Type hinting in code

[penguinolog]

Long story short

While working on big project, typing check helps to exclude shot in the leg scenario.

Expected behaviour

Type annotations on public methods.

Actual behaviour

No any type annotations, including the most docstrings.

Note

Need a decision: typing in global requirements or use version-specific requirements (PEP496/PEP508)

[penguinolog]

@fafhrd91 Please comment decision, if possible.

[fafhrd91]

Personally, I do not use type annotations. But I am fine with type annotations on public methods

[penguinolog]

I'm about requirements on typing library. We can require it on every python version and can specify 3.4 only due to it presence in 3.5+ out of box.

[fafhrd91]

That's fine

[iceboy233]

I personally dislike type annotations in performance sensitive code in python. Any footprint bloat would slow things down, especially in interpreted languages.

[fafhrd91]

I thought there should be no runtime penalty?

[iceboy233]

the annotations are accessible at runtime through __annotations__, so it will increase the memory footprint, allocator load (for the hash table for __annotations__), incref for the type objects, parsing time, etc.

[fafhrd91]

Ah, but compared to overall speed of python that doesn't really matter. I guess it is possible to come up with method how to strip annotations data at startup

[iceboy233]

The annotations are not necessarily for type hints and are runtime visible. It's just the IDEs using them for type hints. I have used annotation for other things in my programs, such as param sanitizing.

[asvetlov]

Annotations are perfect concept but before applying it to aiohttp we need reliable type checker tool.
The only existing tool for now is mypy and it's not production ready yet.

At least I had big problems when tried to check annotations for smaller aio-libs projects.
Let's postpone the issue (without closing).

[thehesiod]

extra datapoint: all the code I write has type hints as it greatly helps finding issues in pycharm while developing

[kxepal]

I also had negative experience with type hints as annotations. Tools are not ready indeed and you have to choose between two evils: do additional imports for getting types and accidentally end with circular ones or double number of files to maintain. So far good old sphinx docstrings are more handy and PyCharm can use it for type hinting. You still have to write docs anyway.

However, I think you can start make annotations today as typeshed part or as standalone project with stubs like pyspark and contribute back fixes of any problems you found.

[asvetlov]

python/typing#84 is a blocker: without support for third-party libs in mypy type annotations are useless for aiohttp users.

[asvetlov]

I don't want to support aiohttp stubs in typeshed but want to add annotations into aiohttp itself

[penguinolog]

I have requirements concern:
typing is available starting from Python 3.5, backport can be required using PEP0508 (incorrectly parsed by some packagers), PEP0426 (deprecated way, totally ignored by some packagers) or patching requirements list in setup.py (bad looking way).