ctypes Structure fields of type c_char_p or c_wchar_p used to accept strings with embedded null characters. I noticed that Python 3.6.4 does refuse them. It seems this has been changed in recent version(s).

There ARE use-cases for this: The Windows-API OPENFILENAME structure is one example. The Microsoft docs for the lpstrFilter field:

"""
lpstrFilter

Type: LPCTSTR

A buffer containing pairs of null-terminated filter strings. The last string in the buffer must be terminated by two NULL characters.

"""

I have attached a simple script which demonstrates this new behaviour; the output with Python 3.6.4 is this:

Traceback (most recent call last):
  File "nullchars.py", line 8, in <module>
    t.unicode = u"foo\0bar"
ValueError: embedded null character

PyUnicode_AsWideCharString was updated to raise ValueError for embedded nulls if the size output parameter is NULL. Z_set in cfield.c should be updated to get the size, which can be ignored here. For example:

Py_ssize_t size; 
buffer = PyUnicode_AsWideCharString(value, &size);

The change mentioned was made in #46714 for bpo-13617 and was released in 3.6.3 (and 3.5.4 now in security-fix-only mode).

This is a regression. Eryk's solution LGTM. Do you mind to create a PR?

But u"foo\0bar" is not terminated by two NULL characters. If this is used in real code, it contains a bug. And the getter of this field will return the string only to the first null character. More work is needed for making this more reliable.

New changeset 73766b0 by Zackery Spytz in branch 'master':
bpo-32745: Fix a regression in the handling of ctypes' c_wchar_p type (bpo-8721)
73766b0

New changeset cf6a796 by Miss Islington (bot) in branch '3.9':
bpo-32745: Fix a regression in the handling of ctypes' c_wchar_p type (GH-8721) (bpo-25812)
cf6a796

New changeset db3ce79 by Miss Islington (bot) in branch '3.8':
bpo-32745: Fix a regression in the handling of ctypes' c_wchar_p type (GH-8721) (bpo-25811)
db3ce79