It's more of an accumulation of things ending up making 3.3 way superior to 3.2 in the end. But,
* The feature I like most is definitely generator delegation, it significantly improves more extensive uses of generators and iterators, and makes generator-as-coroutines a much more interesting proposition (before delegation, calling an other "coroutine" would be rather painful, now you essentially just have to tack a `yield from` in front, as you'd tack an `await` in C# 5.0)
* The flexible string representation finally fixes narrow build's issues with astral planes, which is becoming rather important as astral planes include e.g. emoji, and it significantly reduces the possibility of bugs when working with astral planes (as there's no more behavioral difference between "narrow" and "wide" builds)
* We'll have to see how they're used, but namespace support could be used to significantly cleanup of... well, namespaces (and multiple separate libraries living in the same namespace, without having to resort to PYTHONPATH hacks or setuptools tomfoolery)
* A built-in, clean implementation of contexts/scopes (collections.ChainMap) I can already see plenty of use for. Same for signatures, there's high hijinks potential in that one.
* The rest really is about a better experience all around: reduced memory, "unicode literals" (for Python 2 compat), ElementTree fixups, ...
* The feature I like most is definitely generator delegation, it significantly improves more extensive uses of generators and iterators, and makes generator-as-coroutines a much more interesting proposition (before delegation, calling an other "coroutine" would be rather painful, now you essentially just have to tack a `yield from` in front, as you'd tack an `await` in C# 5.0)
* The flexible string representation finally fixes narrow build's issues with astral planes, which is becoming rather important as astral planes include e.g. emoji, and it significantly reduces the possibility of bugs when working with astral planes (as there's no more behavioral difference between "narrow" and "wide" builds)
* We'll have to see how they're used, but namespace support could be used to significantly cleanup of... well, namespaces (and multiple separate libraries living in the same namespace, without having to resort to PYTHONPATH hacks or setuptools tomfoolery)
* A built-in, clean implementation of contexts/scopes (collections.ChainMap) I can already see plenty of use for. Same for signatures, there's high hijinks potential in that one.
* The rest really is about a better experience all around: reduced memory, "unicode literals" (for Python 2 compat), ElementTree fixups, ...