please backport python-numpy

[Timmie]

I filed this bug for hardy:
https://bugs.launchpad.net/ubuntu/+source/python-numpy/+bug/242669

It seems that I now need to ask for a backport.

Some month ago a new version of numpy was released. This is now the state of the art for numerical computing with python. Especially the masked array functions have improved a lot.
Every scientific programmer on Ubuntu would have the need to compile on his own if this is not backported.

Thanks in advance.

[Benjamin Thyreau (benji2)]

Hi,
I second this request. Numpy 1.1 is back-compatible with 1.0.4, and the Intrepid packages build out-of-the-box on hardy. (Tested with prevu). Already some scientific packages requires 1.1 (such as the improved matplotlib 0.98).
Thanks !

[Gael Varoquaux (gael-varoquaux)]

I tested it too and can confirm it works fine. It is indeed a requirement for matplotlib 0.98 (which is the reason why I need it, and built it from source). It also comes with a set of bugfixes. There is value in backporting it.

[Scott Kitterman (kitterman)]

This needs to be tested with its rdepends. Does the Hardy python-scipy work with this or does it need to be backported too? Also the Intrepid version will use Atlas if installed. Does this work on Hardy?

[Janne Blomqvist (blomqvist-janne)]

I don't require numpy 1.1 ATM, but due to

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=489726

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=489253

http://scipy.org/scipy/numpy/ticket/667

numpy dot() is just excruciatingly slow in hardy. The version in intrepid seems to have this fixed; I haven't benchmarked it but at least the package contains the _dotblas.so file which is the dot()<->BLAS interface.

Heck, since matrix multiplication performance for large matrices is two orders of magnitude slower than it ought to be, even though it's only a performance problem one could argue it's severe enough that this fix (not the entire numpy 1.1.x, mind) would be eligible for proposed-updates?

[Scott Kitterman (kitterman)]

motu-sru: Is this kind of performance hit SRU worthy?

This update will still need testing with the Hardy python-scipy regardless of if it's for -backpprts or -updates.

[Luca Falavigna (dktrkranz)]

If changes required does not require rebuilds of rdependencies or API changes, I think having a fix via -updates is good. Can you prepare a debdiff for this, just to see changes to introduce?

[Goyo (goyodiaz)]

This is not going to be fixed.