Bug in countdown DO +LOOP

[hexagon5un]

: foo 1 4 do i . -1 +loop ;

should print 4 3 2 1. Instead, it just prints 4.

[TG9541]

Before implementing +LOOP I couldn't find any clear requirements. I then compared different implementations and found one the addressed the risk of very long loops due to erroneous upper limit and starting point. That's the one I implemented, since it doesn't require looking at the sign of the increment.
I now understand this is a valid use case, and I'll change the code :-)

[TG9541]

How should the following code behave?
: foo 1 4 do i . -2 +loop ;

[hexagon5un]

gforth and mecrisp-stellaris say: 4 2.

My understanding is that the TOS index is increasing the test is run against <, and when it is decreasing, it's run against <=. (https://www.forth.com/starting-forth/6-forth-do-loops/)

(Equivalently, I think it's begin >= until and begin > until loops under the hood.)

This is crappy, b/c 4 1 do i . 1 +loop -> 1 2 3 (doesn't 4 b/c 4<4 is false) and 1 4 do i . -1 +loop -> 4 3 2 1 (1 b/c 1 <= 1 is true), so the loops have different lenghts, but that's the way Forth is defined.

https://github.com/gerryjackson/forth2012-test-suite is written for ANS Forth. This is in core.fs:

T{ : GD2 DO I -1 +LOOP ; -> }T
T{ 1 4 GD2 -> 4 3 2 1 }T
T{ -1 2 GD2 -> 2 1 0 -1 }T

[TG9541]

Thanks! Good links, too!

I fixed it. If you need binaries right away, I updated the 2.2.6.snapshot.

[hexagon5un]

Binaries not necessary, and I can confirm that at least the obvious test cases are working as they should.