Add support for scale-offset filter

[sfc-gh-mrojas]

What is the suggestion?
Well I tried using some files like https://fly.mpi-cbg.de/~pietzsch/bdv-example/drosophila.h5
However when I try to get the dataset i cant because we dont have a scale-offset filter.

How would it be an improvement?

If support for that filter can be added that will be great, so this library can be used in larger scenarios

Additional context

The file is big so download it from: https://fly.mpi-cbg.de/~pietzsch/bdv-example/

[Apollo3zehn]

I tried to implement this once for a different project and gave up eventually. It is a complex, heavily macro-overloaded 1700 LOC monster (https://github.com/HDFGroup/hdf5/blob/develop/src/H5Zscaleoffset.c). However, maybe I was too impatient and in the end it is easier than I found it at that time. It would be great to see a non-C scale offset filter implementation.

[jamesmudd]

Thanks for looking at jHDF and finding this issue. Your right scale and offset filter is not currently implemented. Thanks for providing an example file. I will take a look at how involved an implementation would be when I get a little time.

[jamesmudd]

For ref https://docs.hdfgroup.org/hdf5/develop/group___d_c_p_l.html#ga5c10165b670e0e984db431aee818cc7e

[jamesmudd]

I have taken a bit of a look at this and its not trivial. I am finding it difficult to get a spec about the on disk format so trying to reverse engineer it but that is proving harder than I expected. So just an update I don't expect this to be added quickly.

[jbfaden]

I've run into a file with this, too. ChatGPT says I can use

h5repack -f NONE input.h5 output_unfiltered.h5

to remove the encoding, which works for me as a work-around.

I'd love to see this added, too.

[jamesmudd]

There is now a alpha build https://jitpack.io/#jamesmudd/jhdf/scale-offset-2-ab9f561da9-1 available that should add scale offset filter. If anyone could give it a try feedback much appreciated. The code is mostly AI written and needs some cleanup before it can be merged but functionally looks ok.