Read task out shape does not match PSRFITS data shape

[luojing1211]

In PR #119 I added the data writing functionality. The file handler shape is (nsample, nbin, npol, nchan), but the PSRFITS shape is (nsample, nbin, nchan, npol). So the

read(nsample, out = PSRFITS)

will report the bug of not the same shape. Where should the reshape happen?

[mhvk]

In your text above, the shapes are in fact the same... At long as nsample is first, I think the easiest route for now is somehing like:

reshape_to_fits = Reshape(result, <fits_shape>)
reshape_to_fits.read(nsample, out=...)

Maybe eventually we can adjust the Reshape task so that it works in front of a writer as well?

[luojing1211]

I made a mistake on the file shape and I just fixed it. Do you mean, we can add reshape task as part of the writer? I think it will work.

[mhvk]

OK, now makes more sense. So, yes, we can already do the reshape pre-write, and I agree that we should be able to make it work on the writer as well. Given properly named sample_shape, it could even become automatic...

[luojing1211]

To make it automatic, we should know the sample label from the upstream file hander. However, this information may be lost at some steps of the pipeline.

[mhvk]

Part of it we can infer from .frequency and .polarization, but it would be hard to do this reliably. Probably best for now to just stick with the Reshape option.