1615 PERROR prints a message for an errno, similar to C's perror(3) but we don't have the extensive
1616 list of strerror strings available, so all we can do is print the errno.
1618 : PERROR ( errno addr u -- )
1626 ASSEMBLER CODE ----------------------------------------------------------------------
1628 This is just the outline of a simple assembler, allowing you to write FORTH primitives
1629 in assembly language.
1631 Assembly primitives begin ': NAME' in the normal way, but are ended with ;CODE. ;CODE
1632 updates the header so that the codeword isn't DOCOL, but points instead to the assembled
1633 code (in the DFA part of the word).
1635 We provide a convenience macro NEXT (you guessed what it does). However you don't need to
1636 use it because ;CODE will put a NEXT at the end of your word.
1638 The rest consists of some immediate words which expand into machine code appended to the
1639 definition of the word. Only a very tiny part of the i386 assembly space is covered, just
1640 enough to write a few assembler primitives below.
1645 ( Equivalent to the NEXT macro )
1646 : NEXT IMMEDIATE AD C, FF C, 20 C, ;
1649 [COMPILE] NEXT ( end the word with NEXT macro )
1650 ALIGN ( machine code is assembled in bytes so isn't necessarily aligned at the end )
1652 HIDDEN ( unhide the word )
1653 DUP >DFA SWAP >CFA ! ( change the codeword to point to the data area )
1654 [COMPILE] [ ( go back to immediate mode )
1657 ( The i386 registers )
1667 ( i386 stack instructions )
1668 : PUSH IMMEDIATE 50 + C, ;
1669 : POP IMMEDIATE 58 + C, ;
1671 ( RDTSC instruction )
1672 : RDTSC IMMEDIATE 0F C, 31 C, ;
1677 RDTSC is an assembler primitive which reads the Pentium timestamp counter (a very fine-
1678 grained counter which counts processor clock cycles). Because the TSC is 64 bits wide
1679 we have to push it onto the stack in two slots.
1681 : RDTSC ( -- lsb msb )
1682 RDTSC ( writes the result in %edx:%eax )
1683 EAX PUSH ( push lsb )
1684 EDX PUSH ( push msb )
1688 INLINE can be used to inline an assembler primitive into the current (assembler)
1693 : 2DROP INLINE DROP INLINE DROP ;CODE
1695 will build an efficient assembler word 2DROP which contains the inline assembly code
1696 for DROP followed by DROP (eg. two 'pop %eax' instructions in this case).
1698 Another example. Consider this ordinary FORTH definition:
1700 : C@++ ( addr -- addr+1 byte ) DUP 1+ SWAP C@ ;
1702 (it is equivalent to the C operation '*p++' where p is a pointer to char). If we
1703 notice that all of the words used to define C@++ are in fact assembler primitives,
1704 then we can write a faster (but equivalent) definition like this:
1706 : C@++ INLINE DUP INLINE 1+ INLINE SWAP INLINE C@ ;CODE
1708 There are several conditions that must be met for INLINE to be used successfully:
1710 (1) You must be currently defining an assembler word (ie. : ... ;CODE).
1712 (2) The word that you are inlining must be known to be an assembler word. If you try
1713 to inline a FORTH word, you'll get an error message.
1715 (3) The assembler primitive must be position-independent code and must end with a
1718 Exercises for the reader: (a) Generalise INLINE so that it can inline FORTH words when
1719 building FORTH words. (b) Further generalise INLINE so that it does something sensible
1720 when you try to inline FORTH into assembler and vice versa.
1722 The implementation of INLINE is pretty simple. We find the word in the dictionary,
1723 check it's an assembler word, then copy it into the current definition, byte by byte,
1724 until we reach the NEXT macro (which is not copied).
1727 : =NEXT ( addr -- next? )
1728 DUP C@ AD <> IF DROP FALSE EXIT THEN
1729 1+ DUP C@ FF <> IF DROP FALSE EXIT THEN
1730 1+ C@ 20 <> IF FALSE EXIT THEN
1735 ( (INLINE) is the lowlevel inline function. )
1736 : (INLINE) ( cfa -- )
1737 @ ( codeword points to the code, remember )
1738 BEGIN ( copy bytes until we hit NEXT macro )
1748 WORD FIND ( find the word in the dictionary )
1751 DUP @ DOCOL = IF ( check codeword <> DOCOL (ie. not a FORTH word) )
1752 ." Cannot INLINE FORTH words" CR ABORT
1761 NOTES ----------------------------------------------------------------------
1763 DOES> isn't possible to implement with this FORTH because we don't have a separate
1768 WELCOME MESSAGE ----------------------------------------------------------------------
1770 Print the version and OK prompt.
1774 S" TEST-MODE" FIND NOT IF
1775 ." JONESFORTH VERSION " VERSION . CR
1776 UNUSED . ." CELLS REMAINING" CR