Thanks, Henry.
As I remarked earlier, this apparent error also seems to appear in
earlier versions, so it's
presumably not a new feature in J902.
It doesn't seem to have been noticed before, though, so probably not
too important.
Cheers,
Mike
On 27/10/2020 22:23, Henry Rich wrote:
With this new information I was able to reproduce the problem.
There is an internal rounding mode that that is pushed, popped, & set
in a couple of dozen places. My guess is that <.@^ set it and then
got interrupted. It appears to be set to FLOOR mode when 1%4x=0.
For the next beta, I will reset the rounding mode to the default
whenever control returns to the keyboard.
It feels like a good solution would need to save & restore the
rounding mode over try./catch. and u :: v, but my understanding of the
system is not up to that.
Henry Rich
On 10/27/2020 3:58 PM, 'Michael Day' via Programming wrote:
Thanks also to Devon for having a look.
I did install J902 on the desktop, but as it has only (!) 4GB RAM,
and hangs the whole Windows Session
too easily on a memory-hungry application such as this, I gave up.
(The first time I tried, I turned the
Desktop off, then rebooted to be rewarded by a Windows update that
took ~ half an hour.)
So, sticking to this laptop, with its 16GB ram, running at about
2GHz,
#5 cumSH 3 10 completes in around 36 seconds.
A short while ago, I ran
20 cumSH 3 10
and started interrupts after around 20 seconds - hard to be exact! -
frequently repeating the interrupt
until the break message appeared in the Terminal screen and once
again got
%4x
0
It looks as if one should wait until at least half of the time for 5
cumSH 3 10 before breaking the process for
20 cumSH 3 10 in order to observe the aberrant behaviour.
Here's a relevant excerpt from the session (in fixed width font) :
timer'#5 cumSH 3 10' NB. my timer - sorry forgot to change box
characters ....
┌──────┬─┐
│36.445│5│
└──────┴─┘
%4x NB. CORRECT RESULT.....
1r4
#20 cumSH 3 10 NB. break after ~20+ sec in my set-up
|break
| +/\%b (]*<.@^)c<.@^>:i.n
%4x NB. WRONG RESULT NOW !!!!!
0
cumSH
4 : 0
n =. x
'b c' =. x: y
+/\ % b (]* <.@^ ) c <.@^ >: i.n
)
JVERSION
Engine: j902/j64avx2/windows
Beta-i: commercial/2020-10-20T10:35:42
Library: 9.02.06
Qt IDE: 1.8.7/5.12.7(5.12.7)
Platform: Win 64
Installer: J902 install
InstallPath: c:/d/j902
Contact: www.jsoftware.com
I suppose it's not a problem if noone else can reproduce it. It
shouldn't be a
problem for me, as this function is very clearly not the way to
tackle the
puzzle I'm trying to crack! But it might possibly point to an error
that could occur
for real some day, somewhere, so it's just about worth boring you all
with it!
Cheers,
Mike
On 27/10/2020 08:32, 'Mike Day' via Programming wrote:
Thanks for trying. I’ll clean-install j902 on my desktop, which I
don’t often use for J, and see if I can reproduce the error. Maybe
more later.
Mike
Sent from my iPad
On 26 Oct 2020, at 23:30, Henry Rich <[email protected]> wrote:
I have tried to reproduce this twice: start the test, wait a while,
then interrupt. Both times the system was uncorrupted.
Can you give more info on when to interrupt to show the error?
Henry Rich
On 10/26/2020 11:24 AM, 'Michael Day' via Programming wrote:
PROGRESS!?
The problem reported under "[Jbeta] J902 exended problem" was NOT
with i-beta in particular.
It is a bit strange, and DOES apply to J901, and perhaps older
versions too. Maybe
something to do with memory allocation for very long extended
representations.
Something happens in the session which alters behaviour, whereby
%4x is initially evaluated as 1r4, but after an interrupt, as
shown below, it returns
a value of 0.
I expect an easier example would give rise to the same effect, but
I'll stick with the
latest manifestation, which is fairly concise.
As I explained earlier, and partly mis-stated, I was exploring,
in Maths-ish notation,
sum{k=1, n} 1 / [(b^(c^k)) . (c^k)]
One of my functions to explore this is
cumSH =: 4 : 0
n =. x
'b c' =. x: y
+/\ % b (]* ^ ) c ^ >: i.n
)
Here's a clip from a session in J 901 :
%4x NB. starts ok
1r4
NB. cumSH soon gets slow - run too many cycles ...
20 cumSH 3 10 NB. interrupted using jbreak.bat...
|break: cumSH
| +/\%b (]*<.@^)c<.@^>:i.n
%4x NB. different value now!
0
datatype%4x NB. still extended, though wrong
extended
JVERSION
Engine: j901/j64avx2/windows
Release-f: commercial/2020-06-11T15:31:28
Library: 9.01.23
Qt IDE: 1.8.6/5.12.7(5.12.7)
Platform: Win 64
Installer: J901 install
InstallPath: c:/d/j901
Contact: www.jsoftware.com
Thanks,
Mike
On 26/10/2020 13:09, 'Michael Day' via Beta wrote:
OK - home again and "solved" this difficulty with extended.
Well, chased it away -
I've no idea _why_ the error arose.
So - in the absence of advice to the contrary, I ran the
prescribed update method,
ie
load'pacman'
'upgrade' jpkg 'jengine'
This method had failed when I first ran it to update to beta-i ,
the failure manifesting
itself as an error reported when starting up J902 from its
Windows shortcut. That was
why I had downloaded the zip file and overwritten my J902 folders
with its contents.
However, running the update in the approved manner worked today.
The Windows
JQt shortcut fires up the session without crashing.
Now, I find
1%4x NB. was 0, yesterday !
1r4
1x%4x NB. " " "
1r4
as expected.
Perhaps my attempt on 20/10/20 to overwrite the beta-h
installation with files and folders
from the zip-file went wrong somewhere...
Sorry to have raised this hare, but it was a real anomaly
yesterday!
Mike
--
This email has been checked for viruses by AVG.
https://www.avg.com
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
--
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
