[Numpy-discussion] NumPy & SciPy with Snow Leopard 64-bit Py-2.6.4

[Nils Wagner]

On Mon, 1 Mar 2010 15:49:52 -0500
  josef.pktd at gmail.com wrote:
> On Mon, Mar 1, 2010 at 3:23 PM, Tom Loredo 
><loredo at astro.cornell.edu> wrote:
>>
>> Just wanted to report qualified success installing NumPy 
>>& SciPy under
>> a 64-bit build of Python-2.6.4 (universal framework) on 
>>OS X 10.6.2
>> (current Snow Leopard).  I am using the current SVN 
>>checkouts
>> (numpy r8270, scipy r6250).
>>
>> NumPy has installed successfully for some time now and 
>>the current
>> SVN maintains this:
>>
>>>>> numpy.test()
>> ..
>> Ran 2521 tests in 8.518s
>> OK (KNOWNFAIL=4, SKIP=1)
>>
>> SciPy has been causing me problems for weeks with 
>>segfaults with
>> IFFT tests, as reported on scipy-dev (no one ever 
>>responded to
>> this so I made no progress in diagnosing it):
>>
>> http://mail.scipy.org/pipermail/scipy-dev/2010-February/013921.html
>>
>> However, r6250 now runs scipy.test() without segfault, 
>>though with
>> 22 errors and 2 failures.  The full tests give:
>>
>>>>> scipy.test('full')
>> ..
>> Ran 4982 tests in 652.478s
>> FAILED (KNOWNFAIL=13, SKIP=27, errors=22, failures=6)
>>
>> I am not familiar with what many of the tests are 
>>covering, so I
>> cannot assess the severity of all the errors and 
>>failures.  Some
>> of them seem to be bugs in the tests (e.g., use of an 
>>unexpected
>> keyword "new" in several histogram tests);
> 
> np.histogram(rvs,histsupp,new=True)
> 
> this still needs to be changed in scipy.stats, because 
>new keyword has
> been removed in numpy trunk 2 weeks ago.
> 
> Josef
> 
> others are innocuous
>> (e.g., missing PIL, which I haven't installed yet). 
>> I've posted
>> the report here:
>>
>> http://www.pastie.org/848651
>>
>> I'd appreciate comments on which issues are nontrivial 
>>and
>> deserve attention to as a SciPy user.  E.g., the first 
>>error
>> is in an lapack test and involves a ValueError where 
>>infs or
>> NaNs appear where they shouldn't.  Is this a bug in the 
>>test,
>> or does it indicate a 64-bit issue that is making 
>>inf/NaN
>> appear where it shouldn't?  E.g., there are arpack 
>>errors,
>> but I don't know what the "Error info=-8" message 
>>signifies.
>> Other arpack errors are due to large solution 
>>mismatches,
>> which I presume are serious and deserve attention.
>>
>> Thanks,
>> Tom Loredo
>>
>>

Last but not least there is an annoying bug lurking around 
for 17 months
http://projects.scipy.org/numpy/ticket/937
  
Nils