Question

I’m using h5py to access HDF5 files and store the h5py File objects in a class. But I’m experiencing some strange behavior in attempting to reassign a closed h5py file instance variable with a new one:

class MyClass:
    def __init__(self, filename):
        self.h5file = None
        self.filename = filename

    def vartest(self):
        self.h5file = h5py.File(self.filename, 'r')
        print self.h5file
        self.h5file.close()
        print self.h5file
        newh5file = h5py.File(self.filename, 'r')
        print newh5file
        self.h5file = newh5file
        print self.h5file
        print newh5file

def main():
    filename = sys.argv[1]
    mycls = MyClass(filename)
    mycls.vartest()

Output:

<HDF5 file "test.h5" (mode r, 92.7M)>
<Closed HDF5 file>
<HDF5 file "test.h5" (mode r, 92.7M)>
<Closed HDF5 file>
<Closed HDF5 file>

Attempting to update the instance variable with the newly opened h5py File object appears to have somehow affected the state of the object, closing it. Regardless of the implementation on the h5py side, I don’t see how this behavior makes sense from my understanding of the Python language (i.e., no overloading of the assignment operator).

This example is run with Python 2.6.5 and h5py 1.3.0. If you want to try this example but don’t have an HDF5 file sitting around you can just change the file access mode from ‘r’ to ‘a’.

Answer 1

Yes, this is a known bug in h5py 1.3, which shows up when you use HDF5 1.8.5 or newer. It’s related to changes in the way identifiers are handled in 1.8.5. You can fix it by using HDF5 1.8.4 or earlier, or by upgrading to h5py 2.0.