AT-SPI: Wrong char reported by GetCharAtOffset when code point > 65535

When retrieving the character at the current cursor/caret position using the GetCharAtOffset method from the AT-SPI Text interface, the character is not correctly reported when its UTF-16 representation has 2 surrogate pairs, i.e. it has a Unicode code point above 65535.

Sample steps to reproduce:

1) build the attached sample program "qt-sample-app-character":

cmake .
make

2) run the app, which has a QTextEdit with the text "abc𐐷d" and moves the cursor to before the "𐐷" (U+10437) character
3) run the attached Python/pyatspi script that retrieves the character at the cursor position using the GetCharAtOffset method from the AT-SPI text interface tha the QTextEdit provides on Linux.

Actual result:
An incorrect code point is reported:

$ python3 pyatspi-script-print-char-at-cursor.py 
code point of char at cursor: 0xfffffff0
Traceback (most recent call last):
  File "/home/michi/git/computer-doc/TODO/misc/pyatspi-script-print-char-at-cursor.py", line 42, in <module>
    print('char at cursor: {}'.format(chr(char_at_cursor)))
                                      ^^^^^^^^^^^^^^^^^^^
OverflowError: Python int too large to convert to C int

Expected result:
The script should print the unicode codepoint of the "𐐷" character in hex ("0x10437") and the original character:

$ python3 pyatspi-script-print-char-at-cursor.py 
code point of char at cursor: 0x10437
char at cursor: 𐐷

 
Note: There's a similar/related Gerrit change for LibreOffice:
https://gerrit.libreoffice.org/c/core/+/151303