Open giuspugl opened 2 years ago
Not a bug, just a different design. Now you instantiate the focalplane and use the load_hdf5() method. For an example, see toast_sim_ground.
Although the design is intentional, @giuspugl is correct that the docstring was stale. I just fixed that in the branch where I am working on the documentation.
The reasoning behind the design choice was that it was too easy to accidentally open the same HDF5 file from multiple processes causing file contention. I was trying to make the user decide what they really wanted to do by opening the file and calling the load_hdf5()
method (the load / save methods are also used when writing / reading the focalplane to a Group of a larger HDF5 file). However, this change does make things less convenient for loading data serially from a notebook for example. We could restore "filename" constructor argument, at the risk of it being misused in parallel workflows. Thoughts?
Thanks @keskitalo and @tskisner for reviewing this issue! I am fine with calling load_hdf5()
as you 're recommending, as i agree this is less convenient and intuitive for serial runs from a notebook.
However, I think that if the docstring encodes this detail (perhaps with some example for serial runs as @keskitalo did above) and it is documented then I don't see anything against adopting a unified approach. we will be surely safe by misuses in parallel workflows.
Ok, good point. I will include an example in the docstring. Will leave this open until that is fixed.
the latest toast3 version is missing a file argument in case we want to provide a focalplane file , althouhg it is expected to be there from the docstring . see https://github.com/hpc4cmb/toast/blob/60bcff20775dee83696d44e151f8a8d3ad3b0038/src/toast/instrument.py#L441
is it just a bug ?