Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
192 changes: 192 additions & 0 deletions asyncer/.agents/skills/asyncer/SKILL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,192 @@
---
name: asyncer
description: Use Asyncer correctly in Python code that mixes async functions, blocking sync functions, worker threads, or concurrent async tasks. Use when choosing between Asyncer, AnyIO, and asyncio patterns; replacing asyncio.gather, AnyIO task groups, to_thread/from_thread calls, or anyio.run wrappers; using asyncer.create_task_group(), TaskGroup.soonify(), SoonValue, asyncify(), syncify(), or runnify(); or avoiding common Asyncer mistakes.
---

# Asyncer Patterns

Use this skill when writing Python code that mixes async functions, blocking sync functions, or concurrent async work and the project depends on `asyncer`.

Asyncer is a small typed layer on top of AnyIO. Prefer the Asyncer helpers below when they fit, because they preserve function signatures, keyword arguments, and return types better than the lower-level AnyIO or asyncio calls.

## Imports

Use one of these styles:

```python
import asyncer
```

```python
from asyncer import asyncify, syncify
```

For task groups, prefer `asyncer.create_task_group()` so the returned task group has Asyncer's extra `soonify()` method.

## Concurrent Async Work

Prefer `asyncer.create_task_group()` plus `task_group.soonify()` for concurrent async function calls.

```python
import asyncer


async def fetch_user(user_id: int) -> User:
...


async def load_users(user_ids: list[int]) -> list[User]:
soon_values: list[asyncer.SoonValue[User]] = []
async with asyncer.create_task_group() as task_group:
for user_id in user_ids:
soon_values.append(task_group.soonify(fetch_user)(user_id=user_id))

return [soon_value.value for soon_value in soon_values]
```

Use this instead of:

```python
await asyncio.gather(*(fetch_user(user_id) for user_id in user_ids))
```

or raw task creation when the goal is simply "start these async calls concurrently and collect their results".

Do not use `task_group.soon(...)`; Asyncer's API is `task_group.soonify(...)`.
The call shape is:

```python
soon_value = task_group.soonify(async_function)(*args, **kwargs)
```

`soonify()` returns a `SoonValue[T]`. Read `soon_value.value` after the `async with` block, because exiting the task group waits for the scheduled tasks to finish.

```python
async with asyncer.create_task_group() as task_group:
first = task_group.soonify(fetch_user)(user_id=1)
second = task_group.soonify(fetch_user)(user_id=2)

users = [first.value, second.value]
```

Inside the same task group, `.value` can raise `PendingValueException` if the task is not complete. If there was an `await` inside the block and you truly need to inspect a value before exiting, check `.ready` first:

```python
async with asyncer.create_task_group() as task_group:
soon_value = task_group.soonify(fetch_user)(user_id=1)
await anyio.sleep(0)
if soon_value.ready:
user = soon_value.value
```

If code needs to consume values while producer tasks are still running, use a stream/channel pattern instead of `SoonValue.value` inside the task group.

If a task started with `soonify()` raises, the exception propagates when exiting the task group. Do not write code that assumes all `SoonValue.value` results are available after a failed task group unless the failure is caught and handled intentionally.

## Call Sync Code From Async Code

Use `asyncer.asyncify()` when async code must call a blocking sync function without blocking the event loop.

```python
from asyncer import asyncify


def render_pdf(order_id: int) -> bytes:
...


async def endpoint(order_id: int) -> Response:
pdf = await asyncify(render_pdf)(order_id=order_id)
return Response(pdf, media_type="application/pdf")
```

Use this instead of directly calling blocking functions from async code, and instead of lower-level patterns such as `loop.run_in_executor(...)` or `anyio.to_thread.run_sync(...)` when you want better argument and return typing.

`asyncify()` accepts AnyIO thread options:

```python
result = await asyncify(blocking_call, abandon_on_cancel=True)(item_id=item_id)
```

Use `abandon_on_cancel`, not the deprecated `cancellable` argument. Pass a `limiter=` when the caller needs to limit concurrent worker threads.

## Call Async Code From Sync Code

Use `asyncer.syncify()` from sync code that is running in a worker thread started by async code. A common chain is:

1. An async app or function is running.
2. It calls sync/blocking code with `asyncify()` or a framework-managed worker thread, such as a FastAPI sync endpoint or sync dependency.
3. That sync code needs to call an async function.

```python
from asyncer import asyncify, syncify


async def read_from_async_client(key: str) -> str:
...


def sync_business_logic(key: str) -> str:
value = syncify(read_from_async_client)(key=key)
return value.upper()


async def handler(key: str) -> str:
return await asyncify(sync_business_logic)(key=key)
```

Use this instead of `anyio.from_thread.run(...)` when passing keyword arguments or when type support matters.

Do not use `syncify()` inside an async function. In async code, use plain `await` on the async function directly.

For sync functions that must work both from an async worker thread and from a plain sync program, `syncify(..., raise_sync_error=False)` is available:

```python
def sync_business_logic(key: str) -> str:
return syncify(read_from_async_client, raise_sync_error=False)(key=key)
```

Use this compatibility mode sparingly. Outside a worker thread it starts a new event loop with `anyio.run()`, which is expensive if repeated in a loop. If you need many calls, wrap the loop in one async function and run that once.

## Run An Async Function From Sync Top-Level Code

Use `asyncer.runnify()` when sync top-level code needs to run one async function with positional or keyword arguments and the current thread is not already running an event loop.

```python
import asyncer


async def main(name: str) -> str:
...


result = asyncer.runnify(main)(name="World")
```

Use this instead of manually wrapping arguments for `anyio.run(...)`:

```python
result = anyio.run(lambda: main(name="World"))
```

Do not call `runnify()` from a thread that already has a running event loop; from async code, just `await main(...)`.

## Choosing The Asyncer Alternative

Use these replacements by default when Asyncer is available:

- `asyncio.gather(...)` for a batch of async calls with typed results: `asyncer.create_task_group()` plus `task_group.soonify(...)`.
- `anyio.create_task_group().start_soon(...)` for ordinary async function calls: `asyncer.create_task_group().soonify(...)`, especially when kwargs or return values matter.
- `anyio.to_thread.run_sync(...)`, `asyncio.to_thread(...)`, or `loop.run_in_executor(...)`: `await asyncer.asyncify(sync_func)(...)`.
- `anyio.from_thread.run(...)`: `asyncer.syncify(async_func)(...)`.
- `anyio.run(async_func, ...)` from sync top-level code with arguments: `asyncer.runnify(async_func)(...)`.

Still use lower-level AnyIO primitives when Asyncer does not provide the abstraction, such as streams, locks, events, capacity limiters themselves, cancellation scopes, timeouts, or task-group startup handshakes that need `task_status.started()`.

## Common Mistakes

- Do not call blocking sync functions directly from `async def`; wrap them with `await asyncify(sync_func)(...)`.
- Do not use `syncify()` as a general replacement for `await`; it is for sync code, normally in a worker thread started by async code.
- Do not wrap async functions when already in async code; use plain `await async_func(...)`.
- Do not read `SoonValue.value` before the task has completed unless `soon_value.ready` is true.
- Do not instantiate `asyncer.TaskGroup` directly; use `asyncer.create_task_group()`.
Loading