fromstring/fromfile behave badly with malformed data (Trac #883) #1481

Original ticket http://projects.scipy.org/numpy/ticket/883 on 2008-08-09 by @pv, assigned to unknown.

See the discussion at:

http://permalink.gmane.org/gmane.comp.python.numeric.general/24115

In summary:

1. Empty string is understood as zero (except at the end of a string)

   np.fromstring('xxxxx', sep='x')
   array([ 0., 0., 0., 0., 0.])

   Missing values are treated as zeros. This is probably accidental, it's more likely that the memory allocated just happens to be filled with zeros...

2. If count is given and not enough entries were read, the rest is padded with uninitialized data

   np.fromstring("23", dtype=np.float, sep="x", count=3)
   array([ 2.30000000e+001, 8.19397657e-269, 1.94625366e-312])

3. Fromfile behaves differently from fromstring (Assume "foo.dat" contains string "23xxx")

   np.fromfile("foo.dat", dtype=np.float, sep="x")
   array([ 2.30000000e+001, 8.36783574e-269, 2.12885402e-313])
   np.fromfile("foo.dat", dtype=np.float, count=5, sep="x")
   5 items requested but only 3 read
   array([ 2.30000000e+001, 6.04468244e-269, -1.54891786e-039])

   Again, the last value is ignored, but the former ones are not treated as zero; instead, the array is filled with uninitialized data for the missing entries. Note also that neither warning or error is raised; instead, a message is printed to stdout.

Question: what is the correct behavior:

A) Raise ValueError on malformed data.

B) Raise ValueError on malformed data if count is given, otherwise truncate to valid part.

C) Truncate to valid part of data.