arguments block seems wrong

the arguments block can be used to validate input arguments. Great! Except that it doesn't seem to work as advertised.

According to the documentation (doc arguments)

"The value assigned to the argument in the function call must be compatible with the specified size, or MATLAB throws an error."

However, when calling this function

function a=f(a)
    arguments
        a (1,3) double
    end
end

with a scalar

>> f(1)

ans =

     1     1     1

The function expands the erroneous scalar argument to a 3-element vector instead of casting the error that was promised!

Why would this be? Am I reading it wrong?

[Duijnhouwer]

>> disp(version)
9.8.0.1417392 (R2020a) Update 4

Answer 1

[Walter Roberson]

Documented. The size of the input must be "compatible" taking into account implicit expansion.

https://www.mathworks.com/help/matlab/matlab_prog/function-argument-validation-1.html#mw_48fe4f61-8ee4-424b-ae2b-c82b8d051910

[Duijnhouwer]

Wow. Why would would anyone want that? Seems to completely defeat the purpose.

[Walter Roberson]

With implicit expansion having been the rule since R2016b, compatible is probably the majority of the need now.

[Duijnhouwer]

But the arguments block syntax was only introduced in R2019a, no?

[Bruno Luong]

Only scalar expansion is tolerated by ARGUMENTS, which exists in MATLAB well before R2016b

function a=f(a)
    arguments
        a (5,3) double
    end
end

I don't know what Walter meant by implicit expansion, but the singleton expansion is not tolerated.

>> f(rand(5,3))

ans =

    0.6229    0.9431    0.6980
    0.2473    0.5422    0.9468
    0.9195    0.5459    0.2302
    0.0582    0.1777    0.2003
    0.9456    0.8473    0.7904

>> f(rand(1))

ans =

    0.0317    0.0317    0.0317
    0.0317    0.0317    0.0317
    0.0317    0.0317    0.0317
    0.0317    0.0317    0.0317
    0.0317    0.0317    0.0317

>> f(rand(5,1))
Error using f
Invalid argument at position 1. Value must be a matrix of size 5-by-3.
 
>> f(rand(1,3))
Error using f
Invalid argument at position 1. Value must be a matrix of size 5-by-3.
 
>> 

[Walter Roberson]

Ah, this part seems to be the key:

MATLAB indexed assignment rules apply to size specifications. For example, a 1-by-1 value is compatible with the size specified as (5,3) because MATLAB applies scalar expansion. Also, MATLAB row-column conversion applies so that a size specified as (1,:) can accept a size of 1-by-n and n-by-1.