framecache writer yaml output

[donald-e-boyce]

The framecache writer was recently updated to write a single npz file as well as the older yaml files. However, the option to write yaml was not included in the standard save.write() interface. This is easily corrected, but unit tests should updated appropritately and new ones should be developed for the npz writer.

[donald-e-boyce]

The current framecache writer uses a keyword arg to write yaml, but in standard usage the writer doesn't called directly. It gets called through the save.write() function.

def write(self, output_yaml=False):

Instead, the save.write() function saves the keyword args, which are now accessible via the options property (new, formerly _opts).

        if self.options.get('output_yaml', False):
            self._write_yml()

[donald-e-boyce]

At this point, I have fixed the unit tests for the yaml-based frame cache and I have added new tests for npz-based frame cache. However, the npz-based frame cache has issues handling general metadata. It can save data of various python types to the array cache, but it converts them to a numpy array with a special dtype. When loading the file, there is no obvious way to recover the original type of the metadata.

In any case, the npz-based frame cache will need to be revisited if we want to use more general metadata. For now, it works fine with numpy arrays as metadata, and we could just merge in the changes and move on.

Any thoughts?

[joelvbernier]

The dtypes in the numpy arrays preserve the original data type (e.g. string vs int vs float), don't they?