xts and return addresses (was: "Back & Forth" - Local variables)

albert@spenarnc.xs4all.nl writes:

R EXECUTE doesn't work for ciforth.

There is no guarantee that a saved interpreter pointer on the
stack is an execution token.

In development Gforth, if you use the xt of a colon definition as a
return address, you get the same effect as EXECUTE.  E.g.

: colon-execute >r ;

1 ' . colon-execute \ prints "1"

I have recently had a puzzling bug syndrome in the following word
(buggy version shown):

: reverse-sections-execute ( xt -- )
    >r sections $@ cell mem-do
        j i @ section-execute
    loop ;

[: current-section @ cr h. ;] reverse-sections-execute

The output is:

$7FF5F428F000
$7FF5F3F1A000
$7FF5F3F3B000
$7FF5E5BFD000
$7FF5E5DFE040
$7FF5F4298000
$7FF5E5DFE040  ok

The last line should not be output.  What happens is that the xt
passed to REVERSE-SECTIONS-EXECUTE is pushed on the return stack, and
I then forgot to drop it.  But instead of that resulting in a crash
(as you would get in Gforth up to around 2020), with the new xt
implementation the ; at the end COLON-EXECUTEs the xt.  And because
the xt is a colon definition, this calls the xt again.

xt=pfa has it's advantages, but here a disadvantage showed up.

- anton
--
M. Anton Ertl  http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
     New standard: https://forth-standard.org/
EuroForth 2023 proceedings: http://www.euroforth.org/ef23/papers/
EuroForth 2024 proceedings: http://www.euroforth.org/ef24/papers/