What about:

jd.('key1').('subkey2')(1)

I am trying to write a general hierarchical data container, jdict, to provide a uniform access and modification interface for struct/containers.Map/dictionary data for both matlab and octave, the current code is at

https://github.com/fangq/jsonlab/blob/afce1de01e0c0e7deed687281bba10a190a86c40/jdict.m

for example, for a complex structure such as

obj = struct('key1', struct('subkey1',1, 'subkey2',[1,2,3]), 'subkey2', 'str');
obj.key1.subkey3 = {8,'test',containers.Map('subsubkey1',0)};

jd = jdict(obj);

I could use various ways to various ways to retrieve and modify sub-element values.

However, on MATLAB, my overloading subsref and subsasgn operators could also use .(i) to read and write element values in Octave.

For example, the below call using the above data

jd.('key1').('subkey2').(1)

is expected to return the 1st element of obj.key1.subkey2, which is 1, but Octave directly threw the following error without even calling either of these indexing operator functions

error: dynamic structure field names must be strings

In comparison, MATALB successfully pass the operator struct('type','()', 'subs', 1) to subsref or subsasgn operators, and allow me to read/write such values.

It appears that the interpreter is doing some error checking before passing on to subsref or subsasgn, but it seems more appropriate to handle this inside the indexing operators themselves.

The expected behavior is to allow accepting .(i) in a classdef's overloading subsref or subsasgn operators.