Re: [Numpy-discussion] Array bug with character (regression from 1.4.0)
Ryan May wrote: The following code, which works with numpy 1.4.0, results in an error: Python 2.6, I presume? In [1]: import numpy as np In [2]: v = 'm' In [3]: dt = np.dtype('c') In [4]: a = np.asarray(v, dt) On SVN trunk: ValueError: assignment to 0-d array In [5]: np.__version__ Out[5]: '2.0.0.dev8297' Thoughts? Nope, but it's likely my bad. Smells a bit like the dtype 'c' has size 0 that doesn't get automatically adjusted in the array constructor, so you end up assigning size-1 string to size-0 array element... Which is strange sice I don't think this particular code path has been changed at all. One would have to follow the C execution path with gdb to find out what goes wrong. -- Pauli Virtanen ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
Re: [Numpy-discussion] Array bug with character (regression from 1.4.0)
On Sun, Mar 21, 2010 at 4:08 PM, Pauli Virtanen p...@iki.fi wrote: Ryan May wrote: The following code, which works with numpy 1.4.0, results in an error: Python 2.6, I presume? In [1]: import numpy as np In [2]: v = 'm' In [3]: dt = np.dtype('c') In [4]: a = np.asarray(v, dt) On SVN trunk: ValueError: assignment to 0-d array In [5]: np.__version__ Out[5]: '2.0.0.dev8297' Thoughts? Nope, but it's likely my bad. Smells a bit like the dtype 'c' has size 0 that doesn't get automatically adjusted in the array constructor, so you end up assigning size-1 string to size-0 array element... Which is strange sice I don't think this particular code path has been changed at all. One would have to follow the C execution path with gdb to find out what goes wrong. I was wondering if this was related to Michael's fixes for character arrays? A little bisection might help localize the problem. Chuck ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
Re: [Numpy-discussion] Array bug with character (regression from 1.4.0)
su, 2010-03-21 kello 16:13 -0600, Charles R Harris kirjoitti: I was wondering if this was related to Michael's fixes for character arrays? A little bisection might help localize the problem. It's a bug I introduced in r8144... I forgot one *can* assign strings to 0-d arrays, and strings are indeed one sequence type. I'm going to back that changeset out, since it was only a cosmetic fix. That particular part of code needs some cleanup (it's a bit too hairy if things like this can slip), but I don't have the time at the moment to come up with a more complete fix. Cheers, Pauli ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
Re: [Numpy-discussion] Array bug with character (regression from 1.4.0)
On Sun, Mar 21, 2010 at 4:29 PM, Pauli Virtanen p...@iki.fi wrote: su, 2010-03-21 kello 16:13 -0600, Charles R Harris kirjoitti: I was wondering if this was related to Michael's fixes for character arrays? A little bisection might help localize the problem. It's a bug I introduced in r8144... I forgot one *can* assign strings to 0-d arrays, and strings are indeed one sequence type. I'm going to back that changeset out, since it was only a cosmetic fix. That particular part of code needs some cleanup (it's a bit too hairy if things like this can slip), but I don't have the time at the moment to come up with a more complete fix. Lots of the code needs some cleanup ;) The first step was to reformat it -- still ongoing, I've got changes on hold for after the release -- and break it up into smaller files with some similarity of function. At some point the code needs to be documented and some macros relaced with inline functions. All macros that implement jumps should be removed and replaced by something more transparent. I'd also like to see some of the policy type stuff move up to cython, which should be easier to document and understand, not to mention making for a cleaner interface to Python. So on and so on... We could use a few more developers, especially with David getting a real job. Chuck ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
Re: [Numpy-discussion] Array bug with character (regression from 1.4.0)
On Sun, Mar 21, 2010 at 5:29 PM, Pauli Virtanen p...@iki.fi wrote: su, 2010-03-21 kello 16:13 -0600, Charles R Harris kirjoitti: I was wondering if this was related to Michael's fixes for character arrays? A little bisection might help localize the problem. It's a bug I introduced in r8144... I forgot one *can* assign strings to 0-d arrays, and strings are indeed one sequence type. I'm going to back that changeset out, since it was only a cosmetic fix. That particular part of code needs some cleanup (it's a bit too hairy if things like this can slip), but I don't have the time at the moment to come up with a more complete fix. That fixed it for me, thanks for getting done quickly. What's amusing is that I found it because pupynere was failing to write files where a variable had an attribute that consisted of a single letter. Ryan -- Ryan May Graduate Research Assistant School of Meteorology University of Oklahoma ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
[Numpy-discussion] Array bug with character (regression from 1.4.0)
The following code, which works with numpy 1.4.0, results in an error: In [1]: import numpy as np In [2]: v = 'm' In [3]: dt = np.dtype('c') In [4]: a = np.asarray(v, dt) On 1.4.0: In [5]: a Out[5]: array('m', dtype='|S1') In [6]: np.__version__ Out[6]: '1.4.0' On SVN trunk: /home/rmay/.local/lib/python2.6/site-packages/numpy/core/numeric.pyc in asarray(a, dtype, order) 282 283 -- 284 return array(a, dtype, copy=False, order=order) 285 286 def asanyarray(a, dtype=None, order=None): ValueError: assignment to 0-d array In [5]: np.__version__ Out[5]: '2.0.0.dev8297' Thoughts? (Filed at: http://projects.scipy.org/numpy/ticket/1436) Ryan -- Ryan May Graduate Research Assistant School of Meteorology University of Oklahoma ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion