TypeError: bad operand type for unary ~: 'NoneType' in _loadWeights_handler during Predict-only mode

[flomertens]

DDFacet crashes when running in --Output-Mode=Predict with --CF-wmax set to non-zero.

Traceback

File "ClassVisServer.py", line 951, in _loadWeights_handler
uvmax_wavelengths = abs(uv[~rowflags,:]).max() * msfreqs.max() / _cc
TypeError: bad operand type for unary ~: 'NoneType'

Root cause

The bug is a chain of steps:

1. Predict-only mode triggers IgnoreWeights(), setting _ignore_vis_weights=True (this also requires --CF-wmax to be non-zero, as the code explicitly skips IgnoreWeights() when wmax==0)
2. This causes wmax_only=True to be passed to _loadWeights_handler
3. Which passes uvw_only=True to readWeights
4. readWeights has already computed rowflags from the FLAG column, but discards it in the uvw_only early return:

if uvw_only:
    return uvw, None, None, None  # rowflags is thrown away

5. Back in _loadWeights_handler, uvw is valid so the if uvw is None guard is not triggered, and the code crashes on ~rowflags

This affects any dataset with at least one unflagged row, i.e. essentially all real datasets running in Predict mode.

potential Fix?

One line change in readWeights:

if uvw_only:
    return uvw, None, rowflags, None

To reproduce

Any DDF Predict run with non-zero --CF-wmax, e.g.:

[cyriltasse]

Thanks @flomertens for the report. I have fixed it i think in the branch I'm currently trying to merge