Breaking changes to jsondecode between 2018a and 2018b?

Problem: The same code and files that worked for jsondecode in 2018a and earlier stopped working in 2018b and beyond.

Exception: JSON text must be a character vector or a scalar non-missing string.

The attached article.txt file was written by

json = jsonencode(structObj);
fID = fopen('./data/article.txt', 'w', 'n', 'UTF-8');
fwrite(fID, json, 'char');
fclose(fID);

I have attempted to read it back in by:

fileName = './data/article.txt';
fID = fopen(fileName, 'r', 'n', 'UTF-8');
readIn = native2unicode(fread(fID), 'UTF-8');
value = jsondecode(readIn);
fclose(fID);

This works in 2018a but throws the exceptions at the jsondecode line in 2018b and later

Answer 1

[Walter Roberson]

readIn = native2unicode(fread(fID, [1 inf]), 'UTF-8');

That is, you were trying to send a column vector to jsondecode and it needs to be a row vector.

[Zachary David]

Hi Walter,

Thank you very much! That appears to have fixed it.

Do you know the specific change in 2018b that made jsondecode start treating (n,1) and (1,n) vectors differently? Both row and column work in 2018a

[Walter Roberson]

Sorry, I do not have any reference. The input is documented in R2018a as "character vector" and in MATLAB documentation that has always meant character row vector, with N x 1 being called a character array.