Simply hooking the IRQ.

[Whammo]

I was surprised how easy this is.
End your code with $ea31 jmp,
Your code holds off the IRQ routine until your code is finished.

: irqpeek $dc0e c@ ;
: irqpoke $dc0e c! ;
: irqoff irqpeek $fe and irqpoke ;
: irqon irqpeek 1 or irqpoke ;

: irqsub ( addr -- )
irqoff $314 ! irqon ; \ irq vector

code flashborder  \ arbitrary asm code
$d020 inc,
$d020 dec,
$ea31 jmp, \ irq vector default 
;code

: irqtest ['] flashborder irqsub ; 
' flashborder irqsub             
: irqrestore $ea31 irqsub ; 

Fixed!

[Whammo]

Accordingly, this IRQ routine fires after the default vector, rather than before,
by loading pushing another state on the stack over the one created by the kernal,
the IRQ returns to your routine, then your routine returns to forth.
Note: V uses SYS to place the cursor , so V will crash durexForth while this is enabled.

: irqpeek $dc0e c@ ;
: irqpoke $dc0e c! ;
: irqoff irqpeek $fe and irqpoke ;
: irqon irqpeek 1 or irqpoke ;

: irqsub ( addr -- )
irqoff $314 ! irqon ;

code pstate   \ uses SYS register locations
$15 lda, pha, $14 lda, pha,
sr lda, pha, ar lda, pha,
xr lda, pha, yr lda, pha,
$ea31 jmp,
;code

code flashborder
$d020 inc,
$d020 dec,
$ea81 jmp, \ into restore registers, rti
;code
: irqtest
['] flashborder $14 ! \ set SYS address
['] pstate irqsub
;

[lonetech]

@Whammo Starts out right, but ['] doesn't leave a stack effect, it leaves compiled code by calling literal. That shifts the stack effect from compilation time of e.g. irqtest to its run time. This is why the postponement recipe compiled the literal call into [']. The only time the execution token of literal was on the stack was while ['] was being compiled. So:

1. When compiling ['], literal was postponed to when ['] runs, after the call to '. This counters the property that literal is immediate.
2. ['] was made immediate, so it runs during compilation of e.g. irqtest.
3. When compiling irqtest, ['] runs ' immediately but postpones the stack effect (its result) using literal.
4. When irqtest runs, the execution token for flashborder is put on the stack by the lit call literal compiled into irqtest.

literal works like: : literal ['] lit compile, , ; immediate

@jkotlinski Feel free to use it; it compiled to the same code when I tried it. The same recipe also occurs in [char] (which seems redundant, as we have character literals).

However, you'll need to define postpone earlier, and that means making it not require ['] for non-immediate words. Something like changing ] ['] compile, to ' compile, literal ].

By the way, postpone undefined-word somehow compiled an infinite loop. This is an ambiguous condition in the standard, so that's not a bug, though ' does check for this error.

[Whammo]

If I'm going to get this, I suppose I'm going to have to start writing compiling words.
I had a similar problem grasping the assembler words, but one day, I could just do it.

[lonetech]

Then you have a grasp of most of the compiling words already! In fact, I think jsr, and compile, are synonyms. The new mechanism here is immediate execution, either by immediate marks in the dictionary or [ ] state switches ([ works like : [ 0 state ! ; immediate). Took me a while to understand what postpone was doing to non-immediate words (since postpone itself is immediate, it had to wrap one compile in another). code starts a definition but doesn't switch to compilation, which is why inline assembly is wrapped in [ ].

[jkotlinski]

#446 looks fine?

[jkotlinski]

A note on "ambiguous condition". I thought it meant same thing as C undefined behavior, which means all bets are off. But on reading the standard, I found "ambiguous condition" has a more specific meaning.

• "Ambiguous condition: A circumstance for which this standard does not prescribe a specific behavior."
• A system shall document the system action taken upon each of the general or specific ambiguous conditions identified in this standard.

It seems like there is some homework to do, especially on the documentation part.
(created #447)

[Whammo]

This one leverages the BRK instruction to push the stack frame.
The IRQ routine runs first, re enables the interrupt, then your code executes.
If your code takes too long it could overflow the processor stack.

: irqpeek $dc0e c@ ;
: irqpoke $dc0e c! ;
: irqoff irqpeek $fe and irqpoke ;
: irqon irqpeek 1 or irqpoke ;

: irqsub ( addr -- )
$314 ! irqon ; \ irq vector

: brksub ( addr -- )
irqoff $316 ! ; \ brk vector

code flashborder  \ not so arbitrary asm code
$314 lda, pha, $315 lda, pha,     \ flashborder to stack
$31 lda,# $314 sta, $ea lda,# $315 sta, \ restore irq vector
      brk, nop,   \ stack for me, do irq routine
\ user code
$d020 inc,        \ return from irq routine here
$d020 dec,        \ then exit
\ end user code
sei, \ cia timer IRQs were held off until reset by irq routine
pla, $315 sta, pla, $314 sta, \ flashborder to  irq -mind the stack
$ea81 jmp,  \ restore regs, rti
;code

: irqtest
$ea31 brksub           \ call irq routine on a brk
['] flashborder irqsub \ call my code on irq
;

[ekipan]

Just a small thing, you can replace irqon irqoff with 1 irq! 0 irq!:

: irq! ( flag -- ) $dc0e c@ swap
 if 1 or else $fe and then $dc0e c! ;

And a pair of more general words for masking bit registers:

: bits@ ( mask addr -- bits ) c@ and ;
: bits! ( flag mask addr -- ) 
 rot >r dup c@ rot ( addr val mask ) r> 
 if or else invert and then swap c! ;

Edit: Whoops, bits! was wrong.

[Whammo]

Wow, you're really good at this stuff!
🥇

[ekipan]

You say, moments after I edit the correction :P

I had actually written these bit masking words for my game of life program, but it was godly slow so I switched to storing whole bytes.

[Whammo]

I guess I kinda write in the idiom of the machine and brute force it. My amateur-ity shows, but
hey, it works! 😄

[ekipan]

Your direct code has the advantage of being faster. Being more general means more faffing about swapping and rotting the stack, using up cycles.

[Whammo]

There is that too. It's lots of fun most of all.
My c64 knows who's boss. 😄