Nested ' fails

[David C. Norris]

Here's a short repro for a problem I'm now encountering with the circular buffer I posted a few days ago. Whereas the words toX and toY below work as expected interactively, using them inside another word fails. Have I misunderstand something important about '?

marker -repro
decimal ram

: point: ( "name" x y -- ) ram create swap , ( x) , ( y) ;
: xy. ( a-addr -- ) @+ . @ . ;

: toX ( "name" x -- ) ' execute ! ;
: toY ( "name" y -- ) ' execute 2+ ! ;

2 3 point: A
A xy.
4 toX A
5 toY A
A xy.

: fine ( x -- ) A ! A xy. ;
7 fine

: hmm ( x -- ) toX A ;
\ FF replies '?' to the following:
8 hmm

-repro

I'm doing this on PIC24.

[Mikael Nordman]

In hmm you would need toX to use ['] for it to work. ' is not state smart.
Something like this might work.

: toX state if postpone ['] else ' then execute  ! ; immediate

[Mikael Nordman]

I tend to avoid words that use TICK and have to interpret. So I would do

4 X A !  \ where X is the index 0
4 A X!   \ Where X! stores to the index offset

There are also wordsets for structs, but I have not used or defined them for FF.

[David C. Norris]

Thank you, Mikael. Appreciating this as a question of state-smartness is most helpful. This looks like an excellent opportunity for me to read Ertl's article, State-smartness — Why it is Evil and How to Exorcise it, cited at the bottom of https://gforth.org/manual/Combined-words.html.

Although the above repro aimed to crystallize the issue into as simple a package as possible, it may have obscured my underlying motivation a bit. My actual trouble is with the INTO word of abovementioned circular buffer, which I would have hoped to use in the same manner as TO. (Is TO itself state-smart?)

The Ertl article looks pretty challenging to me, but already I see that I've stumbled on this state-smartness issue in 'the usual way'. In §4.3, he says,

The most frequent reason for wanting a combined word is parsing words : words that read the input stream.

A bit earlier, Ertl says,

The general solution is to provide two words: one for the interpretation semantics and one for the compilation semantics. Examples: ' and ['], char and [char], and s"-int and s"-comp.

So I could perhaps implement as well an [INTO] word for use during compilation, but that destroys the simplicity of aligning with TO, which needs no companion [TO].

Another solution might be, if I have defined a CIRCBUF: MYBUF, to define also a word !MYBUF following the same, standard naming convention as your !X above.

Still, the dangers Ertl highlights (that users might someday TICK or POSTPONE a state-smart word) seems somehow unlikely in the case of INTO. But then again, my first thought upon initial encounter with the concept of state-smartness was, "Oh, I would never write such a word!"

---

Addendum: Ha! Looking again at that §6.11.1 of the GForth manual, I see the following:

This feature was introduced for implementing TO and S". I recommend that you do not define such words, as cute as they may be: they make it hard to get at both parts of the word in some contexts.

So the design I'll pursue now will require defining a !MYBUF word.

[Mikael Nordman]

I usually would write

3 7 circbuf: mybuffer
mybuffer >circbuf

when I put data into a circular buffer or fifo, as can be seen from the CQ: character fifos in FF PIC24. The CQ: fifos are used by the UART RX and UART buffered TX.

In order to survive a restart CIRCBUF: should store persistent data ( like size) in flash and variable data (like current index) in ram.