umfpack error

[Josh Warner]

This would probably involve an API change, but instead of importing umfpack 
*at all* we could try `import pyamg` first. If pyamg is available, we 
wouldn't even attempt to import umfpack, instead using `'cg_mg'` instead of 
`'cg'`. This sidesteps stderr issues.

The approach only works because the results are equivalent, differing only 
in speed/memory. I don't know why we default to the worst brute force 
option, either. Seems like we could default to `None` and attempt a 
fallback scheme of pyamg -> umfpack -> brute force in order, with a 
different warning structure.

On Friday, October 4, 2013 8:07:39 AM UTC-5, Juan Nunez-Iglesias wrote:
>
> Ok so I'm having trouble actually suppressing this thing. The problem is 
> that to detect whether umfpack is present we try to build the 
> UmfpackContect() object:
>
> try:
>     from scipy.sparse.linalg.dsolve import umfpack
>     UmfpackContext = umfpack.UmfpackContext()except:
>     UmfpackContext = None
>
>
> The constructor raises an ImportError, which we duly catch (although that 
> line should really specify the error type, but whatever). Unfortunately, 
> the UmfpackContext class has a destructor (__del__) which then tries to 
> delete some objects that were never created and raises an AttributeError — 
> which, since we are already in a try: except: clause, gets ignored, hence 
> the message to stderr.
>
> Basically this is a bug in scipy I think?
>
> The standard approach is to temporarily redirect stderr to /dev/null, and 
> this works interactively:
>
> In [1]: from scipy.sparse.linalg.dsolve import umfpack
>
> In [2]: try:
>     UC = umfpack.UmfpackContext()except ImportError:
>     print 'hello'
>    ...:
> hello
> Exception AttributeError: "'UmfpackContext' object has no attribute '_symbolic'" in <bound method UmfpackContext.__del__ of <scipy.sparse.linalg.dsolve.umfpack.umfpack.UmfpackContext object at 0x103938250>> ignored
>
> In [3]: import sys, os
>
> In [4]: null = open(os.devnull, 'w')
>
> In [5]: null.write('hello!')
>
> In [6]: sys.stdout.write('hello!')
> hello!
> In [7]: sys.stderr.write('hello')
> hello
> In [8]: sys.stderr = null
>
> In [9]: sys.stderr.write('hello')
>
> In [10]: try:
>     UC = umfpack.UmfpackContext()except ImportError:
>     print 'hello'
>    ....:
> hello
>
> In [11]: sys.stderr = sys.__stderr__
>
> In [12]: sys.stderr.write('hello')
> hello
>
>
> However, doing this in the code doesn't work, even when I try to flush the 
> redirected stderr! (I'm not even sure that makes sense!)
>
> sys.stderr = open(os.devnull, 'w')try:
>     from scipy.sparse.linalg.dsolve import umfpack
>     UmfpackContext = umfpack.UmfpackContext()except:
>     UmfpackContext = Nonefinally:
>     sys.stderr.flush()
>     sys.stderr = sys.__stderr__
>
>
> Thoughts/ideas?
>
>
>
> On Fri, Oct 4, 2013 at 10:54 PM, Juan Nunez-Iglesias <jni.... at gmail.com<javascript:>
> > wrote:
>
>> Thanks! Although in this case I was using pip install *and* it was on an 
>> EC2 micro instance so I only have 1 CPU. =)
>>
>> But it is noted and very useful going forward. =)
>>
>>
>>
>> On Fri, Oct 4, 2013 at 10:32 PM, Ralf Gommers <ralf.g... at gmail.com<javascript:>
>> > wrote:
>>
>>>
>>>
>>>
>>> On Fri, Oct 4, 2013 at 1:33 PM, Juan Nunez-Iglesias <jni.... at gmail.com<javascript:>
>>> > wrote:
>>>
>>>> On Fri, Oct 4, 2013 at 8:59 PM, Johannes Schönberger <js... at demuc.de<javascript:>
>>>> > wrote:
>>>>  
>>>>> Are you sure you have SWIG installed?
>>>>>
>>>>
>>>> I wasn't and I didn't. =)
>>>>
>>>> Bloody hell scipy takes a long time to build. =)
>>>>
>>>
>>> $ time python setup.py build_ext -i
>>> real    12m6.843s
>>> user    6m3.244s
>>> sys     0m13.124s
>>>
>>> $ time ../Bento/bentomaker build -i -j   # -j == parallel build (on 4 
>>> cores here)
>>> real    1m46.466s
>>> user    2m54.628s
>>> sys     0m10.900s
>>>
>>> Blame distutils (and start using Bento) :)
>>>
>>> Ralf
>>>
>>>
>>>  
>>>
>>>>  Juan.
>>>>
>>>> -- 
>>>> You received this message because you are subscribed to the Google 
>>>> Groups "scikit-image" group.
>>>> To unsubscribe from this group and stop receiving emails from it, send 
>>>> an email to scikit-image... at googlegroups.com <javascript:>.
>>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>>
>>>
>>>  -- 
>>> You received this message because you are subscribed to the Google 
>>> Groups "scikit-image" group.
>>> To unsubscribe from this group and stop receiving emails from it, send 
>>> an email to scikit-image... at googlegroups.com <javascript:>.
>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20131004/c5744249/attachment.html>