I plan to open a report of this problem, but I'm having difficulty
deciding on a pithy, meaningful title. Have you any suggestions?

Feel free to point out how I've misinterpreted something and that
this isn't an actual bug. That's happened to me before. I couldn't
find anything in the documentation suggesting this limit. Does it
fail for other people? Does it fail or work on other REXXes?

If you try it, you may see that the error message on 3.9.3 is even
weirder than on older versions. Regardless of level, notice the
difference between DO I = 1 FOR N and DO I = 1 TO N.

<code>
say 'digits() =' digits()
n = 2**31 /* a 10-digit number */
say format(n,,,0) length(n) '(' || n || ')'
do i = 1 for n
say "In first loop for" n
leave i
end

numeric digits 20
say 'digits() =' digits()
n = 2**31 - 1
say format(n,,,0) length(n) '(' || n || ')'
do i = 1 for n
say "In second loop for" n
leave i
end
n = 2**31
say format(n,,,0) length(n) '(' || n || ')'
do i = 1 to n
say "In third loop TO" n
leave i
end
/*
The following yields:
Error 26.3: Value of FOR expression in DO instruction must be zero
or a positive whole number; found "2147483648"
*/
do i = 1 for n
say "In fourth loop for" n
leave i
end

exit 0

</code>

--
Arthur T. - ar23hur "at" pobox "dot" com

> I plan to open a report of this problem, but I'm having difficulty
> deciding on a pithy, meaningful title. Have you any suggestions?

27 +++ do i = 1 for n
line 27: Invalid whole number
Error 26.3: Value of FOR expression in DO instruction must be zero or a
positive whole number; found "2147483648"

By default Rexx uses nine digits for a number. A whole number therefore cannot be larger than
999999999, which 2**31 is, hence the error.

Here a rexxtry.rex session (using ooRexx but this would not matter for this case):

F:\work\svn\bsf4oorexx\trunk\bsf4oorexx.dev\source_java>rexxtry
REXX-ooRexx_5.0.0(MT)_32-bit 6.05 13 Aug 2021
rexxtry.rex lets you interactively try REXX statements.
Each string is executed when you hit Enter.
Enter 'call tell' for a description of the features.
Go on - try a few... Enter 'exit' to end.
say digits()
9 ........................................... rexxtry.rex on WindowsNT
say 2**32
4.2949673E+9
........................................... rexxtry.rex on WindowsNT
say 2**31
2.14748365E+9
........................................... rexxtry.rex on WindowsNT
numeric digits 10
........................................... rexxtry.rex on WindowsNT
say 2**31
2147483648
........................................... rexxtry.rex on WindowsNT
say 2**32
4294967296
........................................... rexxtry.rex on WindowsNT

As you see, once you use ten digits for a number, 2**31 and 2**32 qualify again as whole numbers.

HTH,

---rony

On 14.09.2021 05:34, Arthur T. wrote:
> I plan to open a report of this problem, but I'm having difficulty
> deciding on a pithy, meaningful title. Have you any suggestions?
>
> Feel free to point out how I've misinterpreted something and that
> this isn't an actual bug. That's happened to me before. I couldn't
> find anything in the documentation suggesting this limit. Does it
> fail for other people? Does it fail or work on other REXXes?
>
> If you try it, you may see that the error message on 3.9.3 is even
> weirder than on older versions. Regardless of level, notice the
> difference between DO I = 1 FOR N and DO I = 1 TO N.
>
> <code>
> say 'digits() =' digits()
> n = 2**31 /* a 10-digit number */
> say format(n,,,0) length(n) '(' || n || ')'
> do i = 1 for n
> say "In first loop for" n
> leave i
> end
>
> numeric digits 20
> say 'digits() =' digits()
> n = 2**31 - 1
> say format(n,,,0) length(n) '(' || n || ')'
> do i = 1 for n
> say "In second loop for" n
> leave i
> end
> n = 2**31
> say format(n,,,0) length(n) '(' || n || ')'
> do i = 1 to n
> say "In third loop TO" n
> leave i
> end
> /*
> The following yields:
> Error 26.3: Value of FOR expression in DO instruction must be zero
> or a positive whole number; found "2147483648"
> */
> do i = 1 for n
> say "In fourth loop for" n
> leave i
> end
>
> exit 0
>
> </code>
>

On Tuesday, September 14, 2021 at 11:48:18 AM UTC-4, Rony wrote:
> By default Rexx uses nine digits for a number. A whole number therefore cannot be larger than
> 999999999, which 2**31 is, hence the error.
>
> Here a rexxtry.rex session (using ooRexx but this would not matter for this case):
>
> F:\work\svn\bsf4oorexx\trunk\bsf4oorexx.dev\source_java>rexxtry
> REXX-ooRexx_5.0.0(MT)_32-bit 6.05 13 Aug 2021
> rexxtry.rex lets you interactively try REXX statements.
> Each string is executed when you hit Enter.
> Enter 'call tell' for a description of the features.
> Go on - try a few... Enter 'exit' to end.
> say digits()
> 9
> ........................................... rexxtry.rex on WindowsNT
> say 2**32
> 4.2949673E+9
> ........................................... rexxtry.rex on WindowsNT
> say 2**31
> 2.14748365E+9
> ........................................... rexxtry.rex on WindowsNT
> numeric digits 10
> ........................................... rexxtry.rex on WindowsNT
> say 2**31
> 2147483648
> ........................................... rexxtry.rex on WindowsNT
> say 2**32
> 4294967296
> ........................................... rexxtry.rex on WindowsNT
>
> As you see, once you use ten digits for a number, 2**31 and 2**32 qualify again as whole numbers.
>
> HTH,
>
> ---rony
> On 14.09.2021 05:34, Arthur T. wrote:
> > I plan to open a report of this problem, but I'm having difficulty
> > deciding on a pithy, meaningful title. Have you any suggestions?
> >
> > Feel free to point out how I've misinterpreted something and that
> > this isn't an actual bug. That's happened to me before. I couldn't
> > find anything in the documentation suggesting this limit. Does it
> > fail for other people? Does it fail or work on other REXXes?
> >
> > If you try it, you may see that the error message on 3.9.3 is even
> > weirder than on older versions. Regardless of level, notice the
> > difference between DO I = 1 FOR N and DO I = 1 TO N.
> >
> > <code>
> > say 'digits() =' digits()
> > n = 2**31 /* a 10-digit number */
> > say format(n,,,0) length(n) '(' || n || ')'
> > do i = 1 for n
> > say "In first loop for" n
> > leave i
> > end
> >
> > numeric digits 20
> > say 'digits() =' digits()
> > n = 2**31 - 1
> > say format(n,,,0) length(n) '(' || n || ')'
> > do i = 1 for n
> > say "In second loop for" n
> > leave i
> > end
> > n = 2**31
> > say format(n,,,0) length(n) '(' || n || ')'
> > do i = 1 to n
> > say "In third loop TO" n
> > leave i
> > end
> > /*
> > The following yields:
> > Error 26.3: Value of FOR expression in DO instruction must be zero
> > or a positive whole number; found "2147483648"
> > */
> > do i = 1 for n
> > say "In fourth loop for" n
> > leave i
> > end
> >
> > exit 0
> >
> > </code>
> >
Even with a higher digits setting, this value is rejected for the FOR argument. This falls into the category of "numbers used directly by REXX" described on page 137 of The REXX Language, second edition. The exprf expression is one of the examples explicitly listed.

Rick

On Mon, 13 Sep 2021 23:34:41 -0400, Arthur T. <arthur@munged.invalid>
declaimed the following:

>Feel free to point out how I've misinterpreted something and that
>this isn't an actual bug. That's happened to me before. I couldn't
>find anything in the documentation suggesting this limit. Does it
>fail for other people? Does it fail or work on other REXXes?
>
>If you try it, you may see that the error message on 3.9.3 is even
>weirder than on older versions. Regardless of level, notice the
>difference between DO I = 1 FOR N and DO I = 1 TO N.
>

> <code>
> say 'digits() =' digits()
> n = 2**31 /* a 10-digit number */
> say format(n,,,0) length(n) '(' || n || ')'
> do i = 1 for n
> say "In first loop for" n
> leave i
> end
>

Regina reference manual (3.9.1 manual, haven't seen a newer one) page
37 "Loop Convergence" might have a part to play. Not sure. You might have
to let the loop run to completion to find out. Might also depend upon how
Regina really implements "for n" termination. I'd probably write the
interpreter to convert to a "to n+start" internally -- and if the statement
already included a "to m" clause, keep the lowest termination.

> numeric digits 20

If the other responder is correct, Regina (and REXX implementations in
general) may ignore this for loop boundaries. So the above documented
convergence may come into play. There is also the consideration of whether
an implementation is using native integer arithmetic for loop indices
(would be fastest, and justify restriction to 32-bit integers).

-=-=-=-
/* REXX */
say "What!"

say 'digits() =' digits()
n = 2**31 /* a 10-digit number */
say "n" n
say "n+1" n+1
do i = 1 for n
say "In first loop for" n
leave i
end

numeric digits 20
say
say 'digits() =' digits()
n = 2**31 - 1
say "n" n
/*say "n+1" n+1*/
do i = 1 for n
say "In second loop for" n
leave i
end

say
n = 2**31
say "n" n
say "No increment for termination needed with a TO"
do i = 1 to n
say "In third loop TO" n
leave i
end

say
say "n" n
/* say "n+1" n+1 */
do i = 1 for n
say "In fourth loop for" n
leave i
end
-=-=-=-

Hmmm... Now this is interesting... Before adding the additional "say
'n+1'..." lines, I achieved

C:\Users\Wulfraed>regina t.rx
What!
digits() = 9
n 2.14748364E+9
In first loop for 2.14748364E+9

digits() = 20
n 2147483647
In second loop for 2147483647

n 2147483648
In third loop TO 2147483648

n 2147483648
32 +++ do i = 1 for n
Error 26 running "C:\Users\Wulfraed\t.rx", line 32: Invalid whole number
Error 26.3: Value of FOR expression in DO instruction must be zero or a
positive whole number; found "2147483648"

But after the additions that changed to...

C:\Users\Wulfraed>regina t.rx
What!
digits() = 9
n 2.14748364E+9
n+1 2.14748364E+9
In first loop for 2.14748364E+9

digits() = 20
n 2147483647
n+1 2147483648
In second loop for 2147483647

n 2147483648
No increment for termination needed with a TO
In third loop TO 2147483648

n 2147483648
n+1 2147483649 <***********************************
36 +++ do i = 1 for n
Error 26 running "C:\Users\Wulfraed\t.rx", line 36: Invalid whole number
Error 26.3: Value of FOR expression in DO instruction must be zero or a
positive whole number; found "21474836484799" <*************

WHICH definitely looks like an error somewhere in the conversion! (I
removed those statements, and the output went back to 2147483648)... Either
"n + 1" with digits 20 causes the error, the one with digits 9 does not. A
problem in the extended math modes?

--
Wulfraed Dennis Lee Bieber AF6VN
wlfraed@ix.netcom.com http://wlfraed.microdiversity.freeddns.org/

In
Message-ID:<56dc90ab-4168-46f8-ba46-240ff749ddb8n@googlegroups.com>,
Rick McGuire <object.rexx@gmail.com> wrote:

>Even with a higher digits setting, this value is rejected for the FOR argument. This falls into the category of "numbers used directly by REXX" described on page 137 of The REXX Language, second edition. The exprf expression is one of the examples explicitly listed.

Thank you!

I'm glad I asked here before opening a problem. And, yes, TRL2 says
that exprf (FOR operand) can be limited, but doesn't put exprt (TO
operand) in the same category. And that exactly matches the results
of my test code.

WAD even if some of us didn't realize it. I hate to think that I may
have to actually read (or at least skim through) TRL2, but this shows
why I should (have).

--
Arthur T. - ar23hur "at" pobox "dot" com

In Message-ID:<mnb4kgpjm6gn62lmhpabi2pqj9mqajkvmd@4ax.com>,
Dennis Lee Bieber <wlfraed@ix.netcom.com> wrote:

>On Mon, 13 Sep 2021 23:34:41 -0400, Arthur T. <arthur@munged.invalid>
>declaimed the following:
>
>> <code>
>> say 'digits() =' digits()
>> n = 2**31 /* a 10-digit number */
>> say format(n,,,0) length(n) '(' || n || ')'
>> do i = 1 for n
>> say "In first loop for" n
>> leave i
>> end
>>
>
> Regina reference manual (3.9.1 manual, haven't seen a newer one) page
>37 "Loop Convergence" might have a part to play. Not sure. You might have
>to let the loop run to completion to find out. Might also depend upon how
>Regina really implements "for n" termination. I'd probably write the
>interpreter to convert to a "to n+start" internally -- and if the statement
>already included a "to m" clause, keep the lowest termination.

That convergence is probably (somehow) why this didn't cause an error
with the default 9-digits of accuracy. And Rick McGuire explained
why, with 20 digits it works with TO but not with FOR.

>Error 26 running "C:\Users\Wulfraed\t.rx", line 36: Invalid whole number
>Error 26.3: Value of FOR expression in DO instruction must be zero or a
>positive whole number; found "21474836484799" <*************
>
> WHICH definitely looks like an error somewhere in the conversion! (I
>removed those statements, and the output went back to 2147483648)... Either
>"n + 1" with digits 20 causes the error, the one with digits 9 does not. A
>problem in the extended math modes?

It's even weirder in Regina 3.9.3 because the error message from my
original example is:
Error 26.3: Value of FOR expression in DO instruction must be zero or
a positive whole number; found "2147483648 109"

--
Arthur T. - ar23hur "at" pobox "dot" com

On Wed, 15 Sep 2021 23:37:48 -0400, Arthur T. <arthur@munged.invalid>
declaimed the following:

>It's even weirder in Regina 3.9.3 because the error message from my
>original example is:
>Error 26.3: Value of FOR expression in DO instruction must be zero or
>a positive whole number; found "2147483648 109"

While I cited the 3.9.1 manual, I'm pretty certain I was running
3.9.3...

C:\Users\Wulfraed>rexx
parse version ver
say ver
^Z
REXX-Regina_3.9.3 5.00 5 Oct 2019

C:\Users\Wulfraed>

Hmmm, your expanded result looks like a garbage attempt at a 10^9 tail
<G>

--
Wulfraed Dennis Lee Bieber AF6VN
wlfraed@ix.netcom.com http://wlfraed.microdiversity.freeddns.org/

On Wed, 15 Sep 2021 05:36:51 -0700 (PDT), Rick McGuire wrote:
>> > Error 26.3: Value of FOR expression in DO instruction must be zero
>> > or a positive whole number; found "2147483648"
>Even with a higher digits setting, this value is rejected for the FOR
>argument. This falls into the category of "numbers used directly by
>REXX" described on page 137 of The REXX Language, second edition. The
>exprf expression is one of the examples explicitly listed.

This seems weird when a smaller value of DIGITS() causes smaller numbers
to be rejected, but I see it says that an implementation restriction
*may* apply, and that 9 is the *mininum* length that must be supported.
So Regina is actually supporting a little *more* than required here.
Trying it in ooRexx, I get this error starting right at 1E10.

¬R