Thu 21 Feb 2013 12:27:27 PM UTC, comment #15:
And, file 27434 should be pushed instead.
And, after that, the report can be closed.
|
Thu 21 Feb 2013 03:14:37 AM UTC, comment #14:
Just to clarify: file #27473 should be removed from this
report and attached to a report on the behavior of the
signbit() function.
|
Thu 21 Feb 2013 02:54:00 AM UTC, comment #13:
OK. I got that (finally)
So, what needs to be done to the changeset you applied
is change the bug #
But, I did not yet mean to submit this as a bug here. There
was discussion on the maintainer's list about how to
deal with this problem which had pretty much settled on
the patch that you found here. I meant to submit it as as a new bug report.
Michael
|
Thu 21 Feb 2013 02:45:10 AM UTC, comment #12:
Check Appendix D.4. There is some information there on writing the ChangeLog.
|
Thu 21 Feb 2013 02:43:23 AM UTC, comment #11:
Rik,
I just send you a separate email explaining
what I did wrong when attaching the last
patch file.
Again, sorry!
Michael
|
Thu 21 Feb 2013 12:42:15 AM UTC, comment #10:
Thanks Rik.
I will try to learn about the changeset format.
When I figure it out should the instructions be added to
the Contributing Guidelines Section of the Manual?
|
Wed 20 Feb 2013 11:37:16 PM UTC, comment #9:
I added a ChangeLog-formatted entry to your second patch and pushed it here (http://hg.savannah.gnu.org/hgweb/octave/rev/7cfb186592de)
|
Sun 17 Feb 2013 03:55:23 AM UTC, comment #8:
New changeset which is the same as the previous
code change. But, this one contains the updated
documentation.
Close after pushing...
(file #27473)
|
Sat 09 Feb 2013 07:00:47 PM UTC, comment #7:
changeset which corrects == 1 to != 0 is attached.
After pushing this change this bug report can be closed.
(file #27434)
|
Sat 09 Feb 2013 05:39:31 PM UTC, comment #6:
Aha. I get
signbit(x)
ans =
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
So, this is something that John Eaton could look at.
He just recently wrote the signbit() code. I think
that it should return 1. But, the documentation says
"non-zero." So, either John will decide 1 is right, or
I will change stemleaf to test for non-zero.
If you need this to work right away, you could just make
the obvious change in stemleaf(). I will probably do that
anyhow. Line 206 needs == 1 replaced by != 0
Thanks for submitting this report.
|
Sat 09 Feb 2013 05:09:24 PM UTC, comment #5:
|
Sat 09 Feb 2013 05:07:30 PM UTC, comment #4:
And, at the same time, what does signbit(x)
yield.
|
Sat 09 Feb 2013 05:06:19 PM UTC, comment #3:
Right. I got it backwards. stemleaf is dropping the
- value in the x vector. There are several possibilities:
What do you get for x after:
x = [22 12 28 52 39 2 12 0 11 11 42 38 44 18 44];
x = -x;
If this produces all negative values (including -0)
then I think the problem may be in signbit() for 32bit
systems.
So, tell me what you get for x as above.
Thanks
|
Sat 09 Feb 2013 04:42:42 PM UTC, comment #2:
Yes, it's the only test that fails (in steamleaf, I mean).
System: Ubuntu 12.10 / 32 bits
Here is what I get after running the test step by step in Octave:
|
Sat 09 Feb 2013 04:10:39 PM UTC, comment #1:
A few questions:
1. Is this the only test that fails?
2. What system are you using?
In any case, this is strange because the error
is in the stored result, not what stemleaf()
returned. But, the display of the stored
result (r2 = ...) is correct.
You could try displaying r2 to see if the traling
"0" did not get stored. You could do this test
directly in Octave, just copy/paste the r2 =
...
|
Sat 09 Feb 2013 03:44:16 PM UTC, original submission:
I get the following test failure when doing "make check" (or "test stemleaf"):
|