fromfile error on windows, not mac

Scott David Daniels Scott.Daniels at Acm.Org
Wed Jul 23 06:06:47 CEST 2008

jadamwil wrote:
> On Jul 22, 2:05 am, Uwe Schmitt <rocksportroc... at>
> wrote:
>> jadamwil schrieb:
>>> Hello,
>>> I am using the numpy fromfile function to read binary data from a file
>>> on disk. The problem is that the program runs fine on a Mac, but gives
>>> an error or warning on windows when trying to read the data. I use it
>>> like this:
>>> Signal = zeros((N, 16), dtype=float32)
>>> for sample in range(0, N):
>>>   # this function gets the next position in the file to seek to
>>>   s = getFilePos(sample)
>>>   # go to the correct location in the file; this IS checked to make
>>> sure it is within the file
>>>   # read the 16 float32 values from the file
>>>   D = fromfile(mFile, dtype=numpy.float32, 16)
>>>   # save D in Signal
>>>   Signal[sample, :] = D
>>> This will fail when sample is ~4. If I change the range to (5,N),
>>> skipping the "bad" file location, it will run fine for a few samples,
>>> and then give another error. The message it gives is:
>>> "16 items requested but only 7 read"
>>> So D is a 7x1 vector, and the program dies when it tries to assign D
>>> to the slice of Signal ("ValueError: shape mismatch: objects cannot be
>>> broadcast to a single shape").
>>> On windows, the Python version is 2.5.2, and the most recent numpy and
>>> scipy are being used as well. I tried using Enthought, but it gave
>>> this error as well, in addition to a c runtime error whenever I
>>> imported scipy (which is another post topic...).
>>> Any ideas on what might be causing this? Is there a way to debug the
>>> fromfile function? And, remember, this works perfectly on a Mac. Would
>>> compiling everything (python, scipy, numpy) potentially solve this?
>> Did you open the file in binary mode ?
>> Greetings, Uwe
> Yes I did. Would that make a difference between a mac and windows?
> Thanks

In a word, yes.
the values for bytes that have issues are 10 and 13.

--Scott David Daniels
Scott.Daniels at Acm.Org

More information about the Python-list mailing list