sub2ind() errors out with NaN inputs - handled differently since R2024a

>> matlabRelease

ans =

matlabRelease with properties:

Release: "R2022b"

Stage: "release"

Update: 3

Date: 17-Nov-2022

>> sub2ind([ 3 5 7], [ 3 3], [2 2], [1 nan])

ans =

6 NaN

versus

>> matlabRelease

ans =

matlabRelease with properties:

Release: "R2024b"

Stage: "release"

Update: 4

Date: 26-Dec-2024

>> sub2ind([ 3 5 7], [ 3 3], [2 2], [1 nan])

Error using sub2ind (line 71)

Out of range subscript.

[Walter Roberson]

That change probably came with R2024a, which had the following documented change to sub2ind:

The sub2ind (R2024a) function now supports scalar expansion and accepts a mix of scalars and vectors for subscript inputs. For example, sub2ind(sz,[1 2 3],2) is now the same as sub2ind(sz,[1 2 3],[2 2 2]). Previously, the subscript inputs were required to be the same size.

[Eric Machorro]

I suspect you are correct - but then the issue is that R2024a added some nice functionality, yet removed a minor aspect of its backward compatibility. If that removal wasn't intentional, maybe it could be undone? Either way I would think it should be documented or fixed in subsequent release.

Answer 1

[Matt J]

You can get the old behavior back if you need it,

Sub2Ind([ 3 5 7], [ 3 3], [2 2], [1 nan])

ans = 1×2

6 NaN

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

function out=Sub2Ind(sz,varargin)

  out=varargin{1};
  for i=2:numel(varargin)

      out=out+(varargin{i}-1)*(sz(i-1));

  end
end

[Eric Machorro]

Thank you - looking at the original code for this , i would've expected something way more complicated. Thank you very much.

Answer 2

[Steven Lord]

This was a bug fix introduced in release R2024a.

[Eric Machorro]

The previous behavior was considered a bug? This is fine - although it might reasonably be expected to be a feature. At least we're just talking abnout definitions - "bug" vs. "feature".

[Steven Lord]

Generally, my mental model for the workflow for sub2ind (from the name) is that what comes out ought to be a valid linear index into an array that selects the same element as the subscripts you passed into the function. NaN is neither a valid subscript nor a valid linear index. Nor is 0.

I = sub2ind([10 10], 1, 0)

Error using sub2ind (line 71)
Out of range subscript.

Perhaps the error message could be more descriptive.

[Eric Machorro]

That is one way to handle it. And I agree that is a rather uninformative error message. But, it is a little deeper than that I think. If, by off chance, it it fed in a nan value, it could (and did prior to R2024a) return a nan for the appropriate index. The newer version removed that functionality, but added another facet of funcationality in being able to handle inputs of different sizes.

This would provide a higher degree of backward compatibility it would seem. But perhaps I am missing something. Not a major deal I suppose.