pyginac-users

[PyGiNaC-users] strange behavior with the test suite

[<pp...@fy...>]

Hi,

while playing around with pyginac, I found some strange behavior with
the regression test suite. The reason to this behavior escapes me. 

When running ./run python2.3 bin/checkall.py most of the time 
everything goes ok, but every now and then an error occurs as follows.

*****************************************************************
Failure in example: exam_normal2()
from line #1 of exam_normalization.exam_normal2
Expected: 0
Got:
normal form of  (-y+x)**(-3)*(x**2-y**2)  returned  (y-x)**(-2)*(y+x)  while 
it should be  (-y+x)**(-2)*(y+x)
normal form of  (x**2-y**2)*(-1/2*y+1/2*x)**(-3)  returned  
(y-x)**(-2)*(8*y+8*x)  while it should be  (-y+x)**(-2)*(8*y+8*x)
2
*****************************************************************
1 items had failures:
   1 of   1 in exam_normalization.exam_normal2
***Test Failed*** 1 failures.

To get an idea how often this happens, I ran the test 100 times, giving 43 
fails and 57 clean passes. The interesting part is that when I run

./run python2.3 exam_normalization.py

I am totally unable to reproduce the error above, leading to a guess that this 
might have something to do with bin/checkall.py and/or the doctest module. 
However, I have not been able to point any other reason as to why this would 
be so. I also checked with python 2.4 (Debian-packaged 2.4.1-2) with the same 
results. 

I am using an up-to-date cvs of PyGiNaC, GiNaC 1.3.2, boost.python 1.33.0,
gcc 4.0.2 and python 2.3.5 (all of which except PyGiNaC are the 
Debian-packaged versions). 

So, does anybody have any idea? Can anyone else reproduce this?

Best Regards,
Matti

-- 
Matti Peltomaki, Graduate Student
Laboratory of Physics 
Helsinki University of Technology
040 7576977, room Y308B

Re: [PyGiNaC-users] strange behavior with the test suite

[Ondrej C. <on...@ce...>]

Yes, we created another ginac bindings in swig, swiginac.berlios.de, and we
have the same problems - they seem to be in ginac.

Ondrej

On Mon, Oct 03, 2005 at 08:27:02PM +0300, Matti Peltom=E4ki wrote:
>=20
> Hi,
>=20
> while playing around with pyginac, I found some strange behavior with
> the regression test suite. The reason to this behavior escapes me.=20
>=20
> When running ./run python2.3 bin/checkall.py most of the time=20
> everything goes ok, but every now and then an error occurs as follows.
>=20
> *****************************************************************
> Failure in example: exam_normal2()
> from line #1 of exam_normalization.exam_normal2
> Expected: 0
> Got:
> normal form of  (-y+x)**(-3)*(x**2-y**2)  returned  (y-x)**(-2)*(y+x)  wh=
ile=20
> it should be  (-y+x)**(-2)*(y+x)
> normal form of  (x**2-y**2)*(-1/2*y+1/2*x)**(-3)  returned =20
> (y-x)**(-2)*(8*y+8*x)  while it should be  (-y+x)**(-2)*(8*y+8*x)
> 2
> *****************************************************************
> 1 items had failures:
>    1 of   1 in exam_normalization.exam_normal2
> ***Test Failed*** 1 failures.
>=20
> To get an idea how often this happens, I ran the test 100 times, giving 4=
3=20
> fails and 57 clean passes. The interesting part is that when I run
>=20
> ./run python2.3 exam_normalization.py
>=20
> I am totally unable to reproduce the error above, leading to a guess that=
 this=20
> might have something to do with bin/checkall.py and/or the doctest module=
=2E=20
> However, I have not been able to point any other reason as to why this wo=
uld=20
> be so. I also checked with python 2.4 (Debian-packaged 2.4.1-2) with the =
same=20
> results.=20
>=20
> I am using an up-to-date cvs of PyGiNaC, GiNaC 1.3.2, boost.python 1.33.0,
> gcc 4.0.2 and python 2.3.5 (all of which except PyGiNaC are the=20
> Debian-packaged versions).=20
>=20
> So, does anybody have any idea? Can anyone else reproduce this?
>=20
> Best Regards,
> Matti
>=20
> --=20
> Matti Peltomaki, Graduate Student
> Laboratory of Physics=20
> Helsinki University of Technology
> 040 7576977, room Y308B
>=20
>=20
>=20
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Power Architecture Resource Center: Free content, downloads, discussions,
> and more. http://solutions.newsforge.com/ibmarch.tmpl
> _______________________________________________
> Pyginac-users mailing list
> Pyg...@li...
> https://lists.sourceforge.net/lists/listinfo/pyginac-users

Re: [PyGiNaC-users] strange behavior with the test suite

[Jonathan B. <jbr...@ea...>]

On Mon, 2005-10-03 at 20:27 +0300, Matti Peltom=E4ki wrote:
> Hi,
>=20
> while playing around with pyginac, I found some strange behavior with
> the regression test suite. The reason to this behavior escapes me.=20
>=20
> When running ./run python2.3 bin/checkall.py most of the time=20
> everything goes ok, but every now and then an error occurs as follows.
>=20
> *****************************************************************
> Failure in example: exam_normal2()
> from line #1 of exam_normalization.exam_normal2
> Expected: 0
> Got:
> normal form of  (-y+x)**(-3)*(x**2-y**2)  returned  (y-x)**(-2)*(y+x)  wh=
ile=20
> it should be  (-y+x)**(-2)*(y+x)
> normal form of  (x**2-y**2)*(-1/2*y+1/2*x)**(-3)  returned =20
> (y-x)**(-2)*(8*y+8*x)  while it should be  (-y+x)**(-2)*(8*y+8*x)

I think that this is a canonicalization issue.  If hash(x) is greater
than hash(y) (when reinterpret_cast'ed back to unsigned ints), then
pow(x - y, 2) normalizes to pow(y - x, 2)

I'm not sure if this behavior is deliberate so I'm bringing it up on
ginac-list.  In the meantime, a patch has been committed to CVS that
works around the error.

-Jonathan

Re: [PyGiNaC-users] strange behavior with the test suite

[<pp...@fy...>]

Hi, 

> I think that this is a canonicalization issue.  If hash(x) is greater
> than hash(y) (when reinterpret_cast'ed back to unsigned ints), then
> pow(x - y, 2) normalizes to pow(y - x, 2)
> 
> I'm not sure if this behavior is deliberate so I'm bringing it up on
> ginac-list.  In the meantime, a patch has been committed to CVS that
> works around the error.

I'm afraid there is something wrong in the workaround. I did a cvs update 
getting your changes (bin/exam_normalization.py and ChangeLog) and rerun 
the test (again several times since it does not fail every time), and I 
still get the error sometimes. 

Are you still able to reproduce the error or is it just me? 

Regards,
Matti

-- 
Matti Peltomaki, Graduate Student
Laboratory of Physics 
Helsinki University of Technology
040 7576977, room Y308B

Re: [PyGiNaC-users] strange behavior with the test suite

[Jonathan B. <jbr...@ea...>]

On Tue, 2005-10-04 at 20:14 +0300, Matti Peltom=E4ki wrote:
> Hi,=20
>=20
> > I think that this is a canonicalization issue.  If hash(x) is greater
> > than hash(y) (when reinterpret_cast'ed back to unsigned ints), then
> > pow(x - y, 2) normalizes to pow(y - x, 2)
> >=20
> > I'm not sure if this behavior is deliberate so I'm bringing it up on
> > ginac-list.  In the meantime, a patch has been committed to CVS that
> > works around the error.
>=20
> I'm afraid there is something wrong in the workaround. I did a cvs update=
=20
> getting your changes (bin/exam_normalization.py and ChangeLog) and rerun=20
> the test (again several times since it does not fail every time), and I=20
> still get the error sometimes.=20
>=20
> Are you still able to reproduce the error or is it just me?=20

I guess I didn't run the test often enough before I committed the fix
last night, because you are right.  Whatever the cause of the error, I'm
inclined to let it slide since this is mathematically correct, and
because (pow(x-y, 2)/pow(y-x,2)).normal() evaluates to numeric(1).

In the current fix, if the test fails one way, it is retried with the
second option.

-Jonathan