Logic error in smooth[_units] function argument parsing

[nickbattam-tessella]

The sqw/dnd functions smooth and smooth_units call smooth_dnd.

The argument parsing logic in this function means it's impossible to call with the resolution shape function without an error being raised:

if strcmp(shape,'resolution') && ~(ndim==2 && numel(width)==3)
    error ('Smoothing option ''resolution'' only available for 2D data sets with three width parameters')
elseif ~(isa_size(width,[1,ndim],'double') || isa_size(width,[1,1],'double'))
    error ('ERROR: argument ''width'' must be a scalar or vector with length equal to the dimensions of the dataset')

The first if requires that the data is 2d with 3 width parameters (for resolution), the second if requires that width is an array of size [1, 2], or scalar for all shape functions in the 2d case.

These are mutually exclusive requirements.

---

update CM:22/09/2023: confirmed that this issue continues. The test test_dnd_class/test_smooth::test_smooth_resolution_shape_arg fails due to this issue and is currently skipped. PR #1291 has been made to fix the need to replace the d2d call with a read_dnd call. With this fix the call to smooth_dnd_ (via d2d.smooth) fails as described above.
smooth_dnd_ is a private function of @DnDBase and contains the above code for if strcmp(shape,'resolution') . As described above, it is called from smooth and other places. Note that the method declaration for smooth_dnd in SQWDnDBase.m (without the underscore) is not defined; only the private method with the underscore has an implementation.
For the elseif branch of the error detection above, this error is deliberately set off at L.126 of test_smooth.m in an AssertExceptionThrown. Otherwise the other tests in test_smooth, mostly hats with one gaussian, do not have either of these errors.
The failure caused by the problem described above occurs with shape=='resolution'. ndim==2 (it is a d2d) and numel(width)==3 (as is apparently required for resolution) so the first error does not occur. But as numel(width) is neither ==2 or ==1 the second error does. The problem described above is real.
My suspicion is that the second error should also be triggered if ~strcmp('resolution') (obviously there are better ways of packaging the if structure). At this point an expert on smooth and its arguments should have a look; I have assigned Toby and Alex and will @ them below.

[cmarooney-stfc]

@cmarooney-stfc check if this code is still present

[cmarooney-stfc]

@abuts @tgperring Toby, Alex - I was charged to see if this problem still exists - it does. I have proposed a fix above, but need a smoothing expert to verify - could you both have a look please? thanks Chris