> LGTM. But it is too late for beta 4. I'll commit the patch either after releasing 3.6.0 or in the 3.7 branch only.

Right now, I suggest to only commit into 3.7. Such minor bug can wait for Python 3.6.1.


> And while we are here I noticed that handling non-ASCII replacement string could be simpler

I also suggest to first commit unicode_encode_ucs1_error_pos.patch and then commit the other part of unicode_encode_ucs1_error_pos2.patch in a separated commit. I will be easy to backport the fix to the 3.6 branch later.

Serhiy: Xiang became a core developer, are you ok if he push himself  unicode_encode_ucs1_error_pos.patch to default tomorrow, and later you rebase your patch on top of that?

I'm not super confident because the fix doesn't come with an unit test, but it's ok if Serhiy reviewed it :-)