Serhiy, nosying you because Ammar identified your commit as relevant.
https://github.com/python/cpython/commit/e036ef8fa29f27d57fe9f8cef8d931d4122d8223
---

3.6 is also security-fix only.

Normally, code bug reports need a minimal, reproducible, initially-failing test case that only uses the stdlib, not 3rd party code.  A test case would have to include a simplified version of a multidict.

The problem here is that a multidict with duplicate keys is not a proper mapping, in spite of having a MutableMapping interface.  (Just curious, what does d['a'] return?). 

The purpose of the package is to meet the needs of HTTP Headers and URL query strings (and other situations) where the 'keys' are value-type tags, not true mapping keys.  Another example would be a bibliography entry that tags each of multiple author names with 'Author:'.  (Aside: A deficiency of git (github) is allowing only 1 author key and only one github username as the value.)

Is an object with duplicate keys legal for **expression in calls?
https://docs.python.org/3/reference/expressions.html#index-48
says that expression must evaluate to a 'mapping'.  The glossary entry
https://docs.python.org/3/glossary.html#term-mapping says
"A container object that supports arbitrary key lookups and implements the methods specified in the Mapping or MutableMapping abstract base classes." followed by unique-key examples.  To me, 'key lookup' implies unique keys.  The Mapping functions include 'keys' and 'items'.  What signature?  To return set-like views, the keys should be unique.

If we take the call to be a bug, is CPython *obligated* to immediately raise an exception?  In other words, must every call with **mapping take the time to check for duplicates because someone might pass a dup-key 'mapping'.