API documentation¶

Async generators¶

In Python 3.6+, you can write a native async generator like this:

async def load_json_lines(stream_reader):
    async for line in stream_reader:
        yield json.loads(line)

Here’s the same thing written with this library, which works on Python 3.5+:

from async_generator import async_generator, yield

@async_generator
async def load_json_lines(stream_reader):
    async for line in stream_reader:
        await yield_(json.loads(line))

Basically:

decorate your function with @async_generator
replace yield with await yield_()
replace yield X with await yield_(X)

That’s it!

Yield from¶

Native async generators don’t support yield from:

# Doesn't work!
async def wrap_load_json_lines(stream_reader):
    # This is a SyntaxError
    yield from load_json_lines(stream_reader)

But we do:

from async_generator import async_generator, yield_from_

# This works!
@async_generator
async def wrap_load_json_lines(stream_reader):
    await yield_from_(load_json_lines(stream_reader))

You can only use yield_from_ inside an @async_generator function, BUT the thing you PASS to yield_from_ can be any kind of async iterator, including native async generators.

Our yield_from_ fully supports the classic yield from semantics, including forwarding asend and athrow calls into the delegated async generator, and returning values:

from async_generator import async_generator, yield_, yield_from_

@async_generator
async def agen1():
    await yield_(1)
    await yield_(2)
    return "great!"

@async_generator
async def agen2():
    value = await yield_from_(agen1())
    assert value == "great!"

Introspection¶

For introspection purposes, we also export the following functions:

isasyncgen(agen_obj)¶: Returns true if passed either an async generator object created by this library, or a native Python 3.6+ async generator object. Analogous to inspect.isasyncgen() in 3.6+.

isasyncgenfunction(agen_func)¶: Returns true if passed either an async generator function created by this library, or a native Python 3.6+ async generator function. Analogous to inspect.isasyncgenfunction() in 3.6+.

Example:

>>> isasyncgenfunction(load_json_lines)
True
>>> gen_object = load_json_lines(asyncio_stream_reader)
>>> isasyncgen(gen_object)
True

In addition, this library’s async generator objects are registered with the collections.abc.AsyncGenerator abstract base class (if available):

>>> isinstance(gen_object, collections.abc.AsyncGenerator)
True

Semantics¶

This library generally tries hard to match the semantics of Python 3.6’s native async generators in every detail (PEP 525), with additional support for yield from and for returning non-None values from an async generator (under the theory that these may well be added to native async generators one day).

Garbage collection hooks¶

This library fully supports the native async generator finalization semantics, including the per-thread firstiter and finalizer hooks. You can use async_generator.set_asyncgen_hooks() exactly like you would use sys.set_asyncgen_hooks() with native generators. On Python 3.6+, the former is an alias for the latter, so libraries that use the native mechanism should work seamlessly with @async_generator functions. On Python 3.5, where there is no sys.set_asyncgen_hooks(), most libraries probably won’t know about async_generator.set_asyncgen_hooks(), so you’ll need to exercise more care with explicit cleanup, or install appropriate hooks yourself.

While finishing cleanup of an async generator is better than dropping it on the floor at the first await, it’s still not a perfect solution; in addition to the unpredictability of GC timing, the finalizer hook has no practical way to determine the context in which the generator was being iterated, so an exception thrown from the generator during aclose() must either crash the program or get discarded. It’s much better to close your generators explicitly when you’re done with them, perhaps using the aclosing context manager. See this discussion and PEP 533 for more details.

Context managers¶

As discussed above, you should always explicitly call aclose on async generators. To make this more convenient, this library also includes an aclosing async context manager. It acts just like the closing context manager included in the stdlib contextlib module, but does await obj.aclose() instead of obj.close(). Use it like this:

from async_generator import aclosing

async with aclosing(load_json_lines(asyncio_stream_reader)) as agen:
    async for json_obj in agen:
        ...

Or if you want to write your own async context managers, we’ve got you covered:

@asynccontextmanager¶: This is a backport of contextlib.asynccontextmanager(), which wasn’t added to the standard library until Python 3.7.

You can use @asynccontextmanager with either native async generators, or the ones from this package. If you use it with the ones from this package, remember that @asynccontextmanager goes on top of @async_generator:

# Correct!
@asynccontextmanager
@async_generator
async def my_async_context_manager():
    ...

# This won't work :-(
@async_generator
@asynccontextmanager
async def my_async_context_manager():
    ...