Ok. sympy does not support AD, so that was a bad idea.
Scientific Python is licensed under the CeCILL-C license. It looks lke
it is a French law adapted version of LGPL. I guess that license would
be sufficient to bundle the part of Scientific Python we need in FIAT,
and then try to port that portion of it.
Johan
On 05/17/2012 12:53 PM, Kirby, Robert C wrote:
I had thought the scientific license allowed me to drop in the ad
file from it, which I had done at one point. Doing this is far
preferable to the dependency. I have company in town& baby due any
day, so if one of you or a delegate could investigate, this would be
great.
Sympy would be quite difficult, as it doesn't support AD (I think).
It would be a significant change, and also have a performance
penalty, I think.
Rob ________________________________________ From: Anders Logg
[[email protected]] Sent: Thursday, May 17, 2012 2:50 AM To: Johan Hake
Cc: Garth N. Wells; DOLFIN Mailing List; FIAT Mailing List; Kirby,
Robert C Subject: Re: [Dolfin] Problems with DOLFIN and SWIG 2.0.5 on
Debian ia64 platform
On Wed, May 16, 2012 at 12:37:08PM +0200, Johan Hake wrote:
On 05/16/2012 11:12 AM, Garth N. Wells wrote:
On 16 May 2012 10:05, Johan Hake<[email protected]> wrote:
I get a taste of the path we have tried to avoid...
A python3 port of all our packages :(
I had a look a while ago, and the hang-up was python-scientific,
on which FIAT depends. python-scientific doesn't look to be
actively maintained.
Looks like FIAT only uses automatic differentiation from
Scientific. Could that be accomplished with sympy? The next release
of sympy should support python3.
I don't know. It's something Rob would need to look at.
Rob, can this be done with SymPy? We're thinking about the
possibilities of running FEniCS with Python3 and the
python-scientific module used in FIAT seems to be the major hangup.
-- Anders
Johan
Garth
Johan
On 05/16/2012 10:27 AM, Johannes Ring wrote:
On Wed, May 16, 2012 at 10:16 AM, Johan
Hake<[email protected]> wrote:
On 05/16/2012 10:02 AM, Johannes Ring wrote:
You mean to replace PyInt_AS_LONG with PyInt_As_Long in
both UFC and DOLFIN?
Well, yes, but is really silly. It looks like there are
some other things that does not work. Is Python.h really
found?
Yes, but I just found this in the log [1]:
-- Found PythonInterp: /usr/bin/python (found version
"2.7.3") -- Found PythonLibs:
/usr/lib/python3.2/config/libpython3.2.so (found version
"3.2.3rc2")
It seems that Python 3.2 is also installed on the build
machine and I guess DOLFIN picks up Python.h from this
version. I will fix this in the Debian package. Thanks for
the help.
[1]
https://buildd.debian.org/status/fetch.php?pkg=dolfin&arch=ia64&ver=1.0.0-5&stamp=1337006055
Johannes
_______________________________________________ Mailing list:
https://launchpad.net/~dolfin Post to :
[email protected] Unsubscribe :
https://launchpad.net/~dolfin More help :
https://help.launchpad.net/ListHelp
_______________________________________________ Mailing list:
https://launchpad.net/~dolfin Post to :
[email protected] Unsubscribe :
https://launchpad.net/~dolfin More help :
https://help.launchpad.net/ListHelp
_______________________________________________
Mailing list: https://launchpad.net/~dolfin
Post to : [email protected]
Unsubscribe : https://launchpad.net/~dolfin
More help : https://help.launchpad.net/ListHelp