The issue #22110 enabled more compiler warnings. Attached patch tries to fix most of them on Linux.

clinic.patch: modify clinic.py to generate "return_value == (type)-1" instead of "return_value == -1" to avoid a warning if return_value is unsigned.

I think this should be done only for unsigned integer types. Otherwise it just dirty sources and can hide actual bugs.

It would be better to only modify clinic for unsigned types, but how do you
check if a type is signed or not?

Either override render() for unsigned type converters, or add new converter 
attribute (in additional to "type", "cast", "conversion_fn", etc).

Hum, I forgot the attach the most important patch: fix_warnings.patch.

The title for this bug was almost offensively generic; it was hard to find in my email.  If this new one is inaccurate, fix it--but please pick something more specific.

Attached is a patch for the "_return_value == -1" behavior for unsigned.

larry.fix.unsigned.int.return.-1.converter.cast.1.txt LGTM.

Victor, looks as you should open new issue for fix_warnings.patch. This is large patch and requires long time for careful review and may be discussion.

New changeset 9c949e0115e2 by Larry Hastings in branch 'default':
Issue #22120: For functions using an unsigned integer return converter,
http://hg.python.org/cpython/rev/9c949e0115e2

Fixed and closed.  I'm assuming Victor will break the other patch out into its own issue.

Since IIUC there's no code in 3.4 that uses an unsigned integer return converter, I'm not backporting the fix.

Since my argument clinic patch hijacked this issue, I created a new one for fix_warnings.patch: issue #22156.