A Python debug build is ABI compatible with a Python release build since Python 3.8 on most platforms (except Windows, Cygwin and Android):
https://docs.python.org/3.8/whatsnew/3.8.html#debug-build-uses-the-same-abi-as-release-build

It should now be way easier to debug an application with a debug build.

I propose to remove runtime debug checks in release mode. assert() is already widely used in the C code base, but there are many runtime checks using PyErr_BadInternalCall() or PyErr_BadArgument. Example:

Py_ssize_t
PyUnicode_GetLength(PyObject *unicode)
{
    if (!PyUnicode_Check(unicode)) {
        PyErr_BadArgument();
        return -1;
    }
    if (PyUnicode_READY(unicode) == -1)
        return -1;
    return PyUnicode_GET_LENGTH(unicode);
}

Attached PR removes these checks when Python is compiled in release mode: when Py_DEBUG is not defined.

Even if I marked this issue as "performance", I don't expect a significant speedup.

New changeset c6a2320 by Victor Stinner in branch 'master':
bpo-37406: sqlite3 raises TypeError for wrong operation type (GH-14386)
c6a2320

Given that extensions call these APIs, I find it highly risky to
disable these checks in any version of the Python runtime and
am -1 on such a change.

Using assert() in C is a pretty bad alternative, since this crashes
the whole process. It should really only be used where no other
means of error handling are possible. Python's exception mechanism
is a much better way to signal and handle such errors at the
application level.

--
Marc-Andre Lemburg
eGenix.com

Professional Python Services directly from the Experts (#1, Jun 26 2019)
>>> Python Projects, Coaching and Consulting ...  http://www.egenix.com/
>>> Python Database Interfaces ...           http://products.egenix.com/
>>> Plone/Zope Database Interfaces ...           http://zope.egenix.com/
________________________________________________________________________

::: We implement business ideas - efficiently in both time and costs :::

eGenix.com Software, Skills and Services GmbH Pastor-Loeh-Str.48
D-40764 Langenfeld, Germany. CEO Dipl.-Math. Marc-Andre Lemburg
Registered at Amtsgericht Duesseldorf: HRB 46611
http://www.egenix.com/company/contact/
http://www.malemburg.com/

I think that the public C API should have runtime checks for its arguments, except performance sensitive functions and macros (like PyTuple_GET_ITEM). The internal C API can use just asserts.

I have a work-in-progress patch which adds tests for the public C API, including calls with invalid arguments.

Python's exception mechanism is a much better way to signal and handle such errors at the application level.

I disagree. There is a difference between exceptions which are possible in normal usage of the code and real bugs, which shouldn't ever happen. A NULL argument to a C API function is not an ordinary exception, it's a real bug. Even if we raise a Python exception for that, it should never be handled in a try/exception block like an ordinary exception.

Using assert() in C is a pretty bad alternative, since this crashes the whole process.

I *prefer* debugging an assert() failure (which is pretty easy with a decent debugger) rather than debugging a Python exception raised from C (which typically does NOT give a useful traceback).

FYI I'm finishing an article to explain the problem that I am trying to solve here. Sorry for the delay, I will reply later :-)

I didn't find time to write a proper a rationale for this change, and I will not be available next weeks. So I prefer to close this issue. I may reopen it later once I have a strong rationale :-) In the meanwhile, have a look at the new PyHandle API idea which will start without sanity check in release mode since day 1 :-)
https://github.com/pyhandle/pyhandle/